HYPOTHESIS A selected postpositional phrase is merged inside VP; an overt pronoun stays inside it, which is why a relative on it has no gap: bichʼįʼ yáshtiʼ=yę́ę (NX-314).
HYPOTHESIS Question and focus enclitics (=sh, -ísh) are interpreted in C⁰ and pronounced on a host; the host is not the scope position (H12): Díí=sh háí yiztał (NX-085). Results §3.
OPEN Whether H12 is syntactic movement or phonology only; whether the polar particle is C⁰ or in Spec,CP (headedness); how the inverse on a postposition fits the inverse movement (clause spine, L5). The word-edge analyses in the legacy sections below are kept as contrast.
Restated on the unified structure (2026-11-15). Postpositions are restated as [PP DP P⁰] with left specifiers: a grammatical P agrees with the DP in its specifier, a lexical P has no probe (H64); a selected PP is merged in VP, and Willie’s “cannot be interrupted” follows from the default, HS-1 and left specifiers. Light d, used below, is withdrawn (Grant, 2026-11-14); light n is an OPEN structural question. Position-class template statements below describe the order of prefixes; they are not the derivation (§3). Flat-S and other pre-VPISH notation below is labelled contrast. Status on the spine: see unified structure, §2b and §4, postposition row. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
2026-11-11 audit pass 2. Wording that used VPISH as the name of a spine, tree, skeleton or set of slots named a structure after the hypothesis. VPISH is a hypothesis name, never a node; the structure is a vP/TP spine with the subject first-merged in left Spec,vP. Only the wording changed.
2026-11-11 audit pass 3. Status badges now use only the label set SOURCED, HYPOTHESIS, SPECULATIVE, OPEN, NONE and GRANT'S CONJECTURE; verdict or task words that had been styled as badges (for example Opened, PRE-VPISH, DONE, SUPPORTS, PARTLY ANSWERED) are now plain text or a label with a qualifier. See the pass 3 report.
Grant asked two things. (1) What is a postposition as a head: a functional head (T? C?), or P? (2) Are Navajo verb + postposition units “phrasal verbs”, given that lexicalization happens at the phrase level as much as at the word level? This page collects what the sources we opened say, with page numbers; shows VPISH-aware derivations of verb + postposition as one unit and of the inverse bi- on the postposition; grades the lexicalization evidence; and adds the sourced verb + postposition frames to the verb dictionary (new group M).
Update 2026-10-22, 8:46pm (T50f).HYPOTHESIS When two postpositions come before one verb, the one that forms a unit with the verb is next to it, and the added one is outside: yá ʼataʼ halneʼ ‘translating for’ (NLA 13), yaa bił hoozhǫǫhgo (1958, with bił hózhǫ́ ‘be happy’, 792), shíká bił ʼoolwoł (15), Willie (43) bił béé hózin (NLA: bił bééhózin). That gives rule IP-8 on Instrument and voice §3.7. Four cases, no counterexample.
Update 2026-10-22, 12:46pm (T49f).SOURCED data, HYPOTHESIS reading. Willie's (29) yá yoołóós and (85)/(86) used on this page are now part of a verb-by-verb check of her whole list (28) (postpositions by verb). Being “grammatical” is a property of the postposition's form; being inside the verb complex depends on the verb and the role, and the voice alternation shows it: yá/bá with ‘lead’ is inside even though ‘for’ is free (not an idiom). So “inside the MVS” and “lexicalized with the verb” are separate questions. Willie's own data also give counterexamples to the animacy column of (28) ((83) jooł ‘ball’ with -aa; (36) the corral with -tsʼą́ą́ʼ).
Update 2026-10-20, 12:46pm. One new datum bears on question (2). The yes/no enclitic can go on the postposition word of a verb + postposition unit: NLA yes/no (6) Neésh hólǫ́? ‘2-with-Q exist.NI’, ‘Do you have any?’, and, on the NLA front page, Nił ísh béého sin doodaiiʼ bił ísh béého zin? (spacing as displayed) ‘Do you know or does s/he know?’. So such a unit is two spelled-out units. If it is lexicalized, it is lexicalized at the phrase level, which fits Grant's suggestion. When the postposition is not first, the NLA marks the placement “?” ((5)). See where the yes/no enclitic -ísh can go (PE1, EN-5).
Update 2026-10-21, 4:46am (T43e). Two points from attached or separate postpositions. (a) L7's middle step (separate word → proclitic) is pronunciation over one syntax (Willie pp. 64, 224; Hale 1997 PDF p. 31), not a degree of lexicalization; the ‘know’ unit is printed Shił béé hózin by Willie (42) and shił bééhózin by the NLA. Hale's causative P, which goes into the word, is a different operation (PC-3, PC-4). (b) For L9 (“phrasisized”), the NLA Postposition Lexicon supplies particular pairs whose translation departs from the glosses, all with the postposition written as a separate word: bídin nishłį́ ‘3-without 1-be’ = ‘I really want (it)’, niyaa hazlį́į́ʼ ‘(you) grew up’, bichʼįʼ sidziil ‘I can withstand (it)’, and -ká + ‘help’ (two NLA sentences besides Chee & Yazzie (14b)). It also shows a regular pattern, -ił ‘with’ + a vehicle-subject motion verb = ‘travel by’ (tagged “transportation”). Storage is still GRANT'S CONJECTURE (T32f, T45c).
Update 2026-10-21, 12:46pm (T45f). The -ił ‘with’ travel pattern is checked against the NLA's full “transportation” list on the -ił travel pattern. It is a pattern built from the parts (‘X moved, with Y’ = ‘Y travelled by X’), not a stored verb + postposition pair: six or more stems, five kinds of carrier. So for L9 (“phrasisized”) it is not a case; the one listed piece is a word-level sense, ʼoolwoł ‘drive’ (TP-0, TP-5; HYPOTHESIS). L9 stays a conjecture about the particular pairs (kind (c)).
Ground rules (Grant's). No analysis is assumed to be fully right; each may be partly right. Silence counts against nothing; only contradicting data count. Grant does not speak Navajo. Every Navajo form below is copied from a source opened this run, with its page and example number. No Navajo form, gloss or judgment is invented. Where a source gives no category, the table says so. The ideas “P under vP”, “light p” and “p-shell” are general-syntax proposals, not Navajo sources. No source I opened states them for Navajo, so nothing on this page is attributed to them.
Bottom line, in plain words
SOURCEDThe sources that say anything treat the Navajo postposition as P, a lexical head, not as T or C. Hale 2003 puts postpositional projections with the verb and the noun as head-final lexical projections (p. 32). Hale 1997 draws a PP with a specifier and a complement, with the pronoun prefix on P as “AGRpo”, and calls P “the relevant agreement-bearing head” (PDF pp. 31–33, (23)–(24)). Hale & Keyser name “adpositional (preposition or postposition)” as the default category of their basic two-argument type (PDF p. 30, (42)). Willie 1991 and Willie & Jelinek 2000 say “postposition” and “postpositional phrase” and give no head category. Nobody I opened calls the postposition T or C.
SOURCEDWhat is unusual about it. It carries its own pronoun prefix, as a verb does (Willie p. 39: “Postpositions resemble verbs in Navajo in being inflected for their object arguments”; Hale 2003 p. 46: “the pronoun is the true object of the postposition”). So P hosts agreement the way Mode/Subject (= T, Hale 2003 p. 10) does. That is a likeness, not a claim anyone made.
HYPOTHESISFunctional or lexical may not be one answer. Willie splits postpositions into grammatical (pronoun objects; part of the MVS) and lexical (noun objects). Hale 2003 (pp. 22, 59) records that the oblique case on a derived object is “traditionally” a null postposition, and Hale 1997 gives the causative -y- as a postposition that behaves unlike ordinary proclitic ones (PDF pp. 52–55). Reading the grammatical ones as functional (case-like) and the rest as lexical is ours. One datum cuts against a fixed classification: the same form -ii’ is lexical with an inanimate subject and grammatical with an animate one (Willie p. 98).
SOURCEDHYPOTHESISVerb + postposition units. Sourced: they are a syntactic unit, Willie's MVS (not interruptible; may be one phonological unit; the domain of the verb's subcategorization, p. 50 (44)), and Chee & Yazzie 2025 call them “phrasal verbs”, with five attested pairs where the postposition shifts the meaning (pp. 22, 34–35). Willie states phrase-level lexicalization outright for noun + lexical postposition (p. 43 (35)). Hypothesis: those verb + postposition pairs are stored units. Open: whether Young & Morgan list any of them (I did not open Young & Morgan).
GRANT'S CONJECTURE “Phrasisized”: lexicalization (storing a unit with its own meaning) applies to phrases such as verb + postposition just as it does to words. This is Grant's coinage, not a field term, and the page uses it only in quotation marks as his label. It is compatible with everything sourced here and proved by none of it.
Intro examples (1)–(4); pronoun prefixes on postpositions; enclitics have none; based on Young & Morgan 1987:25–26
Speas 1986, ch. 3–4
OCR text of the box copy searched for “postposition”, “PP”, “oblique”, and typical postpositional words
Nothing found on the postposition's category (the OCR is unreliable and I did not re-read every Navajo page image; see T32g)
Not opened:
Young & Morgan 1987 and Young, Morgan & Midgette 1992 (no open copy; the Internet Archive record has no readable files); Speas 1990 (Phrase Structure in Natural Language); Hale & Keyser 1993 in its published form; Hale & Platero 1996; Chee 2025 (single-author paper, known only from a prior-run note). A search-engine summary said Speas 1990 treats postpositions as functional projections. I did not open anything that says that and I do not use it. A search snippet from an unidentified Hale paper (studyres.com) says a postposition “requires overt agreement” and mentions Speas on b- as an incorporated pronoun; the page text could not be retrieved, so it stays a lead (T32g).
2. Where each source puts the postposition
Source and page
What it says (quoted or paraphrased closely)
Category it implies
Status
Willie 1991:39–40, (27)–(28)
“Postpositions resemble verbs in Navajo in being inflected for their object arguments. The postpositional object pronouns and the possessive pronouns are identical in form. The postpositional phrase corresponds to an oblique argument of the verb.” Some are “case-marked arguments that the verb is subcategorized for”, some “more like adjuncts”: the grammatical and the lexical postpositions. Grammatical ones “occur only with pronominal objects and are a component of the MVS”; lexical ones “only with nominal (lexical) objects”.
Oblique argument (PP). No head label. “Grammatical” means case-marked and subcategorized for, not a stated functional head.
SOURCED
Willie 1991:43–44, (35)–(36)
Lexical postpositions and their nominal objects “can become lexicalized”; a lexical postposition “can also appear after a grammatical postpositional phrase” ((36)); “it is not possible to stack two grammatical or two lexical postpositions”.
An outer lexical P over an inner grammatical PP is allowed; no head label.
SOURCED
Willie 1991:50, (44)
The verb and its postpositional phrases form the MVS: “it cannot be interrupted; it may constitute a single phonological unit; it is the domain of the expression of the subcategorizational features of the verb; it behaves as a unit under certain syntactic processes”.
The PP is inside the verb's subcategorization domain.
SOURCED
Willie 1991:63–65, 93–95, 97–98
A single nominal is coindexed with the postpositional object if the verb is intransitive (p. 63 (26)); the inverse “has scope over” an oblique and makes it the external argument (p. 94); the same -ii’ is lexical with an inanimate subject and “must be interpreted as a grammatical postposition” with an animate one (p. 98).
Grammatical vs lexical depends on the arguments, not only on the morpheme.
SOURCED
Hale 2003:32
Lexical projections are head-final, “an arrangement which extends to postpositional and nominal (possessive) projections as well”; the functional heads (Qual, M/S) are head-initial.
P is a lexical head (head-final, like V and N).
SOURCED
Hale 2003:46–47, 51
(39a,b): the pronoun is “the true object of the postposition”; the overt noun “an adjunct (to the PP as a whole, or to the clause as a whole)”. With a bare noun, “the object of the postposition is a nominal, in complementary distribution with the pronoun … some languages permit nominals to appear as the complement of a lexical head”. Bare nominal arguments of postpositions appear “in their base positions”.
P is a lexical head that takes a pronoun or a bare noun as complement.
SOURCED
Hale 2003:22, 59
The derived oblique object is “traditionally held to be in a postpositional construction (with non-overt postposition)”, placed “alternatively to positions Ia and IV by Young and Morgan (1987:G42, et passim)”. Hale's own account is Case Binding (pp. 44–45), with Obj(obl) in a specifier and no P head drawn.
A null postposition as the traditional name for oblique case; Hale does not draw it.
SOURCED (the report); the analysis is open
Hale 1997, PDF pp. 31–33, (23)–(24)
“The postposition (P) projects the positions just enumerated, forming the postpositional phrase (PP); the latter bears the complement relation to the verb.” Tree: PP = [DP (specifier, ‘boat’) [P′ DP (‘me’) [P AGRpo sh- + -tsʼá]]], PP a sister of V inside VP. “P, the relevant agreement-bearing head, is the closest c-governing head”; “the specifier of PP, which is also the grammatical object of the transitive verb”, raises to agree with E (M in the intransitive). The word si-tsʼá “is phonologically dependent upon the following word; it is a proclitic”.
P is a head that projects a specifier and a complement and bears agreement. GB-era notation (c-government, E/M).
SOURCED
Hale 1997, PDF p. 43
In the lexical argument-structure types, “In Navajo, the (b) type [head + specifier + complement] is headed by members of the postpositional category which has noun-like characteristics in that language; postpositions inflect like nouns, and a few postpositional stems also belong to the nominal category”. “Morphosyntactic category and structural type are independent variables.”
P is noun-like and lexical.
SOURCED
Hale 1997, PDF pp. 52–55
The causative -y- is a postposition “Pcaus”, an “agreement-bearing head” that realizes the oblique case of the Case-bound subject. Unlike an ordinary proclitic PP complement of V, it “actually incorporates into E” (PDF p. 53).
One P that behaves like a case-realizing functional element.
SOURCED
Hale & Keyser, PDF pp. 11–12, 30 (42)
The basic dyadic structure (head + specifier + complement) has as its default categorial realization “adpositional (preposition or postposition)”; the other, composite dyadic, is verbal. The evidence in the paper is overwhelmingly English; the Navajo passages concern verbs and the ł-classifier, not postpositions.
P is a lexical category (V, N, A, P), not light p.
SOURCED for the statement; OPEN for Navajo P
Willie & Jelinek 2000:270, (50a,b)
Gloss “3-PP”; “postpositional phrase”; “Some postpositions and their objects in Navajo are procliticized to the inflected verb; others remain detached. The verb and any preceding post-positional phrases form a noninterruptable constituent, the Maximal Verb Complex (Willie 1991)”. The noun “does not form a syntactic constituent with the postpositional phrase”.
No head category. No PP node is drawn (the trees (43)–(49) carry only S, S′, IP, V′ and COMP labels).
SOURCED that no category is given
Speas 1986 (ch. 3–4)
No passage on the postposition found in the OCR text.
None found.
OPEN
Young & Morgan 1987 (via McDonough p. 32, Table 2)
Position 0 “Direct object of postposition. Possessive prefix with nouns”; Ia “Null postposition”; Ib “Adverbial – Thematic (‘postpositional stems’)”. McDonough adds that the template “is problematic on several levels, and Y&M do not use it in the grammar except as a point of reference”.
A position-class description, not a syntactic category.
SOURCED as McDonough's report; OPEN for Y&M itself
HYPOTHESISDating. Hale 1997, Willie 1991 and W&J 2000 are pre-VPISH or transitional in notation (see the framework audit). Hale 2003 is VPISH, but draws no PP tree. Nobody in the opened set draws a PP in a vP/TP spine. The derivations below are therefore our restatements, built from the site's spine (Qual > M/S > v > V) plus Hale 1997's P.
3. T, C, P or something else?
Option
Who says it (opened)
What fits
What cuts against
Status
PU-L P is a lexical head: head-final PP, inside VP, complement = pronoun or bare noun, may project a specifier
Hale 2003:32, 46–47; Hale 1997 PDF 31–33, 43; Hale & Keyser PDF 30
Pronoun prefix on P; bare noun as complement ((39b)); “inflect like nouns”; PP sister of V
Nothing printed contradicts it. The (b)-type specifier is only argued for the “drift away” verbs
SOURCED
PU-A P is an agreement-bearing head (the pronoun is valued inside the PP)
Hale 1997 PDF 32 (AGRpo); Hale 2003:46
shí sitsʼ…: the P's own prefix; W&J (50a,b) yi-/bi- on P
—
SOURCED
PU-G “Grammatical” postpositions are functional (case-realizing, applicative-like), lexical ones are P
None. Hooks only: Willie p. 39 (“case-marked arguments that the verb is subcategorized for”); Hale 2003:22, 59 (null postposition = oblique case); Hale 1997 PDF 53 (Pcaus)
Grammatical Ps take only pronouns and sit in the MVS (Willie p. 40); Pcaus incorporates into E unlike other PPs
Willie p. 98: the same-ii’ is lexical or grammatical by the animacy of the arguments. A functional-head analysis must then pick the head by the arguments, not by the morpheme
HYPOTHESIS
PU-S The lexical (enclitic) postpositions are outer heads over a PP, a split-P-like layer
None. Hook: Willie p. 44 (36): [baghan bitsʼą́ą́]-jįʼ, lexical P outside grammatical P
Order fixed outer lexical, inner grammatical; no stacking of two of a kind; enclitics “do not have pronoun prefixes” (Fernald et al.)
One example. Split-P is general syntax (not opened)
HYPOTHESIS
PU-T P is T-like
None
Both P and M/S (= T, Hale 2003:10) host agreement with a pronoun
M/S is above v in the verb's spine (Hale 2003); P is inside VP and heads its own phrase. No data tying P to tense or finiteness
OPEN
PU-C P is C-like (a clause-linking head)
None
W&J gloss the enclitic =go as COMP ((49)); Willie has a lexical postposition =góó ‘to’ ((31), (35))
Nothing opened says =go and =góó are one morpheme: the Swarthmore lexicon lists the “verb and subordinating enclitic” -go separately from -góó ‘to, toward, along’. I do not claim they are related
OPEN (T32c)
PU-p Light p / p-shell above P (P under vP)
None opened
—
General-syntax proposal. No Navajo source opened states or tests it, so I make no claim
OPEN (T32a)
HYPOTHESISWhat would separate the options. (i) PU-L vs PU-G: find any P that pronounces a case-like function with no adjunct meaning and cannot take a nominal complement or an extra specifier; Pcaus is the best candidate in the sources. (ii) PU-S: more outer-over-inner strings like (36). (iii) PU-T and PU-C need data on finiteness or clause-linking by P itself, and a check of whether =go and =góó are one form (Young & Morgan 1987:25–26 lists the postpositions; I did not open it).
4. Data (copied with example numbers)
Form
Gloss and translation as printed
Source
ʼatʼééd yichʼįʼ yáłtiʼ
girl 3.to 3.speaking ‘He is speaking to the girl.’
Willie 1991:63 (26)
tsé yikááʼ neezdá; ʼashkii yikááʼ neezdá
rock 3.on 3.sat ‘He sat down on the rock.’; boy 3.on 3.sat ‘The boy sat down on it./He sat on the boy.’
Willie 1991:63 (27); 64 (28)
yá yoołóós; ʼashkii ʼatʼééd yá néidííłtsʼin
3.for 3.3.lead [sing. +animate] ‘She is leading it for him.’; boy girl 3s.for 3.3.fist-hit ‘He hit the boy for the girl.’
Willie 1991:64 (29); 65 (30)
ʼawééʼ bichʼįʼ yishááł; *ʼawééʼ-góó yishááł
baby 3.to 1.walking ‘I am walking to the baby.’; (the second is starred)
Willie 1991:43 (33), (34)
hooghan=góó déyá
‘I am going home.’ *‘I am going to the hoghan.’
Willie 1991:43 (35)
dibé baghan bitsʼą́ą́jįʼ yishááł
sheep 3s.home 3s.from.at 1s.walking ‘I am walking away from the corral.’
1sg.with 3.nice ‘With me, it is nice. I like it.’; 1sg.with ho-nice ‘I like it there (the area).’; 1.with 3.about ho-known [about X, knowing is with me.] ‘I know it.’; ‘He/she knows that Kee ran away.’
Willie 1991:47 (41a, b); 48 (42); 49 (43)
ʼashkii ʼatʼééd yichʼįʼ yáłtiʼ; … bichʼįʼ yáłtiʼ
3.to 3.speaking ‘The boy is speaking to the girl.’; ‘The boy is being spoken to by the girl.’
Willie 1991:94 (79), (80)
ʼAwééʼ jooł yaa nímááz; … baa nímááz
3.to 3.rolled ‘The baby rolled to the ball.’; 3-to 3-rolled ‘The ball rolled to the baby.’
Willie 1991:95 (83), (84)
ʼashkii ʼatʼééd łį́į́ʼ yá yoołóós; … bá yoołóós
3.for 3.3.lead [sing. +animate] ‘The boy is leading the horse for the girl.’; ‘The girl is leading the horse for the boy.’
3.in 3.lies ‘The ball is in the car.’; ‘The woman is sitting in the car.’; INV3.in 3.sits [The car is sitting in the woman]; my mother 3.with car 3.in DUAL.sitting ‘I sit with my mother in the car.’
Willie 1991:97 (89a, b); 98 (90), (91), (92)
dibé łį́į́ʼ yitah yíghááh; … bitah yíghááh
sheep horse 3-PP 3-join (Direct) ‘The sheep went among the horses.’; (Inverse) ‘The horse went among the sheep.’
W&J 2000:270 (50a,b)
łééchááʼí tó y-iih yilwod; łééchááʼí taah yilwod
dog water 3o-into run:PERF ‘The dog ran into the water.’; dog water-into run:PERF ‘The dog ran into water.’ (spelling as printed, without nasal hooks)
Hale 2003:46 (39a,b)
ʼAwééʼ bi-ʼnii-sh-hóósh
baby 3-INCH-1s-tickle ‘I start to tickle the baby.’ (the overt bi- is the object marker of the oblique, “traditionally regarded as a null postposition”, and is overt because the subject is a local person, p. 59)
‘I play’ → ‘I make him cry’; ‘frightening’ → ‘You are lazy.’; ‘The area became hotter.’ → ‘He became sick, ill.’; ‘I usually walk away out of sight.’ → ‘I usually help him.’; ‘They … are continuously walking around, about.’ → ‘All of them are doing it.’
car almost 1-with down off-3-go.P ‘The car almost ran off (the road) with me.’
Fernald et al., Postposition Lexicon intro (2)
háníshtʼááh … Bił háníshtʼááh
‘to fly it after it (a plane)’ …; ‘to take him after it by plane, to accompany him after it (with the subject of the verb as pilot)’
McDonough 2019:47 (20), reproducing Young & Morgan 1987:d415 (secondhand)
5. VPISH-aware derivations PU1–PU7
Format as on the clause spine: Merge, Agree, Move, Link, Spell-out, tagged Σ (role schema), A (word level, D1 = MVS) or B (clause level, D2). Spine: Qual > M/S (= T) > v > V (Hale 2003 pp. 10, 32). The PP is placed inside VP as in Hale 1997 (23), with P head-final (Hale 2003:32). Everything about where things merge is ours (HYPOTHESIS); the strings and the pieces they are made of are sourced. Roles are read off the translations. Two versions of P's site are carried through: PU-L (lexical P in VP) and, where it changes anything, PU-G (a grammatical P that is a head in the verb's own spine, applicative-like).
PU1. Willie (26) ʼatʼééd yichʼįʼ yáłtiʼ: intransitive verb + grammatical P as one unit
Σ Σ(√SPEAK, with P ‘to’) = {ρA (speaker), ρX (the one spoken to, object of -chʼįʼ, Willie's “Goal; Allative”, p. 41 (28b))}. Both π = 3.
AMerge P (-chʼįʼ) with ρX as its complement (P is head-final: object to its left). Agree P ⟶ ρX [3]; with no local person present the default form is y- (direct). CYC (pronoun ⌢ P) → yichʼįʼ. Willie: the pronoun is the postpositional object pronoun (27); Hale 2003:46: it is “the true object”.
AMerge V (yáłtiʼ, unsegmented: Willie does not segment it) with the PP as its sister inside VP (Hale 1997 (23): the PP “bears the complement relation to the verb”). Merge v with ρA in Spec,v; Merge M/S; Agree M/S ⟶ ρA [3]; Qual.
ASpell-out D1 = the MVS: C1: yichʼįʼ ⌢ yáłtiʼ. Willie (44a): the unit “cannot be interrupted”; (44b): “may constitute a single phonological unit” (compare W&J p. 270: some postpositions are procliticized, others detached).
B Linking (as in nominal adjunction): the one noun is linked to ρX because the verb is intransitive (Willie p. 63, (26)); if its semantics make that impossible it goes to the subject instead ((27), p. 63). Spell-out D2: ʼatʼééd ⌢ [yichʼįʼ yáłtiʼ] ✓.
HYPOTHESISPU-G version. P is not a phrase in VP but a head in the spine between v and V (applicative-like), with ρX in its specifier. The string and the linking are the same. What differs: under PU-L there can be two PPs (nothing printed shows two grammatical PPs in one clause; Willie p. 44 rules out stacking two of a kind on one PP); under PU-G a clause would have one slot for a grammatical P. Willie's (36) puts one grammatical and one lexical P in one phrase, which both versions allow. No opened datum separates them (T32b).
PU2. Hale 1997 (22b) Ashkii tsinaaʼeeł (shí) sitsʼéiníłʼéél: transitive verb with a PP whose specifier is the theme
Σ Σ(√FLOAT, with P ‘away from’) = {ρA (the boy), ρTh (the boat), ρG (‘me’)}. The transitive alternant adds the ł- classifier and an agent (Hale 1997, PDF p. 31).
AMerge P (-tsʼá) with ρG as complement; Agree P ⟶ ρG [1sg]: sh- (Hale's AGRpo). Spec,PP ← ρTh (the PP's specifier is “the grammatical object of the transitive verb”, PDF p. 32).
AMerge V with the PP; Merge v (ł-); ρA in Spec,v; M/S; Agree M/S ⟶ ρA. Hale 1997: ρTh in Spec,PP “cannot find an agreement partner locally and must, therefore raise to Δ2, i.e., Spec of MP, in order to enter into the agreement relation with E” (y-, “AGRo”; PDF p. 32). In our terms: the object prefix yi- is valued by v/M/S, not by P.
ASpell-out D1: si-tsʼá (P with its object agreement) is a proclitic on the verb (Hale 1997, PDF p. 33: “written together with the following word”, “traditionally classified as belonging to the disjunct system”). So the MVS here is one written word. The contractions inside sitsʼéiníłʼéél are not segmented by Hale beyond si-tsʼá; I do not segment them.
BAshkii and tsinaaʼeeł are linked nouns; shí is an optional emphatic pronoun (bracketed in (22)). Spell-out D2 ✓.
SOURCED What Hale 1997 claims here: P is a head with a specifier, a complement and an agreement element, and the PP is the verb's complement. HYPOTHESIS Our restatement in M/S terms. OPEN Whether this “theme in Spec,PP” analysis extends beyond the “drift away” verbs (T32d). Hale 1997 is pre-minimalist (E, M, c-government).
PU3. Hale 2003 (39a) vs (39b): pronoun or bare noun as the object of P
Σ Σ(√RUN, with P ‘into’) = {ρA (the dog), ρX (the endpoint)}.
A(a)Merge P (-iih) with ρX, a pronoun [3o]: y-iih (Hale glosses “3o-into”). tó ‘water’ is a nominal adjunct linked at clause level (Hale p. 46: “an adjunct (to the PP as a whole, or to the clause as a whole)”).
A(b)Merge P (-iih) with the bare noun tó as its complement. There is no ρX pronoun slot, so no y- (“in complementary distribution with the pronoun”, p. 46). The noun and P are spelled as one word: taah (“simply the product of Merge”, p. 47; Hale does not segment taah into noun and postposition).
B (a) łééchááʼí and tó are both linked adjuncts; (b) only łééchááʼí is. Spell-out D2 ✓ both.
SOURCED Hale's translations carry the contrast: ‘the water’ in (a), ‘water’ in (b); and his general point that nominals linked to arguments are definite while incorporates are generally indefinite (pp. 47–48). HYPOTHESIS Prediction of PU-L: only the (b)-type noun is inside D1, so only it is under the MVS's “cannot be interrupted” condition. That is exactly the kind of datum T31f (are noun-object PPs inside D1?) needs: no opened source prints a noun between a bare-noun P and the verb.
PU4. The inverse bi- on the postposition: W&J (50b), Willie (80), (84), (86)
Σ Σ(√JOIN, with P ‘among’) = {ρA, ρX (object of -tah)}; both π = 3. As on L5: in the inverse (50b) ρX carries [top] and ρA [foc]; in the direct (50a) the two features are the other way round.
AMerge P with ρX. The postposition's own pronoun prefix is the inverse marker: bitah (50b) vs yitah (50a). Willie's pairs are the same switch on other Ps: yichʼįʼ / bichʼįʼ (79)/(80), yaa / baa (83)/(84), yá / bá (85)/(86). Note that in each pair the verb is unchanged: the verb has no bi- of its own.
IA-3 (Hale 2003: no A-movement; [top] on the slot from the start): b- is spelled where ρX was merged, inside the PP. No extra assumption. Fits all four pairs.
IA-1* (the topic slot raises to Spec,M/SP in the inverse): ρX would have to leave the PP. But its only exponent is a prefix on P, so what moves must be the feature bundle, with the lower copy pronounced, or the PP must be spelled before the move. This is T31e again, now with four pairs rather than one.
IA-0 (inverse as linking only): bi- on P is accounted for only if bi- is agreement (e3). It makes no prediction about where the PP is.
ASpell-out D1: bitah ⌢ yíghááh. The PP is a separate phonological word inside D1 (MVS-D1).
B Linking: Willie p. 94: the inverse “has scope over an argument with an ‘oblique’ theta role”, making it “the external argument”; W&J p. 270: dibé is coindexed with the bi- object but is not a constituent of the PP.
HYPOTHESISWhat the extra pairs add. (a) The inverse marker on P is not special to (50b); four independent pairs show it on four different postpositions. (b) Because the verb does not change, the inverse of an oblique is a property of the P's pronoun, which supports reading the voice alternation as a switch in the pronoun (Hale 2003-style), not in the verb. (c) Willie p. 98 (92) shows bii’ with a first-person subject, where “Inverse cannot apply”: so on P the b- form is not always the inverse. That needs the separate statement that b- is the general third-person form after a local person (as Hale 2003:59 says for the verb's bi- in (50)). OPEN Willie p. 97 says intransitive sentences have “no Inverse forms, and no grammatical postpositional phrases”, yet (79)/(80) (p. 94) have an intransitive verb with an inverse PP. Reconciling them is T32e. Update 12:46pm: see one-place verb + P §2.2. Willie uses “intransitive” for the bare verb on pp. 57 and 63 and for the whole MVS on p. 97. On the MVS reading the tension goes away. b- on P has three sources (lexical P, inverse, local person), and y- has one (E7).
PU5. Willie (89)–(91): one form -ii’, two heads?
Σ With a ball (inanimate) as the one that lies: bii’ (89a), *yii’ (89b). With a woman (animate) as the one that sits: yii’ (90), and ???bii’ (91) has the “nonsense” inverse reading ‘the car is sitting in the woman’.
APU-L, one head, selected by features: the same P -ii’ always. The choice between y- and b- would then depend on the argument's [animate] feature, with the inverse reading available only when the P object can be an agent-like topic.
APU-G, two heads: with an animate subject a grammatical P merges (applicative-like, in the verb's spine, MVS member) and the voice alternation can apply; with an inanimate one a lexical P merges (adjunct-like, outside the voice domain). Willie's own conclusion: “when the subject is animate, the postposition yii must be interpreted as a grammatical postposition, and the Inverse alternation applies” (p. 98).
A Both give the same strings. They differ in what they claim: PU-L says -ii’ is one lexical item and the inverse is a feature effect; PU-G says there are two syntactic heads spelled alike.
HYPOTHESIS The datum favors neither. It shows that “grammatical” is not a property you can read off the morpheme, which is the strongest sourced reason to doubt a simple “P = functional” label. Test: other Ps that are both, e.g. -tah, -ta’ (Willie p. 44 (37)–(39)); the next step is T32b. Update 12:46pm: the conditions for -ii’ being grammatical are tested on one-place verb + P §5. As worded (“the subject is animate”), Willie's condition fails on (84). An extension of Dowty (some proto-agent property, including movement relative to the P's object) fits all of her data.
Σ Σ(√GO) = {ρA}. The goal is not a role slot: =góó is a lexical P with a noun object (Willie p. 40).
BMergehooghan with =góó (noun complement, the enclitic's host). In the stored version, hooghan=góó is a listed phrase with the narrow meaning ‘home’, and it merges as a unit; in the unstored version it is built by Merge and means ‘to the hoghan’.
B Willie's judgment: with hooghan=góó déyá the reading ‘I am going to the hoghan’ is starred. So the stored version is the only one available for that string.
BSpell-out D2: hooghan=góó ⌢ déyá. It sits outside the MVS (lexical Ps are not MVS members, p. 40), which is T31f's question (see below).
SOURCED Willie's statement and judgment. HYPOTHESIS The listed-phrase mechanism: it is how a phrase-level lexical entry would enter the derivation, and is ours.
PU7. Chee & Yazzie (15b) bitah honiigaii: a verb + P pair with a shifted meaning
Σ Base: honiigaii ‘The area became hotter’ (with areal ho-). Pair: ‘He became sick, ill.’ The authors' gloss: 3sg.among.
ACompositional parse (PU-L): Merge P -tah ‘among’ with ρX [3]: bitah; Merge V honiigaii with the PP inside VP. The meaning would be ‘it became hot among him/it’. The reported meaning is ‘he became sick’.
AListed-pair parse (“phrasisized”): the entry is the pair [-tah + honiigaii] ↔ ‘get sick’, with the P's object slot linked to the sick person. Merge takes the pair as one chunk and still spells P's pronoun in the PP (bitah), so the pronoun slot stays visible to agreement and linking.
ASpell-out D1: bitah ⌢ honiigaii. Same string under both parses.
SOURCED The pair and its translation (Chee & Yazzie p. 35). HYPOTHESIS That the meaning is not predictable from ‘area became hot’ + ‘among it’ is our reading of their translations; they themselves say the postposition “extends or shifts” meaning (p. 32). OPEN Whether speakers store the pair (T32f).
6. Phrase-level lexicalization: what is sourced, what is not
#
Claim
Evidence (page)
Grade
L1
A noun + lexical postposition can be lexicalized with a narrow meaning
Willie p. 43 (35): hooghan=góó déyá ‘I am going home’, *‘to the hoghan’. Her words: “Lexical postpositions and their nominal objects can become lexicalized, that is, have a narrow interpretation.” One example
SOURCED
L2
Verb + grammatical postposition is a syntactic unit
Willie p. 50 (44): not interruptible; may be one phonological unit; domain of the verb's subcategorization. W&J p. 270: “noninterruptable constituent”. Hale 1997 PDF p. 31: “the proclitic portion of the verb word consists of the postposition”
SOURCED (unit-hood; not idiomaticity)
L3
Particular verb + postposition pairs with shifted meaning are attested
Chee & Yazzie pp. 22, 34–35: (8) ‘play’ → ‘make cry’; (13b) ‘frightening’ → ‘lazy’; (14b) ‘walk away out of sight’ → ‘help’; (15b) ‘area became hotter’ → ‘sick’; (16b) ‘walking around’ → ‘doing it’. They call them “phrasal verbs” that “may occasionally extend or shift to a new meaning” (p. 32); 17.4% of the children's and 19.3% of the caregivers' verb tokens occur with a postposition (Table 8, p. 32)
SOURCED as attested in four children's and their caregivers' speech
L4
Their own category is two-way: the postposition “can also add an argument” (compositional) or give a “new or extended meaning”
Chee & Yazzie p. 22; “this is not the only function of postpositions” (p. 35). The one passive + baa token had “no extended meaning” (p. 36)
SOURCED
L5
Willie's psych and ‘know’ verbs need their postposition and resist plain glossing
pp. 47–49 (41)–(43): the PP “is obligatory with these ‘psych’ verbs”; of (42) “an enlightening gloss is not easy to find”
SOURCED as her statements; that the pairs are stored is HYPOTHESIS
L6
Word-level lexicalization of a postposition-like element: thematic prefixes, “verb themes”
McDonough p. 32 (Table 2): Ib “postpositional stems”; p. 44: “verb theme” addresses “the non-compositional or non-transparent nature of many of the combinations of morphemes”; “not a technical term and Y&M do not use it in word formation”; p. 48: “‘verb theme’ per se does not play a role in word formation in the dictionary and grammar”
Secondhand: SOURCED as McDonough's report of Young & Morgan; Y&M itself OPEN
L7
A continuum: separate word → proclitic → thematic prefix
W&J p. 270: “Some postpositions and their objects in Navajo are procliticized to the inflected verb; others remain detached.” Hale 1997 PDF p. 33: P “is a proclitic … traditionally classified as belonging to the disjunct system”; McDonough Table 2 (positions 0, Ia, Ib)
The attachment facts SOURCED; reading them as a lexicalization path HYPOTHESIS
L8
Young & Morgan list particular verb + postposition frames
Only one entry seen, through McDonough p. 47 (20): the headword háníshtʼááh has the thematic prefix há-; bił appears only in an example sentence. Format: words with themes, not phrases
OPEN; HYPOTHESIS that the format hides phrase-level units
L9
Lexicalization applies to phrases such as verb + postposition as much as to words (“phrasisized”)
Compatible with L1, L3 and L5. Not tested: no opened source asks whether the pairs are stored
GRANT'S CONJECTURE
HYPOTHESISEnglish phrasal verbs, for comparison only. General background, not Navajo evidence and not opened here. English V + particle (give up) is often non-literal and allows give it up; V + preposition (look after) takes a noun phrase object. A Navajo verb + P pair differs in two ways that matter. The P carries its own agreement for its object (so it is closer to the preposition kind, which has a complement), and Willie's MVS is not interruptible (so it lacks the separability of English particles). Chee & Yazzie's use of “phrasal verb” covers both cases (argument-adding and meaning-shifting), so it is broader than the English term. “Phrasisized” is Grant's label for the step from a phrase to a stored unit; the nearest field notions are the Athabaskanist's verb theme (word-level) and the general notion of an idiom or multiword expression (not opened). No claim is made that the English term and the Navajo facts coincide.
HYPOTHESISWhat would count against “phrase-level lexicalization of verb + P”. Only data. For instance, if every meaning-shifting pair turned out to be derivable from a regular meaning of the P (then L3 would not need storage), or if the shifted meanings were all restricted to one speaker or one conversation (then they would be productive coinages, which Chee & Yazzie's “innovative” wording, p. 36, allows). Silence in the sources counts against nothing.
7. Dictionary entries added (group M)
All forms are copied with their source and example number; frames and entailments are marked ours. Fifteen new entries; the five older verbs that already carry postposition frames are linked, not duplicated.
Already in the dictionary with postposition frames: yáłtiʼ (yichʼįʼ/bichʼįʼ), yoolóós (yá/bá), hózin (shił béé), nizhóní (shił), yishááł (bichʼįʼ, (36)).
8. Results
Question
Answer from the sources
Grade
Is the Navajo postposition T or C?
No opened source says so. It is treated as P
OPEN for T and C
Is it P, a lexical head?
Yes in Hale 2003:32, Hale 1997 (23)–(24) and PDF 43, Hale & Keyser PDF 30
SOURCED
Is it a light p or a p-shell, or P under vP?
Not stated or tested in any opened source for Navajo
OPEN
Are some postpositions functional?
Hooks: Willie's “grammatical”, the null postposition (Hale 2003:22, 59), Pcaus (Hale 1997 PDF 53). Counter-datum: Willie p. 98
HYPOTHESIS
Is verb + P one unit?
Syntactically yes (Willie (44); W&J p. 270); phonologically sometimes
SOURCED
Are verb + P units “phrasal verbs”?
Chee & Yazzie call them so, with meaning shifts in 5 pairs (and one no-shift contrast)
SOURCED (their term and data); stored units HYPOTHESIS
Is there phrase-level lexicalization?
Explicit for noun + lexical P (Willie (35)); suggestive for verb + P
L1 SOURCED; L9 GRANT'S CONJECTURE
Does the inverse bi- sit on P?
Yes in four pairs (W&J (50); Willie (80), (84), (86))
SOURCED; spell-out under IA-1* OPEN (T31e, T32h)
For Grant's thesis.HYPOTHESIS Nothing here contradicts the weak thesis that the role-bearing pronoun slots are the verb's arguments and nouns are linked add-ons. The postposition's pronoun is another such slot, valued inside the PP (Hale 2003:46). The new point is that a verb's argument frame can include a postposition (Willie p. 50 (44c); the psych verbs), so the verb's “frame” may be a verb + P pair, which matches the phrase-level idea.
9. Tasks T32a–j and next steps
ID
Question
What would settle it
T32a
OPEN Does any Navajo source place P under a light p or v, or say P is functional? (Speas 1990; Hale & Keyser 1993 in print; Hale & Platero 1996)
Open them and quote the pages
T32b
OPEN PU-L or PU-G? Are there two grammatical PPs in one clause? Do -tah/-ta’ pattern as both kinds (Willie p. 44 (37)–(39))?
Willie Ch. 3–4 re-read for two-P strings; Young & Morgan 1987:25–26. Update 12:46pm: Willie (93) bimá yił chidí yiiʼ siké has two y- (grammatical) PPs with one verb, which a single-slot PU-G cannot derive (E6, T33h)
T32c
OPEN Is =go (COMP in W&J (49)) related to the postposition =góó?
Young & Morgan; historical Athabaskan work; a speaker
T32d
OPEN Does “theme in Spec,PP” (Hale 1997 (23)–(24)) extend beyond the drift-away verbs?
Hale 1997 PDF pp. 33–40; Hale & Platero 1996
T32e
Done 2026-10-17 12:46pm Willie p. 97 (“no grammatical postpositional phrases” in intransitives) vs p. 94 (79)/(80)
Worked on one-place verb + P: p. 97 holds for the whole MVS, not the bare verb (T-MVS, HYPOTHESIS). Willie's animacy condition fails on (84) as worded. LIC-EXT and LIC-PROTO both fit; T33a decides between them
T32f
OPEN Are the Chee & Yazzie pairs stored, or built productively? Do Young & Morgan list them?
Young & Morgan 1987 entries for hóyééʼ (p. 458) and honiigaii (p. 667); Chee 2025; speakers
T32g
OPEN Speas 1986 and 1990 on the postposition and on b- as an incorporated pronoun (a search snippet attributes this to Speas); the unidentified Hale paper on the inchoative (identified 2026-10-19: Hale, Munro & Platero 2000, “The Navajo inchoative verb form”, MITWPL 3, pp. 97–116, from Hale 2001's references and the PA paper p. 60; not on the box; T38g; what Hale reports of it is on the ‘start to’ test)Update 2026-10-19, 12:46pm: the search snippet is the Hale, Munro & Platero inchoative manuscript (hale028). Its text says the inchoative “behaves like an ‘incorporated’ postposition—by definition, a postposition requires overt agreement”, and n. 3 cites Speas (1988, as HMP print it) for b- as an incorporated pronoun. That passage is now read; Speas herself is still not opened (HMP page §3.1).
Read Speas 1986 ch. 4 page images; find the Hale paper
T32h
OPEN How does IA-1* spell bi- on P in all four pairs? (= T31e)
Theory-internal; more oblique-inverse data
T32i
OPEN Does a noun ever stand between a bare-noun P and the verb, and is a lexical PP inside D1? (= T31f)
Willie pp. 39–70; texts
T32j
OPEN Test “phrasisized” (L9): elicit or find verb + P pairs whose meaning is not predictable, and ask whether they behave as units (no interruption, fixed order, stored forms)
Grant's texts; a speaker; Young & Morgan examples
Next steps (suggested). (1) Get Young & Morgan 1987 (or Young, Morgan & Midgette 1992) and check the entries for the five Chee & Yazzie bases and for háníshtʼááh. (2) Read Speas 1986 ch. 4 on page images for any statement on postpositions and b-. (3) Look for one more oblique-inverse datum with a lexical P. (4) Decide with Grant whether to promote PU-G to a rule or leave it as a hypothesis.
Changes made elsewhere for this page. Links added from the clause spine (T31e and T31f rows) and from open questions (T32a–j); the verb dictionary has a new group M with 15 entries; sources has a new block; the index lists the page.
Can a one-participant verb have a “grammatical” postposition and an inverse? Willie 1991 p. 97 against p. 94, and what decides it (T32e)
2026-11-11 audit pass 2. Apostrophe fixes only. Navajo glottal stops typed as ' or ’ (for example in prefix names like ʼa- and ʼeʼe-, or words like Awééʼ) now print as ʼ (U+02BC), and a build step no longer turns the closing quote of an English translation inside a Navajo row into a glottal stop (‘…the girls.’, not ‘…the girls.ʼ). No form, gloss or translation wording changed.
Restated on the unified structure (2026-11-15). Postpositions are restated as [PP DP P⁰] with left specifiers: a grammatical P agrees with the DP in its specifier, a lexical P has no probe (H64); a selected PP is merged in VP, and Willie’s “cannot be interrupted” follows from the default, HS-1 and left specifiers. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Status on the spine: see unified structure, §2b and §4, postposition row. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
The postposition-units page left a tension open (T32e). Willie 1991 says on p. 97: “In intransitive sentences, there are no Inverse forms, and no grammatical postpositional phrases, only lexical ones.” Yet on p. 94 her (79)/(80) have the intransitive verb yáłtiʼ ‘is speaking’ with an inverse postposition (yichʼįʼ / bichʼįʼ). This page re-reads Willie pp. 37–50, 57, 63–64 and 93–100 on page images, collects every one-participant verb + postposition string she prints, states the rules formally, tests five candidate conditions against the data, derives the key sentences step by step, and asks what the result means for Grant's point that lexicalization happens at the phrase level as much as at the word level.
Ground rules (Grant's). No analysis is assumed to be fully right; each may be partly right. Silence counts against nothing; only contradicting data count. Grant does not speak Navajo. Every Navajo form below is copied from Willie 1991 page images read this run, or from a site page that records it with page and example number. No Navajo form, gloss, judgment or index is invented; where a test needs a sentence nobody printed, it is given in letters only. Dowty's proto-roles are the baseline; the licensing condition proposed below is framed as an extension of Dowty. “Role in the action” means a Dowty entailment (movement, sentience …), not an event-structure role.
Rule-set letters and notation. Site convention: Σ = role-schema step, A = word level (the verb word and its postpositions, spelled out in cycle D1 = the MVS; see clause spine §4), B = clause level (nouns, lexical adjuncts, linking, cycle D2). Spine: Qual > M/S (= T) > v > V (Hale 2003 pp. 10, 32). ρ = a pronoun slot; ρX = the slot that is the object of a postposition. Slot features [top]/[foc] as in VOICE-TF (clause spine): direct = agent-like slot [top], other slot [foc]; inverse = the reverse. No notation conventions file from the Syntax · Semantics · Interface bot has arrived (checked: its BRIEF.md and VERB-QUEUE.md have no notation section), so the site's own notation is kept.
Bottom line, in plain words
SOURCEDThe contradiction is real on the page, and Willie's own next paragraph limits it. Willie defines an intransitive as a verb “that ha[s] only a single (direct) argument, the subject” (p. 57), and on p. 63 she explicitly allows “an intransitive verb and a postpositional phrase” to form an MVS, with yichʼįʼ yáłtiʼ as the example. So taken at face value, p. 97 contradicts pp. 63 and 94, and also (83)/(84) on p. 95 and (90) and (93) on pp. 98–99. But p. 97's examples are introduced to show a split: with a ball as the one that lies, only the lexical form bii’ is possible ((89)); with a woman as the one who sits, yii’ appears and “the Inverse alternation applies” ((90), (91); her conclusion, p. 98). On p. 99 an impersonal passive also forces the lexical form ((94), *(95)), while the active ‘climb’ allows the inverse ((96), (97)).
HYPOTHESISReconciliation (T32e). Read p. 97 at the level of the whole verb-plus-postpositions unit (the MVS), not the bare verb. It then says: an MVS with only one argument slot has no inverse and no grammatical PP. That is consistent with every example in the book. The real content sits in pp. 98–99: which one-participant verbs let the postposition's object join the argument frame. Willie's wording shifts from “intransitive verb” (p. 63) to “intransitive sentences” (p. 97) and calls the ‘climb’ verb with yaah “transitive” (p. 99). That shift supports the MVS-level reading but does not prove it.
HYPOTHESISWhat decides it. Willie's own condition, an animate subject (p. 98), fails on (84) ‘The ball rolled to the baby’ unless “subject” means the subject after inversion. Two conditions fit all of Willie's data. (i) Some definite argument is animate. (ii) The one participant has at least one of Dowty's “active” proto-agent properties (volition, sentience, causation, movement relative to another participant) in the action, relative to the postposition's object (LIC-PROTO, ours; an extension of Dowty 1991:572). The two come apart only for an inanimate mover with an inanimate goal. No Willie example has that. Witherspoon's (11) ‘the rock rolled upon the tree’, already on the site, is exactly that case, and it has the yi- form. It counts for LIC-PROTO only if its ‘upon’ is a postposition (T20c, open).
HYPOTHESISFor Grant's phrase-level lexicalization. Willie Ch. 3 never calls a verb + postposition pair idiomatic. Every pair on pp. 93–100 means what its parts mean (speak to, roll to, lead for, sit in, lie in, sit with, climb on). Her only lexicalization statement is about a noun + lexical postposition (p. 43 (35)). More to the point, whether a postposition counts as grammatical changes within one verb + postposition pair: with voice ((94) vs (96), same ‘climb’ + -ah ‘on’) and with the reading ((90) vs (91), same ‘sit’ + -ii’). So if verb + postposition pairs are stored, as Grant's “phrasisized” idea says, what is stored is the pair's form and meaning, not whether the postposition's object is one of the verb's participants. That part is worked out fresh in each sentence. Chee & Yazzie's meaning-shifted bitah honiigaii ‘he became sick’ fits this split (§7).
SOURCEDOPENTwo new problems found while re-reading. (a) In (92) and (93), a noun stands between a grammatical postposition and the verb (bimá yił chidí yiiʼ siké). Willie p. 40 says grammatical postpositions are “a component of the MVS”, and p. 50 (44a) says the MVS “cannot be interrupted”. These cannot all hold unless only the postposition next to the verb is inside the MVS (T33d, which joins T31f/T32i). (b) Willie lists -aa ‘to’ as occurring “only with animates” (p. 41), but in (83) its object is a ball (T33c). Also, (93) has twoy- postpositions with one verb, which matters for PU-L vs PU-G (T32b).
HYPOTHESISFor Grant's thesis and STRATA. Every condition that survives refers to properties of pronoun slots: definiteness, Dowty entailments, and the number of slots in the MVS. None refers to nouns. In (89)/(90) the verb stems already differ (Willie glosses ‘lies’ vs ‘sits’), so the deciding property can be read off the verb before any noun is chosen. Nothing here contradicts the weak thesis. Under STRATA (Grant's conjecture), LIC-PROTO is a Layer-1 condition, the voice is decided on slots, and nouns arrive last. The condition stated as noun animacy would sit badly with STRATA unless animacy is a feature of the slot.
1. Sources opened or re-read this run (2026-10-17, 12:46pm PT)
Willie, MaryAnn 1991.Navajo Pronouns and Obviation. PhD dissertation, University of Arizona. Box scan (metal-doping-theta-roles/lit/_pdfs/willie1991.pdf, 243 PDF pp.). Re-read on page images (110 dpi; 300 dpi crops for (89)–(97)): printed pp. 37–44 (PDF = printed + 5), §2.3.2 “The Impersonals” p. 38, §2.4 pp. 39–44 incl. (27), (28), (30), (35)–(39); p. 50 (44); p. 57 §3.1.3; pp. 63–64 (26)–(29) (PDF = printed + 6); §3.3.4–3.3.5 pp. 93–100, (77b)–(97). The OCR text was used only to find passages and to search the whole thesis for “idiom”, “lexicaliz”, “intransitive” and “Impersonal”.
Not opened: Young & Morgan 1987 (needed for T33b, T33e); Fernald et al.'s individual postposition entries for -kʼi and -ii’.
2. The tension, stated exactly
2.1 Two uses of “intransitive”, two tests for “grammatical”
#
Willie's words (page)
What it commits her to
W1
“intransitive verbs that have only a single (direct) argument, the subject” (p. 57); “The single argument of an intransitive may have an agent, patient or theme theta role” (p. 57)
“Intransitive” counts only direct arguments (verb prefixes), not postpositional objects
W2
“sentences where the MVS is composed of an intransitive verb and a postpositional phrase”, example (26) ʼatʼééd yichʼįʼ yáłtiʼ (p. 63)
An intransitive verb can come with a PP inside the MVS; -chʼįʼ is on her list of grammatical postpositions ((28b), p. 41)
W3
“The postpositional phrase corresponds to an oblique argument of the verb … some are case-marked arguments that the verb is subcategorized for, and some are more like adjuncts. I will call the former the grammatical postpositions, and the latter the lexical postpositions” (pp. 39–40)
Test 1 for “grammatical”: argumenthood (subcategorized vs adjunct-like)
W4
“The grammatical postpositions occur only with pronominal objects and are a component of the MVS. The lexical postpositions occur only with nominal (lexical) objects” (p. 40); repeated p. 50
Test 2: object type (pronoun vs noun). Fits -tah: Tsintah (noun object) vs Nihitah (pronoun object) (p. 44 (38), (39))
W5
“The Direct/Inverse voice alternation applies to these arguments also, since they are a part of the MVS” (p. 93)
Voice reaches exactly the grammatical PPs
W6
“In intransitive sentences, there are no Inverse forms, and no grammatical postpositional phrases, only lexical ones. The third person object pronoun of a lexical PP is bi-” (p. 97)
The disputed statement; also: a lexical PP can have a pronoun object, b-
W7
“I conclude that when the subject is animate, the postposition yii must be interpreted as a grammatical postposition, and the Inverse alternation applies” (p. 98)
The split inside intransitives, by animacy
W8
“the verb is a simple passive, an Impersonal; therefore, the postpositional phrase must be lexical” (p. 99); “With a corresponding transitive verb, the Inverse can apply” (p. 99)
A second condition (voice); ‘climb’ with yaah is called “transitive”
SOURCEDThe clashes inside the text. (i) W6 vs W1 + W2: with W1's definition, (26), (79)/(80), (83)/(84), (90) and (93) are intransitive verbs with grammatical PPs, which W6 excludes. (ii) W6 vs W4: bii’ in (89a) has a pronoun object, so W4 makes it grammatical, while W6 makes it lexical. On p. 97 Willie uses test 1 (argumenthood), not test 2. Note also that -ii’ ‘in’ and -ah ‘on’ appear on neither of her lists ((28), (30)).
2.2 Three ways to read p. 97
Reading
What “intransitive sentence” means
Consistent with Willie's data?
Grade
T-STEM
A sentence whose verb has one direct argument (W1)
No: (26), (79)/(80), (83)/(84), (90), (93) are counterexamples printed by Willie herself, (90) on the very next page
Rejected as a reading of her intent; SOURCED counterexamples
T-MVS
A sentence whose MVS has one argument slot, i.e. the verb's one direct argument and no grammatical PP
Yes. The PP half is then close to definitional (W3 + W5), and the inverse half is substantive: one slot gives nothing to rank. The question “which one-participant verbs get a grammatical PP?” passes to pp. 98–99
HYPOTHESIS (adopted)
T-DEFAULT
The general case, which p. 98 then restricts (inanimate subjects only)
Yes for (89)/(90), but needs a second exception for (84), whose roller is inanimate
HYPOTHESIS
HYPOTHESISWhy T-MVS. Three wording clues point the same way. Willie writes “intransitive sentences” on p. 97, against “intransitive verb” on pp. 57 and 63. She calls haasʼnaʼ with yaah (96) a “corresponding transitive verb” (p. 99), although no object noun or object prefix is shown (she does not segment it; whether the stem is transitive is T33e). Her p. 100 summary says “with constructions containing an oblique argument, the Inverse produces a goal or benefactive subject”, which counts the oblique as an argument of the construction. T-MVS and T-DEFAULT give the same rules once the condition of §5 is added. They differ only in what Willie meant, which a text cannot settle (OPEN as to intent).
3. Data (Willie 1991, copied from page images)
Gloss lines are Willie's, copied as printed (her order for postpositions is object.P, e.g. “3.to”). The Leipzig column is ours. It splits off only the postposition's pronoun prefix, which Willie herself identifies ((27), p. 40), uses - for that boundary and . for unsegmented material. Verbs are not segmented beyond her glosses. ✓ = printed without a mark; * / ??? as printed.
Ex.
Navajo (as printed)
Willie's gloss and translation
Leipzig (ours)
Verb's one participant
P form
Page
(26)
ʼatʼééd yichʼįʼ yáłtiʼ
girl 3.to 3.speaking ‘He is speaking to the girl.’
girl 3-to 3.speak
speaker (‘he’)
y-
63
(79)
ʼashkii ʼatʼééd yichʼįʼ yáłtiʼ
boy girl 3.to 3.speaking ‘The boy is speaking to the girl.’
boy girl 3-to 3.speak
the boy
y- (direct)
94
(80)
ʼashkii ʼatʼééd bichʼįʼ yáłtiʼ
boy girl 3.to 3.speaking ‘The boy is being spoken to by the girl.’
boy girl 3-to 3.speak
the girl
b- (inverse)
94
(83)
ʼAwééʼ jooł yaa nímááz
baby ball 3.to 3.rolled ‘The baby rolled to the ball.’
baby ball 3-to 3.roll
the baby
y- (direct)
95
(84)
ʼAwééʼ jooł baa nímááz
baby ball 3-to 3-rolled ‘The ball rolled to the baby.’
baby ball 3-to 3.roll
the ball (inanimate)
b- (inverse)
95
(89a)
jooł chidí biiʼ siʼą́
ball car 3.in 3.lies ‘The ball is in the car.’
ball car 3-in 3.lie
the ball (inanimate)
b- (lexical)
97
*(89b)
jooł chidí yiiʼ siʼą́
(The ball is in the car.)
ball car 3-in 3.lie
the ball
*y-
97
(90)
ʼasdzą́ą́ chidí yiiʼ sidá
3.in 3.sits ‘The woman is sitting in the car.’
woman car 3-in 3.sit
the woman
y- (direct)
98
???(91)
ʼasdzą́ą́ chidí biiʼ sidá
INV3.in 3.sits [The car is sitting in the woman]
woman car INV.3-in 3.sit
the car
b- (inverse)
98
(92)
shimá bił chidí biiʼ séké
my mother 3.with car 3.in DUAL.sitting ‘I sit with my mother in the car.’
1.mother 3-with car 3-in DU.sit
‘I’ (+ mother)
b-, b- (local person present)
98
(93)
bimá yił chidí yiiʼ siké
my mother [sic: bi-] 3.with car 3.in DUAL.sitting ‘He sits with his mother in the car.’
3.mother 3-with car 3-in DU.sit
‘he’ (+ mother)
y-, y-
99
(94)
tsin baah haʼasʼnaʼ
tree 3-on it was climbed ‘The tree was climbed.’
tree 3-on climb.IMPERS (Willie: “simple passive, an Impersonal”)
none definite
b- (lexical)
99
*(95)
tsin yaah haʼasʼnaʼ
(no gloss)
tree 3-on climb.IMPERS
none definite
*y-
99
(96)
tsin yaah haasʼnaʼ
‘He climbed the tree.’ (no gloss line)
tree 3-on 3.climb
climber (‘he’)
y- (direct)
99
???(97)
tsin baah haasʼnaʼ
[The tree climbed up on him.]
tree 3-on 3.climb
the tree
b- (inverse)
100
Fidelity. The site's earlier copy of (89) and (90)–(91) on the postposition-units page and in the dictionary dropped the nasal hooks (“siʼá”, “ʼasdząą́”). The page images have siʼą́ and ʼasdzą́ą́, and both places are corrected this run. In (93) Willie's gloss line says “my mother” under bimá, while her translation has ‘his mother’; this is copied with [sic]. (92) séké and (93) siké are both glossed “DUAL.sitting” although the translations have a single subject. Willie's (29) shimá bił déʼáázh ‘I am going with my mother’ (p. 41, “dl.walk”) shows the same pattern. That the dual counts the comitative partner is our reading of her translations (HYPOTHESIS). The Leipzig label “IMPERS” for (94) uses Willie's own word; I do not segment the form. That its impersonal is the ʼa- type of her §2.3.2 (p. 38) is suggested by the spelling (haʼasʼnaʼ vs haasʼnaʼ) but not stated (HYPOTHESIS).
4. Formal rules (stated before the tests)
Rule
Statement
Basis
Grade
GP-ARG
If P is grammatical, its object slot ρX is a member of Σ(MVS), merges inside the verb's projection (rule set A), and is spelled in D1. If P is lexical, ρX ∉ Σ(MVS), and the PP is a clause-level adjunct (rule set B, D2)
Willie W3, W4 (“component of the MVS”), W5; MVS-D1 (clause spine)
Content SOURCED; A/B and D1/D2 placement HYPOTHESIS
B-DEFAULT
The 3rd-person object pronoun of a lexical P is b-. Corollary: a 3rd-person y- on a P shows that the P is grammatical
W6 (p. 97); (89b), (95)
Rule SOURCED; corollary HYPOTHESIS
VOICE-MIN
The direct/inverse choice (VOICE-TF) is defined only when Σ(MVS) has at least two definite 3rd-person slots. With fewer, a 3rd-person slot on P gets the default b- (with a local person present, b- as in (92); Hale 2003:59)
W5, W6, p. 100 (inverse subject “definite”)
HYPOTHESIS
DEF
Only definite, referential slots count for VOICE-MIN and for the licensing condition. The impersonal argument does not count
Willie p. 38: the impersonal ʼa- “marks an indefinite … argument” and “never permits an adjoined coreferent nominal”; p. 50 (two classes of argument prefixes); p. 99 (94)
HYPOTHESIS from SOURCED statements
LIC-?
For a verb with one direct slot ρ1 and a P whose object could be a slot ρX: P is grammatical iff condition C holds. Candidate conditions in §5
W7, W8
HYPOTHESIS
HYPOTHESISWhat follows without any LIC. GP-ARG + B-DEFAULT + VOICE-MIN derive *(89b) and *(95) once the P is known to be lexical. A lexical P has no voice, so y- cannot appear. They also derive p. 97 under T-MVS: one slot means no inverse, and no grammatical PP by definition. The open part is LIC.
5. Five candidate conditions, tested against every one-participant string
Name
Condition C (P is grammatical iff …)
Source of the idea
LIC-AN
… the verb's subject ρ1 (the one that sits, lies, rolls) is animate and definite
Willie p. 98, read literally
LIC-EXT
… the resulting MVS has an animate, definite slot that can be the external argument (the direct subject, or the inverse subject)
Willie p. 98 + p. 100 (“the Inverse produces a goal or benefactive subject”; inverse subjects are “high on the animacy scale”)
LIC-PROTO
… ρ1 is definite and has at least one of Dowty's Proto-Agent properties (a) volition, (b) sentience/perception, (c) causing an event or change in another participant, (d) “movement (relative to the position of another participant)” in the action, with ρX as the other participant
Ours: an extension of Dowty 1991:572 (27a–d). The parenthesized (e) “exists independently” is excluded, since every participant here has it. Witherspoon's movement point (p. 69, as recorded)
LIC-PAIR
… the stored verb + P pair is listed as grammatical
Grant's phrase-level lexicalization, applied to argument status
LIC-VOICE
… the verb is not impersonal (needed in addition to one of the above)
Willie p. 99
Datum
Observed
LIC-AN
LIC-EXT
LIC-PROTO
LIC-PAIR
(26), (79)/(80)
grammatical; inverse possible
✓ (speaker animate)
✓
✓ (volition, sentience)
✓ (listed)
(83)
grammatical, direct
✓ (baby)
✓
✓ (movement toward the ball; sentience)
✓
(84)
grammatical, inverse; roller = ball
✗ (roller inanimate) unless “subject” = inverse subject, which is LIC-EXT
✓ (baby = inverse subject)
✓ (the ball moves relative to the baby: (d))
✓
(89a), *(89b)
lexical only
✓ (ball inanimate)
✓ (no animate slot)
✓ (lying entails none of (a)–(d))
✓ (listed as lexical)
(90), ???(91)
grammatical; inverse gives a nonsense reading
✓
✓
✓ (sentience; volition if sitting is volitional)
✓
(92), (93)
grammatical (both Ps y- in (93))
✓
✓
✓
✓
(94), *(95)
lexical only (impersonal)
✓ with DEF
✓ with DEF
✓ with DEF (the climber is not a definite slot)
✗ if haʼasʼnaʼ and haasʼnaʼ are one verb + -ah pair; ✓ only if the impersonal is listed separately
(96), ???(97)
grammatical; inverse gives a nonsense reading
✓
✓
✓ (volition, movement)
✓
Witherspoon (11) Tsé tʼiis yikʼichʼínímááz ‘the rock rolled upon the tree’, acceptable; *(12) the bi- form (1977:66, as recorded)
yi- with two inanimates
✗ if ‘upon’ is a P whose object is the yi-
✗ if so (no animate slot)
✓ (the rock moves relative to the tree)
no prediction
Witherspoon (13) Tʼiis tsé yikʼiikę́ę́z ‘the tree fell upon the rock’ (as recorded)
same
✗ if P
✗ if P
✓ (movement)
no prediction
HYPOTHESISResults. (1) LIC-AN read literally is contradicted by (84). Its repaired form is LIC-EXT. (2) LIC-EXT and LIC-PROTO both fit all of Willie's strings. They come apart only when an inanimate participant moves relative to an inanimate postpositional object: schematically, [Ninan1 Ninan2y-P Vmotion]. LIC-EXT predicts *y-, LIC-PROTO predicts y- is fine. (3) Witherspoon's (11) and (13) are such strings and have yi-. But Witherspoon glosses the whole word ‘it-upon-…-it-rolled’ without segmenting it. Whether ‘upon’ is a postposition procliticized to the verb (as Hale 1997 says of such proclitics, PU2) or part of a transitive verb word is T20c. So they favour LIC-PROTO only conditionally (OPEN, T33a, T33b). (4) LIC-PAIR alone fails on (94)/(96) if the impersonal and active ‘climb’ count as one pair. It then needs LIC-VOICE or DEF, and those do the work that pair-listing was meant to do. (5) DEF is needed by every candidate. It is independently motivated, because voice ranks definite slots (p. 100).
HYPOTHESISWhy LIC-PROTO is the more economical statement. Dowty's (d) is relational (“relative to the position of another participant”), so it needs exactly the second participant that the postposition supplies. In (89) neither the ball nor the car moves, and the car is “stationary” in Dowty's sense (28d) only relative to a mover, of which there is none. LIC-PROTO also says the same thing as Witherspoon's “movement and action are part of the considerations that decide who can control or act upon whom” (p. 69, as recorded), now applied to whether a postposition's object joins the frame. That is an extension of Dowty's baseline: Dowty uses the properties to choose subject and object, and here they also decide whether an oblique counts as an argument at all. Willie's animacy (p. 100: the inverse features “are associated with referents that are high on the animacy scale”) is then a frequent correlate, not the condition.
6. VPISH-aware derivations E1–E7
Format as on the clause spine and postposition units. The PP is in VP, P head-final (PU-L), unless noted. Roles are read off Willie's translations. Every statement about where things merge is ours (HYPOTHESIS); the strings, the glosses and the judgments are Willie's (SOURCED). VPISH: the subject slot is merged inside vP (Spec,v) and is not raised to Spec,TP by any step below. Hale 2003 posits no A-movement, and IA-1* raising applies only in the inverse (inverse as Agree).
E1. (89a) jooł chidí biiʼ siʼą́ ‘The ball is in the car’: one-slot MVS, lexical P
Σ Σ(√LIE) = {ρTh} (the thing that lies; Willie p. 57 allows a theme as the single argument). Candidate ρX (the container). LIC-PROTO: ρTh has none of (a)–(d) relative to ρX. LIC-EXT: no animate slot. Both conditions fail, so P is lexical and ρX ∉ Σ.
AMerge V siʼą́ (unsegmented) with ρTh; Merge v, M/S; Agree M/S ⟶ ρTh [3]; Qual. VOICE-MIN: one slot, so no [top]/[foc] split inside Σ.
ASpell-out D1 = MVS = siʼą́ alone.
BMerge P -ii’ with a pronoun ρX [3] (outside Σ). Agree P ⟶ ρX: no voice on a lexical P, so B-DEFAULT gives b-: bii’. Merge (adjoin) the PP at clause level.
BLinkchidí ⟶ ρX (P's object; next to the PP) and jooł ⟶ ρTh. Spell-out D2: jooł ⌢ chidí ⌢ bii’ ⌢ siʼą́ ✓.
*(89b): y- on P would need VOICE-TF, which needs ρX ∈ Σ(MVS) (GP-ARG + VOICE-MIN). ρX ∉ Σ, so the derivation crashes ✓.
HYPOTHESISPrediction. The lexical PP is spelled in D2, the grammatical one in D1 (E2). The strings look alike, so the difference should show up, if at all, in prosody or interruptibility: something could stand between bii’ and siʼą́ in (89a) but not between yii’ and sidá in (90). Nothing printed tests this (T33d).
E2. (90) ʼasdzą́ą́ chidí yiiʼ sidá and ???(91) … biiʼ sidá: grammatical P, direct and inverse
Σ Σ(√SIT) = {ρA (the sitter: sentience; Willie: “the subject … is animate”, p. 97)}. LIC-PROTO ✓ and LIC-EXT ✓, so P is grammatical: Σ(MVS) = {ρA, ρX}. Two definite 3rd-person slots, so VOICE-MIN allows the voice choice. (90) direct: ρA [top], ρX [foc]. (91) inverse: ρX [top], ρA [foc].
AMerge P -ii’ with ρX (Comp,P); Merge V sidá with the PP (inside VP, Hale 1997 (23)); Merge v; ρA in Spec,v (VPISH); Merge M/S; Agree M/S ⟶ ρA; Qual. IA-1*, (91) only: v[EPP] attracts ρX[top]. Its only exponent is the prefix on P, so the lower copy is pronounced, as in PU4 (T31e).
BLink by FOC-LINK (the noun right before the MVS goes to the [foc] slot; W&J p. 268, as on the clause spine). (90): chidí ⟶ ρX, ʼasdzą́ą́ ⟶ ρA: ‘the woman sits in the car’ ✓. (91): chidí ⟶ ρA, ʼasdzą́ą́ ⟶ ρX: ‘the car sits in the woman’. The grammar converges and the meaning is absurd. That matches Willie's “total nonsense … but it has the interpretation shown” (p. 98).
BSpell-out D2: ʼasdzą́ą́ ⌢ chidí ⌢ [yiiʼ sidá] ✓.
HYPOTHESIS In (91) the absurd sitter is assigned by linking (rule set B), after Σ has been fixed with a sentient ρA. So “???” is a clash between the slot's entailments and the noun linked to it, not a failure of LIC. This is the same pattern as Witherspoon's *(12) and *(14) (unacceptable inverse with two inanimates), and it fits the capacity reading of bi- on who acts on whom.
E3. (84) ʼAwééʼ jooł baa nímááz ‘The ball rolled to the baby’: an inanimate mover with an inverse P
Σ Σ(√ROLL) = {ρ1 (the roller)}. Candidate ρX (the goal of -aa). LIC-PROTO: ρ1 moves relative to ρX (Dowty (d)) ✓. LIC-EXT ✓ (ρX = the baby, animate). LIC-AN read literally: ρ1 = the ball, inanimate ✗. So P is grammatical; Σ(MVS) = {ρ1, ρX}. Inverse: ρX [top], ρ1 [foc].
AMerge P -aa + ρX; Merge V nímááz; v; ρ1 in Spec,v; M/S; Agree M/S ⟶ ρ1; Qual. (IA-1*: as E2, lower copy pronounced.)
B FOC-LINK: jooł ⟶ ρ1 [foc]; ʼAwééʼ ⟶ ρX [top]. This is Willie's interpretive principle in a different form: in the inverse, “the nominal immediately preceding the PP is not coreferential with the postpositional object pronoun” (p. 96). Reading: ‘the ball rolled to the baby’ ✓.
BSpell-out D2: ʼAwééʼ ⌢ jooł ⌢ [baa nímááz] ✓.
OPEN Willie lists -aa ‘to’ (28a) among the postpositions that “occur only with animates” (p. 41). In (83) the object of yaa is the ball. Either (83)'s -aa is a different item, or the restriction is a tendency (T33c).
E4. (94) tsin baah haʼasʼnaʼ ‘The tree was climbed’ and *(95): impersonal, lexical P
Σ Σ(√CLIMB, impersonal): the climber is not a definite slot (DEF; Willie p. 99 “an Impersonal”; p. 38: the impersonal argument “never permits an adjoined coreferent nominal”). Candidate ρX (the tree). No definite slot carries (a)–(d), so LIC-PROTO fails; LIC-EXT fails too (the tree is inanimate). P is lexical.
AMerge V haʼasʼnaʼ (unsegmented), v (impersonal), M/S, Qual. Spell-out D1 = haʼasʼnaʼ.
*(95) yaah: needs the voice, which needs two definite slots. Crash ✓.
HYPOTHESIS Note what is not claimed: that ‘was climbed’ promotes the tree to subject. Willie calls it a “simple passive” but glosses ‘tree 3-on’, so the tree stays the postposition's object. On this analysis the impersonal has no definite subject at all, which is why nothing can be ranked.
E5. (96) tsin yaah haasʼnaʼ ‘He climbed the tree’ and ???(97) tsin baah haasʼnaʼ
Σ Σ(√CLIMB) = {ρA (the climber: volition, movement relative to the tree)}. LIC-PROTO ✓, LIC-EXT ✓ (‘he’). P is grammatical: Σ(MVS) = {ρA, ρX}. (96) direct, (97) inverse.
AMerge P -ah + ρX; V haasʼnaʼ; v; ρA in Spec,v; M/S; Agree; Qual. Agree P ⟶ ρX: [foc] → yaah (96); [top] → baah (97). Spell-out D1.
B (96): one noun; by Willie's translation it is the P's object: ‘he climbed the tree’ ✓. (97): FOC-LINK puts tsin on ρA [foc], giving ‘the tree climbed up on him’, the absurd reading Willie reports ✓.
HYPOTHESIS (94) and (96) share the root ‘climb’ and the postposition -ah, and differ in voice. The postposition's status changes with the voice. That is the strongest sourced reason not to store “grammatical vs lexical” with the verb + postposition pair (§7).
E6. (93) bimá yił chidí yiiʼ siké ‘He sits with his mother in the car’: two grammatical Ps
Σ Σ(√SIT.DU) = {ρA (the sitters; dual)}. Two candidate obliques: ρCom (object of -ił ‘with’, grammatical (28g), animate-only) and ρX (object of -ii’). LIC ✓ for both (an animate, sentient sitter). Σ(MVS) = {ρA, ρCom, ρX}, all 3rd person, all direct.
AMerge-ii’ + ρX; Merge V siké; Merge-ił + ρCom (PU-L: a second PP in VP; PU-G: a second applicative-like head). v; ρA; M/S; Agree; Qual. Agree on each P: yii’, yił (direct; y- on both means both are grammatical, by the B-DEFAULT corollary).
A/BSpell-out: the printed order is bimá ⌢ yił ⌢ chidí ⌢ yii’ ⌢ siké. The noun chidí stands between yił and the verb. If D1 contained both PPs, a D2 noun could not land inside it. So either yił leaves D1 (a rule-set-B Move of a grammatical PP to the clause level) or it was never in D1. See §8.
BLinkbimá ⟶ ρCom (next to yił), chidí ⟶ ρX (next to yii’) ✓.
HYPOTHESIS For T32b (PU-L vs PU-G): (93) has two y- postpositions with one verb. PU-G, with a single applicative-like slot per clause, cannot derive it. It would need two such heads or one head with two specifiers. PU-L derives it with no extra assumption. This is the first printed string found that bears on T32b. Willie's ban on stacking (p. 44) concerns two postpositions on one phrase, not two phrases in one clause, so it does not apply.
E7. (92) shimá bił chidí biiʼ séké: why b- here is not the inverse
Σ As E6, with ρA including 1st person (‘I’, dual with the mother).
A Willie: “Inverse cannot apply when there is a first or second person argument in the clause” (p. 98). VOICE-MIN needs two definite 3rd-person slots for the y/b choice, and the 3rd-person P objects get b-: bił, bii’ (Hale 2003:59's statement for the verb's bi- after a local subject, as recorded).
B Linking and spell-out as E6. The same noun-between-PPs order: shimá ⌢ bił ⌢ chidí ⌢ bii’ ⌢ séké.
HYPOTHESIS So a b- on a postposition has three sources: lexical P (B-DEFAULT), the inverse (VOICE-TF), and a local person in the clause. A y- has only one: a grammatical P in the direct voice with all 3rd persons. That asymmetry is why y-, not b-, is the diagnostic Willie uses (“Note that the PP object pronoun here is yii; this is evidence that the Inverse alternation applies”, p. 98).
7. Phrase-level lexicalization: what Willie Ch. 3 shows (Grant's reminder)
#
Question
What the opened pages show
Grade
X1
Does Willie call any verb + postposition pair idiomatic or lexicalized?
No. Whole-thesis search for “idiom”, “lexicaliz” and “narrow interp…”: the only hit is p. 43, about noun + lexical postposition ((35) hooghan=góó déyá). pp. 93–100 re-read on images: no such statement
SOURCED (silence, which counts against nothing)
X2
Are her Ch. 3 verb + P pairs compositional?
By her translations, yes: speak to (79), roll to (83), lead for (85), lie in (89), sit in (90), sit with (93), climb on (96). All of them add an argument; none shifts the meaning
SOURCED translations; “compositional” is our reading
X3
Is the grammatical/lexical status fixed per verb + P pair?
No: ‘climb’ + -ah is lexical in the impersonal (94) and grammatical in the active (96). ‘sit’ + -ii’ gives both a direct (90) and an inverse (91). Only (89)/(90) differ in the verb (‘lies’ vs ‘sits’)
Data SOURCED; “not fixed per pair” HYPOTHESIS
X4
So what can “phrasisized” (L9) store?
The pair's form and meaning: which P goes with which verb, and any shifted meaning (Chee & Yazzie's five pairs). Not the P's argument status, which is computed by LIC + DEF + voice in each sentence
HYPOTHESIS, refining GRANT'S CONJECTURE L9
X5
Does a meaning-shifted pair obey the computed rule?
Chee & Yazzie (15b) bitah honiigaii ‘He became sick, ill’ (base ‘The area became hotter’, areal ho-; p. 35, as recorded). Its subject is the area, which has none of (a)–(d). LIC-PROTO makes the P lexical, so B-DEFAULT gives b-, which is what is printed. (16b) yaa naakaii ‘All of them are doing it’ (base ‘walking around’) has y- with an agentive 3rd-person subject, so a grammatical P, direct. The other three pairs have a local person, so they tell us nothing
Consistent, not a test: bitah could also be an inverse under other assumptions. The prediction *yitah honiigaii (in this sense) is OPEN (T33f)
HYPOTHESISWhat this adds to Grant's conjecture. Grant's reminder was that lexicalization happens at the phrase level as much as at the word level. Willie's data do not test whether verb + P pairs are stored, but they do constrain what a stored pair can contain. Storage can fix the pairing and the meaning. It cannot fix whether the postposition's object is one of the verb's participants, because that changes with voice inside one pair. A word-level parallel is the Athabaskanists' “verb theme”: a non-compositional combination of morphemes (McDonough p. 44, as recorded). The comparison is ours. The phrase-level version would be a stored verb + P frame with computed argument status. What would count against it is data: a meaning-shifted pair whose P keeps y- (or b-) whatever the voice and the participants.
8. The noun between the postpositions: (92), (93) and the MVS
SOURCED Three statements of Willie's and two of her examples cannot all hold as stated. (i) “The grammatical postpositions … are a component of the MVS” (p. 40). (ii) The MVS “cannot be interrupted” (p. 50 (44a)); W&J p. 270: “noninterruptable constituent”. (iii) -ił ‘with’ is a grammatical postposition ((28g), p. 41). (92) and (93) put the noun chidí between bił/yił and the verb.
Option
What it says
Cost
Grade
M1 Only the PP next to the verb is in D1
The MVS (D1) is the verb plus one grammatical PP; a second grammatical PP is merged or spelled in D2
Gives up (i) for the outer PP. Then “grammatical” (in Σ, takes voice) and “MVS member” come apart
HYPOTHESIS
M2 The outer PP moves out
Both PPs are merged in VP (rule set A); at clause level the outer PP moves left past the noun (rule set B Move, like a topic)
Needs a trigger for the move and a reason why the noun can't then follow it into the gap. (44a) becomes a condition on D1 after extraction
HYPOTHESIS
M3chidí is inside the MVS
The noun belongs to the second PP as its object
Contradicts W4 (grammatical Ps take pronoun objects only) and the y- prefix on yii’. Hale 2003 (39b) taah shows that a noun object replaces the pronoun
Disfavoured by SOURCED data
M4 The comitative is a different kind of P
Comitative -ił is not an MVS member despite being “grammatical”, perhaps because it adds to the subject's number (the dual) rather than adding a participant to the action
Needs the dual-counting link (our reading of (29), (92), (93)) to be confirmed
HYPOTHESIS; OPEN
HYPOTHESISConsequence for Grant's changeover. “The MVS is the changeover” (clause spine) survives under M1, M2 and M4. But the MVS then has to be the verb plus its closest grammatical PP, or the verb plus the PPs that are left in place. This sharpens T31f/T32i: (92)/(93) are the first printed strings on this site with a noun between a grammatical PP and the verb. Willie does not comment on the order.
9. What each analysis gets right and misses
Analysis
Gets right
Misses or leaves open
Willie 1991
The grammatical/lexical split and its link to voice (p. 93); the one-participant split ((89)/(90)); the impersonal effect ((94)/(96)); y- as the diagnostic (p. 98); the absurd inverse readings reported as readings, not as ungrammatical strings
“Intransitive” used at two levels (p. 57/63 vs p. 97); two tests for “grammatical” that disagree on (89a) (W3 vs W4); the animacy condition as worded fails on (84); (28a) “only with animates” vs (83); MVS non-interruptibility vs (92)/(93); -ii’ and -ah on neither list
Hale 2003 / 1997 (as recorded)
The P's pronoun is its “true object” (2003:46); a noun object replaces it ((39b) taah), which fits W4; P proclitics in the disjunct zone (1997)
Silent on lexical Ps with pronoun objects (W6) and on one-participant verbs + P. Silence is not counted
Witherspoon 1977 (as recorded)
Movement decides who can act on whom (p. 69). LIC-PROTO carries this over from the verb's yi-/bi- to the argument status of an oblique
Does not segment ‘upon’ (T20c); not about postpositions as such
Dowty 1991 (baseline)
The proto-agent list (a)–(d) states the licensing condition with no Navajo-specific vocabulary; (d) is relational, as an oblique licence needs
Dowty selects subject and object; using the list to decide whether an oblique counts as an argument at all is our extension
Chee & Yazzie 2025 (as recorded)
Verb + P pairs can shift meaning; their two-way category (adds an argument vs shifts the meaning, p. 22) matches X2 vs X4
Say nothing on grammatical vs lexical status
10. Grant's thesis, Dowty, and STRATA
HYPOTHESISWeak thesis (role-bearing pronoun slots are the verb's arguments; nouns are linked add-ons). Every rule above (GP-ARG, B-DEFAULT, VOICE-MIN, DEF, LIC-PROTO) refers to slots and their entailments. Nouns enter only at linking (E1–E7, rule set B). Nothing contradicts the weak thesis. A refinement: pronoun slots also occur outside the verb's frame, on lexical Ps (W6, bii’ in (89a)). So “is a pronoun slot” does not imply “is a verb argument”, and the thesis has to be stated as “the verb's arguments are pronoun slots”, not the converse.
HYPOTHESISDowty as baseline. LIC-PROTO is an extension of Dowty in Grant's sense. It uses the standard proto-agent entailments to decide argument membership of an oblique, with Navajo's voice and definiteness (DEF) on top. Event-structure roles are not in conflict with it. ‘Roll to’ and ‘climb on’ are motion events whose goal is the P's object. That is an event-structure description of the same participant that (d) picks out as the reference point of movement.
GRANT'S CONJECTURESTRATA as a third option (theta roles first; event-structure roles next; lexical noun selection later), beside “nouns are adjuncts” and “nouns are inside the clause”. Test result here: HYPOTHESIS consistent. LIC-PROTO is computed on Layer 1 from the verb's entailments, and DEF and voice on slots. In (89)/(90) the deciding difference is already in the verb (‘lies’ vs ‘sits’) before any noun is chosen. LIC-EXT stated over noun animacy would need nouns before argument structure, which goes against the STRATA order. It is compatible only if animacy is a slot feature supplied by the verb or the P (e.g. -ił “only with animates”, p. 41). The adjunct view handles §5 equally well, because linking comes after Σ in both. The inside-the-clause view (Speas, as recorded) is silent on LIC, which counts against nothing. So the data here do not decide between the three. They show that the deciding step comes before the nouns, and STRATA states that explicitly.
11. Tasks T33a–h and next steps
ID
Question
What would settle it
T33a
OPEN LIC-EXT vs LIC-PROTO: with an inanimate mover and an inanimate P object, is y- on a separate-word postposition acceptable? Schematic: [Ninan1 Ninan2y-P Vmotion] vs [… b-P …]
A speaker; Young & Morgan example sentences
T33b
OPEN Is Witherspoon's ‘upon’ in yikʼichʼínímááz / yikʼiikę́ę́z a postposition whose object is the yi-? (= T20c)
Fernald et al.'s entry for -kʼi; Young & Morgan
T33c
OPEN Willie (28a) -aa “only with animates” vs (83) yaa with a ball as its object
Young & Morgan; the Postposition Lexicon entry for -aa
T33d
OPEN (92)/(93): which PPs are inside the MVS (M1–M4)? Can anything stand between bii’ and siʼą́ (lexical, E1) but not between yii’ and sidá (grammatical, E2)? (joins T31f, T32i)
Speaker judgments; prosody; texts with two PPs
T33e
OPEN Is haasʼnaʼ transitive at the stem (Willie p. 99 “corresponding transitive verb”), or only as verb + yaah?
Young & Morgan's entry for ‘climb’
T33f
OPEN Prediction of LIC-PROTO + B-DEFAULT: no yitah honiigaii in the sense ‘he became sick’ (all 3rd person)
Chee 2025 or Chee & Yazzie's corpus; a speaker
T33g
OPEN Does the comitative's object count in the verb's dual ((29), (92), (93))? If so, is the comitative a different kind of grammatical P (M4)?
Willie pp. 41, 98–99 have only translations; Young & Morgan; a speaker
T33h
OPEN PU-G with two applicative heads, or PU-L, for (93)? (updates T32b)
More two-PP strings; whether both PPs can take the inverse
Status of T32e.HYPOTHESIS resolved as a terminology clash: p. 97 holds at the MVS level (T-MVS), and the substantive question moves to LIC (§5), where LIC-EXT and LIC-PROTO remain and T33a decides between them. Willie's intent stays OPEN.
Changes made elsewhere for this page.Postposition units: T32e row marked done with a link here, notes on PU4 (c), PU5 and the T32b row, and the nasal hooks in (89)–(91) corrected. The dictionary entry for sidá has the same correction. Clause spine: note at the T31f row. Open questions: new row T33a–h. Sources: new block and forms-index rows for (93)–(97). The index lists the page.
Where the yes/no enclitic -ísh can go, and what it shows about where the Navajo word ends (T41f)
2026-11-11 audit pass 2. Apostrophe fixes only. Navajo glottal stops typed as ' or ’ (for example in prefix names like ʼa- and ʼeʼe-, or words like Awééʼ) now print as ʼ (U+02BC), and a build step no longer turns the closing quote of an English translation inside a Navajo row into a glottal stop (‘…the girls.’, not ‘…the girls.ʼ). No form, gloss or translation wording changed.
2026-11-11 audit pass 3. Local box file paths were removed from this page; files inside the repository are given by repo-relative path, and the source citation next to each removed path stays as the reference. See the pass 3 report.
Restated on the unified structure (2026-11-15). Questions are restated with Q in C⁰ above NegP: the wh word stays in its theta position as indefinite + Q (H11), and =sh/-ísh are interpreted in C⁰ and pronounced on a host (H12; H12-move vs H12-pf OPEN). Whether daʼ is a head is OPEN. Light d, used below, is withdrawn (Grant, 2026-11-14); light n is an OPEN structural question. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Position-class template statements below describe the order of prefixes; they are not the derivation (§3). Status on the spine: see unified structure, §4, question rows. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
The page the borrowed terms checked against their originals §6 called the prefix zone at the far left of the verb word “[−MFS]-like”, in Rudin's sense. The call rested on one datum, Hale's report that the yes/no enclitic can split the preverb ka ‘sick’ off from the rest of the verb (2003 n. 12). The Navajo forms in that note print in a broken font, and Rudin's test needs an enclitic that attaches to the first constituent of the clause. Neither point had been checked (T41f). This page reads the two forms with a decoding key taken from the manuscript itself and checked against a second source. It then sets out where -ísh goes, using the Navajo Language Academy's question grammar, opened live this run, and states the placement as rules. The derivations are re-run step by step. The result goes back to the routine's first phenomenon: the word as the point where rule set A (the word) hands over to rule set B (the clause). Two operations stay apart throughout. Merge builds structure: it puts an element at a position. Agree values features such as person and number: a probe copies a value from the closest matching goal. Placing -ísh is neither. It is a spell-out (pronunciation) rule of the clause cycle. Written 2026-10-20, 12:46pm PT.
Update 2026-10-20, 8:46pm (T43f). T43f is answered on the two question endings compared. Willie writes “first word” for this ending as well (1991 p. 92: “In a yes/no question, the clitic -ish usually follows the first word in the sentence”), so his “first word” for =sh is wording, not a contrast with EN-2. On its own host =sh also closes a whole phrase (Willie (19), (20)). Only the shifted =sh with a long first phrase is untested (T44a). The inverse restriction applies to -ísh too, but only to fronted hosts: (76b) łííʼ-ísh iiʼniʼ biisxį́ is fine and *(76a) is out (rule QC-5). *(76a) and *(13) *Daʼísh fall under one domain rule (QC-3). EN-0 to EN-6 are unchanged.
Update 2026-10-21, 4:46am (T43e). T43e is worked on attached or separate postpositions. Of the two readings of D1 in §5, neither survives as stated. The sealed unit of rule set A is the verb word W, and each postpositional phrase is sealed on its own. The MVS is the phrase in which they are ordered and the voice is chosen, not a sealed block (PC-1). That is why -ísh can stand between a postposition word and the verb: (6) Neésh hólǫ́?, and a new NLA case, … dził bąąhgóósh chʼétʼaʼ? ‘… or along the mountainside?’. Whether a postposition that is written onto the verb can carry -ísh is still unattested; PC-5 predicts yes (it would then appear as a separate word, like ka-[á]sh), and predicts no for the causative postposition that Hale puts inside the word (T45a, T45e).
Ground rules (Grant's). Every Navajo form below comes from a source opened this run (Fernald et al., the NLA question grammar; Hale 2003 n. 12 and (9)–(10); Willie & Jelinek 2000 p. 270), or is already on this site with its source number. Each is quoted as printed, with the source's own gloss line, and none is segmented beyond what its source prints. Two characters in Hale's note are restored. They are bracketed, and §2 gives the key. No argument rests on the restored vowels: only the position of the enclitic matters, and the space between ka-[á]sh and nanin[á] is legible. What the sources say is graded SOURCED. Rules and mappings are ours (HYPOTHESIS). Only contradicting data count against an analysis; a source's silence does not. No analysis, the site's own included, is assumed to be fully right.
Notation (the site's own; the Syntax · Semantics · Interface bot's files are still dated Oct 18, 10:05pm PT, with no conventions for this site). Σ = role-schema step (which slots the verb has, and each one's role in the action); A = word level (cycle D1, rule set A: the prefixal, left-headed spine Qual > M/S > v > V, spelled into one word); B = clause level (cycle D2, rule set B: right-headed clause syntax, postpositional phrases, noun linking, particles and enclitics). W = the verb word built on Hale's skeleton (candidate B2). The site's 2026-10-17 update to CYC reads D1 as the Maximal Verb Complex, W plus its preceding postpositional phrases (B3). The enclitic separates the two (§5). CYC, C1, C2: spell-out rules §3. SB1–SB5: spell-out rules §2.2. B1–B3 (the three candidate word edges): rule set A vs B §3. Glosses: the NLA's gloss lines are given as printed (NI = neuter imperfective, P = perfective, SP = si-perfective, CI = continuative imperfective, Q = question). In our own Leipzig lines “=” marks a clitic boundary, where the NLA prints a hyphen.
Bottom line, in plain words
SOURCEDThe two forms in Hale's note can now be read. They are ka naash[á] ‘I am sick’ (lit. ‘I walk about sick’) and ka-[á]sh nanin[á] ‘are you sick?’ (Paul Platero p.c., in Hale 2003 n. 12, p. 69). The broken glyph “¡” is á. Hale's own text uses it for yisdá ‘to safety’ (p. 15), the same preverb his table (10) prints legibly as yisdá (p. 14). The NLA grammar prints the verb naniná ‘2-go.around.CI’, and it cites the enclitic as -ísh, whose vowel copies the vowel before it (“the vowel in the enclitic has assimilated”, yes/no (4)).
SOURCED-ísh is a first-constituent enclitic. That is the kind of enclitic Rudin's test needs. NLA: “Placing it on the first phrase results in a neutral question with no particular focus”; “When the enclitic is not attached to the first phrase of the sentence, the phrase it is attached to is focused”. It occurs only once ((8)), never on da’ ((13)), and counts its “first syntactic constituent … (after da’)” ((9)–(10)).
HYPOTHESISThe split in Hale's note is a split between two words, not inside one. Hale gives ka as his example of a preverb that can “appear as separate words”, and the enclitic split is offered in support. So ka-[á]sh nanin[á] shows the enclitic after the first clause-level unit, which here happens to be a preverb. It does not show an enclitic inside a verb word. The site's “[−MFS]-like” verdict on the prefix zone therefore loses its only datum and goes back to OPEN. On the rules below it cannot be tested with this enclitic at all.
HYPOTHESISWhat it gives instead: a test for where the word ends. Every host of -ísh in the opened data is a clause-level unit: a noun, a postposition word, a whole verb word, a preverb that stands as its own word, or a whole phrase. No opened form puts it inside a verb word. That is silence, so it counts for nothing either way, but it is a clean prediction (EN-1′). If the prediction holds, “can -ísh follow it?” is an independent diagnostic for Grant's changeover point. It also gives Hale's own open problem (“when to stop building onto the skeleton”, n. 12) a testable answer.
HYPOTHESISRight-headedness has a visible effect. In NLA (10a)/(10b) one string gets two readings. The enclitic can be read as closing the whole first phrase (‘the rope my father braided’), which gives a neutral question, or as closing only its last word, the verb ‘braided’, which focuses the verb. The two edges coincide only because the clause-level phrase ends with its verb, and rule set B's head-final order guarantees that.
GRANT'S CONJECTUREVerb + postposition units are split by the enclitic. In Neésh hólǫ́? ‘Do you have any?’ the enclitic goes on the postposition word nee ‘2-with’, inside the ‘have’ unit nee hólǫ́. If such units are lexicalized, then, they are lexicalized at the phrase level, two clause-level units with one meaning, not as one word. That fits Grant's phrase-level lexicalization idea. It also sharpens Willie & Jelinek's “noninterruptable” verb complex (p. 270): no word can break into it, but an enclitic can.
HYPOTHESISTwo word edges, two jobs. So the enclitic stops at the verb word itself (Hale's skeleton, candidate B2), not at the larger verb-plus-postposition complex (B3), which is the domain where the yi-/bi- choice is made. The site had folded the two together on Oct 17. They now come apart: the larger unit is a domain for the voice choice, and the smaller one is the unit the clause-level enclitic treats as solid.
Fernald, Theodore B., Leslie Chavez, Bernadine Cody, Dana Desiderio, Natalie R. Desiderio, Jeremy Fahringer, Ronald Gene, Betsy H. Horner, Lorene Legah, Sharon Nelson & Tyler Tinhorn. 2019. Navajo Questions: grammar and examples, version 3.2022. Navajo Language Academy (NLA)
Opened live (web fetch, 2026-10-20): the yes/no page http://www.talkingdictionary.org/navajo_questions/?q-binary, and the front page http://www.talkingdictionary.org/navajo_questions (its example list, 50 entries). The live yes/no text was compared with the box copy saved on Oct 18 ([local working copy, not in the repository]): the wording is the same. The live page leaves (5) and (7) unnumbered, and the box copy numbers them. The underlining that marks the focused element in (10)–(11) is not in the HTML of either copy. The box itself could not reach the host (curl timed out after 90 s), so no search query could be run
Yes/no (4)–(8), (9), (10a, b), (11a, b), (12), (13) and the prose on placement (§3). From the example list, only two items: the verb naniná ‘2-go.around.CI’ (key check, §2) and one item with -ísh on a postposition (§3, F9)
Hale, Ken. 2003 ms. “On the significance of Eloise Jelinek's Pronominal Argument Hypothesis” (box copy [local working copy, not in the repository]; pages cited by the manuscript's printed numbers)
Re-opened. N. 12 (pp. 68–69) was read in the text layer and on a 300 dpi page image of p. 69. The broken glyphs are in the PDF's embedded font, so they also appear on the image. P. 14 ((9), table (10)) was re-read on a 150 dpi page image, and p. 15 (prose yisd¡ ‘to safety’) in the text layer. Every word with “¡” or “¶” in the text layer was listed (8 with “¡”, 1 with “¶”)
The two n. 12 forms and the decoding key (§2). Hale's “wording up” proposal and its open problem (n. 12). yisdá as preverb (position I) inside the word (9)
Willie, MaryAnn & Eloise Jelinek 2000. “Navajo as a discourse configurational language.” In Fernald & Platero (eds.), The Athabaskan Languages (OUP)
Re-read p. 270 on a 160 dpi page image of the Jelinek collection scan ([local working copy, not in the repository], PDF p. 149)
“Some postpositions and their objects in Navajo are procliticized to the inflected verb; others remain detached. The verb and any preceding post-positional phrases form a noninterruptable constituent, the Maximal Verb Complex (Willie 1991), that is the domain of the voice alternation.”
Rudin 1988; Willie 1991 pp. 210–212; Willie & Jelinek 2000 p. 272
Young & Morgan 1980/1987. The Internet Archive copy of the 1980 edition (navajolanguagegr0000youn) is lending-only (access-restricted). The unofficial text dump already on the box is not used (site policy since Oct 18). So T42d, T42e and T42g are not attempted. Also not opened: Hale, Munro & Platero 2000 (printed); the MIT host for page images (not retried)
2. Reading Hale's n. 12
SOURCEDThe note (n. 12, pp. 68–69, text layer and page image; broken glyphs as printed): “A possible solution to the problem of ‘wording up’ is hinted at in Hale (2001). The idea is to regard the bipartite skeleton as a minimum that must be satisfied. But it is a minimum which can be augmented with additional CV(C) syllables added to the left, to be filled in by left over material … The main problem would be to determine when to stop building onto the skeleton, in order to permit some preverbs, for example, to appear as separate words (e.g., the preverb ka ‘sick’ in ka naash¡ ‘I am sick’ (lit. ‘I walk about sick’). In this construction, and in many comparable cases, the preverb can be separated form the rest of the verb by the polar interrogative enclitiuc -¶sh, as in ka-¡sh nanin¡ ‘are you sick?’ (Paul Platero p.c.)).” (Hale's typos “form”, “enclitiuc” kept.)
Glyph
Every token in the manuscript's prose
Independent check
Reading
¡
yisd¡ ‘to safety’ (twice: pp. 15, 18); g¡¡l ‘locomotion’; m¡¡z (a root “listed … in Young, Morgan, and …”); bin¡’iishdlªªh; naash¡; nanin¡; ka-¡sh
(a) The prose yisd¡ refers to “(9, 10)”, and table (10) on p. 14 prints the same preverb legibly as yisdá (page image). (b) The NLA prints naniná ‘2-go.around.CI’ (front-page item “Kin Łání di ísh … naniná?”, ‘Are you at Flagstaff or at Albuquerque?’). (c) ka-¡sh: the NLA says the enclitic's vowel “has assimilated to match the final vowel” of its host ((4) dibé → Dibéésh); after ka that predicts á
¡ = áSOURCED by (a). (b) and (c) agree
¶
-¶sh (the citation form of the enclitic; one token)
The NLA's citation form is -ísh
¶ = íHYPOTHESIS (one token, but the form is otherwise fixed)
The two forms, restored (restored characters bracketed; our Leipzig lines, not segmented beyond Hale's word division):
#
Form
Gloss (ours, from Hale's translation and the NLA gloss of naniná)
Translation
H12a
ka naash[á]
sick 1-go.around
‘I am sick’ (lit. ‘I walk about sick’)
H12b
ka-[á]sh nanin[á]
sick=Q 2-go.around
‘are you sick?’ (Platero p.c.)
Line break. On p. 69 the hyphen of ka- falls at the end of a line, so we cannot tell whether Hale wrote ka-ásh or kaásh. Either way ka and the enclitic form one unit, separated by a space from nanin[á], and nothing below depends on the hyphen.
OPENbin¡’iishdlªªh ‘I repeatedly feed him liquid’ (n. 13, p. 70) would read bináʼiishdl[ą́ą́]h if “ª” is ą́. That is plausible (compare HMP's -dlą́ą́ʼ ‘drink’ on the HMP page) but unchecked, and it is not used (T43h).
3. Where -ísh goes: the facts
NLA yes/no section, quoted as printed (gloss line as printed; numbering of the box copy, with the live page's unnumbered items marked).
#
Form
NLA gloss line
Translation / NLA comment
On the site before?
(4)
Dibéésh nee hólǫ́?
sheep-Q 2-with exist.NI
‘Do you have sheep?’ “the vowel in the enclitic has assimilated to match the final vowel in dibé”
Same; “When the enclitic is not attached to the first phrase of the sentence, the phrase it is attached to is focused”
new
(12)
Daʼ shééhonísinísh?
Q 1-2-know.NI-Q
‘Do you know me?’
yes
(13)
*Daʼísh shééhonísin?
Q-Q 1-2-know.NI
“the enclitic -ísh cannot appear on the particle da’”
yes
F9 (front page)
Nił ísh béého sin doodaiiʼ bił ísh béého zin? (spacing as displayed by the database, which also splits single words elsewhere, e.g. “nanil nish” glossed as one word)
2-with-Q 3-know.CI or 3-with-Q 3-know.CI
‘Do you know or does s/he know?’ (speaker Maryann Sam)
new
Printed variation kept as is: (9) siníloh, (10)–(11) síníloh; (9) yizhbizhę́ę], (10a) yizhbizhę́]ę́sh (the bracket's position is the NLA's).
SOURCEDSummary of the facts. (i) Hosts: a noun (4), (11b); a postposition word (5), (6), F9; the verb word (7), (12); a whole bracketed phrase (10a); a noun inside that phrase (11a); and, from Hale, a preverb that stands as its own word (H12b). (ii) The neutral position is the first phrase after da’. Elsewhere the host is focused. (iii) Only once per question ((8); F9 has one per disjoined clause). (iv) Never on da’. (v) Two placements are marked “?”: on the postposition (5) and on the verb (7) when neither is first. For comparison, Willie 1991 p. 211 (as recorded): the content-question =sh “can be moved away from the Wh- word to a position following the first word in the sentence”.
4. Rules EN-0 to EN-6 (ours; all HYPOTHESIS unless marked)
These replace the one-line SB5 on the spell-out page (“-ísh / -sh attach after a phrase … Placement is phonological”). SB5 is left as written there, with an update note.
EN-0 EXPONENT -ish spells the clause's polar-question feature [Q] (on C). One exponent
per [Q]: *(8). Whether -ish is itself a head (Foc/Q) or a phrasal enclitic
spelled at C's level: OPEN (headedness section 4). (Facts SOURCED: NLA (8).)
EN-1 HOST -ish is spelled at the RIGHT EDGE of a host X, where X is a SPELLED-OUT UNIT:
the verb word W (already spelled out by rule set A), a postposition word
(its own PP cycle), a noun, a preverb not built into W, or a phrase built
from such units. (Host types SOURCED: NLA (4)-(12), F9; Hale n. 12.
The restriction to spelled-out units is ours.)
EN-1' SEAL Corollary of CYC (the word W is spelled out once, then ordered as one block):
-ish is never spelled inside W: not between two prefixes, not
between a prefix and the stem. No opened form contradicts it; none tests it.
PREDICTION, OPEN (T43a, T43b).
EN-2 NEUTRAL With no narrow focus, X = the FIRST CONSTITUENT of C's complement: the
leftmost maximal phrase below da' (NLA "the first syntactic constituent ...
(after da')"; (4), (6), (10a), (12)).
da' sits above that domain (Spec,CP, as CYC already requires for da'; spell-out
rules N6), so *(13) da'-ish follows: da' is never "first" in C's complement.
EN-3 FOCUS -ish on any other host X marks X as the focus: (10b), (11a), (11b).
The "?" of (5) and (7) (postposition word, verb word, non-initial) is not
derived: OPEN (T43d).
EN-4 EDGES (derived from SB2 + EN-1) In rule set B the head comes last, so the right edge
of a head-final phrase XP is also the right edge of its last word Y. -ish at
that edge is ambiguous between X = XP (neutral, by EN-2) and X = Y (focus on
Y, by EN-3). NLA (10a)/(10b): one string, two readings.
EN-5 TEST If X can host -ish, X is not inside the verb word W. (From EN-1'.)
Hale uses the split as evidence that ka can be a separate word (n. 12: SOURCED
as his argument); the general test is ours. Applied:
ka 'sick' (H12b) -> its own unit (Hale: "separate words")
nee '2-with' (6), (5) -> its own unit, apart from hólǫ́
nił, bił 'with' (F9) -> their own units, apart from the verb
the relative DP (10a) -> a phrase
EN-6 RUDIN Rudin's clitic diagnostic needs "encliticization to the first constituent of
the clause" ((34), p. 463). -ish meets that condition at the clause level
(EN-2). But by EN-1' its host can never be inside the verb word W, so it cannot
split the prefix zone of a single verb word. It can only show where the word
ends. So the diagnostic does not apply to the word-internal disjunct zone,
and that zone's [+/-MFS] status is OPEN (section 7).
Rule
Status
Rests on
EN-0
SOURCED facts; OPEN category
NLA (8)
EN-1
HYPOTHESIS
All hosts in §3 are spelled-out units outside W, or W itself
EN-1′
OPEN prediction
CYC; silence of the sources
EN-2
HYPOTHESIS (statement), SOURCED (facts)
NLA prose and (4), (6), (10a), (12), *(13)
EN-3
HYPOTHESIS; “?” OPEN
NLA (10b), (11a, b); (5), (7)
EN-4
HYPOTHESIS (derived)
SB2 + EN-1; NLA (10a)/(10b)
EN-5
HYPOTHESIS
EN-1′; Hale n. 12
EN-6
HYPOTHESIS
Rudin (34) as recorded; EN-1′
5. What changes
Item
Before
After
Why
Originals page §6, clitic row
HYPOTHESIS “[−MFS]-like on Rudin's decisive test” (n. 12), two caveats
OPEN: the test does not apply inside the word (EN-6). Caveat 2 is answered: -ísh does attach to the first constituent (EN-2). Caveat 1 (“the first of several disjuncts?”) turns out to be moot, because ka is not in the word
Hale: ka appears “as separate words”; NLA placement facts
Merge/Agree in the word §6
“OPEN, leaning ‘not a cluster’”, with n. 12 pointing to divisibility
OPEN, with no lean: n. 12 is not word-internal evidence
Same
SB5
“attach after a phrase … Placement is phonological … not ordered by SB2–SB3”
EN-1 to EN-4. Placement is counted in spelled-out units, and its neutral position comes from C's complement. EN-4 shows that SB2's head-final order does matter for it
NLA (10a)/(10b), (12), *(13)
CYC update of 2026-10-17: D1 = the MVS, with grammatical PPs as already-spelled non-heads inside D1
D1 = W plus its preceding PPs (B3), sealed as one block
The seal that -ísh respects is W (B2), not the MVS. In (6) Neésh hólǫ́ the enclitic sits inside the MVS, between the PP and W. So either D1 is W and PPs are clause-level (SB1 as first written), or D1 is the MVS and the enclitic can still reach its already-spelled non-heads. Either way the two boundaries do different jobs: the MVS (B3) for the voice alternation (W&J), W (B2) for the enclitic. HYPOTHESIS; which of the two readings of D1 is right is OPEN (T43e)
NLA (6), (5), F9; W&J p. 270
Spell-out rules, n. 12 forms
“printed with a broken font … not reproduced”
Reproduced with the key (§2): ka naash[á], ka-[á]sh nanin[á]
Hale p. 14 vs p. 15; NLA naniná
CYC's cost “the cycle boundary itself is stipulated”
Stipulated (Hale: unsolved)
Still stipulated, but now testable: EN-5 says which material Hale's skeleton stopped before. ka is outside W in H12b
EN-5
B1 vs B2 (headedness §3)
Three nested candidates
ka is a position-I preverb (Hale's table (10) puts preverbs in I, left of “#”) that can stand outside the word. yisdá, also position I, is printed inside the word in (9). If yisdá cannot host -ísh, B1 (“#”) is not the changeover and B2 is (T43c)
Hale (9), (10), n. 12
B3 (MVS) “noninterruptable”
As W&J state it
No word can interrupt it, but an enclitic can: (6) Neésh hólǫ́. W&J's word “noninterruptable” is read as “by words”. This is our reading; W&J do not discuss enclitics there
W&J p. 270; NLA (6)
6. Derivations PE1–PE7
Steps are labelled by rule set: Σ role schema, A word (D1), B clause (D2). Each step names its operation: Merge (builds structure), Agree (values person/number), Move, Link (a noun linked to a slot), Spell-out. Verb words are not segmented beyond the source's gloss line. VPISH: every participant slot is first-merged inside the verb's phrase. A noun, if there is one, links to that slot later, in D2.
PE1. NLA (6) Neésh hólǫ́? ‘Do you have any?’: no noun, enclitic on the postposition
Σ Σ(hólǫ́ ‘exist’) = {ρth} (what exists), as on D6. The postposition -ee ‘with’ has its own pronoun slot ρP, valued 2 (‘you’). Its role in the action is the possessor/location contributed by ‘with’.
AMerge V and ρth inside the verb's phrase (VPISH), then the spine. Agree: M/S finds ρth → 3 (the NLA's (15) prints “3-exist.NI”, (1) “exist.NI”). Spell-out W → hólǫ́ (unsegmented).
B PP cycle: Merge P -ee with ρP. Agree: P's pronoun takes φ = 2. Spell-out C1 pronoun ⌢ P → nee (SB4). On the 2026-10-17 reading this PP is an already-spelled non-head inside D1, and the string comes out the same.
BMerge PP and the verb word in D2. Order: PP ⌢ W (C2/SB2). Merge C[Q]. No noun links to ρth or ρP.
BSpell-out [Q] by EN-2: the first constituent of C's complement is the PP nee → printed Neésh. Result Neésh hólǫ́? ✓. Moves: 0.
Consequence (EN-5): nee is a unit separate from hólǫ́, so the ‘have’ unit nee hólǫ́ is two units, not one word.
PE2. NLA (4), ?(5), ?(7), *(8): one noun, four placements
ΣA As PE1, steps 1–3.
BMerge the noun dibé in D2. Link it to ρth (‘sheep’ is what exists). Order dibé ⌢ PP ⌢ W (C2).
BSpell-out [Q]. (4) EN-2: the first constituent is dibé → Dibéésh nee hólǫ́? ✓ (neutral). (5) EN-3 on the PP → Dibé neésh hólǫ́ with focus on ‘with you’: generated, but the NLA marks it “?”. (7) EN-3 on W → Dibé nee hólǫ́nísh: generated, also “?”. Both are OPEN, because EN-3 does not predict a difference from (11b), which is unmarked (T43d). *(8): two exponents of one [Q], excluded by EN-0 ✓.
PE3. NLA (12) Daʼ shééhonísinísh? ‘Do you know me?’ and *(13)
Σ Σ(‘know’) = {ρknower, ρknown}. From the translation, knower = 2 and known = 1. On Dowty's list only the knower has sentience about the knowing. Nothing on this page depends on that ranking.
AMerge: ρknown inside VP; ρknower in Spec,v (VPISH: the subject starts inside the verb's phrase). Agree: M/S finds ρknower → 2. The object exponent spells ρknown = 1. Spell-out W → shééhonísin (the NLA's gloss line “1-2-know.NI” only; not segmented here).
BMerge the verb word in D2. Merge C[Q]. Mergeda’ in Spec,CP (CYC requires it to be a phrase: N6).
BSpell-out [Q] by EN-2: the first constituent of C's complement is the verb word itself → shééhonísin-ísh → Daʼ shééhonísinísh? ✓. *(13): da’ is in Spec,CP, outside C's complement, so it is never “first” → *Daʼísh ✓ (derived, not listed).
The enclitic attaches at the right edge of the whole word, after the stem. It is never inside the word (EN-1′). The whole question has no noun: both participants are slots inside one word.
PE4. NLA (10a)/(10b): one string, two readings (EN-4)
A Relative verb word yizhbizh(ę́) (NLA “3-3-braid.P”), unsegmented, as on N7. Inside its own word cycle: Agree values both 3's; Spell-out W.
B Relative clause, as N7: Mergeshizhéʼé (possessed noun, own cycle: C1 possessor ⌢ N) and tłʼóół. Link each to a slot of ‘braid’. The relative's final head -ę́ę ‘past’ comes last (C1).
B PP bee ‘3-with’; the relative DP is its object's linked phrase; then dibé and the main verb word síníloh (“3-2-rope.SP”). Merge C[Q] and da’ in Spec,CP.
BSpell-out [Q]. (a) EN-2: X = the relative DP, the first constituent after da’. Its right edge is after -ę́ę → neutral reading ✓ (10a). (b) EN-3: X = the relative's verb word, the last word in the DP. Its right edge is the same point → focus on ‘braided’ ✓ (10b).
Same string ✓✓ (the NLA prints only the bracket differently). The coincidence needs the verb to be last in its phrase, which rule set B (head-final) guarantees. Moves: 0.
PE5. NLA (11a): the enclitic on a word inside the first phrase
As PE4, steps 1–3.
BSpell-out [Q] by EN-3 with X = shizhéʼé, the possessed noun inside the relative clause. It is the first word after da’, but not the first constituent (that is the whole DP). So EN-2 does not apply, and EN-3 gives focus. The NLA presents (11a) as a focusing example ✓. Which element is underlined (the father?) is not recoverable from the HTML (T43g).
Point: neutral placement counts constituents, not words. Willie's statement for the content-question =sh (“following the first word”) is worded in words. Whether the two enclitics differ, or the wording does: OPEN (T43f).
PE6. Hale n. 12 H12b ka-[á]sh nanin[á] ‘are you sick?’: the preverb as its own unit
Σ Σ(‘go around’) = {ρmover} (the one who walks about), valued 2.
AMerge ρmover inside the verb's phrase (VPISH); spine. Agree: M/S finds ρmover → 2. Spell-out W → nanin[á] (NLA “2-go.around.CI”; not segmented).
BMergeka ‘sick’ in D2, adjoined to the verb word. It modifies the event (‘walk about sick’) and adds no slot and no role in the action. Order ka ⌢ W (non-head first, C2). Merge C[Q].
BSpell-out [Q] by EN-2: the first constituent is ka → ka-[á]sh nanin[á] ✓ (Platero p.c.). Moves: 0.
Failed alternative (schematic): suppose ka had been built onto the skeleton inside W (SA4 PREFIX, Hale's “CV(C) syllables added to the left”). Then W would be one unit, ka…nanin[á], and EN-1′ would forbid -ísh between ka and the rest. The neutral question would then put -ísh after the whole word. The attested split shows that ka is outside W in H12b. Whether the unsplit form is also possible is not reported (T43c).
H12a ka naash[á] ‘I am sick’: the same with ρmover = 1 and no [Q].
PE7. Hale (9) Yisdánídashizhdoołtééł ‘They (3a = 4th person) will get me back to safety’: three disjuncts inside one word (prediction only)
As recorded on the spell-out page (yisdá-ní-da-sh-d-w-ł-tééł, positions I, II, III … X; re-read on the p. 14 image this run). Here the preverb yisdá (I) is printed inside the word, together with It/Rev ní (II) and plural da (III).
This is the “two or more disjunct prefixes” form that T41f asked for, but no source gives it with -ísh. EN-5 makes the predictions explicit (schematic, no Navajo string is claimed). (a) If yisdá is in W, then -ísh cannot follow yisdá, ní or da, and the neutral question has it after the whole word. (b) If yisdá, like ka, can stand outside, then -ísh after yisdá is possible, and the rest (ní-da-…) is the word W. Either result is informative. A form with -ísh after ní or da would falsify EN-1′ (T43a, T43c).
7. Rudin's test, re-applied
Rudin's requirement (as recorded)
Navajo -ísh
Reading
An enclitic placed by “encliticization to the first constituent of the clause” ((34), p. 463)
Yes at the clause level: first constituent after da’ = neutral (EN-2; NLA prose and (4), (6), (10a), (12))
SOURCED facts; mapping HYPOTHESIS
A group of fronted elements that the clitic can split ([−MFS]) or must follow ([+MFS]) (p. 461)
Inside the word: no opened datum. The one split (H12b) is between two units, neither inside the other
OPEN (EN-6). The verdict “[−MFS]-like” is withdrawn
At clause level, by analogy: several nouns/phrases before the verb
The enclitic may go on any of them, with focus (EN-3). That is Rudin's [−MFS]-type freedom for the clause, which no one has claimed is a wh-cluster
HYPOTHESIS, and not about the word
HYPOTHESISNet. T41f is answered as far as the opened data go. The enclitic is of the right kind. The datum the site relied on is not word-internal. So the word-internal [±MFS] question is OPEN, and with this enclitic it cannot be tested at all if EN-1′ is right. If EN-1′ is wrong (T43b), the test becomes possible.
8. What each side gets right and misses
Side
Gets right
Misses or costs
Hale 2003 n. 12
States the open problem precisely (“when to stop building onto the skeleton”). Gives a real case of a preverb that can stand as its own word, with the enclitic split as evidence
The forms print in a broken font. Gives no rule for which preverbs stand outside (“many comparable cases”). Says nothing about where the enclitic goes in general
NLA Navajo Questions
First phrase = neutral; elsewhere = focus; once; never on da’; and the (10a)/(10b) ambiguity, which turns out to follow from the clause's head-final order
The “?” on (5) and (7) is reported but not explained. There are no preverb or multi-disjunct examples and no word-internal test. The focus underlining is not visible in the HTML
Willie & Jelinek 2000 p. 270
The MVS as a unit that no word can break into, and the distinction between procliticized and detached postpositions
“Noninterruptable” is not qualified for enclitics, and (6) shows that the enclitic can interrupt it
Willie 1991 p. 211 (as recorded)
The content-question =sh is a second-position enclitic; the inverse restriction
Stated as “first word”. For the polar -ísh, the NLA's (11a) shows that the neutral count is by constituent (T43f)
This site, before today
SB5 kept enclitics out of CYC's head ordering. The originals page flagged both caveats
It counted a split between two words as a split inside one. It called placement “not ordered by SB2–SB3”, but EN-4 shows that head-final order matters to it
Rudin 1988 applied to the word
The diagnostic's own condition (first-constituent enclitic) is met at the clause level
It needs a word-internal host. The only Navajo enclitic tested so far does not reach inside the word
Update 2026-10-20, 8:46pm (T43f). The Willie row above was written from the site's record. On the page images (pp. 92, 211), Willie uses “first word” for both endings, so the row's “Stated as ‘first word’” is not a difference between them: the two question endings compared §3.
9. Grant's frame
GRANT'S CONJECTUREThe word as the changeover point (phenomenon 1). The enclitic gives the changeover a test that does not come from the template or from phonology. What -ísh can follow is a clause-level unit, ordered by rule set B. What it cannot reach into, if EN-1′ holds, is a word built by rule set A. On the opened data the changeover lies at Hale's skeleton edge (B2): ka falls outside even though it is a position-I preverb. Whether B1 (“#”) can be ruled out depends on yisdá (T43c). The conjecture is now stated so that it can fail: one -ísh inside a verb word would break it in this form. That no opened source shows such a form is silence and counts for nothing.
GRANT'S CONJECTUREWord-internal Merge or Agree sites. The clause-level enclitic cannot see them. That fits a word that is spelled out once and then sealed (CYC; on the 2000/2001 page, D1 as a phase-like cycle). It neither confirms nor contradicts the sites themselves. The Agree steps on this page (M/S taking 2 or 3; P's pronoun taking 2) are all inside W or the PP cycle, before the enclitic is placed. The Merge steps build both cycles.
GRANT'S CONJECTUREVerb + postposition units as phrase-level lexicalization.nee hólǫ́ ‘have’ (‘with you exists’) and nił/bił … bééhózin ‘know’ (F9) are split by -ísh on the postposition. So if they are lexicalized, the lexicalization is phrase-level, with two units, as Grant suggested (2026-10-17 reminder). W&J's procliticized postpositions (p. 270) are the next test: can a procliticized P still host -ísh (T43e)?
HYPOTHESISTheta roles and slots (phenomenon 2). (6) Neésh hólǫ́? and (12) Daʼ shééhonísinísh? are complete questions with no noun at all. Every participant is a slot (a pronoun prefix on P or in the verb word), and the enclitic finds a host anyway. This fits the weak thesis (role-bearing slots are the arguments; nouns are optional and link later). It is not a test of the strong thesis.
GRANT'S CONJECTURESTRATA. Placing -ísh is the last step in every derivation here. It comes after the roles (Σ), after Agree inside the word, and after nouns link in D2. It changes no participant's role in the action, and focus does not change the role either: in (11b) ‘sheep’ is still what is roped. That is the order STRATA expects (roles first, the event layer next, nouns later), with the focus/question layer at the very end. As before, it does not separate STRATA from its rivals. Theta roles and event-structure roles are untouched and stay compatible layers.
10. Tasks T43a–h and statuses
Task
What it needs
T43a
A form with two or more disjunct prefixes and -ísh (Young & Morgan, NLA, speakers). Where can the enclitic go? After the whole word only, or after a preverb? Any placement between ní/da and the rest would falsify EN-1′
T43b
Any form with -ísh (or content =sh) inside a verb word, between two prefixes or before the stem. This falsifies EN-1′ and CYC's seal
T43c
Can yisdá (Hale (9)) or another preverb that Hale prints inside the word host -ísh? Is an unsplit ka… one-word form also possible? Which preverbs are Hale's “many comparable cases”? This decides B1 vs B2
T43d
Why are (5) neésh and (7) hólǫ́nísh only “?” while (11b) dibéésh is fine? (Heads of their phrases vs nouns? the PP inside the MVS?)
T43e
W&J p. 270, procliticized vs detached postpositions: can a procliticized P host -ísh? EN-5 says that if it can, it is not inside the word. This also decides whether D1 is W or the whole MVS (§5)
T43f
Polar -ísh vs content =sh: first word (Willie p. 211; W&J p. 272 “usually … after the first word”) or first constituent (NLA)? Does -ísh show Willie's inverse restriction?
T43g
The NLA's focus underlining in (10)–(11): a printed or PDF version of Navajo Questions (which element is focused in (11a))
T43h
Apply the ¡/¶ key to the other broken-font prose forms in Hale 2003 (g¡¡l, m¡¡z, bin¡’iishdlªªh; is “ª” ą́?), and check them against legible example-font copies
Statuses: T41f answered (this page). The enclitic attaches to the first constituent, but the n. 12 split is not inside the word, so the word-internal Rudin test is OPEN and moves to T43a/T43b. T40a is closed into T43a. T42d, T42e, T42g are not attempted: Young & Morgan is still unreachable (the Internet Archive copy is lending-only), and HMP 2000 is not on the box. All other T41/T42 items are unchanged.
Attached or separate? Procliticized vs detached postpositions, where the word ends, and which verb + postposition pairs are idioms (T43e)
2026-11-11 audit pass 2. Apostrophe fixes only. Navajo glottal stops typed as ' or ’ (for example in prefix names like ʼa- and ʼeʼe-, or words like Awééʼ) now print as ʼ (U+02BC), and a build step no longer turns the closing quote of an English translation inside a Navajo row into a glottal stop (‘…the girls.’, not ‘…the girls.ʼ). No form, gloss or translation wording changed.
2026-11-11 audit pass 3. Local box file paths were removed from this page; files inside the repository are given by repo-relative path, and the source citation next to each removed path stays as the reference. See the pass 3 report.
Restated on the unified structure (2026-11-15). Postpositions are restated as [PP DP P⁰] with left specifiers: a grammatical P agrees with the DP in its specifier, a lexical P has no probe (H64); a selected PP is merged in VP, and Willie’s “cannot be interrupted” follows from the default, HS-1 and left specifiers. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Flat-S and other pre-VPISH notation below is labelled contrast. Derivations below in which the verb or subject does not raise are labelled contrast; the working default raises V→v→T and the subject to Spec,TP. Status on the spine: see unified structure, §2b and §4, postposition row. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
Some Navajo postpositions are written onto the front of the verb (“procliticized”); others stand as a separate word right before it (“detached”). The yes/no page left three questions about them (T43e). (1) Is the first spell-out unit D1 the verb word W, or the whole verb-plus-postpositions complex (Willie's MVS)? (2) Can a procliticized postposition carry the yes/no ending -ísh? (3) Following Grant's note of 2026-10-17, do verb + postposition pairs look idiomatic, i.e. lexicalized at the phrase level, like English phrasal verbs? This run re-read Willie 1991 on page images (pp. 24, 25, 48, 49, 64, 100, 221, 224; Ch. 2–3 searched for printed attached forms), re-read Hale 1997 on the proclitic postposition and the causative postposition (PDF pp. 2, 30–34, 53), re-read Hale 2003 n. 12–13, and opened two live pages of the Navajo Language Academy: the yes/no question grammar and the Postposition Lexicon, whose example lists supplied new forms. The work is about the routine's first phenomenon, the word as the point where rule set A (the word) hands over to rule set B (the clause), and about the second: what a verb + postposition pair contributes to the verb's participant slots. Two operations are kept apart throughout. Merge builds structure: it puts an element at a position. Agree values features such as person and number: a probe copies a value from the closest matching goal. Whether a postposition is written onto the verb is neither of these. It is a pronunciation (spell-out, prosody) fact. Written 2026-10-21, 4:46am PT.
Update 2026-10-22, 4:46am (T49).HYPOTHESIS PC-2 is falsified as written. NLA (9)–(11b) on the yes/no page print … bee dibé síníloh? ‘Did you rope the sheep with the rope …?’, with the object noun between bee “3-with” and the verb. Willie 1991 p. 41 (28f) lists -ee ‘with’ Instrumental as a grammatical postposition (page image). PC-2 is restated as IP-3: nothing from rule set B stands between a Σ-PP (a PP whose object is a participant of the verb's role schema) and W. A means phrase like the rope's is merged on the clause side (IP-1). T45d is answered yes for this bee, and is still open for Σ-PPs (T49a). See can a noun stand between a postposition and its verb?
Update 2026-10-22, 12:46pm (T49f).HYPOTHESIS PC-2′ (= IP-3) now refers to IP-1′: the PPs that must stand next to the verb are those whose slot the yi-/bi- voice ranks with the verb's own (Willie (79)–(88): ‘to’ with ‘speak’ and ‘roll’, ‘for’ with ‘lead’), not only those whose object the stem singles out. Verb by verb, with every postposition of Willie's list (28): postpositions by verb.
Update 2026-10-21, 12:46pm (T45f). The transportation pattern of §2.5 and PD7 is followed up on the -ił travel pattern, from the NLA's full “transportation” list (read in the Internet Archive's copies; the live site did not answer). HYPOTHESIS (1) PC-7b holds as a pattern: at least six verb stems (‘fly’, ‘float’, ‘crawl’, ‘move.SFO’, ‘run/drive/return’, ‘drive/push.wheel’), five kinds of carrier (lightning included) and six speakers; the ‘travel by’ meaning is built from ‘with’ + the verb, and the only listed piece is a word-level ‘drive’ sense of ʼoolwoł. (2) “The vehicle is the subject” is too narrow: the subject is whoever the stem ranks first, the carrier with movement stems, the human driver with the ‘drive’ stem -bąs (Háí lá kintah góó sh ił ʼadooł bąs? ‘Who can drive me to town?’), the traveller with -ee ‘by means of’ (TP-1 to TP-4). PD7's point stands, widened: control that the stem does not describe never picks the subject (sh ił dah ʼadiilwod ‘so I just drove that way’ has the driver in the ‘with’ phrase). (3) PC-2 restated: the preverbs dah ‘up’ and yah ‘into’ are written apart between a PP and the rest of the verb, but they are the verb word's left edge (Hale 2003 pp. 14–17; Hale 1997 p. 2), so nothing from rule set B intervenes (TP-6; T45d answered for preverbs).
Ground rules (Grant's). Every Navajo form below comes from a source opened this run, or is already on this site with its source number. Forms are quoted as printed, with the source's own gloss line, and nothing is segmented beyond what the source prints. What the sources say is graded SOURCED. Rules and mappings are ours (HYPOTHESIS). Grant's own ideas are GRANT'S CONJECTURE. Only contradicting data count against an analysis; a source's silence does not. No analysis, the site's own included, is assumed to be fully right. Grant does not speak Navajo, and nothing here is a native-speaker judgment.
Notation (the site's own; the Syntax · Semantics · Interface bot's BRIEF.md and VERB-QUEUE.md are still dated Oct 18, 10:05pm PT, with no conventions for this site). Σ = role-schema step (which slots the verb has, and each one's role in the action); A = word level (rule set A: the prefixal, left-headed spine, spelled into one word W); B = clause level (rule set B: right-headed clause syntax, noun linking, particles and enclitics). W = the verb word as assembled by head movement (Hale 1997's “syntactically assembled verb word”, Hale 2003's skeleton; candidate boundary B2 on rule set A vs B). MVS = Willie's “Maximal Verb-Sentence”, the verb plus its grammatical postpositional phrases (W&J 2000: “Maximal Verb Complex”; boundary B3). CYC, D1, D2: A/B as spell-out §3. EN-0 to EN-6: the yes/no page §4. PU-L, PU-G, L1–L9: postposition units. In our own lines “=” marks a clitic boundary.
Bottom line, in plain words
SOURCEDAttached or separate is a matter of pronunciation, not of grammar. Willie says the same kind of postpositional phrase “may be phonologically integrated into the MVS, or [it] may form the first word in the MVS” (p. 64), and “Some grammatical postpositional phrases are phonologically incorporated into the verb; some are not” (p. 224). Hale 1997 says of the loosely attached elements, the postposition included, that they are “phonologically dependent upon whatever follows to the right. This is a phonological matter, not a syntactic one” (PDF p. 31). One verb + postposition unit is even printed both ways by two sources: Willie (42) Shił béé hózin ‘I know it’ (three words), and the NLA's shił bééhózin (the postposition written into the verb word, with the verb listed as bééhózin ‘there is knowledge’).
HYPOTHESISAnswer to (1): the sealed unit is the verb word W, not the MVS. The MVS is real, but it is a phrase (the verb and the postpositional phrase it takes as complement), not a sealed spell-out block. Each postpositional phrase is spelled out and sealed on its own, and so is W. The MVS's “cannot be interrupted” follows from height: nouns and particles are all merged higher, and with head-final order they end up to its left (seven NLA sentences opened this run put doo, daatsʼí or ʼayóo before the postposition, and none puts anything between it and the verb). The yes/no ending can still go between the postposition and the verb, because it attaches to the right edge of an already spelled unit: NLA (6) Neésh hólǫ́? and, new this run, … dził bąąhgóósh chʼétʼaʼ? ‘… or along the mountainside?’. This settles the question the yes/no page left open in favour of its second option, and it revises the site's 2026-10-17 note “D1 = MVS”.
OPENAnswer to (2): not attested either way. No opened source prints -ísh on a postposition that is written onto the verb. Our rules predict that it is possible, and that the postposition then surfaces as a separate word, as Hale's preverb ka does in ka-[á]sh nanin[á] ‘are you sick?’ (Hale 2003 n. 12). They predict that it is impossible for the one postposition Hale says truly goes into the verb, the causative -y- of (41b) ʼAwééʼ bi-di-y-é-saʼ ‘I burped the baby’ (Hale 1997 PDF p. 53). Both are predictions (T45a, T45e).
HYPOTHESISAnswer to (3): some pairs are idioms, and being an idiom does not depend on being attached. The NLA's examples show three kinds of verb + postposition pairing. (a) Plain, meaning built from the parts: Naatʼáanii bichʼįʼ yáshtiʼ ‘I'm talking to the leader’. (b) A regular pattern with its own meaning: ‘with’ + a motion verb whose subject is a vehicle means ‘travel by it’ (tsinaaʼeeł nił níʼéél ‘boat 2-with 3-float’ = ‘you came by boat’; the NLA tags these “transportation”). (c) Particular pairs whose meaning is not the sum of the glosses: bídin nishłį́ ‘3-without 1-be’ = ‘I really want (it)’; niyaa hazlį́į́ʼ ‘2-under area-grow’ = ‘(you) grew up’, where the NLA notes the verb “means ’space came into being’”; bichʼįʼ sidziil ‘3-towards 1-strength’ = ‘I can withstand (it)’; -ká ‘after, for’ + a ‘go/run’-type verb glossed ‘help’, now in two NLA sentences besides Chee & Yazzie's (14b). Every kind-(c) postposition is written as a separate word. So the idioms are phrase-level, as Grant suggested, and being written onto the verb is neither needed for an idiom nor a sign of one.
GRANT'S CONJECTUREFor Grant's two phenomena. The word edge comes out sharper: rule set A builds and seals W (and, separately, each postpositional phrase). Writing a postposition onto the front of W happens after sealing, as pronunciation. The one postposition that goes inside W (the causative) is the one that does a grammatical job (realizing case), which fits Willie's statement that all “closed class, functional elements are incorporated” into words (p. 221). For the participant slots: a verb + postposition pair adds a slot (the postposition's object) to the verb's own slots, and in the transportation pattern the person who travels is that added slot, while the vehicle (the one that moves) is the verb's subject. Who is the verb's subject goes by the verb's own role in the action, not by who is in charge of the trip.
Willie, MaryAnn 1991. Navajo Pronouns and Obviation. PhD dissertation, University of Arizona (UMI 9210298; box scan)
Re-read on page images (110 dpi) of printed pp. 24, 25 (PDF 29, 30), 64 (PDF 70), 100 (PDF 106), 221 (PDF 227), 224 (PDF 230); (42) and (43) on 300 dpi crops of pp. 48–49 (PDF 53–54). Offsets: printed = PDF − 5 in Ch. 2 (pp. 24–50), PDF − 6 from Ch. 3 on. The text layer of pp. 39–104 (PDF 44–110) was searched for any printed form or gloss with a postposition written onto a verb: none found (OCR-based, so a near miss is possible; T45g). Whole-thesis search for “idiom”: 0 hits; “lexicaliz”: p. 43 only (as recorded on 10-17)
p. 24 “some of which also appear phonologically attached to left margin of the inflected verb, while others immediately precede the inflected verb”; p. 25 MVS = “a complex that behaves as a unified constituent under fundamental rules of the syntax”; p. 64 “phonologically integrated into the MVS, or … the first word in the MVS”, (29); p. 100 “the Inverse has scope over the entire MVS”; p. 221 “all closed class, functional elements are incorporated into the lexical items with which they are associated by government”; p. 224 “Some grammatical postpositional phrases are phonologically incorporated into the verb; some are not”; (42), (43)
Hale, Kenneth 1997. “Remarks on the syntax of the Navajo verb” (NLA workshop; box PDF hale023.pdf; cited by PDF page)
Re-read: text layer of PDF pp. 2, 30–34, 43, 53–54; page images of PDF pp. 31, 33, 34, 53 (120 dpi). Hale prints “¬” for ł in this PDF; it is restored as ł below and marked
p. 2: disjunct elements = “proclitics”, “more loosely attached”, including “postpositional proclitics”; p. 31: “a phonological matter, not a syntactic one”, (22a, b); p. 33: si-tsʼá “is phonologically dependent upon the following word; it is a proclitic”; p. 34: “only weakly integrated with the assembled verb word”; p. 43: postpositions “inflect like nouns”; p. 53: “Normally, when the PP complement of a verb is phonologically dependent, i.e., a proclitic, it merely attaches to the fully derived verb to its right, binding with it rather loosely … The causative P … actually incorporates into E”, (41a, b)
Hale, Kenneth 2003 ms. “The significance of the Pronominal Argument Hypothesis” (box PDF halejel.pdf)
Re-read, text layer of pp. 68–71 (n. 12–15)
n. 12: the preverb ka can “appear as separate words” and be “separated … by the polar interrogative enclitic” (restored forms as on the yes/no page); n. 13: an oblique object takes the leftmost prefix position unless a preverb competes (form not restored: T43h)
Fernald et al., Navajo Postposition Lexicon, version 10.2023 (Navajo Language Academy)
Opened live (web fetch, 2026-10-21, about 5am PT), twice: https://talkingdictionary.swarthmore.edu/navajo_postposition/ and the same page with ?intro. Each load shows a random list of 50 example sentences, so the two loads differ; every form used is listed in §2 with load, speaker and Young & Morgan page as printed. A box note records them ([local working copy, not in the repository], not committed). curl from the box returned nothing
17 example sentences (§2.3–2.5); the verb selector's entry “bééhózin there is knowledge”; the semantic-domain tag “transportation”
Fernald et al. 2019, Navajo Questions: grammar and examples, v3.2022 (NLA)
Opened live (web fetch, 2026-10-21): the yes/no page http://www.talkingdictionary.org/navajo_questions/?q-binary. Same wording as on 2026-10-20
(4), ?(5), (6), ?(7), (8) and the prose “Can we put the enclitic on the postposition? … If the postposition nee is the first word in the sentence, however, -ísh can appear there”
Young & Morgan 1987 itself (lending-only; the unofficial dump on the box is not used, by site policy). The NLA examples cite its pages (YM 1987:185, 410, 415, 661); those are the NLA's citations, not ours. Willie & Jelinek 2000 p. 270 was not re-opened this run (re-read on 10-20; used as recorded). Chee & Yazzie 2025 used as recorded on 10-17
2. The facts
2.1 The same kind of postpositional phrase, attached or separate
Hale 1997, PDF p. 31 (page image), (22); “¬” in the PDF restored as ł
(22a)
Tsinaaʼeeł
(shí)
sitsʼáníʼéél.
‘The boat drifted away from me.’ (no gloss line in the source)
(22b)
Ashkii
tsinaaʼeeł
(shí)
sitsʼéiníłʼéél.
‘The boy rowed (lit., floated) the boat away from me.’
PDF p. 31: “The proclitic portion of the verb word consists of the postposition -tsʼá; its complement (shí ‘me’, in this instance) and its specifier (tsinaaʼeeł ‘boat’) are independent words and are not phonologically dependent.” PDF p. 33: “the word formed by this head together with its object agreement inflection (in this case si-tsʼá ‘away from me’) is phonologically dependent upon the following word; it is a proclitic to that word. It is written together with the following word and is traditionally classified as belonging to the disjunct system of the verb word”.
Willie 1991:64 (page image), (29)
(29)
yá
yoołóós
3.for
3.3.lead [sing. +animate]
‘She is leading it for him.’
P. 64: “Grammatical postpositional phrases may be phonologically integrated into the MVS, or they may form the first word in the MVS as in the following Benefactive construction.” P. 24: “goal arguments are postpositional phrases, some of which also appear phonologically attached to left margin of the inflected verb, while others immediately precede the inflected verb.” P. 224: “Some grammatical postpositional phrases are phonologically incorporated into the verb; some are not. The grammatical postpositional phrases fall under the scope of Inverse voice alternation.”
‘I don't know the White Mountain Apache language.’
K3 (L2; Lorene B Legah)
ʼAtʼééd ʼéí chʼiyáán ʼálʼį́įgi łáʼí nidi doo bił bééhózin da.
girl TOP food 3-make.NI-at one but 3-with 3-know.CI NEG
‘The girl does not know how to cook food.’
K4 (L2; Lorene B Legah)
Ni daatsʼí nił bééhózin ʼéí doodaiiʼ nicheii daatsʼí bił bééhózin?
2 perhaps 2-with 3-know.CI TOP or 2-grandfather perhaps 3-with 3-know.CI
‘Do you know or does your grandfather know?’
The lexicon's verb selector lists “bééhózin there is knowledge” as a verb (both loads). The NLA question front page's “Nił ísh béého sin doodaii’ bił ísh béého zin?” (F9, spacing as displayed by the database, recorded on 10-20) puts -ísh on nił/bił, not on béé.
2.3 The yes/no ending between a postposition word and the verb
airplane-Q hogan 3-over-along 3-fly.P or mountain 3-side-along-Q 3-fly.P
‘Did the airplane fly over the house or along the mountainside?’
Q2 (L2; Lorene B Legah)
Daʼ chidíísh bee hólǫ́ doodaiiʼ ni daatsʼí chidí nee hólǫ́?
Q car-Q 3-with 4-exist.NI or 2 maybe car 2-with 4-exist.NI
‘Does s/he have a car or do you have a car?’
In Q1 each half of the ‘or’ question has its own question ending: on the subject chidí naatʼaʼí ‘airplane’ in the first half, and on the postpositional phrase dził bąąhgóó ‘along the mountain's side’ in the second, where it is the first constituent (the subject is not repeated). The ending sits between that phrase and the verb chʼétʼaʼ.
2.4 Verb + postposition pairs whose translation departs from the glosses
NLA Postposition Lexicon, example lists (live, 2026-10-21)
I1 (L2; YM 1987:661; Natalie Desiderio)
Chidí naaʼnaʼí ʼayóo bídin nishłį́.
tractor really 3-without 1-be.NI
‘I really want a tractor.’
I2 (L2; YM 1987:410; Tyler Tinhorn)
Háadi niyaa hazlį́į́ʼ?
where-at 2-under area-grow.P
‘Where did you grow up?’ Lexical notes: “Verb means ’space came into being’”
1-after indef-help.R or just-neg 1-about 1-2-bother.NI-neg
‘Keep helping me or else don't bother me!’
Already on the site: Chee & Yazzie 2025 (14b) ánáshwoʼ ‘I usually walk away out of sight’ → biká ánáshwoʼ ‘I usually help him’ (p. 34; postposition units §4). The lexicon lists -ká as “after, for”. In I4 and I5 the NLA glosses the verb itself as ‘help’; how the meaning is split between postposition and verb is the NLA's choice, and we do not re-segment.
2.5 A regular pattern and a plain pairing
NLA Postposition Lexicon, example lists (live, 2026-10-21); T1–T3 are tagged “transportation” under -ił ‘with’
T1 (L1; Bernadine M. Cody)
Daʼ Cubadę́ę́ʼ tsinaaʼeeł nił níʼéél?
Q Cuba-from boat 2-with 3-float.P
‘Did you come from Cuba by boat?’
T2 (L1; Wallace Dale)
Hádą́ą́ʼ ʼáłtsé chidí naatʼaʼí nił nastʼaʼ?
When past first airplane 2-with 3-fly.P
‘When did you first fly in an airplane?’
T3 (L2; John Harvey)
Kintahdę́ę́ʼ łį́į́ʼ shił nááldlosh.
town-from horse 1-with 3-crawl.CI
‘I'm riding back from town on horseback.’
P1 (L2; Lorene B Legah)
Naatʼáanii bichʼįʼ yáshtiʼ.
leader 3-to1-talk.I
‘I'm talking to the leader.’ (gloss spacing as printed)
P2 (L1; Maryann Sam)
Shizhéʼé bichíditsoh yaa naashnish.
1-father 3-truck-large on.it 3-3-worked.P
‘My father worked on his truck.’ (the database prints the gloss line twice; given once)
2.6 The one postposition that goes inside the verb word
Hale 1997, PDF p. 53 (page image), (41); “¬” restored as ł
(41a)
ʼAwééʼ
d-ee-zaʼ.
baby
d-SPF:3-belch:PERF
‘The baby burped.’
(41b)
ʼAwééʼ
bi-di-y-é-saʼ. (< b-d-y-s-´-sh-ł-zaʼ)
baby
3-d-y-SPF:1s-ł-belch:PERF
‘I burped the baby.’
PDF p. 53: “Normally, when the PP complement of a verb is phonologically dependent, i.e., a proclitic, it merely attaches to the fully derived verb to its right, binding with it rather loosely, like any other disjunct proclitic element. The causative P, realizing the oblique case of the Case-Bound argument awééʼ in (40), exhibits a different behavior morphologically. It actually incorporates into E, appearing to the right of the first overt E, if there is one. … In the process of incorporation, the agreement morphology (b- in (41b)) is left in initial position”. The same form is already on the site from another source (plain ł-causer, as Gessner (11); derivations); here Hale 1997's own page is used, and only for where the postposition ends up.
3. Attached vs separate: what each source says
Source
Attached and separate postpositions are …
What decides which
Status
Willie 1991 pp. 24, 64, 224
the same grammatical PPs, all in the MVS, all under the inverse
Not stated. “may be phonologically integrated … or … the first word”
SOURCED
Willie 1991 p. 50 (44b)
The MVS “may constitute a single phonological unit”: so it is not always one
—
SOURCED (as recorded)
Hale 1997 PDF pp. 2, 30–31, 33–34
proclitics (“disjunct prefixes”), “rather loosely attached”, outside the head-movement word; the attachment is “a phonological matter, not a syntactic one”
“prosodic structure”, perhaps (p. 34: “follows perhaps from factors having to do with the prosodic structure of the resulting sequence”)
SOURCED
Hale 1997 PDF p. 53
One exception: the causative P incorporates into E, inside the word, leaving its agreement b- initial
It realizes the oblique case of the causee
SOURCED
W&J 2000 p. 270 (as recorded)
“Some postpositions and their objects … are procliticized to the inflected verb; others remain detached”
Not stated
SOURCED
Willie (42) vs NLA K1–K4
The ‘know’ unit: béé hózin (Willie) and bééhózin (NLA)
Orthography, or speaker/period variation: the sources do not say
SOURCED (both spellings); cause OPEN
HYPOTHESISReading. All five agree that procliticization is a property of pronunciation laid over one syntax. The only source that names a syntactic difference (Hale's causative P) treats it as a different operation (incorporation into a head of the word), not as a stronger degree of procliticization. So there are two things, not a scale: leaning (post-syntactic, outside W) and incorporation (syntactic, inside W). The continuum the site recorded on 10-17 (L7: separate word → proclitic → thematic prefix) is kept as a description of the attested range, but its middle step is now read as prosody.
4. Rules PC-0 to PC-7 (ours; all HYPOTHESIS unless marked)
These refine CYC, EN-1/EN-5 and PU-L without rewriting them.
PC-0 ONE SYNTAX A grammatical PP is the verb's complement inside VP (Hale 1997 (23); PU-L).
Whether it is later written onto W does not change its syntax. (Willie
pp. 24, 64, 224; Hale 1997 PDF p. 31 "a phonological matter, not a syntactic
one": SOURCED. Stated as a rule: ours.)
PC-1 TWO SEALS Spell-out seals (a) each PP (pronoun + P: the PP sub-cycle of CYC) and
(b) the verb word W (the head-movement word). The MVS itself is NOT a sealed
unit. So the sealed spell-out unit of rule set A is W, and D1 = W.
Revises the 2026-10-17 note "D1 = MVS". The MVS is kept as the PHRASE
[VP PP V] + the heads above it, i.e. the domain where CYC orders the PP before
W (C1) and where the voice choice is made (Willie p. 100: "the Inverse has
scope over the entire MVS").
PC-2 NO INTRUDER Nothing that is merged is ever pronounced between a grammatical PP and W.
Derived, not stipulated: nouns are linked from higher positions (Top/Foc,
clause spine), and particles and adverbs (doo, daats'í, 'ayóo) merge
above VP; with head-final order they end up to the LEFT of [PP W].
(Facts SOURCED: NLA K2, K3, K4, I1, I3, I4 and Q2's second half; no opened form
shows an intruder: silence, not proof.) This is Willie's (44a) "it cannot be
interrupted", restated.
PC-3 LEAN After sealing, a phonologically dependent PP attaches to the next unit to its
right (Hale 1997 PDF p. 53 "merely attaches to the fully derived verb to its
right, binding with it rather loosely"). Post-syntactic (prosody). Not fixed
per P: Willie (42) béé hózin vs NLA bééhózin. Changes no seal.
PC-4 INCORPORATE A P that realizes a case (Hale's Pcaus) is merged INTO a head of W (E) by head
movement before W is sealed; its agreement b- stays at W's left edge (41b).
(SOURCED as Hale's analysis; the link to Willie p. 221 "all closed class,
functional elements are incorporated" is ours.)
PC-5 HOST -ísh may attach at the right edge of any sealed unit (EN-1). So:
detached PP-word + -ísh: (6) Neésh hólǫ́; Q1 dził bąąhgóósh OK (SOURCED)
leaning PP + -ísh: predicted OK; the P then surfaces as a
separate word (like ka-[á]sh, Hale n. 12)
NOT ATTESTED (T45a)
incorporated P (inside W) + -ísh: predicted impossible (EN-1') NOT ATTESTED (T45e)
PC-6 NO SIGN Attachment is no evidence of lexicalization, and lexicalization does not need
attachment: the particular pairs of 2.4 (bídin nishłį́, niyaa hazlį́į́',
bich'į' sidziil, -ká + 'help') are all written with a separate P.
PC-7 THREE KINDS Verb + P pairings come as (a) built (meaning = parts: P1); (b) a pattern with
its own meaning over a class of verbs (-ił + vehicle-subject motion verb =
'travel by': T1-T3); (c) particular pairs with an unpredictable meaning (I1-I5).
Only (c) needs a stored phrase-level entry (Grant's "phrasisized", L9).
Rule
Status
Rests on
PC-0
SOURCED (statements), HYPOTHESIS (as a rule)
Willie pp. 24, 64, 224; Hale 1997 PDF pp. 31, 33
PC-1
HYPOTHESIS
PC-0; NLA (6), Q1 (an ending between PP and W); CYC; Willie p. 100
SOURCED (Hale's statement), HYPOTHESIS (timing after the seal)
Hale 1997 PDF pp. 31, 34, 53; Willie (42) vs NLA K1–K4
PC-4
SOURCED (Hale's analysis); link to p. 221 HYPOTHESIS
Hale 1997 PDF p. 53, (41b); Willie p. 221
PC-5
detached: SOURCED; leaning and incorporated: OPEN predictions
NLA (6), Q1; Hale 2003 n. 12; EN-1, EN-1′
PC-6
HYPOTHESIS on SOURCED spellings
NLA I1–I5; Chee & Yazzie (14b)
PC-7
HYPOTHESIS; storage of (c) GRANT'S CONJECTURE
NLA T1–T3 (the “transportation” tag), I1–I5, P1
HYPOTHESISWhy not “D1 = MVS, sealed”? If the whole MVS were one sealed block, EN-1′ (no ending inside a sealed unit) would forbid (6) Neésh hólǫ́ and Q1 dził bąąhgóósh chʼétʼaʼ. Both are printed as fine. Keeping “MVS sealed” would need a special clause letting the ending reach inside, which is the option the yes/no page left open and which nothing else supports. PC-1 needs no such clause.
HYPOTHESISWhy not “D1 = W, and PPs are clause-level (SB1 as first written)”? That would put grammatical PPs in the same domain as nouns and particles, and then something would have to stop nouns and particles from landing between the PP and the verb. Under PC-1 the PP stays inside the verb's phrase (the MVS), and PC-2 gets the order from height. So PC-1 keeps the 10-17 insight (PPs belong with the verb), and drops only the claim that the MVS is sealed.
OPENWhat PC-1 does not explain. The “?” on NLA (5) ?Dibé neésh hólǫ́? (ending on the PP when a noun comes first). EN-3 treats a non-first ending as focus, and the NLA marks (5) doubtful rather than starred. Whether focusing a grammatical PP is degraded, and why, is still T43d.
5. Derivations PD1–PD8
Steps are labelled by rule set: Σ role schema, A word (W, and the PP sub-cycle), B clause. Each step names its operation: Merge (builds structure), Agree (values person/number), Move, Link (a nominal linked to a slot), Seal (spell-out of a unit), Lean (PC-3, prosody). VPISH: every participant slot is first-merged inside the verb's phrase; in the V-P variant the subject slot raises inside the word to Spec,M/SP, in V-H it stays. The strings are the same either way. Hale 1997's trees are transitional (E, M, c-government; framework audit); they are restated in the site's spine Qual > M/S > v > V (Hale 2003). No flat-S tree is used. Roles in the action are ours, read from the translations.
PD1. Hale 1997 (22a) Tsinaaʼeeł (shí) sitsʼáníʼéél ‘The boat drifted away from me’: a leaning PP
Σ Σ(√FLOAT + P ‘away from’) = {ρTh (movement), ρG (the one moved away from; no role-in-the-action entailment beyond being a reference point)}. No agent: the intransitive alternant is unaccusative (Hale PDF p. 32).
A PP sub-cycle: Merge P -tsʼá with ρG as complement. Agree P ⟶ ρG [1sg] → s(h)-. Seal PP → si-tsʼá. ρTh is in Spec,PP (Hale's analysis).
AMerge V with the PP (complement). M/S. ρTh needs a partner: Agree M/S ⟶ ρTh [3] (Hale: it raises to agree with M; under IA-3-style no-movement the Agree is enough for the string). Head movement builds W. Seal W → …áníʼéél (Hale does not segment it; neither do we).
A/B Order inside the MVS by CYC C1: PP ⌢ W. Lean (PC-3): si-tsʼá is phonologically dependent → written together: sitsʼáníʼéél.
BMergetsinaaʼeeł ‘boat’ and optional emphatic shí; Linktsinaaʼeeł ↔ ρTh, shí ↔ ρG. They are “independent words and are not phonologically dependent” (p. 31). Result (22a) ✓.
PD2. Willie (29) yá yoołóós ‘She is leading it for him’: a separate PP, same syntax
Σ Σ(√LEAD + P ‘for’) = {ρA (volition, causation of movement), ρP (moved, animate), ρBen (the one it is done for)}. All 3rd person.
A PP sub-cycle: Merge P -á with ρBen; Agree [3], direct → y-; Seal → yá.
AMerge V with the PP and ρP; v with ρA; M/S; Agree 3.3; Seal W → yoołóós (“3.3.lead”).
A/B CYC C1: yá ⌢ yoołóós. No lean: yá is “the first word in the MVS” (p. 64). Steps 1–3 are identical in kind to PD1; only this step differs (PC-0).
PD3. Willie (42) Shił béé hózin vs NLA K1 … shił bééhózin: one unit, two spellings
Σ Σ(√KNOW, with -ił ‘with’ and -ee ‘about’) = {ρExp (sentience: the knower; object of ‘with’), ρTh (what is known; object of ‘about’)} plus areal/pleonastic ho- in the verb (Willie p. 47–49; pleonastic ho- & psych).
A Two PP sub-cycles: Merge-ił with ρExp [1sg], Agree → shił, Seal. Merge-ee with ρTh [3], Agree → b- (Willie “3.about”), Seal → béé.
ASeal W → hózin (Willie “ho-known”). CYC C1, higher non-head first: shił ⌢ béé ⌢ hózin (which PP is higher is ours; the order is printed).
A/BLean (PC-3): Willie: none → Shił béé hózin (42) ✓. NLA: the inner PP leans → shił bééhózin (K1) ✓. The outer PP shił never leans in any opened form, and in F9 the ending goes on it: Nił ísh béého sin.
B K1: MergeNaashtʼézhí bizaad ‘Zuni language’ (gloss as printed: “English 3-language”); Link ↔ ρTh. Shí hanii tʼéiyá ‘I, contrary, only’ outside. Nothing is pronounced between shił and bééhózin (PC-2).
Note. The NLA lists bééhózin as a verb, ‘there is knowledge’. That is a word-level listing of [béé + hózin] in a lexicon. The experiencer's -ił is listed separately, under the postposition. So the ‘know’ unit is listed in two layers: inner P + verb as a word, outer P as a phrase-level partner. This is the NLA's arrangement (SOURCED); reading it as two layers of lexicalization is ours (HYPOTHESIS).
PD4. Hale 1997 (41b) ʼAwééʼ bi-di-y-é-saʼ ‘I burped the baby’: a P inside W
Σ Σ(√BELCH + CAUSE) = {ρA (causation; the speaker), ρCausee (the one who burps: the bodily process happens to it)}.
AMerge Pcaus-y- with ρCausee; Agree P ⟶ ρCausee [3] → b- (Hale: “finds its agreement partner within PP”).
A Build the word: V, v (ł), M/S with ρA [1s] (Agree → SPF:1s), E (d-). Move (head movement) Pcausinto E: “It actually incorporates into E, appearing to the right of the first overt E”. Its agreement b- stays at the left edge.
ASeal W → bi-di-y-é-saʼ ✓. The P is inside the sealed unit. PC-5 therefore predicts that -ísh cannot separate b- or -y- from the rest (T45e). Contrast PD1, where the P is sealed in its own unit and only leans.
BLinkʼawééʼ ↔ ρCausee.
PD5. NLA (6) Neésh hólǫ́? and Q1 … dził bąąhgóósh chʼétʼaʼ?: an ending between the PP and W
A (6): PP sub-cycle Seal → nee ‘2-with’; W Seal → hólǫ́ (‘exist.NI’). CYC: nee ⌢ hólǫ́.
B (6): Merge C[Q]. EN-2: the first constituent of C's complement is the PP (no noun). EN-1/PC-5: -ísh at its right edge → Neésh hólǫ́ ✓. If the MVS were sealed, this would be blocked (§4).
A Q1, second half: inner PP Merge-ąąh ‘on, beside’ with a 3rd-person slot, Agree → b- (gloss “3-side”); outer lexical P -góó ‘along’ over it (Willie p. 44's grammatical + lexical layering, as on PU1/PU-S). The noun dził ‘mountain’ is inside the same phrase: the ending follows bąąhgóó, not dził, so [dził bąąhgóó] is one constituent (‘along the mountainside’; noun + postposition like possessor + possessed, which Hale 1997 PDF p. 43 notes for postpositions: they “inflect like nouns”). Seal → dził bąąhgóó. W Seal → chʼétʼaʼ (“3-fly.P”).
B Q1: each half of the doodaii’ ‘or’ question gets its own ending on its first constituent (EN-2 per half). Half 2 has no subject noun, so its first constituent is the PP → dził bąąhgóósh chʼétʼaʼ ✓. Half 1 → Chidí naatʼaʼíísh on the subject noun phrase.
Two cautions. (i) Here the noun forms a constituent with the PP, whereas W&J (50) say the noun “does not form a syntactic constituent with the postpositional phrase” for the grammatical -tah. The difference may be the outer lexical -góó (Willie p. 40: lexical Ps take nominal objects) or the locative ‘side’ postposition; not decided (T31f/T32i). (ii) For the same reason, whether this PP is inside the MVS is open. (6) is the clean case: nee hólǫ́ ‘have’ is a grammatical-P unit with no noun.
BMerge negation doo … da above the MVS: doo is head-initial-looking, da clause-final. Wherever exactly they sit, both are higher than VP, so with head-final order doo lands left of shił, never between shił and bééhózin (PC-2). The same holds for daatsʼí in K4, ʼayóo in I1 and I3, and łáʼí nidi doo in K3.
BMergeDziłghą́ą́ʼí bizaad ‘White Mountain Apache language’; Link ↔ ρTh ✓.
Where doo and da sit in the spine is not settled here (no opened source places them; T45d asks only whether anything ever intrudes).
PD7. NLA T1 Daʼ Cubadę́ę́ʼ tsinaaʼeeł nił níʼéél? ‘Did you come from Cuba by boat?’: the transportation pattern
Σ The verb's own schema: Σ(√FLOAT) = {ρTh (movement)}: the boat floats. The pattern (PC-7b) adds P -ił ‘with’ with its slot ρCom: the one who goes along. Role in the action for ρCom: change of location together with ρTh; nothing in the gloss says it causes the movement.
A PP sub-cycle: Merge-ił with ρCom [2]; Agree → nił; Seal. W: Agree M/S ⟶ ρTh [3]; Seal → níʼéél (“3-float.P”). CYC: nił ⌢ níʼéél.
BMergetsinaaʼeeł ‘boat’; Link ↔ ρTh, the verb's subject. MergeCubadę́ę́ʼ ‘from Cuba’ (lexical P, adjunct). Mergeda’ in Spec,CP. Result T1 ✓.
Observation. The human, the one who chose the trip, is not the verb's subject; the vehicle is, because the verb ‘float’ selects the mover. T2 (airplane ‘flew’ with you) and T3 (horse ‘crawl.CI’ with me = ‘riding on horseback’) repeat this. Subject choice follows the verb's own role in the action (movement), not the overall event's instigator. Hale 1997 (22a) Tsinaaʼeeł (shí) sitsʼáníʼéél has a ‘float’ verb with the boat as subject too, but with a different postposition and no travel meaning: the pattern's meaning comes from -ił + a vehicle subject, which is why PC-7 calls it a pattern and not an idiom of one verb.
PD8. NLA I1 Chidí naaʼnaʼí ʼayóo bídin nishłį́ ‘I really want a tractor’: a particular pair
Σ Built reading (gloss “3-without 1-be.NI”): Σ(√BE) = {ρSub [1]} + P -ídin ‘without’ with ρX [3]: ‘I am without it’. Stored reading (PC-7c): the pair [-ídin + √BE] ↔ ‘want’, with ρSub = the one who wants (sentience) and ρX = the thing wanted.
A Either way the same steps: PP sub-cycle Merge-ídin with ρX, Agree [3], Seal → bídin; W Agree M/S ⟶ [1], Seal → nishłį́. CYC: bídin ⌢ nishłį́. No lean: written apart.
BMergeʼayóo ‘really’ above the MVS (PC-2: left of bídin). Mergechidí naaʼnaʼí ‘tractor’; Link ↔ ρX ✓.
What the stored reading adds: nothing to the syntax, and no change to which slot each noun links to. It changes only the meaning assigned to the pair. That is what “phrase-level lexicalization” means here: one listed meaning spread over two sealed units. The same holds for I2 niyaa hazlį́į́ʼ (‘2-under’ + ‘space came into being’ = ‘grew up’) and I3 bichʼįʼ sidziil (‘3-towards’ + ‘1-strength’ = ‘can withstand’).
6. Which pairs are idioms? (Grant's 2026-10-17 note)
Pair
Literal from the glosses
Printed translation
Kind (PC-7)
P written
Source
bichʼįʼ yáshtiʼ
3-to 1-talk
‘I'm talking to (the leader)’
(a) built
apart
NLA P1
yaa naashnish
on.it worked
‘worked on (his truck)’
(a) or close to it
apart
NLA P2
-ił + vehicle-subject motion verb
(vehicle) floated/flew/crawled with X
‘X came by boat / flew in an airplane / rode on horseback’
(b) pattern (NLA tag “transportation”)
apart
NLA T1–T3
-ił + bééhózin
with X, there is knowledge (about it)
‘X knows (it)’
(c) at the outer layer; inner béé + hózin listed as one verb
outer apart; inner both ways
Willie (42), (43); NLA K1–K4
bídin nishłį́
3-without 1-be
‘I really want (it)’
(c)
apart
NLA I1 (YM 1987:661)
niyaa hazlį́į́ʼ
2-under area-grow (“space came into being”)
‘(you) grew up’
(c)
apart
NLA I2 (YM 1987:410)
bichʼįʼ sidziil
3-towards 1-strength
‘I can withstand (it)’
(c)
apart
NLA I3 (YM 1987:185)
-ká + ‘go/run’-type verb
after/for X + verb glossed ‘help’ by the NLA; ‘walk away out of sight’ for the base in Chee & Yazzie
‘help X’
(c), recurring across three sources/speakers
apart
NLA I4, I5; Chee & Yazzie (14b)
SOURCED The forms, glosses, translations and the NLA's “transportation” tag and lexical note. HYPOTHESIS The three-way sorting, and the judgment that a translation “departs from the glosses”. That judgment is ours, made from English glosses; a speaker might find a literal path we cannot see. GRANT'S CONJECTURE That kind (c) pairs are stored as phrase-level units (“phrasisized”). The new evidence fits it in two ways: the same pair recurs with the same meaning across independent sources and speakers (-ká + ‘help’), and the NLA itself flags one verb's literal meaning as different from the translation (I2). Neither proves storage; a productive metaphor would also recur.
HYPOTHESISCompared with English phrasal verbs. (General background, not Navajo evidence.) English give up is a verb + particle idiom; English look after is a verb + preposition idiom with an object. Navajo kind (c) is closer to the second: the postposition has its own object slot with its own agreement (b-, ni-), and the noun linked to that slot is an ordinary participant (‘a tractor’, ‘a lot of cold’). Unlike English particles, the Navajo postposition cannot be separated from the verb by another word (PC-2), only by a clitic.
OPEN Whether Young & Morgan list the kind (c) pairs as units (T32f, T45c). The NLA cites YM pages for I1–I3, so the sentences come from that dictionary, but how YM file them (under the verb, under the postposition, or as a theme) was not opened.
7. What changes
Item
Before
After
Why
T43e (yes/no page)
“can a procliticized P host -ísh, and is D1 = W or the MVS?”
D1 half answered: sealed unit = W (and each PP); the MVS is the phrase in which they are ordered and the voice is chosen (PC-1). Host half: still unattested; now a two-way prediction (PC-5; T45a, T45e)
Willie pp. 24, 64, 224; Hale 1997 PDF pp. 31, 53; NLA (6), Q1
Spell-out rules, 10-17 note “D1 = MVS”
The MVS taken as D1, sealed, with PPs as already-spelled non-heads inside it
Kept: PPs are inside the MVS and ordered there by CYC. Dropped: the MVS as a sealed unit. D1 (the sealed A-unit) = W
PC-1; the ending in (6) and Q1
Clause spine MVS-D1
“The first spell-out cycle D1 is QualP together with the VP-internal PPs”
Read as: the first ordering domain is QualP with its PPs; the sealed units are W and each PP. No derivation on the clause spine changes its string
PC-1
Yes/no page §5 “two word edges, two jobs”
Either D1 = W with PPs clause-level, or D1 = MVS with the enclitic reaching inside (OPEN)
Neither as stated: D1 = W, PPs inside the MVS phrase but sealed separately (PC-1, PC-2)
§4
Rule set A vs B, B2 vs B3
B3 (MVS) = domain of the voice alternation; B2 (W) = the enclitic's edge
Confirmed and sharpened: B3 is a syntactic domain without a seal; B2 is the seal. Procliticization is not a third boundary (PC-3)
PC-0, PC-3
Postposition units L7 (continuum)
separate word → proclitic → thematic prefix, read as a possible lexicalization path
The middle step is prosody (PC-3), not a degree of lexicalization (PC-6). Incorporation (Pcaus) is a different, syntactic operation (PC-4)
Hale 1997 PDF pp. 31, 53; NLA I1–I5
Postposition units L9 (“phrasisized”)
Compatible with the data, untested
Still a conjecture, now with five particular pairs from the NLA and one pair recurring across three sources
§6
8. What each source gets right and misses
Source
Gets right
Misses or costs
Willie 1991
One MVS whether the PP is attached or not (pp. 64, 224); the inverse scopes over the whole MVS (p. 100); “may constitute a single phonological unit” (p. 50), i.e. not always
“It cannot be interrupted” does not say by what; the yes/no ending does interrupt it ((6), Q1). No printed example of an attached grammatical P in Ch. 2–3 (gloss search), so “phonologically integrated” has no example of its own in the chapters read
Hale 1997
Attachment is phonological, not syntactic (p. 31); the proclitic “merely attaches … rather loosely” (p. 53); the causative P is the exception that goes inside (p. 53)
GB-era machinery (E, M, c-government, raising for agreement); restated here. The prosodic reason for leaning is left as “perhaps” (p. 34)
NLA Postposition Lexicon
Sentences with speaker and YM source; a “transportation” tag that picks out a pattern; a lexical note on I2's verb; bééhózin listed as a verb
Random example lists, so a cited sentence may not be on the page at the next load. Gloss/translation mismatch in K1 (“English” vs “Zuni”). The gloss line in P2 is printed twice
This site before today
PPs belong with the verb (10-17, MVS-D1); the enclitic's seal is W (10-20)
Treated the MVS as sealed; read the attached/separate range as a lexicalization path (L7)
9. Grant's frame
GRANT'S CONJECTUREThe word as the changeover point (phenomenon 1). This run gives the changeover a cleaner shape. Rule set A builds and seals W, and, as separate small units, each postpositional phrase. Rule set B then orders sealed units head-finally and places clitics at right edges. Writing a postposition onto the front of W (leaning) happens after sealing and belongs to neither rule set's syntax: it is pronunciation. The single exception, the causative postposition, is moved into W before sealing, and it is the case-realizing one, which fits Willie's “all closed class, functional elements are incorporated” (p. 221). So the word edge is where the functional material stops, not where the written word stops. That is a fit, and it can fail: one ordinary (non-causative) postposition that is provably inside W, e.g. separated from its verb by nothing while a conjunct prefix of W precedes it, would break PC-1.
HYPOTHESISTheta roles and slots (phenomenon 2). In every derivation the postposition adds one slot with its own agreement, and nouns link to slots afterwards. In the transportation pattern (PD7) the person who travels is the postposition's slot and the vehicle is the verb's subject, because the verb selects the mover. That is a case where Dowty-style role-in-the-action entailments of the verb, not the overall event's instigator, decide the subject. It fits the weak thesis (the verb's arguments are role-bearing slots, here extended by the postposition's slot). It is not a test of the strong thesis. Theta roles and event-structure roles stay compatible here: at the level of the whole trip, the traveller and the vehicle both change location (one event-structure layer), while the verb's own theta roles say which one moves under its own description.
GRANT'S CONJECTURESTRATA (theta roles first, event structure next, noun selection later). The order in PD1–PD8 is Σ (roles) → Agree and sealing of the PPs and W → ordering in the MVS → linking of nouns → clitic placement and leaning. The stored meaning of a kind (c) pair (PD8) attaches to the pair without changing any step. Where it enters (at Σ, as a listed role frame, or later, as an interpretation of the finished MVS) is not decided by these data. A fit, not a separating test.
GRANT'S CONJECTUREWord-internal Merge and Agree sites. PD4 is the one opened case that bears on it. In (41b), the causative postposition's Agree happens in its own phrase (b- with the baby), and then the postposition itself is merged into E by head movement, inside the word, while b- stays at the word's left edge. So inside one word, a head's Agree exponent and its final Merge site are in different places (Hale's description, SOURCED; read as separate Merge and Agree sites, HYPOTHESIS). This is consistent with Grant's idea that a word can contain separate Merge sites and Agree sites. It does not show that such sites are general. Merge and Agree are kept apart here as always: Merge put -y- into E; Agree gave b- its person.
GRANT'S CONJECTUREPhrase-level lexicalization. Supported as a description of kind (c) (§6): the idioms are phrase-level, made of two sealed units with one listed meaning, and attachment plays no part in them. As before, “phrasisized” is Grant's label, not a field term.
10. Tasks T45a–h and statuses
Task
What it needs
T45a
A yes/no ending on a postposition that is normally written onto the verb (e.g. the béé of bééhózin, or si-tsʼá): is it possible, and does the postposition then appear as a separate word (PC-5)?
T45b
One source or one speaker writing the same grammatical postposition both attached and apart: free variation, or conditioned (by prosody, by the verb)?
T45c
Do Young & Morgan 1987 list the kind (c) pairs (-ídin + ‘be’, -yaa + hazlį́į́ʼ, -chʼįʼ + sidziil, -ká + ‘help’) as units? (blocked: YM lending-only)
T45d
Any word (noun, particle, adverb) printed between a grammatical PP and its verb? One would falsify PC-2's height account
T45e
Can -ísh follow the b- of a causative like (41b), or split the word anywhere? Predicted no (PC-4, PC-5, EN-1′)
T45f
Is the -ił “transportation” pattern productive (new vehicles, new motion verbs), as kind (b) predicts? The NLA's tagged list for -ił would give more cases
T45g
A printed example of Willie's “phonologically integrated” grammatical PP in Willie 1991 itself (the OCR gloss search of pp. 39–104 found none; a page-image check of Ch. 4–8 would settle it)
T45h
Where do doo … da, daatsʼí and ʼayóo sit in the clause spine (PD6)?
Statuses: T43e: the D1 half is answered (PC-1); the host half is now T45a/T45e. T32f (are Chee & Yazzie's pairs stored?) gains one recurring pair (-ká + ‘help’); still OPEN. T32i/T31f (lexical PPs in D1?) unchanged; Q1 does not decide it. T43c, T43d, T43h unchanged. T42d, T42e, T42g still blocked on Young & Morgan.
What else changed on the site
Update notes on where the yes/no enclitic -ísh can go (lede), A/B as spell-out (after the 10-17 note on where D1 stops), clause spine (after MVS-D1), rule set A vs B (after the B1–B3 note) and postposition units (lede). There is a new row on open questions (T45a–h), a sources section (added 2026-10-21, 4:46am), and a home-page entry. No verb-dictionary entries changed. Navajo forms new to the site (all with sources): NLA K1–K4, Q1, Q2, I1–I5, T1–T3, P1, P2. Willie (29), (42), (43), Hale 1997 (22a, b), (41a, b) and NLA yes/no (4)–(6) were already on the site and are re-read here.
Two question endings compared: content-question =sh and yes/no -ísh. First word or first phrase, and what the inverse forbids (T43f)
2026-11-11 audit pass 3. Local box file paths were removed from this page; files inside the repository are given by repo-relative path, and the source citation next to each removed path stays as the reference. See the pass 3 report.
2026-11-11 audit pass 4. Interlinear rows whose form and gloss do not line up one-to-one were re-checked against the source page. Each is marked † with a note under its table: SOURCED where the source prints it that way, OPEN where the source cannot confirm it. See the pass 4 report.
Restated on the unified structure (2026-11-15). Questions are restated with Q in C⁰ above NegP: the wh word stays in its theta position as indefinite + Q (H11), and =sh/-ísh are interpreted in C⁰ and pronounced on a host (H12; H12-move vs H12-pf OPEN). Whether daʼ is a head is OPEN. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Status on the spine: see unified structure, §4, question rows. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
Navajo has two question endings. The yes/no ending -ísh was set out as rules EN-0 to EN-6 on where the yes/no enclitic -ísh can go. The content-question ending =sh (the one in háísh ‘who’) was recorded on asking about a role §8 from Willie 1991, who says it can move to “a position following the first word in the sentence” and cannot do so in the inverse. That page left two questions (T43f). Does =sh count by words while -ísh counts by phrases? Does -ísh have the same restriction in the inverse? This run re-read Willie 1991 pp. 210–212 and p. 92 and Willie & Jelinek 2000 pp. 272–273 on page images, and opened the Navajo Language Academy's content-question page live. The two endings are compared rule by rule, and the derivations are run step by step. The work is about the routine's first phenomenon, the word as the point where rule set A (the word) hands over to rule set B (the clause), and it touches the second: which participant slot a question asks about is fixed before any ending is placed. Two operations are kept apart. Merge builds structure: it puts an element at a position. Agree values features such as person and number: a probe copies a value from the closest matching goal. Placing either question ending is neither of these. It is a spell-out (pronunciation) step of the clause cycle. Written 2026-10-20, 8:46pm PT.
Ground rules (Grant's). Every Navajo form below comes from a source opened this run (Willie 1991 pp. 92, 210–212; Willie & Jelinek 2000 pp. 272–273; the NLA content-question page), or is already on this site with its source number. Forms are quoted as printed, with the source's own gloss line, and nothing is segmented beyond what the source prints. What the sources say is graded SOURCED. Rules and mappings are ours (HYPOTHESIS). Only contradicting data count against an analysis; a source's silence does not. No analysis, the site's own included, is assumed to be fully right.
Notation (the site's own; the Syntax · Semantics · Interface bot's BRIEF.md and VERB-QUEUE.md are still dated Oct 18, 10:05pm PT, with no conventions for this site). Σ = role-schema step (which slots the verb has, and each one's role in the action); A = word level (cycle D1, rule set A: the prefixal, left-headed spine, spelled into one word W); B = clause level (cycle D2, rule set B: right-headed clause syntax, noun linking, particles and enclitics). EXT = the slot the voice prefix makes the clause's subject (yi-: the agent; bi-, the inverse: the patient). INV-6 (the nominal next to the verb is the internal participant) and INV-7 (focus fronting under bi-): the inverse page. EN-0 to EN-6: the yes/no page §4. SH-FOC, SH-MOVE, WH-SPLIT: asking about a role §11. Glosses are the sources' own (Willie: “3.3.kicked”, “INV.3.3.kicked”, “=Q”, “=FOCUS”; W&J: “3ACC-3NOM-kissed (Inverse)”; NLA: “where-to-Q”). In our own lines “=” marks a clitic boundary.
Bottom line, in plain words
SOURCEDBoth endings close a whole phrase when they sit on their own host. In Willie's (19) ʼashiiké díkwíí=sh yiztsʼǫs ‘She kissed how many boys?’ and (20) hastiin łį́į́ʼ háídí=ígíí=sh yizloh ‘The man roped which horse?’, =sh comes at the end of the two- or three-word question phrase that Willie calls “a complex adjunct” (p. 212), after the determiner =ígíí in (20). The NLA says the same in general terms: the particle goes “after the constituent (phrase) that is being emphasized”. This is how -ísh behaves too (NLA yes/no (10a)).
HYPOTHESIS“First word” is Willie's wording for both endings, so it does not show a difference. Willie writes “first word” for the moved =sh (p. 211), and also for the yes/no -ísh: “In a yes/no question, the clitic -ish usually follows the first word in the sentence” (p. 92, opened this run). For -ísh, the NLA's (10a) shows that a whole first phrase can be closed. The one opened example of a moved =sh, (16) Díí=sh háí yiztał, has a first phrase one word long, so it cannot tell word from phrase. That one point stays OPEN (T44a). Our working rule is that the moved =sh lands where the plain yes/no -ísh lands.
HYPOTHESISThe inverse limits shifting a question ending, not the place it lands. Compare two inverse sentences that both put a question ending on the first word, which is the inverse's subject. Willie (76b) łííʼ-ísh iiʼniʼ biisxį́ ‘Did the lightning kill the horse?’ is fine. The yes/no ending is simply in its plain first position. *(17) Díí=sh háí biztał is out. Here the content ending has been shifted off its question word ‘who’. The same holds for the other inverse examples. The content ending is fine on its own question word wherever that word sits ((14b), W&J (57b), (59), (60a)). It may not be shifted (*(17)), and it may not be fronted with its word (*W&J (60b) reading 2). The yes/no ending may not go on a word that has itself been fronted (*(76a)). So in the inverse a question is built only on the unshifted, unfronted order (rule QC-5).
HYPOTHESISWhat it means for the word/clause changeover. Both endings are counted in clause-level units and attach at right edges, as the clause's head-final order predicts. No opened example puts either one inside a verb word (silence, which counts for nothing either way). The inverse restriction runs one way only. The voice is decided and spelled inside the verb word (rule set A), and the clause level (rule set B) can only read it when it places the ending. Nothing at the clause level changes which slot a question asks about, or anyone's role in the action.
Willie, MaryAnn 1991. Navajo Pronouns and Obviation. PhD dissertation, University of Arizona (UMI 9210298; box scan)
Re-read on page images of printed pp. 210, 211, 212 (PDF pp. 216–218, 150 dpi), with (16), (17), (19), (20) re-read on 300 dpi crops. (20) had been read at 110 dpi only, and the 300 dpi crop confirms the site's transcription. Printed p. 92 (PDF p. 98, 110 dpi): (76) and the prose after it. The older 300 dpi check of (76) is recorded on the inverse page
(14a, b), (15)–(20) and the prose of §8.4–8.5; (76a, b) and “In a yes/no question, the clitic -ish usually follows the first word in the sentence …”
Willie, MaryAnn & Eloise Jelinek 2000. “Navajo as a discourse configurational language.” In Fernald & Platero (eds.), The Athabaskan Languages (OUP)
Re-read on page images of pp. 272–273 (Jelinek collection scan [local working copy, not in the repository], PDF pp. 151–152, 120 dpi)
(57a, b), (58a, b), (59), (60a, b) and the prose: “The particle -sh encliticized to Wh-words has inherent focus, and usually occurs after the first word in the sentence”; “The bi-pronoun marks a topic/focus inversion that is incompatible with focus movement of the Wh-word”
Fernald, Theodore B. et al. 2019. Navajo Questions: grammar and examples, version 3.2022. Navajo Language Academy (NLA)
Opened live (web fetch, 2026-10-20): the content-question page http://www.talkingdictionary.org/navajo_questions/?q-content. Compared with the box copy saved on Oct 18 ([local working copy, not in the repository]): same wording. The live page numbers (1), (2), (5) and leaves (3), (4) unnumbered. The box copy numbers all five, and that numbering is used here
(1)–(5) and the prose on -sh, lá, -shąʼ: “the emphasis is generally carried out by adding a particle after the constituent (phrase) that is being emphasized”; “they frequently appear just after the h-word or cliticize to it, although they can appear in other places”; “the content question enclitic -sh is not the same as the yes/no question enclitic -ísh”
Young & Morgan 1980/1987 (Internet Archive copy lending-only; the unofficial dump on the box is not used, by site policy). NLA search results for the “-sh enclitic for content questions” filter (the filter is a client-side selector; earlier box requests to the host timed out). Willie 1991 Ch. 3 on -ísh beyond p. 92
P. 211: “The =sh clitic can be moved away from the Wh- word to a position following the first word in the sentence. … But in Inverse constructions, it is impossible to attach the question enclitic to the first word in the sentence. … I assume this is excluded since the Inverse is already a marked construction type that places focus on the patient as subject.” P. 212: “In these phrases, we see a Wh- word combined with an ordinary nominal to build a complex adjunct; extraction from these phrases is not possible.”
Willie 1991:92 (page image), (76)
(76a)
*iiʼniʼish
łííʼ
biisxį́
(76b)
łííʼ-ísh
iiʼniʼ
biisxį́
‘Did the lightning kill the horse?’ (the only translation printed for (76); no gloss line in the source)
P. 92: “In a yes/no question, the clitic -ish usually follows the first word in the sentence. When there is focus movement, as in (76a), this clitic does not attach to the moved element. The question clitic cannot attach anywhere in this structure, which simply cannot be used to form a yes/no question.” (The inverse page reads (74a) iiʼniʼ łį́į́ʼ biisxį́ as the focus-fronted order, with iiʼniʼ ‘lightning’ (the agent) fronted over łííʼ ‘horse’, the inverse subject: I5.)
Willie & Jelinek 2000:272–273 (page images)
(57a)
háí-sh
yiztsʼǫs
who-Q
3ACC-3NOM-kissed (Direct)
‘Who did he kiss?’
(57b)
háí-sh
biztsʼǫs
who-Q
3ACC-3NOM-kissed (Inverse)
‘Who kissed him?’
(58a)
Jáan
háí-sh
yiztsʼǫs
John
who-Q
3ACC-3NOM-kissed (Direct)
‘As for John, he kissed who?’
(58b)
háí-sh
Jáan
yiztsʼǫs
who-Q
John
3ACC-3NOM-kissed (Direct)
1) ‘Who was it, that kissed John?’ Or: 2) ‘Who was it, that John kissed?’
(59)
háí-sh
Jáan
biztsʼǫs
who-Q
John
3ACC-3NOM-kissed (Inverse)
‘Which one (of you) was it, that was kissed by John?’
(60a)
Jáan
háí-sh
biztsʼǫs
John
who-Q
3ACC-3NOM-kissed (Inverse)
‘As for John, he was kissed by whom?’
(60b)
háí-sh
Jáan
biztsʼǫs
who†
John
3ACC-3NOM-kissed (Inverse)
1) ‘Which one (of you) was kissed by John?’ 2) *‘John was kissed by whom/which one?’
SOURCED † Segmentation note (2026-11-11 audit pass 4): Willie & Jelinek 2000:273 (60b) glosses háí-sh as “who”, although (57)–(60a) on the facing pages gloss the same word “who-Q”. Kept exactly as printed.
P. 272: “There is no obligatory Wh-movement in Navajo, but there may be focus (leftward) movement of the Wh-word. The particle -sh encliticized to Wh-words has inherent focus, and usually occurs after the first word in the sentence.” “When a Wh-word and a second nominal are adjoined, both adjuncts have focus.” “In (58a), the first adjunct is a topic … The second reading of (58b) shows optional focus movement of the Wh-word + Q to the leftmost position.” Pp. 272–273: “The inverse does not permit optional focus movement of the Wh-word”; “The bi-pronoun marks a topic/focus inversion that is incompatible with focus movement of the Wh-word.” (60b) is glossed “who”, not “who-Q”, as printed.
‘Who’s that old lady over there?’ (topic -shąʼ at the end of the three-word phrase ‘that old woman’)
SOURCEDSummary. (i) =sh marks focus (Willie p. 210; W&J “inherent focus”; NLA “used to focus on a phrase”). -ísh marks a yes/no question and is neutral on the first phrase (NLA yes/no). The NLA says the two are “not the same”. (ii) On its own host, =sh follows the whole wh-phrase: after the second word in (19), after the third word and the determiner in (20). (iii) =sh can be shifted off the wh-word to after the first word (16), but not in the inverse *(17). (iv) Under bi-, =sh on its own unmoved wh-word is fine wherever the word stands: first (14b), (57b), (59), or second (60a). A wh-word fronted with =sh is out *(60b-2). (v) The yes/no -ísh under bi- is fine on the unmoved first phrase (76b), but out on a fronted one *(76a), and then there is nowhere else to put it (p. 92). (vi) Willie writes “first word” for both endings (pp. 92, 211), and so do W&J for =sh (p. 272). The NLA writes “phrase”/“constituent” for both.
3. The two endings side by side
Property
Yes/no -ísh
Content =sh
Same?
What it marks
Yes/no question; neutral on the first phrase, focus elsewhere (NLA yes/no)
Focus, always (Willie p. 210; W&J p. 272; NLA)
No SOURCED
Basic host
The first constituent after da’ (EN-2)
The wh-phrase (Willie p. 210; NLA “just after the h-word”)
No SOURCED
Edge counted
Right edge of a whole phrase (NLA (10a)); a word inside it only with focus ((11a))
Right edge of the whole wh-phrase ((19), (20); NLA “after the constituent (phrase)”)
Yes SOURCED facts
Off its basic host
On any other phrase = focus (EN-3)
Shifted to “the first word” (16)
Different directions SOURCED
Shifted to the first unit: a word or a phrase?
Phrase ((10a)), though Willie says “word” (p. 92)
Only a one-word case is opened (16)
OPEN (T44a)
In the inverse
On the unmoved first phrase ✓ (76b); on a fronted phrase ✗, and then nowhere ✗ (76a)
On its unmoved wh-word ✓ ((14b), (57b), (59), (60a)); shifted ✗ (17); fronted with the wh-word ✗ (60b-2)
Same ban, different cases (QC-5) HYPOTHESIS
Inside a verb word
No opened form
No opened form
Silence (EN-1′ stays a prediction)
4. Rules QC-0 to QC-6 (ours; all HYPOTHESIS unless marked)
These add to EN-0–EN-6 and SH-FOC/SH-MOVE without rewriting them. QC-1 extends EN-1/EN-4 to =sh, and QC-2 restates SH-MOVE in EN-2's terms.
QC-0 TWO EXPONENTS -ish spells the clause's polar [Q] on C (EN-0). =sh spells [Q, FOC] on the
wh-phrase that carries the question variable. Different morphemes:
SOURCED (NLA "not the same"; W&J "inherent focus").
QC-1 PHRASE EDGE =sh is spelled at the RIGHT EDGE of its host, the whole wh-phrase, outside any
closing element of that phrase: [[łį́į́’ háídí]=ígíí]=sh (20); [’ashiiké díkwíí]=sh
(19). (Facts SOURCED: (19), (20), Willie p. 212 "complex adjunct"; NLA "after
the constituent (phrase)".) Same edge rule as EN-1 + EN-4 for -ish. Since the
wh-word is last in its phrase (rule set B, head-final), the word edge and the
phrase edge coincide in (19); (20) separates them, because =ígíí closes the
phrase before =sh.
QC-2 SHIFT =sh may instead be spelled at the clause's first unit (SH-MOVE, (16)). Working
rule: the first unit is the one EN-2 picks for neutral -ish, i.e. the first
constituent of C's complement. "First word" (Willie p. 211; W&J p. 272) is
read as wording, because Willie uses it for -ish too (p. 92), where NLA (10a)
shows a phrase. Word vs constituent for the shifted =sh: OPEN (T44a).
QC-3 DOMAIN Neither -ish (by EN-2) nor a shifted =sh (by QC-2) lands on anything above C's
complement. da’ is there (Spec,CP): *(13) *Da’ísh. A focus-fronted phrase
is also above the counted domain: *(76a) ii’ni’ish "does not attach to the
moved element". Whether the fronted phrase is in Spec,CP or in a focus
position just below it: OPEN.
QC-4 NO POLAR FRONT A yes/no question cannot contain a focus-fronted phrase (Willie p. 92: "this
structure ... simply cannot be used to form a yes/no question": SOURCED for
the inverse case (76a)). So no other host is possible either. Stated, not
derived. Whether it also holds for fronting under yi-: OPEN (T44c).
QC-5 INVERSE FREEZE Under bi-, a question ending may only sit on a host in its unmoved position,
by its basic rule:
=sh on its own wh-word wherever INV-6 puts it: (14b), (57b), (59), (60a) OK
=sh shifted off the wh-word (QC-2): *(17)
=sh fronted with its wh-word: *(60b) reading 2
-ish on the unmoved first phrase (EN-2): (76b) OK
-ish on a fronted phrase: *(76a)
Key pair: (76b) and *(17) both put an ending on the clause-initial inverse
subject. So the ban is on SHIFTING or FRONTING, not on the landing site.
(Facts SOURCED; the single statement is ours. Willie's reason, "the Inverse is
already a marked construction type that places focus on the patient as
subject", and W&J's "topic/focus inversion that is incompatible with focus
movement", are SOURCED.)
QC-6 SEAL No change: neither ending is inside a verb word in any opened form (EN-1',
still an OPEN prediction, now for both endings).
Rule
Status
Rests on
QC-0
SOURCED
NLA content page; W&J p. 272; Willie p. 210
QC-1
HYPOTHESIS (statement), SOURCED (facts)
Willie (19), (20), p. 212; NLA content prose; EN-1, EN-4
QC-2
HYPOTHESIS; word vs constituent OPEN
Willie (16), pp. 92, 211; NLA yes/no (10a)
QC-3
HYPOTHESIS
EN-2; NLA *(13); Willie *(76a), p. 92
QC-4
SOURCED (inverse case); generality OPEN
Willie p. 92
QC-5
HYPOTHESIS (statement), SOURCED (facts and both authors' reasons)
Willie (14b), *(17), (76); W&J (57b), (59), (60)
QC-6
OPEN prediction
Silence of the sources
HYPOTHESISWhy not “no question ending on the inverse subject”? That would be the simplest reading of *(17), and of Willie's reason (the inverse subject already carries focus). It fails twice: (76b) puts -ísh on the inverse subject łííʼ, and (59) puts =sh on the wh-word that is the inverse subject (‘which one … was kissed by John’). Both are fine. So the landing-site version is contradicted by opened data, and QC-5 is stated over shifting and fronting instead.
OPENA cost QC-5 does not remove. Plain focus fronting under bi- is fine in a statement ((74a), INV-7). QC-5 bans it only in questions. That is a stipulation, as Willie's own wording (“simply cannot be used to form a yes/no question”) also is. Two ways it might follow are listed as T44d.
5. Derivations QD1–QD7
Steps are labelled by rule set: Σ role schema, A word (D1), B clause (D2). Each step names its operation: Merge (builds structure), Agree (values person/number), Move, Link (a nominal linked to a slot), Spell-out. VPISH: every participant slot is first-merged inside the verb's phrase. In the V-P variant the subject slot then raises inside the word to Spec,M/SP; V-H leaves it in place. The strings are the same either way. Verb words are not segmented beyond the source's gloss line. Entailment labels are ours, from the translations.
QD1. Willie (19) ʼashiiké díkwíí=sh yiztsʼǫs ‘She kissed how many boys?’: =sh at the edge of a two-word phrase
Σ Σ(√KISS) = {ρA (volition, sentience), ρP (sentience)}, as on W2. Both are 3rd person. Choose yi-: EXT = ρA.
AMerge ρP as complement of V and ρA in Spec,v (VPISH). Agree: 3.3. (V-P) Move ρA to Spec,M/SP. Spell-out W → yiztsʼǫs (“3.3.kissed”). The word is sealed.
BMergeʼashiiké ‘boys’ with díkwíí ‘how many’ → the wh-phrase. The question word comes last, so the phrase is head-final, as rule set B expects. Link: one adjunct, so it links to ρP (ADJ-OBJ): ‘how many boys’ is who was kissed.
BMerge C[Q]. Spell-out [Q, FOC] by QC-1 at the right edge of the wh-phrase → ʼashiiké díkwíí=sh. Order wh-phrase ⌢ W. Result (19) ✓. Moves at clause level: 0.
The ending follows the second word. “After the first word” would have given *ʼashiiké=sh díkwíí (schematic; not attested or reported). Because the wh-word is last, word edge and phrase edge coincide here.
QD2. Willie (20) hastiin łį́į́ʼ háídí=ígíí=sh yizloh ‘The man roped which horse?’: =sh outside the phrase's determiner
BMergełį́į́ʼ ‘horse’ with háídí ‘which’. Merge the determiner =ígíí, which closes the phrase: [[łį́į́ʼ háídí]=ígíí]. This is Willie's “complex adjunct”.
BMergehastiin ‘man’. Link by INV-6: the inner adjunct (the wh-phrase, next to W) ↔ ρP and the outer adjunct hastiin ↔ EXT = ρA.
BSpell-out [Q, FOC] by QC-1 at the right edge of the whole wh-phrase, outside =ígíí → háídí=ígíí=sh ✓. The clitic order, determiner inside and question outside, is the nesting order of a head-final phrase read from the inside out.
QC-2 is optional and not used here: =sh stays after the third word, not after the first (hastiin). That fits Willie's “can be moved”.
A (16) yi-, EXT = ρA; (17) bi-, EXT = ρP. Merge as in QD1. Agree 3.3. (V-P) Move the EXT slot to Spec,M/SP. Spell-out W → yiztał / biztał. The voice is now fixed in the word.
BMergeháí ‘who’ (inner) and díí ‘this one’ (outer). Link by INV-6: (16) háí ↔ ρP, díí ↔ ρA; (17) háí ↔ ρA, díí ↔ EXT = ρP. The variable (‘who’) is in its linked position in both.
BSpell-out [Q, FOC]. (16): QC-2. The first constituent of C's complement is díí (one word) → Díí=sh ✓. (17): QC-2 would give the same string, but QC-5 bans shifting under bi- ✗. The basic placement, Díí háísh biztał, is not given by Willie: it is a prediction of QC-5 and QC-1, schematic and unconfirmed (T44b).
The roles in the action are the same with or without the shift. Only where the ending is pronounced changes.
QD4. Willie (76b) łííʼ-ísh iiʼniʼ biisxį́ ‘Did the lightning kill the horse?’ and *(76a): the yes/no ending under bi-
Σ Σ(√KILL) = {ρA (causation), ρP (affected, change of state)}. Lightning (ρA) is lower than the horse on Willie's scale, so the inverse is used (p. 91, as recorded on I5). bi-: EXT = ρP.
AMerge ρP and ρA (VPISH). Agree 3.3. (V-P) Move EXT to Spec,M/SP. Spell-out W → biisxį́.
BMergeiiʼniʼ (inner) and łííʼ (outer). Link by INV-6: iiʼniʼ ↔ ρA, łííʼ ↔ EXT = ρP. Merge C[Q].
B (76b) Spell-out [Q] by EN-2. The first constituent of C's complement is łííʼ → łííʼ-ísh ✓. The ending is not shifted, so QC-5 is satisfied, even though it sits on the inverse subject.
B (76a) INV-7: Moveiiʼniʼ to the left edge (A-bar, leaving a copy), as in (74a). Its link to ρA is unchanged. Spell-out [Q]: not on iiʼniʼ, because it is above the counted domain (QC-3) and is a fronted host (QC-5) ✗. Not anywhere else either (QC-4, Willie: “cannot attach anywhere in this structure”) ✗.
Compare QD3 (17): there the ending also lands on the inverse subject, but by shifting, and that is out. Here it lands there by its basic rule, and that is fine. This pair is the evidence for stating QC-5 over shifting rather than over the landing site.
QD5. W&J (59) háí-sh Jáan biztsʼǫs and (60a) Jáan háí-sh biztsʼǫs: =sh on the unmoved wh-word under bi-
ΣA Σ(√KISS) as QD1, but with bi-: EXT = ρP. Spell-out W → biztsʼǫs (“3ACC-3NOM-kissed (Inverse)”).
B (59): Link by INV-6: inner Jáan ↔ ρA, outer háí ↔ EXT = ρP (“which one … was kissed by John”; the D-linked reading W&J give for a wh-word on a bi-pronoun). Spell-out [Q, FOC] by QC-1 on háí → háí-sh, first in the string but not shifted ✓.
B (60a): inner háí ↔ ρA, outer Jáan ↔ ρP (‘As for John, he was kissed by whom?’). QC-1 → háí-sh in second position ✓.
B (60b) reading 2 needs háí ↔ ρA, so it must be the inner adjunct, and in first position it would have to be fronted over Jáan (INV-7-style). Fronting a =sh-marked wh-word under bi- is banned by QC-5 ✗. Reading 1 is (59) ✓.
QD6. Willie (14b) háísh bichʼįʼ yáłtiʼ ‘He is being talked to by whom?’: the inverse on a postposition
Σ Σ(√SPEAK) = {ρA} plus the postposition -chʼįʼ ‘to’ with its own slot. The voice bi- is on the postposition (W6): EXT = the PP's object.
ASpell-outyáłtiʼ (“3.speaks”); PP cycle: pronoun ⌢ P → bichʼįʼ (“INV3.to”).
BMergeháí. Link to ρA (the speaker, as Willie's translation shows). Spell-out [Q, FOC] by QC-1 → háísh, unmoved ✓ (QC-5).
QD7. NLA content (3) Háágóó sh díníyá? and (5) Ńléí ʼasdzą́ą́ sání shąʼ háí ʼátʼį́?: the NLA's own placements
A (3) Spell-out W → díníyá (“2-go.I”). (5) → ʼátʼį́ (“3-be.N”).
B (3) Háágóó ‘where-to’ is an adverbial wh-word and binds no slot (WH-ADV). QC-1 → Háágóó sh. The NLA prints a space before sh and glosses it as one unit, “where-to-Q”. The spacing is not used.
B (5) Mergeńléí ʼasdzą́ą́ sání ‘that old woman (over there)’ as a topic. The topic particle -shąʼ goes at the right edge of the three-word phrase, the same edge rule (EN-4/QC-1). háí is the wh-word, with no =sh. Whether -shąʼ is related to =sh is not taken up here (T44f).
6. What changes
Item
Before
After
Why
T43f (yes/no page)
“first word (Willie) or first constituent (NLA)?”, plus the inverse question
Answered, mostly. On its own host =sh counts by phrase (QC-1). Willie's “first word” is used for both endings, so it is no evidence of a difference (QC-2). The shifted case stays OPEN (T44a). The inverse: -ísh is restricted too, but differently from =sh, and one statement covers both (QC-5)
“=sh may move to after the first word of the sentence, except in the inverse”
Kept as Willie's rule (SOURCED). Sharpened as QC-2 + QC-5. In the inverse it is the shift that is blocked, while =sh on an unmoved inverse-subject wh-word is fine ((59), (57b), (14b))
W&J (59) vs Willie *(17)
WH-SPLIT (asking about a role §11)
“=sh is a clause-edge element of rule set B”
More precise: =sh is spelled at the right edge of its phrase (QC-1). It reaches the clause edge only by the optional shift (QC-2)
Willie (19), (20)
Inverse page §5.3 / T5e
“under bi-, a fronted internal nominal cannot carry the question clitic” (OPEN)
Partly answered. The same freeze also covers *(17), which involves no fronting at all, and (76b) shows that the inverse subject itself can carry -ísh. The fronting-without-clitic half of T5e is still OPEN
QC-5
EN-2 / *(13)
da’ is never “first” because it is in Spec,CP
The same domain rule also excludes -ísh on a fronted phrase (QC-3): *(13) and *(76a) fall under one rule
Willie p. 92
EN-1′ SEAL
A prediction for -ísh
Now a prediction for both endings (QC-6); still no opened form either way
Silence
7. What each source gets right and misses
Source
Gets right
Misses or costs
Willie 1991 Ch. 8 and p. 92
The shift of =sh, its ban in the inverse, the complex wh-phrase with =sh at its end, and the ban on -ísh with focus fronting (“cannot attach anywhere”)
“First word” for both endings, where the NLA shows a phrase for -ísh. The reason given for *(17) (“the Inverse … places focus on the patient as subject”) would also exclude (76b) and (59) if read as a ban on the landing site. It works only as a ban on shifting
Willie & Jelinek 2000 pp. 272–273
=sh has “inherent focus”; the inverse wh-paradigm (57)–(60), including fine cases with =sh on the inverse subject (59)
“Both adjuncts have focus” uses “focus” more broadly than a clitic-marked focus, which makes any one-focus-per-clause rule hard to state from their wording. “Incompatible with focus movement of the Wh-word” is about wh-words; Willie (74a) allows non-wh fronting under bi- in statements
NLA content-question page
Particles go “after the constituent (phrase)”; =sh and -ísh are different; they “can appear in other places”
No examples of the “other places” in the prose, and no inverse examples. The examples behind the filter could not be retrieved
This site before tonight
WH-SPLIT separated the variable from the ending. INV-7's note linked *(76a) to W&J
T43f framed the contrast as word vs phrase, but the “word” wording is Willie's for both endings. The inverse ban was stated per ending
8. Grant's frame
GRANT'S CONJECTUREThe word as the changeover point (phenomenon 1). Both endings are placed by rule set B. They count clause-level units and attach at right edges, and in (20) the order =ígíí=sh is the head-final nesting read from the inside out. Neither ending is ever inside a verb word in the opened data. That is silence, so it counts for nothing, but it keeps the prediction (QC-6) that would break the conjecture if one counterexample appeared. The new point is the direction of information flow. The voice (yi-/bi-) is decided at Σ and spelled inside the verb word by rule set A. Rule set B reads it when placing the ending (QC-5), and never changes it. That matches the inverse page's “rule set B only reads the result”, now with two endings as evidence instead of one.
HYPOTHESISTheta roles and slots (phenomenon 2). In every derivation above, which slot the question word asks about is fixed by INV-6 after the voice choice, before the ending is placed. Shifting =sh ((16)) changes nobody's role in the action. In (59) vs (60a) the same string type asks about the patient or the agent depending only on linking. This fits the weak thesis (the question variable ranges over a role-bearing slot; WH-SLOT). It is not a test of the strong thesis.
GRANT'S CONJECTURESTRATA. The order in every derivation is roles (Σ) → Agree and spell-out of the word → linking of nominals → the question/focus layer. QC-5 is a case where the last layer is restricted by an earlier one (the voice), and never the reverse. That is the order STRATA expects. As before, it is a fit, not a test that separates STRATA from its rivals. Theta roles and event-structure roles are untouched here and stay compatible layers.
GRANT'S CONJECTUREWord-internal Merge or Agree sites. Not tested: no ending reaches inside the word. The Agree steps on this page (3.3 inside each verb word) and the Merge steps that build the wh-phrases are on opposite sides of the word edge. Whether the =sh shift is Move (with a copy) or only a choice of where to pronounce the ending is open (T44e). Willie calls it “moved”, but no opened datum shows a copy.
GRANT'S CONJECTUREPhrase-level lexicalization of verb + postposition units. Not tested this run. (14b) has a PP before the verb with =sh on the wh-word, not inside the PP + verb unit.
9. Tasks T44a–g and statuses
Task
What it needs
T44a
A shifted =sh with a first phrase of two or more words: does it follow the first word or the whole phrase (QC-2)?
T44b
The unshifted inverse version of (17), háí with =sh next to biztał: is it fine, as QC-1 + QC-5 predict (schematic)?
T44c
A yes/no question with focus fronting under yi- (where fronting is possible at all, WH-FOC): is QC-4 general or inverse-only?
T44d
Why does the freeze hold only in questions? Plain fronting under bi- is fine in statements ((74a)). Is the voice's prominence and a question's focus one position, or two?
T44e
Is the =sh shift Move (a copy; reconstruction effects) or a spell-out choice?
T44f
The NLA's “other places” for -sh/lá, and the relation of topic -shąʼ to =sh (the NLA filter's example list)
T44g
Can =sh and -ísh occur in the same clause, and if not, is that QC-0 (one [Q]) or a focus limit?
Statuses: T43f answered except for the shifted case, which moves to T44a. T5e is partly answered by QC-5. T43e (procliticized postpositions) is not attempted this run. T42d, T42e, T42g are still blocked on Young & Morgan. All other T43 items are unchanged.
What else changed on the site
Update notes on where the yes/no enclitic -ísh can go (lede and §8), asking about a role (§8 and the rules table) and the inverse (§5.3). There is a new row on open questions (T44a–g), a sources section (added 2026-10-20, 8:46pm), and a home-page entry. No verb-dictionary entries changed. Navajo forms new to the site (all with sources): NLA content (2), (3), (5). Willie (14a, b), (16), (17), (19), (20), (76a, b) and W&J (57)–(60) were already on the site and are re-read here on page images.
Asking about a role: wh-questions, focus, and the permutations Navajo forbids
2026-11-11 audit pass 2. Apostrophe fixes only. Navajo glottal stops typed as ' or ’ (for example in prefix names like ʼa- and ʼeʼe-, or words like Awééʼ) now print as ʼ (U+02BC), and a build step no longer turns the closing quote of an English translation inside a Navajo row into a glottal stop (‘…the girls.’, not ‘…the girls.ʼ). No form, gloss or translation wording changed.
Restated on the unified structure (2026-11-15). Questions are restated with Q in C⁰ above NegP: the wh word stays in its theta position as indefinite + Q (H11), and =sh/-ísh are interpreted in C⁰ and pronounced on a host (H12; H12-move vs H12-pf OPEN). Whether daʼ is a head is OPEN. Derivations below in which the verb or subject does not raise are labelled contrast; the working default raises V→v→T and the subject to Spec,TP. The “roles not words” theory, where it appears below, is dropped. Status on the spine: see unified structure, §4, question rows. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
This page reads Willie's (1991) Chapter 8, “Questions and Focus” (pp. 199–219), which the site had opened before only for its examples (28)–(29) on the possessor-raising page. The site's wh-question derivation (D7) was built from Willie & Jelinek 2000:272–273. Willie's claim is that question words in Navajo are adjuncts exactly like nouns. A word like háísh ‘who’ is “coindexed with (licensed by) a third person pronominal” in the verb (p. 201). It “binds that pronoun” and stands where a noun linked to that pronoun would stand (p. 203). The run gives four results. (a) A question word asks about a role slot, not about a word. It needs a 3rd-person slot to bind, it excludes a noun on the same slot, and which slot it asks about is set by the same order and voice rules as for nouns (§3–§5). This matches Grant's thesis. (b) Willie lists the orders Navajo forbids. A question word that asks about the patient can go first, ahead of the agent's noun, only when the two are far apart in animacy. In ‘What did John see?’ that works, because ‘what’ cannot be the one who sees. With two humans it is impossible: háísh John yiyiiłtsą́ means only ‘Who saw John?’ (p. 208). This is the shape Grant asked about: a lower argument may get ahead of a higher one only when the roles stay recoverable (§6). (c) That ban conflicts with Willie & Jelinek 2000:273, where the same kind of string with ‘kiss’ is given both readings. The site does not decide between them (§6.3, T19a). (d) A clause cannot have both an adverbial question word and an argument question word, *‘where who kissed Mary’ (27). This is the first sourced fact for Grant's multiple-wh test (§9).
The nominal adjunction page found that a Navajo noun is never an argument. It is an optional adjunct “licensed by a pronoun” in the verb (Willie 1991:52). The verb's arguments are role slots, Σ(√) = {ρA, ρP, …} (proto-role schema). This page asks three things of the chapter on questions.
Phenomenon 2 (roles, not words). When you ask “who?”, what do you ask about: a word or a slot? Is the answer fixed by the slot's entailments?
Argument ranking (Grant's original point). The natural order puts the higher argument's noun first. When can a lower argument get ahead of it, and which orders are impossible?
Phenomenon 1 and VPISH. Does a question word ever move, inside the clause or out of it? Does it show anything about where a slot first Merges?
2. Sources opened this run (2026-10-13, ~1:00pm PT)
Willie, MaryAnn 1991.Navajo Pronouns and Obviation. PhD dissertation, University of Arizona (the scan on the box). Fresh 110 dpi page images of printed pp. 200–218 (PDF 206–224), i.e. all of Chapter 8 except its first page (p. 199, whose prose was read from the text layer). Page offset: PDF = printed + 6, as in Chapters 3, 4 and 7 (from the page headers). The diacritics of (7a), (10a), the second “(9)”, the second “(10)”, “(4a)” on p. 208, (11a–b), (14a), (15), (16)–(19), (27a) and (32) were checked on 300 dpi crops. The rest were read at 110 dpi. The OCR layer was used only to find the section starts.
Willie & Jelinek 2000, pp. 272–273, (57)–(60): not re-opened. They are quoted from the site's own D7, where the page images were checked on 2026-10-06.
Transcription and glosses. Navajo is given as printed, with the glottal stop as ’ and the nasal hook as ą/į/ǫ. Willie's gloss lines are copied exactly, misprints included and marked [sic]. Her verb glosses give object.subject (“3.3.kiss”), and under the inverse “INV.3.3” (the bi- object is the subject). Where she leaves a name unglossed, the gloss cell is empty. Verbs are not segmented beyond her glosses. Numbering: Willie reuses “9)” on p. 206 and “10)” on p. 207, and on p. 208 she prints “(4a), repeated here” over a sentence that is not her (4a). This page calls these (9′), (10′) and (4a′).
Cited by Willie in these pages, not opened: Willie 1989 (Direct vs Inverse); Willie, to appear (“inner speech”); Schauber 1979; Speas 1990; Saxon 1986 (Dogrib).
3. Two kinds of question word (§8.0–8.1) SOURCED
P. 199: “I will argue that Wh-words in Navajo always appear as adjuncts in A-bar positions, and thus cannot undergo Wh- movement. … the positions of Wh- words at S-structure in Navajo is the same as that of other adjoined nominals. These adjunct positions follow from the valence of the verb and the Direct/Inverse voice alternation”. P. 200: “there are no argument assymetries [sic] with respect to Wh- words. I assume that Wh- movement takes place at LF in Navajo.” P. 201, (3): “a. Nominal Wh- adjuncts: those that must be coindexed with (licensed by) a third person pronominal, and are used to question an argument. b. Adverbial Wh- adjuncts: Wh- words that cannot be coindexed with a pronoun, and are used to question adverbial notions such as time, place, manner, etc.” “Nominal Wh- adjuncts can never be coindexed with a first, second, or fourth person pronoun, since coindexing depends upon matching phi-features.” P. 202: a nominal wh-adjunct “excludes a referential nominal adjunct to the argument it questions, and takes the place that such a nominal adjunct would occupy in the normal word order of the sentence.”
Willie 1991:200–202, (2), (4), (5)
(2a)
ʼashkii
naalnish
boy
3-working
‘The boy is working.’
(2b)
kweʼé
naalnish
here
3-working
‘He is working here.’
(4a)
háísh
naalnish
who
3-working
‘Who is working?’
(4b)
háadish
naalnish
where
3-working
‘Where is he working?’
(5a)
*háísh
shizhéʼé
naalnish
who
1s-father
3-working
(5b)
háadish
shizhéʼé
naalnish
where
1s-father
3-working
‘Where is my father working?’
GRANT'S CONJECTURE matched by SOURCED. A nominal question word has the three properties of something that asks about a slot. It needs a slot to bind (3a). It must agree with the slot's person value, so it can never ask about a 1st, 2nd or 4th person slot (p. 201). And a slot takes at most one adjunct, so ‘who’ and ‘my father’ cannot both go with the one slot of ‘work’ *(5a). ‘Where’ binds no slot, so it can sit next to ‘my father’ (5b). HYPOTHESIS (ours) WH-SLOT: a nominal question word is a variable over the referent of one slot of Σ. Which word fills the slot plays no part in its licensing. This is ADJ-NOTARG, ADJ-PHI and ADJ-ONE (nominal adjunction §11) applied to question words, and it follows from them without a new rule.
4. Which slot is asked about (§8.2) SOURCED
Pp. 202–203: “In Direct (yi-) constructions, an NP adjunct to the patient appears immediately before the verb, and by default an NP adjunct to the agent appears in the outermost position. … In the Inverse (bi-) construction, this coindexing is reversed, and the patient is the subject.” P. 203: “A Wh- word questioning a pronoun binds that pronoun, and appears in the same adjunct position that a coindexed NP would occupy. That is, the coindexing of the Wh- adjunct is determined by the Direct/Inverse voice alternation.” P. 205: “When only the Wh- word is present, it is interpreted as asking about the agent or the patient according to the same word order conventions.”
Willie 1991:203–205, (6), (7), (9)
(6a)
Sally
John
yiztsʼǫs
3.3.kiss
‘Sally kissed John.’
(6b)
Sally
John
biztsʼǫs
INV3.3.kiss
‘Sally was kissed by John.’
(7a)
háísh
John
yiztsʼǫs
who
3.3.kiss
‘Who kissed John?’
(7b)
John
háísh
yiztsʼǫs
who
3-3-kiss
‘John kissed who?’
(9a)
háísh
yiztsʼǫs
who
3.3.kiss
‘He kissed who?’
Willie gives “echo question” translations “to show that the Wh- words appear where a corresponding non-interrogative NP would be” (p. 204). So the gloss ‘John kissed who?’ for (7b) is an ordinary question, ‘Who did John kiss?’.
HYPOTHESIS (ours) The question word gets its slot by the linking rules for nouns, with no rule of its own: ADJ-OBJ for (9a) (a lone adjunct goes with the object), ADJ-TWO for (7) (inner with the patient, outer with the agent), and INV-6 (inverse page). That is what “no argument asymmetries” (p. 200) comes to in the site's terms. Nothing about questions refers to grammatical functions. Only the slot and the voice matter.
5. The inverse: “which one” (§8.2) SOURCED
Willie 1991:204–205, 208, (8), (9b), (11)
(8a)
háísh
John
biztsʼǫs
who
INV.3.3.kiss
‘Which one (of you) was kissed by John?’
(8b)
John
háísh
biztsʼǫs
who
INV.3.3.kiss
‘John was kissed by who?’
(9b)
háísh
biztsʼǫs
who
INV.3.3.kiss
‘He was kissed by who?’
(11a)
John
Mary
haʼátʼíísh
yeiníʼą́
what
3.3.3.gave
‘John gave Mary what?’
(11b)
John
Mary
haʼátʼíísh
beiníʼą́
what
INV3.3.3.gave
‘John was given what by Mary?’
Pp. 204–205: “The Inverse construction, as I argued Chapter 3 [sic], must have a definite subject. Therefore, when the Wh- word binds the definite pronoun, this is presuppositional, of the kind seen in Wh- phrases such as "which one". The reading of Ex. (8b) is that the subject must be one of a presupposed set, to which the addressee belongs.” P. 208: “In ditransitives, the Inverse switches the agent and goal theta roles assigned to pronouns in the verb-sentence.”
SOURCED A question word that binds the inverse subject inherits INV-4 (definiteness) and is read as ‘which one’. This agrees with W&J 2000 (59)/(60b) as quoted in D7. (Willie's text says “(8b)”, but the ‘which one’ translation is on (8a), where háísh is the subject; T19f.)
SOURCED (11): only agent and goal swap under the inverse. The theme ‘what’ keeps its place next to the verb in both voices. That is INV-5 (“a theme in a three-slot frame never does” compete) with a question word in the theme slot.
HYPOTHESIS (ours) In the site's terms, INV-4 is a condition on a slot (EXT = ρX must be definite). A question word over that slot is forced to be presuppositional, so the question is shaped by the slot's requirement and not by the word.
6. Forbidden orders: when can a lower argument go first? (§8.2) SOURCED
Willie 1991:206–208, (10), (9′), (10′), (4a′)
(10a)
John
haʼátʼíísh
yiyiiłtsą́
what
3.3.saw
‘John saw what?’
(10b)
*haʼátʼíísh
John
biiłtsą́
what
INV.3.3.saw
[‘What was seen by John?’]
(9′)
haʼátʼíísh
John
yiviiłtsą́ [sic]
what
3.3.saw
‘What did John see?’ (printed “9)”, p. 206; yiviiłtsą́ as printed, cf. yiyiiłtsą́ in (10a))
(10′)
ʼashkii
yiyííbį́į́ʼ
boy
3.3.picked [inanimate round objects]
a. ‘The boy picked them [i.e, berries]’ b. *‘They [berries] picked the boy.’ (printed “10)”, p. 207; = Ch. 3 (22))
(4a′)
háísh
John
yiyiiłtsą́
‘Who saw John?’ / *‘John saw who?’ (printed “(4a), repeated here”, p. 208, no gloss line)
P. 206: “Ex. (10b) shows a context where the Inverse cannot be used to place a Wh- adjunct in clause initial position. The Navajo animacy heirarchy [sic] requires that nominals referring to entities higher on the animacy scale precede nominals referring to lower entities.” Pp. 206–207, on (9′): “This sentence can only have the reading shown, where it is the inanimate object that is being questioned, even though the word order violates the animacy hirarchy [sic] and the normal word order for Direct constructions. This is a kind of focus movement of the Wh- word, and is permitted when the semantic features of the lexical items make it impossible to mistake the agent and patient -- because they are so far apart on the animacy scale. Comparable focus movement in non-interrogative sentences is possible when the semantic features exclude unwanted interpretations.” Pp. 207–208: “Focus movement of the Wh- adjunct is not possible when the agent and patient are of equal rank in the hierarchy, and the semantic features of the lexical items does [sic] not prevent misinterpretations.” Then (4a′), “not ambiguous”.
6.1 The shape of the edge
GRANT'S CONJECTURE Grant's original point (2026-10-04) is that the natural order puts the higher, proto-agent argument first. The interesting edge is when a lower argument gets ahead of it, and some permutations are impossible. Willie's §8.2 gives that edge for questions, with four cases, all SOURCED:
Case
Patient (lower)
Order
Voice
Result
(10a)
‘what’
agent ≺ patient (natural)
yi-
✓
*(10b)
‘what’
patient ≺ agent
bi-
✗: the inverse would make the inanimate the subject, against the ranking (INV-3)
(9′)
‘what’
patient ≺ agent
yi-
✓: focus movement; only one reading is possible
(4a′)
‘John’ (patient), ‘who’ (agent)
wh ≺ John
yi-
only ‘Who saw John?’; the focus-moved reading ‘Who did John see?’ is out
So the passive-like route (the inverse) is closed when the lower argument is inanimate, and the movement route is open only when the roles cannot be confused.
6.2 The license is the role schema
HYPOTHESIS (ours) FOCUS-RECOVER: optional focus fronting of an adjunct (a question word or a noun) over another is allowed only if, after fronting, the adjunct–slot links are still fixed by ENT-LINK (nominal adjunction §9), that is, by the slots' entailments and sortal class, without word order. In (9′) the seeing slot of ‘see’ requires a sentient participant (an entailment label of ours, from the translation). ‘What’ is not sentient, so it can only go with the seen slot, and order is free to change. In (4a′) both ‘who’ and ‘John’ fit both slots, so order must keep its job and fronting is blocked. (10′) is the same override with a noun, and Willie says so herself (p. 207). On this reading Willie's “semantic features of the lexical items” are, for the verb, entailments of the slots. The impossible permutation is the one the role schema cannot disambiguate. This is a hypothesis about her statement, not her wording. She speaks of “the animacy scale”, and for ‘pick (round objects)’ the deciding property is the object's shape, a property of the slot, not of rank.
6.3 A conflict with Willie & Jelinek 2000 OPEN
The site's D7 gives W&J 2000:273 (58b) háí-sh Jáan yiztsʼǫs two readings: “1) ‘Who was it, that kissed John?’ Or: 2) ‘Who was it, that John kissed?’”, the second by “optional focus movement of the Wh-word + Q to the leftmost position”. Willie 1991 (4a′) háísh John yiyiiłtsą́ is the same configuration: two humans, direct voice, ‘who’ first. She gives it one reading only, and says that focus movement “is not possible when the agent and patient are of equal rank” (pp. 207–208). Her (7a) háísh John yiztsʼǫs, with the same verb as W&J's (58b), is translated only ‘Who kissed John?’, though she does not star the other reading there. The verbs differ (‘see’ vs ‘kiss’), but nothing in either text makes the verb the reason. Possible explanations, none checked: a change of analysis between 1991 and 2000, speakers who differ (Willie reports “usage known to me”, p. 218, in another connection), or an effect of the verb. The site keeps both as SOURCED, for different sources, and records the clash as T19a. D7's reading 2 should be read with this note.
7. Postpositions, and an answer to T18d (§8.3)
Willie 1991:209–210, (12)–(14)
(12a)
John
Mary
yichʼįʼ
yáłtiʼ
3.to
3.speaks
‘John is talking to Mary.’
(12b)
John
Mary
bichʼįʼ
yáłtiʼ
who [sic]
INV.3.to
3.speaks
‘John is being talked to by Mary.’
(13a)
John
háísh
yichʼįʼ
yáłtiʼ
who
3.to
3.speaks
‘John is talking to who?’
(13b)
John
háísh
bichʼįʼ
yáłtiʼ
who
INV3.to
3.speaks
‘John is being talked to by who?’
(14a)
háísh
yichʼįʼ
yáłtiʼ
who
3.to
3.speaks
‘He is talking to who?’
(14b)
háísh
bichʼįʼ
yáłtiʼ
who
INV3.to
3.speaks
‘He is being talked to by whom?’
P. 208: “The object of a grammatical postposition is questioned without Wh- movement. The Inverse voice alternation has scope over the verb word and any associated postpositonal [sic] phrases -- oblique arguments.” Pp. 209–210: “Note that in (13b), the Wh- word intervenes between the Inverse postpositional phrase and the nominal ("John") that is associated with it, just as in (12b), where an ordinary nominal adjunct coindexed with the postpositional object appears.”
SOURCED Under bi- on the postposition, the noun associated with the postposition's object is the outer one (‘John’), and the noun next to the postposition is the speaker. T18d, partly answered. The nominal adjunction page proposed MIRROR-LINK for the direct voice: the k-th noun out from the MVS goes with the k-th exponent in from its left edge, so the noun next to the postposition goes with the postposition's object. (13b) shows that this does not hold in the inverse. There the noun next to bichʼįʼ goes with ρA, whose exponent is inside the verb word, and the outer noun goes with the postposition's object. HYPOTHESIS MIRROR-LINK is therefore a direct-voice default, and under bi- the INV-6 positions take over: the inner position is the internal-position argument (here ρA) and the outer is EXT. The nesting is by position after voice, not by where the exponents are spelled out. That fits the schema's ordering Σ → VOICE → linking (D7 result).
8. The focus clitic =sh (§8.4–8.5) SOURCED
Willie 1991:210–212, (15)–(20)
(15)
John
háí-shį́į́
yiztał
s.o.-PROB
3.3.kicked
‘John (probably) kicked someone.’
(16)
Díí=sh
háí
yiztał
DEM=Q
who
3.3.kicked
‘This person kicked who?’
(17)
*Díí=sh
háí
biztał
DEM=Q
who
INV.3.3.kicked
[‘This person was kicked by who?’]
(18)
ʼashkii
háí-dí-lá
yiztsʼǫs
boy
which one
3.3.kissed
‘She kissed which boy?’
(19)
ʼashiiké
díkwíí=sh
yiztsʼǫs
boys
how many=FOCUS
3.3.kissed
‘She kissed how many boys?’
(20)
hastiin
łį́į́ʼ
háídí=ígíí=sh
yizloh
man
horse
which=DET=Q
3.3.roped
‘The man roped which horse?’ ((20) read at 110 dpi only)
P. 210: “The Wh-words háísh and haʼátʼíísh both have the interrogative clitic =sh attached; this clitic overtly marks focus. … A Wh-word without the =sh clitic, and with an attached modal clitic, can be interpreted as an indefinite in a non-interrogative sentence.” P. 211: “The =sh clitic can be moved away from the Wh- word to a position following the first word in the sentence. … But in Inverse constructions, it is impossible to attach the question enclitic to the first word in the sentence. … I assume this is excluded since the Inverse is already a marked construction type that places focus on the patient as subject.” P. 212: “we see a Wh- word combined with an ordinary nominal to build a complex adjunct; extraction from these phrases is not possible.”
SOURCED The question force is in the clitic, not in the stem: háí without =sh plus a modal clitic is ‘someone’ (15). The clitic can sit on the first word of the sentence instead (16), but not in the inverse *(17).
SOURCED *(17) agrees with INV-7 on the inverse page (from Willie Ch. 3 (76a)): under bi-, a fronted element cannot carry -ísh. Willie's reason is the same in both places: the inverse already puts focus on the patient.
HYPOTHESIS (ours) The question word itself (the variable over a slot) and the focus clitic (a clause-level marker that goes after the first word) are two pieces. The first belongs to the slot and is licensed by ADJ-PHI. The second is a clause-edge element of rule set B, like the yes/no -ísh. In (16) the variable stays in its linked position (just before the verb, the patient) while =sh goes to the left edge.
Update 2026-10-20, 8:46pm (T43f). Pp. 210–212 were re-read on page images, with (16), (17), (19), (20) at 300 dpi. (20) is confirmed as transcribed above. Two refinements are on the two question endings compared, and the rules here are kept as written. (1) On its own host, =sh closes the whole wh-phrase, after the determiner in (20) (QC-1). It reaches the clause edge only by the optional shift (QC-2). (2) In the inverse it is the shift that is blocked, not =sh on the first word as such: W&J (59) háí-sh Jáan biztsʼǫs has =sh on an unmoved wh-word that is the inverse subject, and it is fine (QC-5).
9. Two question words in one clause (§8.6, p. 215) SOURCED
Willie 1991:214–215, (26), (27)
(26a)
hádáaʼshąʼ
John
Mary
yiztsʼǫs?
when
3.3.kissed?
‘When did John kiss Mary?’
(26b)
háísh
baghan=di
Mary
yiztsʼǫs?
who
3POSS:home=at
3.3.kissed?
‘At whose home did he kiss Mary?’
(27a)
*háa=di
háísh
Mary
yiztsʼǫs
where=at
who
3.3.kissed
(27b)
*háa=di
Mary
háísh
yiztsʼǫs
where=at
who
3.3.kissed
P. 215: “The expression baghan-di [sic, =di in (26b)] in (26b) is an adverbial adjunct, a post-positional phrase in which the noun is directly governed by the postposition. There is a small class of Lexical Postpositions in Navajo that have this syntax. … It is these Grammatical Postpositions that fall under the scope of the Inverse voice alternation”. “It is impossible to have both an adverbial and an argumental wh- word in the same clause.”
SOURCED An adverbial question word (‘where’) and an argument question word (‘who’) cannot share a clause, in either order *(27a, b). Willie gives no example, starred or not, with two argument question words, and gives no reason for *(27).
SOURCED (26b): ‘who’ can bind the possessor b- inside a lexical postpositional phrase. So a nominal question word can be licensed by a pronoun that is not one of the verb's slots. Willie calls this whole phrase an adverbial adjunct.
HYPOTHESIS (ours) The ban is not about the focus clitic. In *(27) only háísh carries =sh (háa=di has none), so “one =sh per clause” does not explain it. Nor is it about slots, since ‘where’ binds no slot. A one-question-per-clause limit would cover it, but that is a restatement, not an explanation (T19c).
For Grant's test sentence “who bet who what at what casino” (open questions §2): “at what casino” is a place question on a lexical (locative) postposition, the class of (26b)/(27). HYPOTHESIS In Willie's usage, the full sentence would be out as one clause by *(27), whatever happens with the three argument questions. Whether two or three argument question words can share a clause is still unknown (T19b). No Navajo version of the sentence is given here.
10. No long-distance questions (§8.6) SOURCED
Willie 1991:213–218, (21), (22), (25), (31), (32)
(21)
háísh
łį́į́ʼ
yidooloh
nínízin
who
horse
3.3.roped
2.think
‘Who will rope the horse, you think?’ (‘Who do you think will rope the horse?’)
(22)
John
łį́į́ʼ
yidooloh
nínízin
horse
3.3.roped
2.think
‘John will rope the horse, you think.’
(25)
John
háísh
yiztsʼǫs
nínízin
who
3.3.kissed
2.think
‘John kissed who, you think?’ (‘Who do you think John kissed?’)
(31)
hádááʼshąʼ
dah diiyá
ní
when-Q
3.left
3.said
‘When did he leave, she said.’
(32)
hádááʼshąʼ
ʼání,
dashdiiyá=hágíí
ʼáhóótʼiid
when-Q
3.3.said,
4.left=COMP
3.happened
‘When did she say it, his leaving took place.’ ‘When did she say that he left?’
P. 213: “propositional clauses that appear with these verbs, when they are without complementizers, are not complement clauses, but loosely adjoined as "inner speech" … This agreement is like direct quotes in English.” P. 214: “the subordinate clause must precede the main clause, and the Wh- word remains in the clause it has scope over.” P. 218: “Wh-words, whether subcategorized for or not, must appear in the clause they have scope over at LF. Previous work by Schauber (1979) and Speas (1990) suggests that there may be some variation across speakers in this respect; I report on usage known to me. … Since Wh- words must be adjuncts to the clause over which they have scope, and there are no complement clauses in Navajo, there is no long-distance Wh- movement. … In this respect, Navajo again differs from Northern Athabaskan. Saxon (1986) has shown long distance Wh-movement of "adverbial" Wh- words in Dogrib.”
SOURCED A question word stays in the clause whose verb has the slot it binds ((21), (25)). With ‘think’ the clause is an adjoined “inner speech” clause, like the reported-speech page's quotative frames. In (32), where ‘when’ asks about the saying, the left clause is a separate clause, ‘she said it’, and the leaving is in a =hágíí clause with a 4th-person subject.
HYPOTHESIS (ours) This follows from WH-SLOT plus Σ closing at B3 (WORD-EDGE′). A question word must bind a slot, and slots belong to one verb's Σ. A slot of the lower verb cannot be bound from the higher clause because the higher clause has no adjunct position linked to it. That gives “no long-distance questions” without a movement constraint.
SOURCED (p. 218) Northern Athabaskan differs: Dogrib has long-distance movement of adverbial question words (Saxon 1986, cited, not opened). Willie says this fits “the different status of nominals in the Northern vs. the Southern branches”. The site keeps this as a lead only (T19h).
11. Rules
Notation as on the nominal adjunction page. Rule names are ours, and the page references are Willie 1991.
WH-ADJ Question words are adjuncts in A-bar positions at S-structure, in the
same positions as adjoined nominals; no S-structure wh-movement.
(pp. 199–200, 219)
WH-NOM A nominal question word must be coindexed with a 3rd-person pronoun
and questions an argument; never 1st, 2nd or 4th. (p. 201, (3a))
WH-ADV An adverbial question word cannot be coindexed with a pronoun.
(p. 201, (3b))
WH-EXCL A nominal question word excludes a referential nominal on the same
pronoun and takes its position. (p. 202, *(5a))
WH-POS The slot a question word asks about is fixed by the nominal linking
rules and the yi-/bi- voice, also when it is alone.
(pp. 203, 205; (7)–(9), (13), (14))
WH-INV-D A question word binding the inverse subject is presuppositional
("which one"). (pp. 204–205; (8))
WH-DITR In ditransitives the inverse swaps agent and goal; a theme question
word stays next to the verb. (p. 208; (11))
WH-FOC Optional focus movement of a question word over the agent's nominal
is possible only when "semantic features" exclude the other reading;
not with equal-rank agent and patient. (pp. 206–208; (9′), (10′), (4a′))
WH-INV-AN The inverse cannot make an inanimate question word the subject.
(p. 206; *(10b))
SH-FOC =sh marks focus; a wh stem without =sh + modal clitic = indefinite.
(p. 210; (15))
SH-MOVE =sh may move to after the first word of the sentence, except in the
inverse. (p. 211; (16), *(17))
WH-PHR Wh-word + noun form one complex adjunct; no extraction from it.
(p. 212; (18)–(20))
WH-ADV-ARG No adverbial and argumental question word in the same clause.
(p. 215; *(27))
WH-LOCAL A question word stays in the clause it scopes over; no long-distance
movement, argument or adverbial. (pp. 212–214, 217–218; (21)–(32))
Rule
Status
Basis
WH-ADJ … WH-LOCAL above
SOURCED
Willie 1991 Ch. 8 as cited. WH-FOC conflicts with W&J 2000 (58b) (§6.3, T19a).
WH-SLOT. A nominal question word is a variable over the referent of one slot of Σ; its licensing ignores lexical content.
FOCUS-RECOVER. Optional fronting of an adjunct over another is allowed only if ENT-LINK alone still fixes every adjunct–slot link.
HYPOTHESIS
WH-FOC, (10′), Ch. 3 ADJ-SEM. §6.2.
MIRROR-LINK′. MIRROR-LINK holds in the direct voice only. Under bi- the inner/outer positions follow INV-6 after voice, whatever the exponent's place in the MVS.
HYPOTHESIS
(12b), (13b), p. 209. Partly answers T18d. §7.
WH-SPLIT. The question variable belongs to the slot; =sh is a clause-edge element of rule set B.
HYPOTHESIS
SH-FOC, SH-MOVE. §8.
LOCAL-Σ. WH-LOCAL follows from WH-SLOT plus Σ closing at B3.
Conventions as on the derivations page: Σ schema step, A word-internal linearization (prefixes left of the stem), B clause level (complements and adjuncts left of their head). V-P is the VPISH variant, in which the subject slot raises to Spec,M/SP inside the word. V-H keeps it in situ (Hale). Entailment labels are ours, from Willie's translations. Verbs are not segmented beyond Willie's glosses. “x” is the question variable, LF is Willie's level for wh-scope.
W1. (4a) háísh naalnish, *(5a), (5b)
Σ Σ(√WORK) = {ρA} (vol; ours).
Merge ρA in Spec,v. Agree: π = 3 (“3-working”). (V-P) Move to Spec,M/SP; (V-H) none. Same string.
A Spell-out naalnish. B2 = B3; Σ closes.
B Adjoin háísh at the left (WH-ADJ). WH-NOM: it needs a 3rd-person pronoun; ρA is 3, so it binds ρA. LF: x = the referent of ρA; ‘Who is working?’.
*(5a): shizhéʼé would also need ρA. Two adjuncts on one slot violate ADJ-ONE / WH-EXCL. ✗
(5b): háadish is adverbial (WH-ADV) and binds no slot. shizhéʼé takes ρA. ✓
W2. Direct: (7a) háísh John yiztsʼǫs, (7b) John háísh yiztsʼǫs, (9a) háísh yiztsʼǫs
Σ Σ(√KISS) = {ρA (vol, sent), ρP (sent)}; both π = 3, so VOICE is defined (INV-0). r(ρA) = r(ρP) (both human), so voice is free (INV-3). Choose yi-: EXT = ρA.
Merge ρP in Comp,V, ρA in Spec,v. Agree 3.3. (V-P) Move ρA to Spec,M/SP.
A Spell-out yiztsʼǫs. Σ closes.
B (7a): two adjuncts. ADJ-TWO / INV-6: inner (John) ↔ ρP, outer (háísh) ↔ EXT = ρA. ‘Who kissed John?’
(7b): inner (háísh) ↔ ρP, outer (John) ↔ ρA. ‘Who did John kiss?’ (Willie's echo gloss ‘John kissed who?’).
(9a): one adjunct. ADJ-OBJ: háísh ↔ ρP. ‘Whom did he kiss?’ (‘He kissed who?’).
W3. Inverse: (8a) háísh John biztsʼǫs, (9b)
Σ As W2, but choose bi-: EXT = ρP, and ρA is in the internal position (INV-2). INV-4: the referent of ρP is definite.
A Spell-out biztsʼǫs.
B INV-6: inner (John) ↔ the internal position = ρA; outer (háísh) ↔ EXT = ρP.
LF: x ranges over referents of ρP, which INV-4 requires to be definite, so x ranges over a presupposed set: ‘Which one (of you) was kissed by John?’ (WH-INV-D).
(9b): one adjunct. Under bi- a lone adjunct links to the internal position (INV-6), here ρA: ‘He was kissed by who?’
W4. Forbidden and allowed orders: (10a), *(10b), (9′), (4a′)
Σ Σ(√SEE) = {ρA (sent; ours), ρP}; π = 3.3.
(10a): the question is about ρP, ‘what’, which is inanimate. INV-3: r(ρA) (John) > r(ρP), so EXT = ρA, yi-. B inner haʼátʼíísh ↔ ρP, outer John ↔ ρA. ✓
*(10b): bi- would make EXT = ρP = ‘what’, lower than John. INV-3 ✗ (WH-INV-AN). The inverse route to putting ‘what’ first is closed.
(9′): yi- as in (10a). Base B order John haʼátʼíísh V. A′-Move (focus): haʼátʼíísh to the left edge, ⟨haʼátʼíísh⟩ left behind. FOCUS-RECOVER check: can ‘John’ link to ρP and ‘what’ to ρA? ρA requires a sentient participant (ours) and ‘what’ is not one, so ENT-LINK fixes the links alone. Fronting is allowed and there is one reading: ‘What did John see?’ ✓
(4a′) háísh John yiyiiłtsą́: both ‘who’ and ‘John’ are sentient and fit both slots. FOCUS-RECOVER fails for a fronted háísh, so the string has only its base parse: outer háísh ↔ ρA, inner John ↔ ρP. ‘Who saw John?’ only (Willie 1991). Under W&J 2000 (58b) the fronted parse would also exist (T19a).
W5. Ditransitive: (11a) John Mary haʼátʼíísh yeiníʼą́, (11b) … beiníʼą́
Σ Σ(√GIVE) = {ρA, ρG, ρTh} (“3.3.3.gave”). INV-5: only ρA and ρG compete for EXT; ρTh never does.
B Innermost haʼátʼíísh ↔ ρTh in both voices (WH-DITR). Then Mary ↔ the internal competitor and John ↔ EXT: (11a) John = giver, Mary = goal; (11b) John = goal (subject), Mary = giver. This matches Willie's translations.
W6. Postposition under the inverse: (13b) John háísh bichʼįʼ yáłtiʼ (T18d)
Σ Σ(√SPEAK) = {ρA} + PP[-chʼįʼ ‘to’], with the PP object as the internal competitor (the inverse “has scope over … postpositional phrases”, p. 208).
VOICE bi- on the postposition: EXT = the PP object, ρA internal. INV-4: the addressee is definite.
A Spell-out bichʼįʼ yáłtiʼ. The MVS closes at B3; the leftmost exponent in the MVS is b- (PP object).
B MIRROR-LINK would link the innermost adjunct (háísh) to b-. Willie (p. 209): John is the noun associated with the inverse PP, so háísh ↔ ρA. MIRROR-LINK ✗; INV-6 after voice ✓ (MIRROR-LINK′). ‘John is being talked to by who?’
W7. =sh on the first word: (16) Díí=sh háí yiztał, *(17)
SH-MOVE: =sh is spelled out after the first word, Díí=sh. The variable stays in place (WH-SPLIT). ‘This person kicked who?’ ✓
*(17): bi-. SH-MOVE is blocked in the inverse (= INV-7's ban on -ísh on a fronted element). ✗
W8. No long-distance question: (25) John háísh yiztsʼǫs nínízin
Σ Clause 1: Σ(√KISS) as W2, yi-. Clause 2: Σ(√THINK) with ρ = 2 (“2.think”).
B In clause 1: inner háísh ↔ ρP, outer John ↔ ρA. Clause 1 is adjoined to clause 2 as “inner speech” and precedes it (p. 214).
LF: háísh scopes over clause 1 only (WH-LOCAL). The English ‘Who do you think John kissed?’ is Willie's paraphrase, but the structure is ‘John kissed who, you think?’. There is no slot of √THINK for háísh to bind, so it cannot be adjoined in clause 2 (LOCAL-Σ).
13. VPISH
SOURCED Question words, like nouns, are never in an argument position (p. 199), so they are never in Spec,vP or Spec,TP. Willie puts wh-scope at LF (p. 200).
HYPOTHESIS (ours) No example here separates V-P from V-H. In both, the slot is valued inside the word and the question word binds it from the outside. The order facts in §4–§7 are about the B-level positions after VOICE, which both variants share.
HYPOTHESIS (ours) One point bears on the movement question Grant raised: the only movement in this chapter is optional A′ focus movement of an adjunct, and it is licensed by the role schema (FOCUS-RECOVER), not by a feature of a functional head. If FOCUS-RECOVER is right, Navajo's “extra” movement depends on the entailments of the slots, which is the level Grant's thesis puts the arguments at.
14. Open
#
Question
What would settle it
T19a
Willie 1991 (4a′) gives one reading for háísh John yiyiiłtsą́ and bans focus movement with equal ranks (pp. 207–208). W&J 2000 (58b) gives two readings for háí-sh Jáan yiztsʼǫs. A change of analysis, speakers who differ, or the verb?
Speaker judgments on both strings; W&J 2000 §11 re-read for a comment on 1991. OPEN
T19b
Can two argument question words share a clause (‘who kissed whom’)? Willie gives no example.
A source with a multiple argument-wh example, or elicitation (multiple-wh test). OPEN
T19c
Why is *(27) out? Not one =sh per clause (háa=di has none) and not one adjunct per slot.
Starred/unstarred pairs with two adverbial question words (‘when … where’). OPEN
T19d
Is FOCUS-RECOVER about slot entailments, or about animacy rank as Willie says? ‘pick (round objects)’ (10′) turns on shape, not rank.
A pair with an animacy gap but no entailment clash (both slots allow inanimates), or the reverse. OPEN
T19e
Does (26b) ‘who’ on a possessor inside a lexical PP count as an argument question for *(27)? Can it combine with ‘who’ on a verb slot?
Speaker judgments. OPEN
T19f
Fidelity: (12b)'s gloss line has “who” with no question word; Willie's text says the ‘which one’ reading is (8b) while the translation is on (8a); (9′)/(10′)/(4a′) are numbered as repeats.
Another copy of the dissertation, or Willie 1989. OPEN
T19g
A nominal question word can never bind a 4th-person slot (p. 201). How is a question asked about a 4th-person participant? Is it always rephrased with a 3rd?
A source on 4th-person questions; cf. fourth person. OPEN
T19h
Dogrib (Saxon 1986) has long-distance adverbial wh. Does Dogrib also have complement clauses and nouns in argument positions, as Willie's North/South contrast implies?
Saxon 1986 (not opened). Do not merge with Navajo. OPEN
Fidelity notes: yiviiłtsą́ (9′, p. 206) vs yiyiiłtsą́ (10a, p. 206; 4a′, p. 208), checked on 300 dpi crops. hádáaʼshąʼ (26a) vs hádááʼshąʼ (31), (32). “baghan-di” (p. 215 text) and “háísh bighan-di” (p. 216 text) vs baghan=di in (26b), (28a). =sh glossed “=Q” (16), (20) and “=FOCUS” (19). “3.3.kiss” vs “3-3-kiss” (7b). (14b) ‘whom’ vs ‘who’ elsewhere. As printed: “assymetries”, “heirarchy”, “hirarchy”, “overriden”, “argued Chapter 3”, “postpositonal”, “the semantic features … does not”. (20) was read at 110 dpi only.
‘With me it went’: is the -ił “transportation” pattern a productive pattern or an idiom, and who is the subject when you travel? (T45f)
2026-11-11 audit pass 2. Apostrophe fixes only. Navajo glottal stops typed as ' or ’ (for example in prefix names like ʼa- and ʼeʼe-, or words like Awééʼ) now print as ʼ (U+02BC), and a build step no longer turns the closing quote of an English translation inside a Navajo row into a glottal stop (‘…the girls.’, not ‘…the girls.ʼ). No form, gloss or translation wording changed.
Restated on the unified structure (2026-11-15). Postpositions are restated as [PP DP P⁰] with left specifiers: a grammatical P agrees with the DP in its specifier, a lexical P has no probe (H64); a selected PP is merged in VP, and Willie’s “cannot be interrupted” follows from the default, HS-1 and left specifiers. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Position-class template statements below describe the order of prefixes; they are not the derivation (§3). Flat-S and other pre-VPISH notation below is labelled contrast. Status on the spine: see unified structure, §2b and §4, postposition row. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
The previous page sorted Navajo verb + postposition pairs into three kinds and put one of them, -ił ‘with’ + a motion verb meaning ‘travel by’, in the middle: a regular pattern, not a one-off idiom (attached or separate postpositions §6, PC-7b). That rested on three sentences. T45f asks whether the pattern really is productive (many verbs, many vehicles, many speakers) or a small listed set, and it re-tests the claim that “the vehicle is the subject”. This run opened the Navajo Language Academy's complete list of sentences that its Postposition Lexicon tags “transportation”, plus random example lists and the lexicon's “instrumental” list, and re-read Hale 1997 p. 2 and Hale 2003 pp. 14–18 on preverbs. The live NLA site did not answer this run, so every NLA page was read in the Internet Archive's copies (details in §1). The work bears on the routine's second phenomenon (which participant a verb takes as its subject, and why: the role schema) and, through word order, on the first (where the verb word starts: the changeover from rule set A to rule set B). Two operations are kept apart throughout. Merge builds structure: it puts an element at a position. Agree values features such as person and number: a probe copies a value from the closest matching goal. Written 2026-10-21, 12:46pm PT.
Update 2026-10-21, 8:46pm (T47). What is used up bit by bit on a trip is followed up on distance as measure. SOURCED Dowty 1991 (pp. 568–569) puts the incremental part of drive (a car) from New York to Chicago on the path and calls the mover a holistic theme; Hay, Kennedy & Levin 1999 call the measure along the path the incremental theme. HYPOTHESIS So the rules here (TP-0 to TP-7: who is the subject) and Grant's claim (the distance, not the vehicle, is the incremental theme) are independent: whether the vehicle is subject (TP-1) or object/unnamed (TP-2), it is a second holistic theme on the path it shares with the rider. No sourced sentence combines this pattern with a distance phrase (T48a).
Ground rules (Grant's). Every Navajo form below comes from a source opened this run, or is already on this site with its source. Forms are quoted as displayed, with the source's own gloss line, and nothing is segmented beyond what the source prints. What the sources say is graded SOURCED. Rules and mappings are ours (HYPOTHESIS). Grant's own ideas are GRANT'S CONJECTURE. Only contradicting data count against an analysis; a source's silence does not. No analysis, the site's own included, is assumed to be fully right. Grant does not speak Navajo, and nothing here is a native-speaker judgment. “Who is in control” and “who moves” are read from the NLA's English translations; that reading is ours.
Notation (the site's own; the Syntax · Semantics · Interface bot's files carry no conventions for this site: VERB-QUEUE.md's only convention is the lemma + sense-subscript display for its own dictionary, unchanged in its repository since Oct 3. The file dates on the box now read Oct 21, about 7:45–7:54am PT, as do several hundred other files touched at the same time). Σ = role-schema step (which slots the verb has, and each one's role in the action); A = word level (rule set A: the prefixal, left-headed spine, spelled into one word W, preverbs included); B = clause level (rule set B: right-headed clause syntax, noun linking, particles and enclitics). ρMv = the mover the stem describes; ρRd = the one carried along (rider, passenger, the person travelling); ρC = a causer; ρIns = a means. PC-0 to PC-7: attached or separate postpositions §4. CYC: A/B as spell-out §3. “Role in the action” = Dowty-style entailments of the verb (movement, causation, volition, sentience), as everywhere on this site.
Bottom line, in plain words
SOURCEDThe NLA's “transportation” list is wider than three sentences, and it is built the same way throughout. All twelve tagged sentences have -ił ‘with’ marking the person who travels (shił ‘1-with’, nił, bił, nihił), and the verb's own subject is a third person. The verbs vary: glossed ‘drive’/‘run’/‘return’, ‘fly’, ‘float’, ‘crawl’, ‘move.SFO’ (a slender flexible object), and ‘drive’/‘push.wheel’. The movers vary: an unnamed vehicle, an airplane, a boat, a horse, and, in a story, lightning (ʼatsiniltłʼish bił dah yizdéél ‘Monster Slayer was carried away eastward by lightning’). Six speakers are named (counting “Sharon Nelson” and “Sharon J. Nelson” as one, and Wallace Dale from the 4:46am live load).
HYPOTHESISAnswer to T45f: a pattern, not an idiom. Six or more verb stems, five kinds of carrier and many speakers share one form and one meaning, and the meaning follows from ‘with’ plus the verb: ‘it (the carrier) moved, with me’ = ‘I travelled by it’. Nothing new has to be stored for each pair. One piece is not in the parts: when the translation says the rider drove (‘so I just drove that way’), the control comes from somewhere else. The NLA lists that stem's verb as ʼoolwoł ‘drive’ and glosses the same stem ‘run’ elsewhere, so the ‘drive’ sense may be listed for that one verb (a word-level listing), not for the pattern. Whether speakers extend the pattern to new verbs cannot be shown from a dictionary (T46c).
HYPOTHESIS“The vehicle is the subject” needs restating. The subject is not always the vehicle. It is whoever the verb stem itself describes as moving, or causing the movement. With ‘fly’, ‘float’, ‘crawl’, ‘run’ the subject is the thing that moves, and the person travels in the ‘with’ phrase, even when that person is the driver: sh ił dah ʼadiilwod ‘1-with start.off 3-drive.P’ = ‘I just drove that way’. With the stem glossed ‘drive’ in ʼadooł bąs and ʼeesbąs, the subject is the human who drives (Háí lá kintah góó sh ił ʼadooł bąs? ‘Who can drive me to town?’), and the ‘with’ phrase holds the passengers. With -ee ‘by means of’ the traveller is the subject and the vehicle is the means (Łééchąąłgaii b ee nááshdááł ‘by Greyhound bus I was returning’). So one trip can be said three ways, and in each the subject is picked by the stem's own role in the action: causing outranks moving, moving outranks riding.
HYPOTHESISA side result for where the word starts (T45d). In the opened sentences the only separately written words that ever stand between a -ił phrase (or another postpositional phrase) and its verb are dah ‘up/start off’ and yah ‘into’. Hale treats such elements as preverbs: the leftmost part of the verb word (Hale 2003 pp. 14–17) and placed after the postpositional proclitics (Hale 1997 p. 2). So they are the left edge of the verb word W, not something from the clause that has got in. The rule that nothing from the clause intervenes (PC-2) survives, stated more exactly.
GRANT'S CONJECTUREFor Grant's frame. This is the clearest case yet for the role-schema view: the same trip, with the same people and the same vehicle, gets a different subject depending only on which stem is chosen, and the subject slot can be filled by nothing at all (in ‘I just drove that way’ the vehicle is never named; the verb's subject is a bare third-person slot). The verb's arguments behave like role-bearing slots chosen by the stem's entailments, not like particular words.
Fernald et al., Navajo Postposition Lexicon, version 10.2023 (Navajo Language Academy), talkingdictionary.swarthmore.edu/navajo_postposition/
Live site unreachable this run (web fetch: 504 on the main page, ?intro and ?semantic_ids=transportation; time-outs on single entries; curl from the box: TLS failure on https, connect time-out on http; about 12:50–1:05pm PT). Opened in the Internet Archive's Wayback Machine copies (web fetch): ?semantic_ids=transportation (the 2025 and 2026 copies, identical); ?semantic_ids=instrumental (2025 copy); single entries ?entry=16 (captured 2025-03-31) and ?entry=1986; the main page in three copies (2024, 2025, 2026), each with its own random 50-sentence list. Exact capture times of the other copies were not retrieved (the archive's lookup service refused with “too many requests”). These are the NLA's own pages as archived, not the live site; a sentence may have been changed since capture
The full “transportation” list S1–S13 (§2.1); the -ił headword's five semantic tags; -ee examples E1, E2; ‘drive’-stem examples D1–D4; preverb-order examples R1–R5; speakers where displayed
Fernald et al., Navajo Verb Modes: grammar and examples, version 3.2022 (NLA)
Wayback copy (2025) of navajo_verb_mode/?semantic_ids=transportation
The same 17-entry “transportation” list (cross-check; identical sentences and glosses); verb links “ʼoolwoł drive”, “na‘oolwoł return”
Hale, Kenneth 1997. “Remarks on the syntax of the Navajo verb” (box PDF hale023.pdf)
Re-read, text layer of PDF p. 2
The order of the “more loosely attached ‘proclitic’ elements”: “adverbial, thematic, and postpositional proclitics. These occur in the reverse order to that in which they were just listed”
Hale, Kenneth 2003 ms. “The significance of the Pronominal Argument Hypothesis” (box PDF halejel.pdf)
Re-read, text layer of printed pp. 14–18 (§2 “Preverbs, Quantifiers, and the Verb Stem”)
Template (10), position I = Preverb; “I claim that preverbs are adjuncts”; productive preverbs adjoin to QualP or M/SP, or lower to vP; Hargus & Tuttle (p.c.) on “truly lexical” preverb-verb idioms; “The adjuncts … correspond to the disjunct sector of the Navajo verb word”
Young & Morgan 1987 (lending-only; the unofficial dump on the box is not used). None of the transportation sentences shows a YM citation in the archived list. The 4:46am live loads are not re-opened; where a speaker was recorded then and is not displayed in the archived list, that is said
Display. The archived pages show some words split by spaces where the database links a piece to a lexicon entry (sh ił, naa t’a’ í n ił, Kintah dę́ę́ʼ). The same sentences on the live site at 4:46am were displayed without those spaces (T2 chidí naatʼaʼí nił nast’a’, T3 Kintahdę́ę́ʼ łį́į́ʼ shił nááldlosh). Forms below are quoted as displayed in the copy that was opened; the spaces are the database's display, not word divisions we claim.
2. The facts
2.1 The complete “transportation” list (NLA Postposition Lexicon)
The list page reads: “Your search for * in transportation has returned 17 entries”. Thirteen are displayed in both copies: the headword -ił ‘with’ (tags: accompaniment, cognitive experiencer, goal, transportation, instrumental) and twelve sentences, one of them twice (S8/S9). What the other four are is not shown (T46h).
NLA Postposition Lexicon, ?semantic_ids=transportation (Wayback copies, 2025 and 2026); speakers as displayed
S1 (Sammie Largo)
ʼÁádóó níléíjigo daaníigo tʼóó ʼáajigo sh ił dah ʼadiilwod.
and.then over.there-toward pl-3-say.I-GO just that.way-toward 1-with start.off 3-drive.P
‘And then they said to go that way, so I just drove that way.’
‘My father drove (a wagon) up out of the canyon with me walking alongside.’
D3
Chahaʼoh bi yaa gi niʼdííł bąs.
shade.house 3-under-at 2-drive.F
‘Park (the car) under the ramada.’
D4
Nichid ísh há adi nííníłbą́ą́z?
2-car-Q where-at 3-2-park.P
‘As for your car, where did you park it?’
D1 is not among the thirteen displayed transportation entries; whether it is one of the four not displayed is unknown. In D4 the car is the verb's object (“3-2-park”); in D2 the wagon is not named at all.
‘Coming back from Phoenix by Greyhound bus I got left behind in the restaurant in Flagstaff.’
E2
Tsinaabąąs bee nidaʼa bąąs go shiʼdizh chį́.
wagon 3-with pl-3-drive.I-GO 1-pass-born.P
‘I was born when wagons were being used.’
The NLA glosses -ee here as “3-with” too, but files it under -ee ‘by means of’, tag “instrumental”. In E1 the bus is the postposition's object and the traveller is the verb's subject (“1-return”). In E2 the wagon is the means and the subject is plural people (“pl-3-drive”). E1's second half also has a ‘run’ verb, shi yaa haalwod ‘1-under 3-run.P’, whose unnamed subject is, by the translation, the bus that left the speaker behind.
2.4 -ił in its other uses
NLA Postposition Lexicon, main page (Wayback copy, 2026)
A1 (Tyler Tinhorn)
Jáan Mary bił baghan.
John Mary 3-with 3-live
‘Mary lives with John.’ (tag: accompaniment)
Already on the site with -ił in other uses: ‘know’ shił bééhózin (cognitive experiencer; K1–K4). A1 has the same shape as the transportation sentences (Y + -ił, then a verb whose subject is someone else), with a verb of living instead of moving.
2.5 What stands between a postpositional phrase and its verb
Doo hanii ni ʼałdóʼ nilį́į́ʼ bi kʼi dah ʼíníił da.
NEG contrary 2 also 2-horse 3-on up saddle 3-2-put.PlO.I NEG
‘Why don't you saddle your horse too?’
R4
ʼÍíshją́ą́ díí ʼatooʼ béésh biiʼ kǫʼí bi kááʼ dah séką́.
don’t forget this stew stove 3-top up 1-put.P.OC
‘Don’t forget that I put the stew on the stove.’
The NLA links dah to its Adverb and Particle Lexicons (“dah up at an elevation, off, static, holding”) and yah to the Particle Lexicon (“yah into”). Hale 1997 PDF p. 2: “There exists in addition a variety of more loosely attached ‘proclitic’ elements preceding the object position, including the distributive plural marker da-, the iterative ná-, and an array … of adverbial, thematic, and postpositional proclitics. These occur in the reverse order to that in which they were just listed”. So, left to right: postpositional, thematic, adverbial, ná-, da-. Hale 2003 p. 14, template (10): position I = “Preverb”; p. 16: “I claim that preverbs are adjuncts … productive preverbs are variable, in keeping with their generally adverbial nature”, adjoined to the projection of Qualifier or M/S; p. 17: “The adjuncts — i.e., the preverbs and the quantifier elements … — correspond to the disjunct sector of the Navajo verb word”; pp. 17–18: sometimes lower, adjoined to vP. Whether dah and yah are preverbs in Hale's sense is our identification (from the NLA's glosses and Hale's ‘adverbial proclitic’), not a statement of either source.
3. Who is the subject? Sentence by sentence
Sentence
Verb (gloss as printed)
Verb's subject (person in gloss)
What moves / what causes it (from the translation)
P object
Who controls (translation)
S1
start.off 3-drive.P
3, no noun
an unnamed vehicle
-ił: 1 (speaker)
the speaker (‘I just drove’)
S2
up 3-move.SFO.P
3: ʼatsiniltłʼish ‘lightning’
lightning carries
-ił: 3 (Monster Slayer)
the lightning
S3
3-return.P
3, no noun
unnamed vehicle
-ił: 1
the speaker (‘I drove back’)
S4
3-run.Prog; 3-drive.P
3, no noun (twice)
unnamed vehicle
-ił: 1 (twice)
the speaker (‘riding around’, ‘I drove’)
S5
up 3-fly.P
3: chidí naatʼaʼí ‘airplane’
the airplane
-ił: 1dpl
not said
S6
3-float.P
3: tsinaaʼeeł ‘boat’
the boat
-ił: 2
not said
S7
3-fly.P
3: chidí naatʼaʼí
the airplane
-ił: 2
not said
S8
3-drive.F
3: háí ‘who’
the driver causes (wheeled) motion
-ił: 1 (gloss “1-me”)
the subject, ‘who’
S10
3-crawl.CI
3: łį́į́ʼ ‘horse’
the horse
-ił: 1
not said (the speaker rides)
S11
3-drive.I
3, no noun
unnamed vehicle
-ił: 3 (s/he); -ká: 1
‘s/he’ (‘S/he is driving’)
S12
1-drive.F
1 (as glossed)
?
-ił: 1
the speaker
S13
in 3-push.wheel.P
3, no noun
? (a person pushing, or the wheeled thing)
-ił: 3 (my father)
not said (English passive)
D1
1-drive.Prog
1 (speaker)
the driver
-ił: 3 (my children)
the speaker
D2
up-3-drive.P
3: shizhéʼé ‘my father’
the driver
none
the father
E1
1-return.Prog
1 (speaker)
the traveller
-ee: 3 (the bus)
not said
E2
pl-3-drive.I
pl-3, no noun
people using wagons
-ee: 3 (wagon)
the people
SOURCED The glosses, translations and the subject's person. HYPOTHESIS The “what moves / what causes it” and “who controls” columns are our reading of the English translations. Two counts follow. (i) In the eleven sentences whose verb is glossed ‘fly’, ‘float’, ‘crawl’, ‘move’, ‘run’, ‘return’ or ‘drive.I/P’ (S1–S7, S10, S11), the subject is always third person and the traveller is always the -ił object, including the four in which the traveller is the one who drives (S1, S3, S4, S11). (ii) With the -bąs/-bą́ą́z stem (S8, D1, D2, D3) the subject is the human driver; the vehicle is unnamed or, in D4, the object. S12 and S13 do not fit cleanly (TP-7).
4. Pattern or idiom?
Test
Kind (b) pattern predicts
Kind (c) idiom predicts
Found
How many verb stems?
many
one, or a few fixed ones
At least six: the stem glossed ‘run/drive/return’ (S1, S3, S4, S11), ‘fly’ (S5, S7), ‘float’ (S6), ‘crawl’ (S10), ‘move.SFO’ (S2), ‘drive/push.wheel’ (S8, S13)
How many kinds of carrier?
any thing that moves and can carry
fixed
Unnamed vehicle, airplane, boat, horse, lightning (a mythic carrier, with a classificatory stem for slender flexible objects)
How many speakers?
many
—
Sammie Largo (S1), John Harvey (S2; S10 on the 4:46am load), Sharon J. Nelson (S3; “Sharon Nelson” on S6 and S11, taken as the same person), Bernadine M. Cody and Devon Denny (S6), Wallace Dale (S7, 4:46am load): six
Is the meaning predictable from the parts?
yes
no
Yes for ‘travel by’: ‘X moved, with Y’. A1 Jáan Mary bił baghan shows the same ‘with’ (accompaniment) on a non-motion verb, and Hale 1997 (22a) Tsinaaʼeeł (shí) sitsʼáníʼéél shows ‘float’ with the boat as subject and another postposition. Not predictable: that the rider is the one who drives (S1, S3, S4, S11)
Is anything listed?
at most a sense of one verb
the pair
The NLA's verb list has “ʼoolwoł drive” (S11) and “na‘oolwoł return” (S3); the same stem is glossed ‘run’ in S4's first clause and in E1's haalwod. So a ‘drive’ sense is listed for that verb, not for the -ił pair
HYPOTHESISVerdict. Kind (b) holds, and in fact sits closer to kind (a) (built from the parts) than PC-7 said. The ‘travel by’ meaning needs no stored pair: it is ‘with’ + a verb whose subject is a carrier. What is stored, if anything, is word-level: a ‘drive’ sense of the verb ʼoolwoł, under which the rider in the ‘with’ phrase is understood to be in control. OPEN Productivity in the strict sense (speakers extending the pattern to new vehicles or new motion verbs) cannot be shown from a dictionary's tagged list (T46c). The breadth found (six stems, a mythic carrier) is what productivity predicts, but a long-established listed set would look the same.
5. Rules TP-0 to TP-7 (ours; all HYPOTHESIS unless marked)
These refine PC-2 and PC-7 without rewriting them.
TP-0 BUILT 'Y travels by X' = [PP Y -ił] + W, where W's stem describes X as moving and
X is W's subject. -ił keeps its 'with' value (accompaniment, A1). No
pair-specific entry. (S1-S7, S10, S11)
TP-1 MOVER With a stem whose role in the action for its subject is MOVEMENT (fly,
float, crawl, move.SFO, run/return), the subject is the carrier. The
traveller is the -ił object, whether or not the traveller controls the trip
(S1, S3, S4, S11 vs S5-S7, S10).
TP-2 CAUSER With the stem glossed 'drive' (-bąs/-bą́ą́z), whose subject has CAUSATION of
wheeled motion, the subject is the driver (S8 háí, D1 1st person, D2 my
father, D3 2nd person). The vehicle is unnamed, the object (D4 'park' 3-2-),
or an -ee means (E2). The -ił object is the one carried (S8 me, D1 my
children).
TP-3 MEANS With -ee 'by means of', the vehicle is the P object and the traveller is the
verb's subject (E1 1-return). A third way to say the same trip.
TP-4 ONE RANKING TP-1 to TP-3 are one rule: the verb's subject is the participant with the
most proto-agent entailments UNDER THE STEM'S OWN DESCRIPTION, with
causation > movement > being carried. The postposition (-ił, -ee) takes a
participant the stem does not rank first. (Dowty-style; ours)
TP-5 LISTED SENSE The 'drive' reading of 'run'-type W + -ił (rider in control) is not in the
parts. It is either a listed sense of the verb ('oolwoł 'drive', NLA) or
an inference. Word level (rule set A's lexicon), not a stored phrase.
TP-6 PREVERB EDGE The only separately written words seen between a PP and its verb are the
preverbs dah 'up/start off' and yah 'into' (S1, S2, S5, S13, R1-R4).
They are W's left edge (Hale 2003 pp. 14-17: position I, adjunct of QualP
or M/SP or vP, "the disjunct sector of the Navajo verb word"), ordered after
the postpositional proclitics (Hale 1997 PDF p. 2). PC-2 restated: nothing
from rule set B ever stands between a PP and W; rule-set-A material at W's
left edge can be written as a separate word.
TP-7 MISFITS S12 "1-with 1-drive.F": rider = subject, against TP-1, if the gloss is
right. S13 "3-with in 3-push.wheel.P": whether the subject is a pusher
(TP-2) or the wheeled thing (TP-1) is not decidable from the translation.
Both OPEN (T46a, T46b).
Rule
Status
Rests on
TP-0
HYPOTHESIS on SOURCED forms
S1–S11; A1; Hale 1997 (22a)
TP-1
HYPOTHESIS (subject's person and translations SOURCED)
S1–S7, S10, S11
TP-2
HYPOTHESIS
S8, D1–D4, E2
TP-3
HYPOTHESIS on SOURCED forms
E1 (E2)
TP-4
HYPOTHESIS; the proto-role link is GRANT'S CONJECTURE
S1, S2, S5, S13, R1–R4; Hale 1997 PDF p. 2; Hale 2003 pp. 14–17
TP-7
OPEN
S12, S13
HYPOTHESISWhy not “the subject is always the vehicle” (the 4:46am wording)? S8 and D1–D3 have human drivers as subjects with the same -ił + motion shape (S8 is on the transportation list itself). The 4:46am wording was right for the three sentences it had, which all have movement stems.
HYPOTHESISWhy not “the subject is whoever is in charge”? S1, S3, S4 and S11 have the person in charge (the driver) in the -ił phrase, not in the subject. Control does not pick the subject; the stem's description does.
6. Derivations TD1–TD7
Steps are labelled by rule set: Σ role schema, A word (W, and the PP sub-cycle), B clause. Each step names its operation: Merge (builds structure), Agree (values person/number), Move, Link (a nominal linked to a slot), Seal (spell-out of a unit). VPISH: every participant slot is first-merged inside the verb's phrase; in the V-P variant the subject slot raises inside the word to Spec,M/SP, in V-H it stays. The strings are the same either way. The spine is Hale 2003's Qual > M/S > v > V; no flat-S tree is used. We do not segment any verb beyond the NLA's gloss.
TD1. S5 Chidí naatʼaʼí nihił dah diit’a’ ‘The plane took off with us’: mover subject, preverb at W's edge
Σ Σ(√FLY) = {ρMv (movement; the thing that flies)}. The postposition -ił adds ρRd (changes location together with ρMv; no causation). TP-4: ρMv outranks ρRd → ρMv is the subject slot.
A PP sub-cycle: Merge-ił with ρRd; Agree P ⟶ ρRd [1, dual/plural] → nih-; Seal → nihił.
AMerge V (√FLY), v, M/S; Agree M/S ⟶ ρMv [3]. Merge the preverb dah as an adjunct (Hale 2003 p. 16: to the projection of the highest functional head, or lower to vP). Head movement assembles W. Seal W = dah diit’a’ (“up 3-fly.P”); dah is written as a separate word (TP-6).
A/B CYC C1: PP ⌢ W → nihił dah diit’a’. The order PP before dah is Hale 1997 p. 2's order (postpositional before adverbial).
BMergechidí naatʼaʼí ‘airplane’; Link ↔ ρMv ✓.
TD2. S1 … tʼóó ʼáajigo sh ił dah ʼadiilwod ‘so I just drove that way’: the driver in the ‘with’ phrase, no noun for the subject
Σ Σ(the stem glossed ‘drive’/‘run’) = {ρMv (movement)} + -ił: ρRd. TP-1: ρMv is the subject. ρRd = the speaker.
A PP: Merge-ił with ρRd; Agree [1sg] → sh-; Seal → shił (displayed sh ił).
A W: Agree M/S ⟶ ρMv [3]; Merge preverb dah (“start.off”); Seal → dah ’adiilwod (“start.off 3-drive.P”).
B No noun is merged for ρMv: the vehicle is never named. The clause is complete with a bare third-person slot. Mergetʼóó ʼáajigo ‘just that way’ and the preceding =go clause (‘they said to go over there’) above.
Interpretation. The translation ‘I just drove’ puts the speaker in control. The syntax does not: the speaker is ρRd. Control comes from TP-5 (a ‘drive’ sense of the verb, or inference). Merge built the slots; Agree gave the subject slot [3] and the P slot [1]; neither assigned control.
TD3. S2 … ʼatsiniltłʼish bił dah yizdéél jiní ‘Monster Slayer was carried away eastward by lightning’: a carrier that is not a vehicle
Σ Σ(the stem glossed ‘move.SFO’: a slender flexible object moves) = {ρMv} + -ił: ρRd. The classificatory stem describes the mover's shape; lightning fits it. TP-1: ρMv = subject.
A PP: Merge-ił with ρRd; Agree [3] → b-; Seal → bił. W: Agree M/S ⟶ ρMv [3]; Merge preverb dah ‘up’; Seal → dah yizdéél.
BMergeNaayééʼ Neizghání ‘Monster Slayer’ and ʼatsiniltłʼish ‘lightning’; Link ‘lightning’ ↔ ρMv, ‘Monster Slayer’ ↔ ρRd. Both nouns are third person, so Agree alone does not tell them apart; the links follow the translation and the stem's shape class. Which linking rule (focus, topic, order) the clause-spine page would use for two third-person nouns here is not tested (T46e). jiní ‘4-say’ (‘it is said’) is a separate clause.
Note. The English is passive (‘was carried away by lightning’); the Navajo has the carrier as subject, as in every TP-1 sentence. No passive morphology is glossed.
TD4. S8 Háí lá kintah góó sh ił ʼadooł bąs? ‘Who can drive me to town?’: a causer stem
Σ Σ(the stem glossed ‘drive’, -bąs) = {ρC (causation of wheeled motion, volition)}, with the vehicle implicit (D4 shows it can be the object). -ił adds ρRd. TP-2/TP-4: ρC outranks ρRd → ρC is the subject.
A PP: Merge-ił with ρRd; Agree [1sg] → sh-; Seal. W: Agree M/S ⟶ ρC [3]; Seal → ʼadooł bąs (“3-drive.F”).
BMergeháí ‘who’; Link ↔ ρC. Mergelá (the NLA's “particle for content questions”) after it; Mergekintah góó ‘to town’ (lexical P -góó). Result S8 ✓. Same trip shape as S1, different stem, different subject.
TD5. D1 Shaʼáłchíní Tooh góó bił ʼeesbąs ‘I am driving my children to Shiprock’
Σ As TD4: {ρC} + -ił: ρRd (the children, carried).
A PP: Merge-ił with ρRd; Agree [3] → b-; Seal → bił. W: Agree M/S ⟶ ρC [1sg]; Seal → ʼeesbąs (“1-drive.Prog”).
BMergeshaʼáłchíní ‘my children’; Link ↔ ρRd (the only third-person slot). MergeTooh góó ‘to Shiprock’. Result D1 ✓.
Contrast with S11 (TD7): there the driver is in the bił phrase; here the passengers are. The postposition holds “the one carried”; who that is depends on what the stem made the subject.
TD6. E1 Łééchąąłgaii b ee nááshdááł ‘returning by Greyhound bus’: the vehicle as means
Σ Σ(√RETURN, glossed “1-return.Prog”) = {ρMv (the traveller moves)} + -ee ‘by means of’: ρIns (the bus). TP-3/TP-4: the stem describes the traveller's movement, so the traveller is the subject; the bus is a means.
A PP: Merge-ee with ρIns; Agree [3] → b-; Seal → bee (displayed b ee). W: Agree M/S ⟶ ρMv [1sg]; Seal → nááshdááł.
BMergeŁééchąąłgaii ‘Greyhound bus’; Link ↔ ρIns. Hoozdo dę́ę́ʼ ‘from Phoenix’ above. The clause ends in -go and is chained to the rest of E1.
Note. In the same sentence, shi yaa haalwod ‘1-under 3-run.P’ (our reading: ‘[it] ran out from under me’ = I got left behind; the NLA has the same verb in bi yaa haalwodgo ‘3-under 3-go.P-GO’, translated ‘it ran out from under him’, main page, 2026 copy) has the unnamed bus as the ‘run’ subject, as TP-1 predicts for a ‘run’ stem.
TD7. S11 Naʼnízhoozhí dę́ę́ʼ shí ká bił ʼool woł ‘S/he is driving from Gallup to get me’: two postpositions, driver in the ‘with’ phrase
Confirmed as a pattern over at least six stems and five carriers, and restated: the subject is the participant the stem ranks first (TP-4), the vehicle only with movement stems. Closer to “built” than first said; the only listed piece is a word-level ‘drive’ sense (TP-5)
§3, §4
PD7 observation (same page)
“Subject choice follows the verb's own role in the action (movement), not the overall event's instigator”
Kept, and widened: with a causer stem (‘drive’, -bąs) the instigator is the subject, because causation is the stem's own role in the action. What never picks the subject is control that the stem does not describe (S1, S3, S4, S11)
TP-1, TP-2
PC-2 (no intruder)
“Nothing that is merged is ever pronounced between a grammatical PP and W”
Restated: nothing from rule set B. Preverbs dah, yah are written apart between PP and the rest of W, and they belong to W (TP-6). T45d is answered for preverbs; it stays open for nouns, negation and adverbs (none found)
S1, S2, S5, S13, R1–R4; Hale 1997 p. 2; Hale 2003 pp. 14–17
T45f
open
Answered as far as a dictionary allows (pattern; breadth shown). Strict productivity → T46c
§4
8. What each source gets right and misses
Source
Gets right
Misses or costs
NLA Postposition Lexicon
A semantic tag that picks out the pattern across many stems and speakers; separate tags for ‘with’ (-ił) and ‘by means of’ (-ee), which is exactly the TP-1 vs TP-3 split; verb links (“ʼoolwoł drive”)
“17 entries” but 13 displayed; S8 listed twice; S12 gloss “two yesterday … 1-drive.F” for ‘in two days, I will drive’ (one word or the person may be off); “1-me” in S8; -ee glossed “3-with”. Speaker not displayed for half the list. Unreachable live this run
Hale 1997, Hale 2003
The order of the disjunct elements (postpositional before adverbial: p. 2); preverbs as adjuncts in the verb word, with variable height (2003 pp. 16–18)
Neither discusses dah or yah by name in the pages read; the identification is ours. GB-era machinery restated in the spine
This site before today
The transportation pattern as kind (b); subject by the verb's role in the action (PD7)
“The vehicle is the subject” was too narrow (TP-2); PC-2 did not mention preverbs
9. Grant's frame
GRANT'S CONJECTURETheta roles as a proto-role schema (phenomenon 2). This is the strongest opened case on the site that the verb's subject is chosen by the stem's role entailments and not by the words in the sentence. The same trip (the same traveller, the same vehicle) gets three different subjects: the carrier with a movement stem (TP-1), the driver with a causing stem (TP-2), the traveller with a returning stem and -ee (TP-3). The ranking causation > movement > being carried is Dowty's proto-agent list applied to the stem's own description. And the subject slot can be a bare third-person slot with no noun ever named (S1, S3, S4, S11), which is what the weak thesis says verbs take: role-bearing slots, with nouns as optional partners (theta-args thesis). It does not test the strong thesis (nouns never arguments): the nouns that do appear link to these slots as usual.
HYPOTHESISTheta roles and event-structure roles together. At the level of the whole trip, both traveller and carrier change location (one event-structure description: a motion event with a co-mover). The theta roles say which of them the stem describes as the mover or causer. The two descriptions do not conflict; they answer different questions, and the subject follows the theta-role one.
GRANT'S CONJECTUREThe word as the changeover point (phenomenon 1). TP-6 sharpens the edge. Rule set A's word W can begin with a preverb that is written as a separate word (dah, yah), and the postpositional phrase stands just before that. So the written word and the grammatical word differ at both ends: a postposition can be written onto W while being outside it (PC-3), and a preverb can be written apart while being inside it (TP-6). The changeover from A to B is at W's grammatical left edge, wherever the spaces fall. This also predicts that -ísh after dah would split W, as Hale's ka-[á]sh does (T43c).
GRANT'S CONJECTURESTRATA (theta roles first, event structure next, noun selection later). The derivations run Σ (the stem ranks its participants) → Agree and sealing → linking of nouns, and in TD2 the last step is empty: no noun at all for the subject. That fits roles before words. It is a fit, not a separating test: a lexicalist account with the same ranking would derive the same strings.
GRANT'S CONJECTUREPhrase-level lexicalization. The transportation pattern turns out not to need a phrase-level entry. Its one listed piece is a word-level verb sense (‘drive’). This narrows L9 (“phrasisized”) to kind (c) pairs, where it still fits (§6 there).
10. Tasks T46a–h and statuses
Task
What it needs
T46a
S12 shił ʼadees wod “1-with 1-drive.F”: is the subject really first person (rider = subject, against TP-1), or is the gloss off? Another printing of the form, or a speaker
T46b
S13 bił yah ʼiʼííbą́ą́z “3-push.wheel.P”: is the subject a person pushing (TP-2) or the wheeled thing (TP-1)?
T46c
Productivity proper: a new vehicle (e.g. a bicycle, dziʼizí, which the NLA has in another sentence) or a new motion stem in the -ił pattern; a dictionary cannot show coinage
T46d
A movement stem with the traveller as subject and -ił holding the vehicle (would break TP-1/TP-4)
T46e
Two third-person nouns with the pattern (S2): which linking rule picks ‘lightning’ for the subject (focus/topic/order on the clause spine)?
T46f
Is dah/yah a preverb in Hale's sense? Can -ísh follow it (B1 vs B2, T43c)? Is a noun, negation or adverb ever printed between a PP and dah (T45d remainder)?
T46g
How Young & Morgan list ʼoolwoł ‘drive’ vs the ‘run’ stem (TP-5; blocked: YM lending-only)
T46h
The four “transportation” entries the list counts but does not display (re-open the live site when it answers)
Statuses: T45f: answered as a pattern (TP-0–TP-5); strict productivity → T46c. T45d: answered for preverbs (TP-6), open for clause-level words (T46f). T45a, T45b, T45c, T45e, T45g, T45h: unchanged. T43h: not attempted this run. T42d/e/g: still blocked on Young & Morgan.
What else changed on the site
Update notes on attached or separate postpositions (lede), postposition units (lede) and the theta-args thesis (after the 10-15 note). There is a new row on open questions (T46a–h), a sources section (added 2026-10-21, 12:46pm), and a home-page entry. No verb-dictionary entries changed. Navajo forms new to the site (all with sources): NLA S1–S5, S8, S11–S13, D1–D4, E1, E2, A1, R1–R4. S6, S7, S10 were already on the site as T1–T3 and are re-read here in the archived list.
‘We ran about two miles’: is the distance the incremental theme? Distance phrases in Navajo motion sentences, the vehicle, and the ‘but didn't get there’ test (T47)
2026-11-11 audit pass 2. Apostrophe fixes only. Navajo glottal stops typed as ' or ’ (for example in prefix names like ʼa- and ʼeʼe-, or words like Awééʼ) now print as ʼ (U+02BC), and a build step no longer turns the closing quote of an English translation inside a Navajo row into a glottal stop (‘…the girls.’, not ‘…the girls.ʼ). No form, gloss or translation wording changed.
Restated on the unified structure (2026-11-15). Postpositions are restated as [PP DP P⁰] with left specifiers: a grammatical P agrees with the DP in its specifier, a lexical P has no probe (H64); a selected PP is merged in VP, and Willie’s “cannot be interrupted” follows from the default, HS-1 and left specifiers. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Flat-S and other pre-VPISH notation below is labelled contrast. Status on the spine: see unified structure, §2b and §4, postposition row. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
Grant's claim today: in a sentence like ‘I ran a mile’ or ‘I drove twenty miles’, the part of the sentence that is used up bit by bit as the event goes on is the distance (‘a mile’), not the vehicle. The car is not consumed piece by piece, so it cannot be what measures out the event. This page checks that claim against the semantics literature it uses (Dowty 1991, Krifka 1998, Hay, Kennedy & Levin 1999), and then looks for sourced Navajo sentences with a distance phrase to test it. It also asks how the distance phrase relates to the -ił travel pattern, where the vehicle is often the subject. The work bears on the routine's second phenomenon: which participant a verb takes as its argument, and which of its roles in the action (Dowty-style entailments) end up as slots in the verb word. Two operations are kept apart throughout. Merge builds structure: it puts an element at a position. Agree values features such as person and number: a probe copies a value from the closest matching goal. Written 2026-10-21, 8:46pm PT.
Ground rules (Grant's). Every Navajo form below comes from a source opened this run, or is already on this site with its source. Forms are quoted as displayed, with the source's own gloss line, and nothing is segmented beyond what the source prints. Where no sourced form exists, the test is stated with schematic forms only (capital-letter placeholders, no Navajo string). What the sources say is graded SOURCED. Rules and mappings are ours (HYPOTHESIS). Grant's own ideas are GRANT'S CONJECTURE. What needs data is OPEN. Only contradicting data count against an analysis; a source's silence does not. Grant does not speak Navajo, and nothing here is a native-speaker judgment. No judgment is reported unless a source gives it.
Notation (the site's own; unchanged from the -ił travel pattern). Σ = role-schema step (which participants the verb's stem describes, and each one's role in the action); A = word level (rule set A: the prefixal, left-headed spine, spelled out as one word W); B = clause level (rule set B: right-headed clause syntax, noun linking, postpositional phrases, particles). W is the changeover point from A to B. ρMv = a mover; ρC = a causer; ρCo = a companion who moves along; π = the path (the stretch of space the motion covers); μ = a measure of π (‘two miles’). “Holistic theme” and “incremental theme” are Dowty's terms (§2.1). TP-0 to TP-7: the -ił travel pattern §5.
Bottom line, in plain words
SOURCEDThe literature agrees with Grant about the car, and nearly word for word about the distance. Dowty 1991 (pp. 568–569) uses Grant's own kind of example: drive (a car) from New York to Chicago, run a mile. He says the thing affected bit by bit there is “the PATH John traverses”, not John and not the car. If the trip is interrupted, part of the path has been covered, but no part of John is in Chicago. Dowty calls the mover a holistic theme: it changes as a whole, not part by part. Krifka (1998 ms., p. 25) adds that “Movement events can be measured by the path that they cover, as in Mary walked two kilometers”. Hay, Kennedy & Levin (1999, §4.2) go one step further: the “incremental theme” is best understood not as a participant at all but as “the measure of change along a path of motion”. On their definition, ‘a mile’ in ‘ran a mile’ is the incremental theme, which is exactly Grant's wording. They also give the English judgment Grant's test needs: “?? She ran a mile, but didn't quite finish it” (their (38a)).
GRANT'S CONJECTUREGrant's claim, stated exactly. In a motion event the distance phrase is the incremental theme; the vehicle is not. Our grade: supported for English by the three sources above. Dowty's own wording is slightly different (the path is the incremental theme, and ‘a mile’ says how much of it is covered), and Hay, Kennedy & Levin's wording is Grant's. The difference matters for Navajo (§5): the path itself is never a slot in the verb word in anything opened, while the distance phrase is a separate phrase on the clause side.
SOURCEDNavajo: one distance-in-miles sentence, and no sentence with both a vehicle and a distance. The only sourced Navajo sentence with a mile phrase is still the NLA's entry 479, re-opened this run: Naakidi tsin sitą́ ądi daatsʼí kinaaldá b ił ʼahééniijééʼ, glossed “two mile 3-run.pl.P-at maybe purberty.ceremony 3-with RECIP-1dpl-run.pl.P”, ‘We (pl.) ran about two miles with the girl who was running in the puberty ceremony’ (N1). It has running, not a vehicle. Nothing like ‘I came twenty miles by car’ turned up in what could be opened. The NLA's live site was down, and its full-text search could not be run (§1). What did turn up are eleven more sentences, new to the site, about how far something goes, how far it gets, or not getting there (N2–N12). The closest to Grant's request is Chidí naatʼaʼí ʼíinilaaígíí háajįʼ ninítʼaʼ? ‘How far did the airplane that you built go when it flew?’ (N3). That sentence has a vehicle as subject and a ‘how far’ phrase built with -jį́ʼ ‘up to, as far as’.
HYPOTHESISWhat the Navajo shows so far. In N1 the distance phrase carries a postposition in the gloss (“…-at”), and the verb's gloss shows no object prefix for it (“RECIP-1dpl-run.pl.P”). So the distance is not one of the verb word's slots. No Agree reaches it, and it is merged on the clause side, like a goal phrase with -góó ‘to’ or -jį́ʼ ‘as far as’. That fits Dowty's remark that the path “is not a syntactically realized argument at all”. It also fits Grant's slot thesis in a new way. The incremental theme is a role in the action (it comes from the verb's meaning) that has no pronoun slot in W. So not every role in the verb's schema becomes a slot (§5, DM-3).
OPENThe ‘but didn't get there’ test has no Navajo answer yet. No source opened gives a judgment for a Navajo motion verb with a distance phrase plus ‘but didn't finish / didn't get there’. Two sourced facts shape the test. First, Smith, Perkins & Fernald (2007, §4.1) say that in Navajo telicity “does not” have consistent distributional correlates. There are not even separate ‘for an hour’ and ‘in an hour’ adverbials, so the English tests do not carry over directly. In their (28b), ‘drive his/her car to town in an hour’, the telic reading “needs the morpheme ni expressing the final endpoint”. Second, a Navajo progressive motion sentence with a goal phrase can be followed by non-arrival. The NLA's N5 is Dził dę́ę́ʼ hoołtį́į́ł ńtʼééʼ tʼáá nihichʼįʼjįʼ ʼásdįįd, ‘A rainstorm was moving along from the mountain but it fizzled out before it reached us’. So the test has to use the perfective, and §6 gives the predictions in schematic form only.
Fernald et al., Navajo Adverb Lexicon, v3.2022, and Navajo Verb Modes, v3.2022 (Navajo Language Academy), talkingdictionary.swarthmore.edu
Live site unreachable this run (web fetch: 504 on navajo_adverb/?entry=479, time-out on an adverb page; curl from the box: no connection; about 8:55–9:20pm PT). Opened in Internet Archive Wayback copies (2025) by web fetch: navajo_adverb/?entry=479, navajo_verb_mode/?entry=479 (identical), navajo_adverb/ (a random 50-sentence list), navajo_adverb/?semantic_ids=location, navajo_verb_mode/?semantic_ids=location, navajo_verb_mode/?semantic_ids=goal, navajo_adverb/?entry=1996. Not archived (404): a “distance” tag, the adverb introduction, the per-verb page for ‘go’, the per-adverb page for chʼééh ‘in vain’, and a progressive page. The NLA's own text search needs the live site and could not be run. These copies are the NLA's pages as archived; a sentence may have changed since
N1 (entry 479, with speaker); N2–N10 (§3)
Smith, Carlota S., Ellavina T. Perkins & Theodore B. Fernald 2007. “Time in Navajo: Direct and indirect interpretation.” IJAL 73(1): 40–71 (Swarthmore Works repository copy)
Opened as text (web fetch of the repository PDF). The PDF's font turns several Navajo letters into other characters in the text layer (ł prints as l; nasal and accented vowels as “ç”, “Ç”, “¢”, “!”). A page-image copy could not be downloaded: the repository answered the box with a Cloudflare “Turnstile” challenge, which was not bypassed. So Smith et al.'s Navajo is quoted only where the text layer is clean, or in their glosses and translations, with the garbled form shown as printed and marked
Progressive “available only for motion verb bases” (§3.1); viewpoints and endpoints (§4.1); “no consistent distributional correlates for telicity” (§4.1); (27), (28b) ‘drive his/her car to town in an hour’ with ni (pp. 64–65); (30a) rainstorm moving along (§4.3)
Dowty, David 1991. “Thematic proto-roles and argument selection.” Language 67: 547–619 (box PDF, scan)
Opened on page images, pp. 567–569 and 572 (PDF pp. 22–24, 27)
Incremental theme as a homomorphism (p. 567); (20), (21) incl. drive (a car) from New York to Chicago, run a mile (p. 568); the PATH; HOLISTIC THEMES; the path “not a syntactically realized argument at all”; path as direct object (cross the desert) (p. 569); “incremental theme” as a Patient Proto-Role property (p. 572, (28b))
Krifka, Manfred 1998. “The origins of telicity.” In S. Rothstein (ed.), Events and Grammar. Manuscript version, author's site
Opened as text (PDF text layer); page numbers are the manuscript's
Telicity is a property of the description (‘running’ vs ‘running a mile’, p. 9); a path argument even with sleep for a train passenger (p. 10); movement relations and path arguments (§4.2, p. 22); measuring movement by path length, walk two kilometers telic (§4.4, p. 25 and n. 14)
Hay, Jennifer, Christopher Kennedy & Beth Levin 1999. “Scalar structure underlies telicity in ‘degree achievements’.” SALT IX, 127–144. Preprint from B. Levin's site
Opened as text; the preprint has its own pagination (PDF pp. 13–15 = §4.1–4.2)
(38a) “?? She ran a mile, but didn't quite finish it” vs (38b) ran a race; measure phrases with motion verbs ((40)); the incremental theme as “the measure of change along a path of motion”; Dowty's holistic theme
Willie 1991 (text layer of the whole thesis), Hale 1997, Hale 2001, Hale 2003 ms., Kibrik 2005, Platero (ed.) 1974, Fernald NLA ch. 3, and the other Navajo texts on the box: no “mile(s)”, “kilometer”, tsin sitą́. Willie and Hale have no distance-phrase sentence in the text that could be searched
Not opened
—
Young & Morgan 1987 (lending-only; the unofficial dump on the box is not used); Tenny 1994 (measuring-out; book, not open access); Levin & Rappaport Hovav 1995 and Rappaport Hovav & Levin's papers on scales (a download returned a web page, not the paper); Midgette 1995, 1996 (“Lexical aspect in Navajo: the telic property”); Smith 1996 (“Aspectual categories in Navajo”). What other works are cited as saying is attributed to the citing source. Web-search result snippets were used only to find addresses, never as sources
Display. The archived NLA pages show some words split by spaces where the database links a piece to an entry (sitą́ ądi, b ił, háa di). Forms are quoted as displayed; the spaces are the database's, not word divisions we claim (as on the -ił travel pattern §1).
SOURCED p. 567: “the meaning of a telic predicate is a homomorphism from its (structured) Theme argument denotations into a (structured) domain of events”; “if x is part of y, then if a telic predicate maps y (as Theme) onto event e, it must map x onto an event e′ which is part of e” (mow the lawn: the state of the lawn shows how far the mowing has got). p. 568, (20): effected and destroyed objects (build a house, eat a sandwich). Then (21): “walk from the bank to the post office, drive (a car) from New York to Chicago, run a mile; grow into an adult, become an architect”. p. 568–569: “If John drives from New York to Chicago, John necessarily undergoes a definite change of location … but if this trip were interrupted before it was finished, we would not infer that part of John has arrived in Chicago while the rest of him is still in New York. Upon reflection, it is clear that what is partially but not totally affected in this case … is the PATH John traverses in driving from NY to Chicago”. The mover is a “HOLISTIC THEME”: it changes, but “not because they undergo a change part by part”. The path “is (like Jackendoff's ‘Theme’ in the verb butter) not a syntactically realized argument at all; the prepositional phrases from New York and to Chicago refer to the beginning and end points of the Path. However, an Incremental Path Theme can be syntactically realized as a Direct Object in semantically parallel telic examples like cross the desert, traverse the United States (in six days)”.
HYPOTHESIS Reading for Grant's case. The car in Dowty's drive (a car) is, at most, a second holistic theme: it moves as a whole along the same path. Dowty does not name a role for the car. He puts it in parentheses, and the incremental part is the path. Dowty does not say on these pages what a mile in run a mile is: it is listed in (21) with the path cases, and the comment covers the whole list.
2.2 Krifka 1998: the path as an argument, measured by distance
SOURCED p. 9: “one and the same event of running can be described by running (i.e., by an atelic predicate), or by running a mile (i.e., a telic, or delimited, predicate)”; Vendler “says that if a person stops in between, then this person did run, but did not run a mile”. p. 10: Bill walked from the university to the capitol relates Bill, an event “and some path in space”; such arguments “may be realized by syntactic arguments … but also by adjuncts”; “sleep may come with a path argument in John slept from Lyon to Paris (think of John being a passenger of a train)”. §4.2, p. 22: movement predicates “have at least three arguments, for an object, a path, and an event. The path argument can be made explicit, even though it often remains implicit”. §4.4, p. 25: “Movement events can be measured by the path that they cover, as in Mary walked two kilometers”; walk a kilometer is shown to be telic. n. 14: “measure functions like km are originally measures for distances, not for paths”, generalised to paths by length.
HYPOTHESIS For Grant's case, Krifka separates two things that ‘a mile’ joins. One is the path, the semantic argument that is used up. The other is the distance measure (km, mile), a function on paths. The passenger example matters for vehicles: a train passenger who sleeps still has a path. The path belongs to whoever moves, whether or not that person causes the motion.
2.3 Hay, Kennedy & Levin 1999: the measure is the incremental theme
SOURCED §4.1: “When a bound on the difference value is linguistically specified, a non-cancellable telic interpretation arises. If, however, a bound is only contextually supplied, then the telicity arises through implicature and is cancellable”. (38a) “?? She ran a mile, but didn't quite finish it.” (38b) “She ran a race, but didn't quite finish it.” (40a) “The plane descended 1000 meters.” Verbs of directed motion “describe a change along a projected scale: the path of movement of the affected argument”. §4.2: “with verbs of motion, the homomorphism is defined by the path traversed by one of its arguments”; “the semantic object that best corresponds to Dowty's incremental theme is in fact the difference value (i.e., the measure of change along a path of motion, in spatial extent, or in some other scalar property). On this view, the ‘incremental theme’ is properly construed as a measure of some property of an argument of a verb, not an argument”; Dowty “recognizes that with verbs of motion the term ‘incremental theme’ is most appropriately applied to the path of motion, even if it is often left unexpressed, rather than to the argument corresponding to the moving object or theme, which Dowty names the ‘holistic theme’”. They also say that “telicity and incremental theme (qua difference value) are independent”: a predicate is telic only if the incremental theme is bounded.
HKL cite the Dowty passage as “Dowty (1991:659)”; the article ends on p. 619, and the passage they describe is on p. 569 (read above). We take “659” to be a misprint (HYPOTHESIS).
OPEN Tenny's “measuring-out” (1987, 1994) and Levin & Rappaport Hovav's work on directed motion and scales are cited by these papers but were not opened. What they say is given here only as HKL and Krifka report it.
2.4 Summary of the theory
Question
Dowty 1991
Krifka 1998
HKL 1999
Grant
What is used up bit by bit in ‘drive from NY to Chicago’ / ‘run a mile’?
the path
the path (adjacency of path parts ↔ event parts)
the path is the scale; the incremental theme is the measure of change along it
the distance, ‘a mile’
The mover (John)?
holistic theme
the object argument of the movement relation
holistic theme (Dowty's term)
—
The car?
in parentheses; not incremental
not discussed; the sleeping passenger has a path
not discussed
not incremental: “not used up bit by bit”
‘a mile’?
listed with the path cases (21)
an extensive measure of the path (km′)
a linguistically specified bound on the difference value → non-cancellable telicity (38a)
the incremental theme
HYPOTHESIS Grant's claim and HKL's redefinition say the same thing. Against Dowty and Krifka it is a terminological refinement, not a conflict: they call the path the incremental theme and ‘a mile’ its measure. No opened source treats the vehicle as an incremental theme.
3. The Navajo facts
3.1 The one sentence with a mile phrase (re-opened)
NLA Adverb Lexicon ?entry=479 (Wayback copy, 2025); the same entry in Verb Modes
N1 (Sharon J. Nelson)
Naakidi tsin sitą́ ądi daatsʼí kinaaldá b ił ʼahééniijééʼ.
two mile 3-run.pl.P-at maybe purberty.ceremony 3-with RECIP-1dpl-run.pl.P
‘We (pl.) ran about two miles with the girl who was running in the puberty ceremony.’
As printed: “purberty” for ‘puberty’; the gloss line has seven items for nine displayed pieces. The two words tsin sitą́ get one gloss, “mile”. The next gloss, “3-run.pl.P-at”, goes with the displayed ądi (possibly together with the end of sitą́). So the gloss gives a verb-like piece ending in “-at”, which reads as ‘where they ran’ inside the distance phrase. We do not segment further (T48f). The English ‘about’ answers to daatsʼí “maybe”. The main verb is glossed “RECIP-1dpl-run.pl.P”: perfective, first person plural subject, and no object person.
3.2 Other sentences about how far (new to the site)
sheep some away 3-1-lose.P-GO almost evening in.vain 3-after 1-run.around.P
‘When I lost some sheep I ran around after them until near sundown without success.’
N10
Chʼééh dé yá a go hooghan góó náásh dááł.
in.vain 1-go.sg.P-GO home-to 1-return.Prog
(No translation is displayed in the archived copy.)
N11
Dáʼákʼeh gó yaa yish woł.
cornfield down.that.way 1-run.Prog
‘I’m running along beside the cornfield.’
N12 (Dana Desiderio, Devon Denny)
Kintah gó yaa yish ááł ńtʼééʼ béeso ńdiiłtsooz
areal-down 1-go.Prog past money 3-1-pick.FFO.P
‘I was walking down the street when I found a dollar bill.’
As printed: N2's verb link is not displayed; N3 “3-2-make.P-ÍGÍÍ”; N5 “Prg”; N6's gloss “1-physical.being” and “pl-3-path.extend.P”; N8 “1-weave.P” after tʼáadoo; N12's gloss has no item for Kintah. The NLA's Adverb Lexicon also lists doo deighánígóó ‘a long ways, far’ (random list, 2025 copy) without a sentence. N10's ‘go in vain’ is left uninterpreted (T48g).
3.3 What Smith, Perkins & Fernald 2007 say about aspect and endpoints
SOURCED §3.1: “Three modes express aspectual viewpoints: the Perfective, Imperfective, and Progressive. The Progressive is a type of imperfective; it is available only for motion verb bases in Navajo.” §4.1: “The Perfective makes events visible for semantic interpretation with endpoints (atelic, durative events have arbitrary final endpoints). The Imperfective and Progressive make events visible for semantic interpretation without endpoints.” The Progressive “focuses internal intervals only” (fn. 33: only verbs with a “Cursive” stem shape, “verbs of literal motion and some verbs of metaphorical motion”). On telicity: “the feature Telic/Atelic does not [have distributional correlates]. Although morphemes indicating telicity appear in telic verb words (Midgette 1996), there are no consistent distributional correlates for telicity. For instance, Navajo does not have distinct adverbials conveying duration vs. completion, e.g., for an hour and in an hour.” pp. 64–65: the one adverbial ‘up to an hour’ reads ‘for an hour’ with atelic verb words and ‘in an hour’ (completive) with telic ones:
Smith, Perkins & Fernald 2007, (28b), p. 65, as the repository PDF's text layer prints it (see §1: “ç”, “Ç” are font artefacts for letters with hooks and accents, and ł prints as l; the line is not restored)
one hour-pass-up.to town-to 3-car cess-3subj-Fut-drive
‘S/he will drive his/her car to town in an hour’ (completive)
SOURCED Their comment: “(27a) needs the adverb ’altso ‘fully, completely’ for the telic interpretation. (28b) needs the morpheme ni expressing the final endpoint.” [The text layer says “(27a)”, but the ’altso example is their (28a) ‘I will write the whole paper in an hour’; the label is printed as given.] §4.3, (30a): ‘Yesterday, while I was out herding, there was a rainstorm moving along the mountainside’, with a progressive glossed “3subj-prog-move” and the past particle; “explicit information is needed to locate a progressive in the past”.
HYPOTHESIS Two things follow for Grant's test. (i) In (28b) the car is the object of the ‘drive’ verb (‘his/her car’), and the endpoint is marked inside the verb word by ni- “cess”, not by the goal phrase alone. (ii) A Navajo motion sentence can be telic without a distance phrase (goal + ni-). So a distance phrase is one way among several to bound the path.
4. Where the vehicle fits
Sentence
Mover(s) (holistic themes)
Vehicle's role
Path / its bound
Measured by a distance?
N1 ran ~two miles
1dpl subject + the girl (-ił)
none
π, bounded by the measure
yes: naakidi tsin sitą́ ‘two mile(s)’
N3 how far did the airplane fly
the (model) airplane, subject
subject (TP-1, movement stem)
π, end asked with háajįʼ ‘where-up.to’
asked (‘how far’)
S1 sh ił dah ʼadiilwod (TP page)
unnamed vehicle (subject) + the speaker (-ił)
subject
π, direction ʼáajigo ‘that way’
no
S8 / D1 (‘drive’ stem -bąs)
the vehicle and the passengers
unnamed; the human driver is subject
π, goal -góó
no
Smith (28b)
the car (and the driver)
object, bichidí ‘his/her car’
π, goal + ni- “cess”
no (time measure ‘in an hour’)
N2 how far is it
none (a state)
none
π itself, from kodóó ‘here-from’ to nighanjįʼ ‘2-home-to’, is what is ‘long’
asked (‘how far’) by the neuter verb ‘long’
HYPOTHESIS In every opened sentence the vehicle is either the subject (with a movement stem), or the object or unnamed (with the ‘drive’ stem). Whichever it is, the vehicle moves along the same path as the traveller: it is a second holistic theme, never the thing that is used up. The distance, when it is said, measures the path, which the vehicle and the traveller share. So the TP rules (who is the subject) and Grant's claim (what is used up bit by bit) are independent: changing the subject from the car to the driver, as in TP-1 vs TP-2, does not change the incremental dimension.
5. Rules DM-0 to DM-7 (ours; all HYPOTHESIS unless marked)
DM-0 TERMS (SOURCED summary) In a motion event: mover(s) = holistic theme(s)
(Dowty p. 569); the path pi = the incremental dimension (Dowty pp. 568-569;
Krifka §4.2); a distance mu = an extensive measure of pi (Krifka §4.4);
HKL §4.2 call mu (the measure of change along pi) the incremental theme.
DM-1 GRANT (GRANT'S CONJECTURE) The distance phrase is the incremental theme; the
vehicle never is. Agrees with HKL's definition; with Dowty/Krifka it
reads "the distance phrase states the incremental theme's extent".
DM-2 NOT A SLOT The incremental dimension pi is a role in the action (an entailment of the
motion stem) with no slot in W: no subject, object or P prefix agrees with
it in any opened sentence (N1 "RECIP-1dpl-run.pl.P"; N3 "3-fly.P").
DM-3 TWO KINDS The role schema Sigma has (i) participant roles, which get slots and are
OF ROLE Agree goals (mover, causer, rider), and (ii) dimension roles (pi; a time
scale), which get no slot and are only Merged by phrases on the B side
(a goal -góó, a limit -jį́', a measure mu).
DM-4 MEASURE IS B A distance phrase is Merged on the clause side (rule set B) as a phrase
with its own postposition ("...-at" in N1's gloss), adjoined to the clause
that contains W. It measures pi; it neither Agrees nor is linked to a slot.
DM-5 PATH CAN BE A neuter verb can predicate a property of pi itself: N2 nízah "3-long.N"
A SUBJECT with source and limit phrases; N6 "pl-3-path.extend.P". Here the
stretch is the verb's 3rd-person subject slot (cf. Dowty's "path as direct
object", cross the desert). Which is the subject in N6 (shigáál "1-
physical.being" or an areal 3) is not decidable from the gloss. OPEN.
DM-6 VIEWPOINT (SOURCED, Smith et al.) Perfective = with endpoints; Imperfective and
Progressive = without; Progressive only for motion verbs. So a
progressive motion clause asserts no arrival, with or without a goal
(N5, N11, N12). The telicity test must use the perfective.
DM-7 ENDPOINT IN W (SOURCED for (28b)) A goal phrase alone does not make the drive telic in
Smith's (28b); ni- "cess" inside W marks the final endpoint. Whether a
distance phrase alone (no ni-) does: OPEN (§6).
Rule
Status
Rests on
DM-0
SOURCED
Dowty 1991 pp. 567–569; Krifka 1998 ms. pp. 9–10, 22, 25; HKL 1999 §4.1–4.2
DM-1
GRANT'S CONJECTURE; supported for English
DM-0; HKL (38a)
DM-2
HYPOTHESIS on SOURCED glosses
N1, N3; S1–S13 (no path prefix anywhere in the TP list)
DM-3
HYPOTHESIS; the link to the slot thesis is GRANT'S CONJECTURE
DM-2; Dowty p. 569 (“not a syntactically realized argument at all”)
DM-4
HYPOTHESIS
N1 (one sentence)
DM-5
HYPOTHESIS; N6's subject OPEN
N2, N6
DM-6
SOURCED (Smith et al. §3.1, §4.1); application HYPOTHESIS
N5, N11, N12
DM-7
SOURCED for (28b); the distance half OPEN
Smith et al. (28b), p. 65
HYPOTHESISWhy not “the distance phrase is the verb's object”, as English ran a mile looks? N1's verb shows no object prefix and the distance phrase has a postposition in its gloss. One sentence is thin evidence, but it is the only one, and nothing contradicts it.
6. The telicity test, schematic (T47 design; no Navajo judgments exist in the sources opened)
Schematic forms only. Vmot.P = a perfective motion verb word such as the ones in S1–S13 or N1; GOAL-góó = a place with -góó ‘to’; MEAS = a distance phrase like N1's naakidi tsin sitą́ …; ‘…but not there’ = a following clause meaning ‘but didn't get there / didn't finish’ (no Navajo string is proposed for it).
Frame
English analogue (HKL)
Prediction if DM-1 holds and a stated distance bounds π
Status
T1 Vmot.P, …but not there
She ran, but … (no bound)
fine: the perfective gives an endpoint, but nothing fixes where it is (Smith: “arbitrary final endpoints”)
OPEN
T2 GOAL-góó Vmot.P, …but not there
She ran to the store, but didn't get there
uncertain. -góó is glossed “to, toward, along” (NLA), so it may give only a direction; Smith's (28b) needed ni- for the endpoint (DM-7)
OPEN
T2′ GOAL ni-Vmot.P, …but not there
—
contradiction expected: ni- marks the final endpoint (Smith et al.)
OPEN (no judgment printed)
T3 MEAS Vmot.P, …but didn't finish
(38a) “?? She ran a mile, but didn't quite finish it”
contradiction expected, as in English, if MEAS states a bound on π (HKL: “non-cancellable”). If MEAS were only an approximation (N1 has daatsʼí ‘maybe/about’), the clash may soften
OPEN
T4 MEAS Vmot.Prog, …but not there
She was running a mile, but stopped
fine (DM-6: no endpoint asserted); attested for a goal-limit with N5
goal-limit case SOURCED (N5); distance case OPEN
T5 vehicle-subject Vmot.P + -ił + MEAS (‘I came twenty miles by car’)
The car took me twenty miles
same as T3: the vehicle being subject (TP-1) should not change telicity, because the vehicle is not the incremental theme (DM-1, §4)
OPEN; no sourced sentence of this shape (T48a)
HYPOTHESIS The decisive pair is T3 vs T5. If a clash appears in both, the distance measures out the event whatever is subject (Grant's claim). If it appears only when the vehicle is subject, the vehicle would be doing measuring work after all. Smith et al.'s warning applies: Navajo telicity has “no consistent distributional correlates”, so the test is a judgment of contradiction, not an adverbial test.
7. Derivations DD1–DD6
Steps are labelled by rule set: Σ role schema, A word (W, and each PP's sub-cycle), B clause. Each step names its operation: Merge (builds structure), Agree (values person/number), Move, Link (a nominal linked to a slot), Spell-out (sealing a unit; “Seal” on earlier pages). VPISH: every participant slot is first-merged inside vP; in the V-P variant the subject slot then moves inside the word to Spec,M/SP, in V-H it stays. The strings come out the same either way. The spine is Hale 2003's Qual > M/S > v > V; no flat-S tree is used. No verb is segmented beyond its source's gloss.
DD1. N1 Naakidi tsin sitą́ ądi daatsʼí kinaaldá b ił ʼahééniijééʼ ‘We ran about two miles with the girl …’
A W: Merge V, v; ρMv first-merged in Spec,vP (VPISH). Merge M/S (perfective, “P”). Agree M/S ⟶ ρMv [1, dual/plural]. (V-P: Move ρMv to Spec,M/SP.) RECIP as glossed. Spell-out W = ʼahééniijééʼ. No probe in W finds π: π has no features to value (DM-2).
A PP sub-cycle: Merge-ił with ρCo; Agree P ⟶ ρCo [3] → b-; Spell-outbił (displayed b ił).
BMergekinaaldá; Link ↔ ρCo. Mergedaatsʼí “maybe” (‘about’).
BMerge the distance phrase [naakidi tsin sitą́ ądi] (“two mile 3-run.pl.P-at”) at the clause edge. No Agree, no Link: it is a measure μ of π (DM-4). Interpretation (Krifka §4.4): km′/mile′ of the event = 2, so the description is bounded. The approximator daatsʼí softens the number, not the boundedness.
Result N1 ✓. Merge built every position. Agree touched only the two participant slots, not the incremental dimension.
DD2. N3 Chidí naatʼaʼí ʼíinilaaígíí háajįʼ ninítʼaʼ? ‘How far did the airplane that you built go?’: vehicle subject, limit asked
Σ Σ(√FLY) = {ρMv} + {π}. ρMv → subject (TP-1). The vehicle is the mover, a holistic theme.
A W: ρMv first-merged in vP; Agree M/S ⟶ ρMv [3]; Spell-outninítʼaʼ (“3-fly.P”).
BMergeháajįʼ (“where-up.to”, -jį́ʼ ‘up to, as far as’): a phrase naming π's far end, questioned. No Agree.
BMerge the relative-marked noun phrase chidí naatʼaʼí ʼíinilaaígíí (“airplane 3-2-make.P-ÍGÍÍ”), built in its own cycle; Link ↔ ρMv. Wh-placement of háajįʼ as on wh-questions (not re-derived here).
Note. The vehicle is subject and the distance is asked about. Neither the vehicle nor the limit phrase Agrees for the path.
DD3. Smith et al. (28b) ‘S/he will drive his/her car to town in an hour’: vehicle object, endpoint in W
Σ Σ(the stem glossed ‘drive’) = {ρC (causation of wheeled motion)} + {ρMv the vehicle (moved as a whole: holistic theme)} + {π}. TP-2: ρC → subject.
A W: Merge V, v; ρC in Spec,vP. Mergeni- “cess” (the authors' gloss; the endpoint of π, DM-7). Merge M/S (Future); Agree M/S ⟶ ρC [3]. Spell-out W, glossed “cess-3subj-Fut-drive”. The authors' gloss shows no object person for the car.
BMerge ‘his/her car’ (“3-car”); Link ↔ the vehicle role. Merge ‘town-to’ (as glossed; π's limit). Merge ‘up to an hour’ (time measure; completive with a telic W, pp. 64–65).
Note. The car is a linked participant; the endpoint is in W (rule set A, ni-); the time measure is in B. Nothing here is used up bit by bit except π, which is covered in an hour.
DD4. N5 Dził dę́ę́ʼ hoołtį́į́ł ńtʼééʼ tʼáá nihichʼįʼjįʼ ʼásdįįd: progressive toward a limit, then no arrival
Σ Σ(√RAIN.move) = {ρMv: an areal subject (“areal-rain.Prg”)} + {π}.
A W1: Merge M/S Progressive (motion verbs only, DM-6); areal subject; Spell-outhoołtį́į́ł. Progressive: no endpoint asserted.
BMergedził dę́ę́ʼ ‘from the mountain’ (π's start); Mergeńtʼééʼ (past particle).
B Second clause: Mergetʼáá nihichʼįʼjįʼ (“just 1pl-toward-up.to”, π's intended limit) + W2ʼásdįįd “3-disappear.P”.
Result. The limit is named, and the translation says it was not reached (‘fizzled out before it reached us’). No contradiction: a progressive motion clause asserts no arrival. This is the T4 cell for a limit, not for a distance.
DD5. N2 Haash nízah kodóó nighanjįʼ? ‘How far is it from here to your place?’: the path as the thing measured
Σ Σ(√LONG, neuter) = {θTh: the thing that is long}. No motion and no mover. Here the thing that is long is the stretch from ‘here’ to ‘your home’.
A W: Agree M/S ⟶ θTh [3] (“3-long.N”); Spell-outnízah.
BMergekodóó ‘here-from’ and nighanjįʼ ‘2-home-to’: π's two ends, as with Dowty's from New York, to Chicago. Mergehaash ‘how-Q’ (degree question).
Note (DM-5). In a state, the path is what the verb's subject slot is about. In a motion event (DD1) the same kind of stretch has no slot. That is the Navajo counterpart of Dowty's point that the path is usually not an argument, but can be one (cross the desert).
DD6. Schematic T5: ‘I came twenty miles by car’ (no Navajo string; T48a)
Σ Movement stem: {ρMv = vehicle} + -ił: {ρRd = speaker} + {π} (TP-1 shape, as in S1/S3).
A W: Agree M/S ⟶ ρMv [3]; PP: Agree P ⟶ ρRd [1] → shił (as in S3); Spell-out both.
BMerge MEAS (‘twenty’ + ‘mile’, schematic) as in DD1 step 5; no Agree.
Prediction (DM-1, §6 T5): the measure bounds π, which the vehicle and the rider share. The subject being the vehicle makes no difference to what is measured out. OPEN until a sourced sentence or a speaker's judgment exists.
Kept. Added: the subject choice does not decide what is used up; the vehicle (subject or object) and the rider are holistic themes on one shared path (§4)
the verb's arguments are role-bearing slots; nouns are optional partners
Narrowed: the slots are the participant roles. A dimension role (the path) is entailed but gets no slot, and is only measured by a clause-level phrase (DM-2, DM-3)
N1, N3; Dowty p. 569
T47 (Grant's question)
new
Answered for the theory (supported; §2). For Navajo: one mile sentence (N1), eleven related sentences (N2–N12), no vehicle + distance sentence; the telicity test is OPEN and designed (§6)
§3, §6
9. What each source gets right and misses
Source
Gets right
Misses or costs
Dowty 1991
Names the exact contrast Grant draws: the path changes part by part, the mover changes as a whole; uses drive (a car) and run a mile
Does not say what a mile is; no role for the car; English only
Krifka 1998
Separates the path (an argument, possibly implicit) from the distance (a measure function); proves walk a kilometer telic; the sleeping passenger's path
No vehicle-as-subject case; English only. Manuscript pagination
Hay, Kennedy & Levin 1999
Defines the incremental theme as the measure, Grant's wording; gives the ‘didn't quite finish’ judgment (38a)
Cites Dowty's page as 659 (misprint); English only
Smith, Perkins & Fernald 2007
Navajo viewpoints and endpoints; the progressive only with motion verbs; no adverbial telicity test; (28b) with ni-
No distance phrases; Navajo text quoted here only from a garbled text layer (page images blocked)
NLA lexicons
N1 (the one mile sentence, with a speaker), N2–N12 with glosses and translations
N1's gloss under-segments the distance phrase; no “distance” tag in the archived copies; live search unavailable; N10 has no translation in the copy
This site before today
TP rules on subject choice
Treated the trip without its path; no account of measure phrases
10. Grant's frame
GRANT'S CONJECTUREThe distance is the incremental theme. Supported in the sense Hay, Kennedy & Levin define, and consistent with Dowty and Krifka, who place the incremental dimension on the path and treat ‘a mile’ as how much of it is covered. The car point is right in all three: nothing in the car is used up as the trip goes on. Navajo cannot yet confirm it: the one mile sentence has no vehicle, and no source gives the ‘didn't finish’ judgment (§6).
GRANT'S CONJECTURETheta roles as a proto-role schema (phenomenon 2). Dowty lists “incremental theme” among the contributing properties for the Patient Proto-Role (p. 572, (28b)), so it is a role in the action. But in Navajo it is the one role in these sentences that never gets a pronoun slot. The verb word agrees with movers, causers and riders, not with the path. So the role schema contains more than the slots: some roles are entailments that only phrases on the clause side can measure (DM-3). This is a refinement of the weak thesis, not a break with it: the slots are still role-bearing, and they are still filled before any noun is chosen.
HYPOTHESISTheta roles and event-structure roles together. The path is where the two meet. As a theta-level entailment (Dowty) it belongs to the stem. As an event-structure ingredient (the scale whose bound makes the event telic, HKL) it is supplied by the clause: a distance (N1), a limit (-jį́ʼ, N3, N5), or the endpoint prefix ni- inside W (Smith (28b)). The event-structure layer here is split between rule set A (ni-) and rule set B (the measure phrase).
GRANT'S CONJECTURESTRATA (theta roles first, event structure next, noun selection later). DD1 runs in that order. Σ gives the mover and the path; W's Agree fills only the mover; and the measure and the nouns come last on the clause side. It is a fit, not a separating test: a lexicalist account that adds the measure as an adjunct derives the same string.
GRANT'S CONJECTUREPhrase-level lexicalization. The NLA glosses two written words, tsin sitą́, as one item, “mile”. That is a phrase-level lexical unit of the kind Grant describes. Whether its parts mean ‘wood/post’ + ‘it lies/is in place’ is not shown in any source opened this run (T48f).
HYPOTHESISThe word as the changeover point (phenomenon 1). The endpoint of a path can be marked on both sides of W's edge. It can be marked inside W (rule set A: ni-), or by a phrase outside W (rule set B: -góó, -jį́ʼ, a measure). This gives a test for the changeover. If ni- and a distance phrase can both appear and must agree in what they say about the endpoint, a B-side measure would be constraining an A-side head. That would be a cross-boundary dependency, and it would have to be stated as interpretation rather than Agree (T48c).
11. Tasks T48a–h and statuses
Task
What it needs
T48a
A sourced Navajo sentence with a vehicle and a distance phrase (‘I came twenty miles by car’ type): the live NLA's text search for tsin sitą́, ‘mile’, naadiin ‘twenty’ (when the site answers); Young & Morgan 1987 (blocked: lending-only)
T48b
The telicity judgments of §6 (T1–T5): a source that prints them, or a speaker's judgment that Grant supplies. Without either, the test stays OPEN
T48c
Can ni- (Smith's “cess”) and a distance phrase appear together, and must they match? (the changeover test in §10)
T48d
Midgette 1996, “Lexical aspect in Navajo: the telic property”, and Smith 1996, “Aspectual categories in Navajo”: which morphemes mark telicity in the verb word, and do they include anything for distance?
T48e
A page-image copy of Smith, Perkins & Fernald 2007 to restore (28b) and (30a) letter by letter (the repository's challenge blocked the download this run)
T48f
N1's distance phrase: what naakidi and ądi are, whether the “-at” is -di ‘at’, and whether tsin sitą́ ‘mile’ is built from ‘post’ + ‘lie/be in place’ (a source that segments it)
T48g
N6 shigáál deiz tʼiʼ “1-physical.being pl-3-path.extend.P”: which participant is the subject of the ‘path extends’ verb? And N10 chʼééh dé yá a go “in.vain 1-go.sg.P-GO”: an idiom, or a non-arrival perfective?
T48h
Tenny 1994 (measuring-out) and Levin & Rappaport Hovav on directed motion and scales, read in the original (cited here only through HKL and Krifka)
Statuses: T47: theory answered (DM-0, DM-1 supported for English); Navajo test designed, data OPEN (T48a, T48b). T46a–h: unchanged. T45a–c, T45e, T45g, T45h: unchanged. T42d/e/g: still blocked on Young & Morgan.
What else changed on the site
Update notes on the -ił travel pattern (lede) and the theta-args thesis (after the 2026-10-21 12:46pm note). New row on open questions (T47, T48a–h), a sources section (added 2026-10-21, 8:46pm), and a home-page entry. No verb-dictionary entries changed. Navajo forms new to the site (all with sources): NLA N1 (entry 479) and N2–N12, plus Smith, Perkins & Fernald (28b) (text-layer form, marked).
Can a noun stand between a postposition and its verb? bee dibé síníloh, Willie's uninterruptible verb complex, and where a postpositional phrase is merged (T49)
2026-11-11 audit pass 3. Local box file paths were removed from this page; files inside the repository are given by repo-relative path, and the source citation next to each removed path stays as the reference. See the pass 3 report.
Restated on the unified structure (2026-11-15). Postpositions are restated as [PP DP P⁰] with left specifiers: a grammatical P agrees with the DP in its specifier, a lexical P has no probe (H64); a selected PP is merged in VP, and Willie’s “cannot be interrupted” follows from the default, HS-1 and left specifiers. Light d, used below, is withdrawn (Grant, 2026-11-14); light n is an OPEN structural question. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Flat-S and other pre-VPISH notation below is labelled contrast. Status on the spine: see unified structure, §2b and §4, postposition row. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
Up to now the site has held that nothing on the clause side ever stands between a grammatical postpositional phrase (PP) and the verb word W (rule PC-2 on attached vs separate postpositions). That follows Willie 1991, who says the verb and its grammatical PPs form one constituent, the MVS, which “cannot be interrupted”. This run found that four sentences already on the site break that order. They are NLA (9)–(11b) on the yes/no page: … bee dibé síníloh? ‘Did you rope the sheep with the rope …?’. There the object noun dibé ‘sheep’ stands between bee ‘3-with’ and the verb. And Willie herself lists -ee ‘with’ (instrumental) as a grammatical postposition. This page states the conflict exactly. It then sets out four ways out, says which one each rule set and Grant's theta-role thesis can live with, and gives a schematic test that would decide between them. It also runs the site's rule book for inconsistencies, and finds two more, one of which gives a new testable prediction about preverbs. Two operations are kept apart throughout. Merge builds structure: it puts an element at a position. Agree values features such as person and number: a probe copies a value from the closest matching goal. Written 2026-10-22, 4:46am PT.
Update 2026-10-22, 8:46pm (T50a, T49b–d).SOURCED data, HYPOTHESIS reading. X2 is attested as an order: object before the instrument, Shimá sání lééchą́ą́ʼí tsin bee nidiiłhaal (NLA 92) and Béésh ʼáłtsʼózí bee ʼatsidí bee ʼaháníłneʼ (NLA 90, YM 1987:41). Every option derives both orders, so this doesn't separate them. X4: in the two 3/3 instrument sentences the form is bee, not yee, as O1/IP-1′ predicts. See Instrument and voice. X3 is still OPEN.
Update 2026-10-22, 12:46pm (T49f, T49h).HYPOTHESIS IP-1 below is revised to IP-1′ on postpositions by verb. Willie 1991 prints yi-/bi- voice pairs on three grammatical postpositions: ‘for’ with ‘lead’ ((85)/(86) p. 95), ‘to’ with ‘roll’ ((83)/(84) p. 95) and ‘to’ with ‘speak’ ((79)/(80) p. 94; (12a, b) p. 209). She says the inverse has scope over “the entire MVS” (p. 100) and over “any associated postpositonal phrases” (p. 208). The beneficiary of ‘lead’ and the goal of ‘roll’ are not participants the stem singles out, but they are inside the MVS. So IP-1 as worded is contradicted. IP-1′: a PP is inside the MVS when the voice ranks its slot together with the verb's own. The rope sentences are unaffected, because the instrument is never shown alternating. The X4 citation below (“Willie p. 224”) is supplemented by pp. 100 and 208, both opened on page images this run. CYC now has D1 = W and C1′ in its rule text (CYC v2).
Ground rules (Grant's). Every Navajo form below comes from a source opened this run, or is already on this site with its source. Forms are quoted as printed or displayed, with the source's own gloss line, and nothing is segmented beyond what the source prints. Where no sourced form exists, a test is stated with schematic forms only (capital-letter placeholders, no Navajo string). What the sources say is graded SOURCED. Rules and mappings are ours (HYPOTHESIS). Grant's own ideas are GRANT'S CONJECTURE. What needs data is OPEN. Only contradicting data count against an analysis; a source's silence does not. No analysis, the site's own included, is assumed to be fully right. Grant does not speak Navajo, and nothing here is a native-speaker judgment.
Notation (the site's own; unchanged). Σ = role-schema step (which participants the verb's stem describes, and each one's role in the action); A = word level (rule set A: the prefixal, left-headed spine, spelled out as one word W, plus the PP sub-cycle); B = clause level (rule set B: right-headed clause syntax, noun linking, particles, enclitics). W is the changeover point from A to B. MVS = Willie's “Maximal Verb-Sentence” (W&J 2000: “Maximal Verb Complex”): the verb plus its grammatical PPs. A Σ-PP is a PP whose object is one of the participants the verb's own stem describes. ρA = the roper (agent); ρTh = the thing roped; ρIns = the rope used. CYC, C1–C3, D1, D2: A/B as spell-out §3. PC-0 to PC-7: attached vs separate P §4. EN-0 to EN-6: the yes/no page §4. TP-0 to TP-7: the -ił travel pattern §5. DM-0 to DM-7: distance as measure §5.
Bottom line, in plain words
SOURCEDThe order is printed four times. The NLA's yes/no page prints (9), (10a), (10b), (11a) and (11b) as one sentence with the enclitic moved around. Four of them have the same order: the rope phrase, then bee “3-with”, then dibé ‘sheep’, then the verb “3-2-rope.SP”. In (11b) the noun between them even carries the question enclitic: … bee dibéésh síníloh? The NLA calls the bracketed rope phrase “the first syntactic constituent”, separate from bee.
SOURCEDAnd Willie's rules forbid it, as written. Willie 1991 (p. 40): “The grammatical postpositions occur only with pronominal objects and are a component of the MVS.” Her list (28) on p. 41 includes “f. -ee ‘with’ Instrumental”, which “occurs only with inanimates”. On p. 50, (44a), the MVS “cannot be interrupted”. Willie & Jelinek 2000 (p. 270) also call the MVS “noninterruptible”. So if the NLA's bee is Willie's (28f), the noun dibé interrupts an MVS. Both pages of Willie were re-read on page images this run.
HYPOTHESISOur reading: a postposition's site depends on the verb, not on the postposition. Willie's “grammatical” is a class of forms: the postpositions that take a pronoun object. Being part of the MVS is a matter of site. We propose that a PP is merged inside the verb's phrase only when its object is one of the participants the verb's stem describes (IP-1). The rope you rope with is not one of them, so its PP is merged on the clause side, above the focus noun, and the focus noun ends up next to the verb. Two sourced contrasts fit this. In W&J (50a, b) dibé łį́į́ʼ bitah yíghááh, where ‘among’ is part of what ‘join’ describes, both nouns come before the PP. Willie p. 98 already shows one postposition counting as grammatical or lexical depending on the arguments. The same form bee is used for the possessor with ‘exist’ (NLA Q2 chidíísh bee hólǫ́ ‘does s/he have a car’), where it sits next to the verb.
OPENWhat would show this is wrong. A sourced sentence with a noun between a Σ-PP and its verb, for example a noun between bee and hózin ‘know about’ or between bitah and yíghááh. That would mean the noun really sits low, next to the verb, as a configurational analysis (Speas 1986: SOV as the basic order, nouns in argument positions) would have it. That outcome would count against the site's version of Grant's thesis (T-MID: nouns are never arguments of V). Until then the datum is compatible, undecided, not contradicting (§4, §7).
HYPOTHESISThe rule-book check found two more problems (§8). (i) CYC as written does not order a PP before W once the MVS is no longer a sealed cycle: the PP and the verb are sisters, so height (C2) says nothing. A one-line fix (C1′) restores it. (ii) The site says both that the preverbs dah and yah belong to W (TP-6) and that a preverb carrying -ísh is outside W (EN-5, Hale's ka-[á]sh nanin[á]). Both can hold only if preverbs have two sites. That gives a new prediction: a preverb can carry -ísh only if it comes before every Σ-PP of its verb. In the order “PP dah W”, -ísh on dah is predicted impossible (PV-2). No opened source tests it.
Fernald et al. 2019, Navajo Questions: grammar and examples, v3.2022 (NLA), yes/no page (?q-binary)
Re-read in the box copy of the live page saved on the 2026-10-20 run ([local working copy, not in the repository]). The live NLA site did not answer this run (web fetch 504; curl no connection, about 4:50am PT).
(9), (10a), (10b), (11a), (11b); (4)–(7)
Willie, MaryAnn 1991. Navajo Pronouns and Obviation. PhD dissertation, University of Arizona (box scan)
Page images at 110 dpi of printed pp. 40, 41 (PDF 45, 46); text layer of pp. 49–50 ((43), (44)) and a whole-thesis search for “instrument”, “bee”, “interrupt”
Grammatical vs lexical postpositions; (27), (28f) -ee Instrumental; (44a)
Speas, Margaret 1986. Adjunctions and Projections in Syntax. PhD dissertation, MIT (box scan)
Page image of p. 248
“The word order shown above, SOV is generally taken to be the unmarked basic word order”; (4.5a, b)
NLA prose: “The first syntactic constituent in this sentence (after da’) is the bracketed phrase shizhéʼé tłʼóół yizhbizhę́ (the rope that my father braided).” “When the enclitic is not attached to the first phrase of the sentence, the phrase it is attached to is focused.” Display notes: (9) prints siníloh, (10a)–(11b) print síníloh; (10a) prints the bracket before ę́sh. All kept as displayed. The focus readings are the NLA's (it underlines the focused gloss).
SOURCED In all five, bee “3-with” stands right after the rope phrase, and dibé ‘sheep’ stands right before the verb. The rope is what bee's pronoun refers to (by the translation), and the sheep is what is roped. No sentence on the page puts dibé before bee. That is silence, not a judgment.
2.2 Willie 1991 on grammatical postpositions (page images)
P. 40: “The grammatical postpositions occur only with pronominal objects and are a component of the MVS. The lexical postpositions occur only with nominal (lexical) objects.” P. 41, (28) “Grammatical Postpositions”: a. -aa ‘to’ Dative; b. -chʼįʼ ‘to’ Goal, Allative; c. -tsʼą́ą́ʼ ‘from’ (take away from; Malefactive); d. -á ‘for’ Benefactive; f. -ee ‘with’ Instrumental; g. -ł ‘with, together’ Comitative; h. -ghá ‘through’ (penetrate through); i. -lááh ‘beyond’ Comparative, Locative. “Of these postpositions, (28 a,c,d,g) occur only with animates, (28 f) occurs only with inanimates, and others occur with both.” (The list skips e; the scan's “wit!)” is ‘with’ on the image.) P. 39–40 (as recorded on postposition units): some oblique phrases “are case-marked arguments that the verb is subcategorized for, and some are more like adjuncts. I will call the former the grammatical postpositions, and the latter the lexical postpositions.” P. 50, (44) (text layer): the MVS “a. … cannot be interrupted; b. … may constitute a single phonological unit; c. … is the domain of the expression of the subcategorizational features of the verb”.
SOURCED Willie gives two criteria for “grammatical” at once: subcategorized for by the verb (p. 39), and taking a pronoun object (p. 40). For most of her examples they coincide. For the NLA's rope sentences they come apart: bee has a pronoun object (b-), but nothing shows that ‘rope’ subcategorizes for an instrument.
SOURCED In all three, every noun precedes every PP, and the PPs are next to the verb. W&J call łį́į́ʼ in (50b) “the preceding focus nominal” (p. 270), so for them the focus noun comes before the PP, not between the PP and the verb. Their p. 266 wording, “A nominal immediately preceding the inflected verb is in the focus position”, is stated for “third-person transitive sentences”, and none of the examples on the site that it is used for has a PP between the noun and the verb.
HYPOTHESIS The difference between these and the rope sentences is the role of the PP's object. ‘Among’ in (50b) states where the joiner goes. In (43) the ‘with’ and ‘about’ phrases are the knower and the thing known. In Q2 the ‘with’ phrase is the possessor. In each case the PP's object is a participant the verb (or the verb + P unit) describes. The rope in (9)–(11b) is the means used, and ‘rope (an animal)’ can be described without naming it. The role readings are ours, from the translations.
2.4 Speas 1986 on basic order (page image)
P. 248: “Navajo NPs are not morphologically marked for case (or even number, in most instances), as shown in (4.5), where word order alone indicates the grammatical function of the overt NP. … The word order shown above, SOV is generally taken to be the unmarked basic word order.” (4.5a) Atʼééd ashkii yiyiiłtsą́ ‘The girl saw the boy’.
HYPOTHESIS Speas does not discuss PPs on this page. That a configurational VP would put the object noun between a higher PP and V is our extrapolation, not her claim (§4, O2).
3. The conflict, exactly
Statement
Source
What it says about (9)–(11b)
Grade
Grammatical Ps take pronoun objects and are in the MVS
Willie p. 40
bee (pronoun object) is in the MVS
SOURCED
-ee ‘with’ Instrumental is grammatical
Willie p. 41 (28f)
the same
SOURCED
The MVS cannot be interrupted
Willie p. 50 (44a); W&J p. 270
dibé may not stand between bee and síníloh
SOURCED
The NLA's printed order
NLA (9)–(11b)
dibé stands there, five times, in one speaker-built paradigm
SOURCED
PC-2: nothing merged is pronounced between a grammatical PP and W
this site, 2026-10-21
falsified as written, if bee is grammatical in Willie's sense
HYPOTHESIS (now revised: IP-3)
HYPOTHESIS So the three sources cannot all be right with the same reading of “grammatical”. The question is whether the NLA's bee here is (a) not in the MVS, (b) in the MVS, with the noun inside the MVS too, or (c) in the MVS, with the MVS interruptible after all.
4. Four ways out
Option
What it says
Derives (9)–(11b)?
Cost
Effect on Grant's thesis (T-MID)
O1 SITE BY Σ (ours)
A PP is merged in VP only if its object is a participant in the verb's role schema Σ. The instrument of ‘rope’ is not, so its PP is merged on the clause side above Foc. dibé is the focus noun, right before the MVS, which here is just W.
Yes (ID1)
Willie's (28) becomes a list of forms, not of sites. But her p. 98 already lets one P (-ii’) be grammatical or lexical by the arguments.
Compatible: no noun in VP
O2 LOW NOUN (configurational; our extrapolation from Speas's SOV)
The object noun is V's sister inside VP, and the PP sits higher in VP: PP ⌢ N ⌢ V.
Yes (ID2)
Must explain (50b) and (43), where the nouns come before the PP; and W&J's inverse linking without nouns in VP
Against T-MID (a noun in Comp,V)
O3 PP OUT
The PP is in the MVS but moves out to a topic position, with its linked noun.
Yes
No trigger visible (no topic reading is given); Willie (44d) says the MVS behaves as a unit
Compatible
O4 LOW FOCUS
The MVS is interruptible by the focus noun. W&J p. 266's “immediately preceding the inflected verb” is literal.
Yes
Contradicts Willie (44a) and W&J p. 270; (50b) would then need a reason why łį́į́ʼ does not stand after bitah
Compatible, if Foc is above VP but below the PP's landing site; PC-2 dropped
O0 PC-2 as written
No intruder after any pronoun-object PP
No
—
—
HYPOTHESISPreferred: O1, because it is the only option that also derives the contrast with (50b), (43) and Q2 without an extra rule. O3 and O4 each need a stipulation for when the PP leaves or the noun goes in. O2 needs nouns low in (9)–(11b) and high in (50b). Not settled. The contrast rests on two sources with different speakers and a person difference: the rope sentences have a 2nd-person subject, so W&J's 3/3 focus rules (p. 267: no direct/inverse alternation with a non-third-person argument) do not strictly apply to them. The deciding test is in §7.
5. Rules IP-0 to IP-6
IP-0 FACT (SOURCED) NLA (9)-(11b): [DP_i] bee_i dibé_j W, with bee "3-with" = Willie's
(28f) -ee 'with' Instrumental (only with inanimates), dibé linked to W's object.
IP-1 SITE BY Σ A PP is first-merged inside VP (the MVS; rule-set-A side of the clause) iff its
object slot is a participant role in the verb's Σ. Otherwise the PP is merged
on the clause side (rule set B), adjoined above FocP, even when its P is one of
Willie's "grammatical" (pronoun-object) forms. The P still Agrees with its own
slot and is sealed by the PP sub-cycle either way. HYPOTHESIS
IP-2 FOCUS EDGE Spec,FocP is the position right above the MVS (W&J p. 266, p. 270 on (50b)).
With a Σ-PP the focus noun precedes the PP; with no Σ-PP it precedes W.
SOURCED as order / HYPOTHESIS as head
IP-3 PC-2' (revises PC-2 and TP-6's restatement) Nothing from rule set B is pronounced
between a Σ-PP and W. A non-Σ PP can be separated from W by the focus noun
(IP-0) and, by height, by anything merged between FocP and its own site.
HYPOTHESIS
IP-4 LINK HIGH A noun linked to a P's slot is merged above that PP: DP_i before bee (IP-0);
the clause before bił béé (Willie (43)); dibé before łį́į́’ bitah (50b).
Follows from C2 if links are made from Top/Foc. SOURCED as order / HYPOTHESIS
IP-5 ROLE-REL Whether a participant gets a slot of W or a Σ-PP is fixed by the stem's own
description, not by the referent or the P's form. The same -ee is a Σ-PP for
the possessor of 'exist' (Q2) and a clause-side PP for the rope of 'rope' (IP-0).
Unifies TP-4 (subject by the stem's description), DM-3/DM-5 (a path is a slot of
'be long' but not of 'run') and IP-1. HYPOTHESIS
IP-6 C1' (fix to CYC; §8) In D2, a sealed unit x precedes the sealed unit y when x is a
complement or specifier of a head that y spells out. Otherwise C2 (height)
orders sealed units. HYPOTHESIS
Rule
Status
Rests on
IP-0
SOURCED
NLA (9)–(11b); Willie pp. 40–41
IP-1
HYPOTHESIS
IP-0 vs (50b), (43), Q2; Willie p. 98 precedent
IP-2
SOURCED order / HYPOTHESIS head
W&J pp. 266, 268, 270
IP-3
HYPOTHESIS
PC-2, TP-6, IP-1
IP-4
SOURCED order / HYPOTHESIS mechanism
three sources, three Ps
IP-5
HYPOTHESIS
TP-4, DM-3, DM-5, IP-1
IP-6
HYPOTHESIS
CYC; PC-1; Hale 2003's lexical heads take their argument on the left (SB4)
6. Derivations ID1–ID6
Steps are labelled by rule set: Σ role schema, A word (W, and the PP sub-cycle), B clause. Each step names its operation: Merge (builds structure), Agree (values person/number), Move, Link (a noun linked to a slot), Seal (spell-out of a unit). VPISH: every participant slot is first-merged inside the verb's phrase. In the V-P variant the subject slot raises inside the word to Spec,M/SP; in V-H it stays. The strings are the same either way. The spine is Hale 2003's Qual > M/S > v > V, with silent Foc and Top above it (clause spine). No flat-S tree is used. No verb is segmented beyond its source's gloss.
Σ Σ(√ROPE) = {ρA (causation, volition: the roper), ρTh (caught, affected: the sheep)}. The rope is entailed only as “some means”. It is not a participant the stem singles out, so it has no place in Σ (IP-1). Role labels ours.
AMerge V √ROPE with ρTh (Comp,V); Merge v, ρA in Spec,v (VPISH); Merge M/S; Agree M/S ⟶ ρA [2sg] (Merge built the slots; Agree only copies person and number). ρTh [3] is spelled on its own slot. V-P: Move ρA → Spec,M/SP (silent). Merge Qual. Seal W = síníloh (“3-2-rope.SP”). Rule set A, CYC on D1: non-heads first, then heads, each higher-first.
A PP sub-cycle: Merge-ee with its own slot ρIns; Agree P ⟶ ρIns [3] → b-; Seal → bee.
BMerge Foc above the MVS (here the MVS is just W). Mergedibé in Spec,FocP; Link ↔ ρTh (person 3, sheep fits the role ✓).
BMerge the PP [bee] adjoined above FocP (IP-1: clause side).
BMerge Top; Merge the relative DP [shizhéʼé tłʼóół yizhbizhę́ę] in Spec,TopP; Link ↔ ρIns (IP-4). The relative is its own CP-derivation (relative as CP), sealed before it is merged here.
BMerge C with [Q]; Mergeda’ in Spec,CP (EN-2, spell-out N6).
BSeal D2 by CYC: C2 (higher first) gives da’ ⌢ DP ⌢ bee ⌢ dibé ⌢ [W]. C1′ (IP-6) is not needed: no Σ-PP. → Daʼ [shizhéʼé tłʼóół yizhbizhę́ę] bee dibé síníloh ✓.
B Post-syntax: -ísh at the right edge of the sealed unit dibé (EN-1; not first, so focus by EN-3) → dibéésh ✓ (vowel copy as in NLA (4)).
HYPOTHESIS No noun is merged inside VP. Movement: none (V-H), or one silent raising inside W (V-P).
ID2. The same string under O2 (low noun)
Σ as ID1, but in O2 the noun dibé itself is the argument (Speas-style).
A? Merge V with dibé in Comp,V. Merge the PP [bee] as a VP adjunct (higher than the noun).
A v, M/S, Qual as in ID1. Head movement builds W, leaving the noun and the PP behind.
B DP in Top, da’ in Spec,CP; Seal: da’ ⌢ DP ⌢ bee ⌢ dibé ⌢ W ✓.
HYPOTHESIS Same string. The difference: a noun sits in a role position inside the verb's own phrase. On the site's terms that puts a B-side unit inside the A-side domain, and the changeover is no longer at W for this noun. Under O2, W&J (50b) needs both nouns raised out of VP, past the PP.
ID3. W&J (50b) dibé łį́į́ʼ bitah yíghááh under IP-1 (a Σ-PP)
Σ Σ(√JOIN + ‘among’) = {ρA (the one who goes: movement, volition), ρX (the group joined: location)}. ρX is a participant, so IP-1 puts its PP in VP.
A PP: Merge-tah with ρX; Agree [3] + inverse [top] → b-; Seal → bitah.
AMerge V with the PP inside VP; v, ρA; M/S Agree ⟶ ρA [3]; Qual. Seal W = yíghááh.
B Foc above the MVS: łį́į́ʼLink ↔ ρA; Top: dibéLink ↔ ρX.
BSeal D2: C2 puts dibé ⌢ łį́į́ʼ before the MVS; inside it C1′ (IP-6) puts bitah ⌢ W (the PP is V's complement). → dibé łį́į́ʼ bitah yíghááh ✓.
ID4. NLA Q2, first clause Daʼ chidíísh bee hólǫ́ (the same -ee as a Σ-PP)
Σ Σ(‘exist’ + ‘with’, the possession unit) = {ρTh (the car that exists), ρPoss (the one it is with)}. Both are participants (IP-5).
A PP: Merge-ee with ρPoss; Agree [3] → b-; Seal → bee. Merge V with the PP in VP; M/S Agree ⟶ the slot the NLA glosses “4-”. Seal W = hólǫ́.
Bchidí in Foc, Link ↔ ρTh; da’ in Spec,CP. Seal D2: da’ ⌢ chidí ⌢ [bee ⌢ W] ✓. -ísh on chidí (first constituent, EN-2) → chidíísh ✓.
HYPOTHESIS ID1 and ID4 use the same postposition form. Only Σ differs, and that decides the PP's site (IP-5).
ID5. What PC-2 as written predicted for (9) (O0)
A as ID1 steps 1–3, but with [bee] merged in VP (grammatical = MVS).
Bdibé in Foc (above the MVS), the DP in Top. Seal: da’ ⌢ DP ⌢ dibé ⌢ [bee ⌢ W]. Predicted string Da’ DP dibé bee W ≠ printed Da’ DP bee dibé W ✗. PC-2 as written is falsified for this P.
ID6. A preverb outside W, with a Σ-PP (schematic; PV-2, §8)
Σ a verb whose Σ has a participant realized by a PP; a preverb PV (schematic).
A PP sub-cycle → [PP] sealed. W sealed without PV (PV adjoined at the clause side, PV-1(b)).
BSeal D2: PV is higher than the PP (adjoined above QualP; the PP is in VP), so C2 puts PV first, and C1′ keeps PP ⌢ W. → PV(=Q) PP W possible; PP PV=Q W impossible.
7. The deciding test (schematic)
Test
Schematic string
O1 (site by Σ)
O2 (low noun)
O4 (low focus)
Status
X1 a noun between a Σ-PP and W
N1 P-Σ N2 W (e.g. with ‘know about’, ‘among … join’)
impossible
possible (N2 the object)
possible (N2 in focus)
OPEN
X2 the instrument after the object noun
N_obj N_ins bee W
possible only with N_obj as topic (above the PP)
possible
possible
OPEN
X3 the rope sentence with two 3rd persons (yi-/bi- active)
N_A N_ins bee N_Th W
N_Th is the focus noun and is read by the voice (W&J p. 266)
N_Th is read by position in VP
as O1
OPEN
X4 inverse on the instrument's P
… yee/bee … alternating with the voice of ‘rope’
unexpected (Willie p. 224; also pp. 100, 208, opened 2026-10-22 12:46pm: the inverse has scope over the MVS's PPs)
expected (PP in VP)
expected
OPEN
HYPOTHESIS X1 is the decisive one. A single sourced X1 sentence would favour O2 or O4 over O1. If its N2 is the verb's object and not a focus, it would count against T-MID. No opened source has one: (50b), (43), Q2 and the transportation sentences S1–S13 all have the nouns before the Σ-PP.
8. Rule-book check: where the site's own rules disagree
The A/B rules now sit on more than ten pages, and many have been revised by update notes rather than rewritten. This section lists every disagreement found by reading the rule blocks side by side this run, and the reading now in force.
The CYC block on two pages still says D1 = MVS; PC-1 replaced it with a notice, not a new rule.
D1 = W; each PP is its own sealed sub-cycle; the MVS is an unsealed phrase (PC-1).
HYPOTHESIS (bookkeeping)
RB-2
CYC C1–C3 vs PC-1
Once the MVS is not a cycle, the PP and W are both sealed units ordered in D2. W is a non-head there (K3). The Σ-PP is V's complement, so it and V are sisters, and C2 (asymmetric c-command) says nothing about their order. The site's derivations kept writing “C1: PP ⌢ W”, which treats W as a head in the same domain where K3 treats it as a non-head.
IP-6 (C1′): a sealed unit precedes the unit that spells out the head it is an argument of. This is Hale's “lexical heads take their argument on the left” (SB4/K2) restated for sealed units, so no new kind of rule is added.
HYPOTHESIS (new)
RB-3
TP-6 (preverbs dah, yah are W's left edge, written apart) vs EN-1/EN-5 (a preverb that hosts -ísh is not inside W; Hale's ka-[á]sh nanin[á])
If ka and dah are the same kind of element, one of the two is wrong.
PV-1: two preverb sites. (a) Adjoined inside D1 (to M/SP or vP; Hale 2003 pp. 14–17): part of W, may be written apart, cannot host -ísh. (b) Adjoined at the clause side (or the edge of QualP, spelled with D2): its own unit, can host -ísh. PV-2 (prediction, from C2 + C1′): a preverb that hosts -ísh precedes every Σ-PP of its verb; in “PP dah W” (S1, S5) dah cannot host -ísh.
HYPOTHESIS; PV-2 OPEN
RB-4
DM-3 (dimension roles such as the path get no slot) vs DM-5 (a path can be the subject of a neuter verb, N2 nízah)
As worded, DM-3 says the path never gets a slot, while DM-5 gives it one.
IP-5: slot-hood is relative to the stem. The path is a participant of ‘be long’ and a dimension of ‘run’.
HYPOTHESIS
RB-5
PC-2 / TP-6's restatement (“nothing from rule set B between a PP and W”) vs NLA (9)–(11b)
HYPOTHESISThe current A/B statement in one place. Rule set A = CYC on each sealed word-level unit: W (heads Qual > M/S > v > V, higher first; non-heads, meaning slot exponents and D1 preverbs, before them, higher first) and each PP (pronoun ⌢ P). Rule set B = CYC on D2 over sealed units: higher first (C2), except that a unit that is an argument of a head precedes that head's unit (C1′). Clause-final particles are heads of D2 and come last (C1). Enclitics are placed after sealing, at the right edge of a sealed unit (EN-1). The changeover is sealing. Everything sealed into W is ordered by A; everything ordered among sealed units is ordered by B. A PP is a unit sealed by A-type ordering (pronoun ⌢ P) that is ordered by B, and Σ decides whether it is ordered next to W (IP-1).
9. What changes
Attached vs separate P: PC-2 is restated as IP-3 (Σ-PPs only); T45d gets its first opened candidate (NLA (9)–(11b)). PC-7's three kinds of verb + P pairing are unaffected.
Clause spine: Foc sits right above the MVS (IP-2), which may be W alone. A clause-side PP can stand between Top and Foc.
Theta-args thesis: one new datum, graded compatible, undecided (it needs IP-1). It is not counted as contradicting T-MID unless an X1 sentence turns up.
Distance as measure: DM-3 is read as relative to the stem (IP-5, RB-4). The instrument joins the path as a role in the action with no slot in W, but unlike the path it has a P slot.
What each analysis gets right and misses (on this question)
Analysis
Gets right
Misses
Willie 1991
The MVS as a unit, with PPs next to the verb in all her examples (43), (44); the grammatical/lexical split; p. 98: the split can depend on the arguments
Lists -ee Instrumental as grammatical without a verb-by-verb test; (9)–(11b) break (44a) on that listing
Willie & Jelinek 2000
The focus noun precedes the MVS ((50b), p. 270)
The p. 266 wording (“immediately preceding the inflected verb”) and the p. 270 wording come apart when a PP is present; no instrument examples
NLA Navajo Questions
A five-sentence paradigm with the enclitic in each place, with glosses and focus readings
No variant with dibé before bee, so the order's flexibility is unknown
Speas 1986
SOV as the basic order; nouns' function read off order (4.5)
Nothing on PPs here; O2 is our extrapolation
This site before today
PPs belong with the verb; the noun-linking positions are high
PC-2 was stated for every pronoun-object P, and the rope sentences were on the site from 10-20 without being checked against it; CYC's C1 was used inconsistently (RB-2)
10. Grant's frame
GRANT'S CONJECTUREThe word as the changeover point (phenomenon 1). Sharpened, not changed. The changeover is the sealing of W. What looked like a fuzzy zone (PPs written onto W, preverbs written apart, a noun between a PP and W) turns out to be three separate things. The first is prosody after sealing (PC-3). The second is a preverb with two sites (PV-1). The third is a PP whose site depends on Σ (IP-1). Each one is either inside the sealed word or ordered among sealed units, never both. That is what a single changeover point predicts.
GRANT'S CONJECTURETheta roles as a proto-role schema, not lexical items (phenomenon 2). IP-5 is the strongest form of the roles-first idea that the data allow so far. Where a participant is pronounced depends on the stem's role schema, never on the word that names the participant or on the postposition's form. The same -ee is next to the verb for the possessor of ‘exist’ and is clause-side for the rope of ‘rope’. This fits Grant's weak thesis (verbs take role-bearing slots; nouns are linked add-ons) and supports it in one respect: the noun dibé is linked to a slot of W from a clause-side position (ID1). But the same string has a configurational derivation (ID2), so the datum does not decide between them.
HYPOTHESISTheta roles and event-structure roles together. The instrument is part of how the action is done (an event-structure ingredient: the means), not one of the participants the stem ranks (theta level). It gets a postposition but no place in the verb's own phrase, as the path gets a measure phrase but no slot (DM-3). So the line between “in Σ” and “added on the clause side” falls where the theta-level participants end. That is Grant's layering, observed once more, not tested.
GRANT'S CONJECTURESTRATA (theta roles first, event structure next, noun selection later). ID1 runs in that order: Σ (two participants), W sealed, the clause-side means PP, then the nouns. It is a fit, not a separating test. A lexicalist grammar that lists the instrument as an optional adjunct derives the same string.
GRANT'S CONJECTUREWord-internal Merge or Agree sites. Nothing here is inside W. The one finding at W's edge is PV-1: a preverb can be merged inside D1 (a word-internal Merge site at the left edge, like Position I in MA-2(a)) or outside it. If PV-2 holds, the Merge site of a preverb, and not its spelling, decides whether it can carry -ísh.
GRANT'S CONJECTUREPhrase-level lexicalization. IP-1 separates two things that L9 (“phrasisized”) could blur. bee hólǫ́ ‘have’ and béé hózin ‘know about’ are Σ-units (the P's participant is part of the verb + P meaning). bee … síníloh is not a unit at all: it is a verb plus a means phrase. Only the first kind is a candidate for a stored phrase.
11. Tasks T49a–h and statuses
Task
What it needs
T49a
X1: any sourced sentence with a noun between a Σ-PP and its verb (bee hólǫ́, béé hózin, bitah yíghááh, the -ił travel pattern). One would favour O2/O4 over O1. Sources: NLA lexicons (live site down this run), Willie Ch. 3–8 page images, W&J 2000
T49b
X2: the rope sentence or another instrument sentence with the object noun before the instrument PP
T49c
X3: an instrument sentence with two 3rd persons (so that yi-/bi- applies): is the noun after bee read by the voice, as W&J's focus rule says?
T49d
X4: does the instrument's P ever alternate yee/bee with the voice of the verb (Willie p. 224: the inverse has scope over the MVS's PPs)?
T49e
PV-2: -ísh after dah/yah following a PP (predicted impossible), and ka-type preverbs with a Σ-PP (predicted before the PP). Ties to T46f, T43c
T49f
Willie's list (28) verb by verb: which of (28a–i) ever occur away from the verb, and with which verbs (is -ee Instrumental the only one?)
T49g
Speas 1990 (revised chapter) and Barss, Hale, Perkins & Speas 1989: where they put PPs relative to the object noun (does any configurational source predict PP N V?)
T49h
Rewrite the CYC blocks on A/B as spell-out and clause spine with D1 = W and C1′ in the rule text itself (RB-1, RB-2), and re-run their derivations
Statuses: T45d (anything between a grammatical PP and its verb): answered yes for an instrumental bee (NLA (9)–(11b)); PC-2 revised to IP-3; open for Σ-PPs (→ T49a). T46f: the dah/-ísh half now has a stated prediction (PV-2, → T49e). T48a–h, T46a–e, T46g, T46h: unchanged (live NLA down). T43h: not attempted. T42d/e/g: still blocked on Young & Morgan.
Willie's grammatical postpositions, verb by verb: which postpositional phrases belong to the verb complex, and the spell-out rule rewritten (T49f, T49h)
Restated on the unified structure (2026-11-15). Postpositions are restated as [PP DP P⁰] with left specifiers: a grammatical P agrees with the DP in its specifier, a lexical P has no probe (H64); a selected PP is merged in VP, and Willie’s “cannot be interrupted” follows from the default, HS-1 and left specifiers. Light d, used below, is withdrawn (Grant, 2026-11-14); light n is an OPEN structural question. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Position-class template statements below describe the order of prefixes; they are not the derivation (§3). Flat-S and other pre-VPISH notation below is labelled contrast. Status on the spine: see unified structure, §2b and §4, postposition row. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
The previous run (noun between P and verb, T49) proposed rule IP-1: a postpositional phrase (PP) is merged inside the verb's own phrase, the MVS, only when its object is one of the participants the verb's stem describes. Otherwise it is merged on the clause side. This run checks that rule against every grammatical postposition in Willie 1991's list (28a–i), one verb at a time, on the page images of her Chapters 3 and 8. The test is a diagnostic both Willie and Willie & Jelinek 2000 state: the MVS is the domain of the yi-/bi- voice alternation, so a postposition that alternates yi-/bi- with the voice is inside it. The result is a conflict with IP-1 as worded, and a revised rule, IP-1′. The second half of the page does the bookkeeping task T49h. It rewrites the site's spell-out rule CYC with the sealed verb word W as the word-level domain and fix C1′ written into the rule itself, and it re-runs the derivations that use the clause level. Two operations are kept apart throughout. Merge builds structure: it puts an element at a position. Agree values features such as person and number: a probe copies a value from the closest matching goal. Written 2026-10-22, 12:46pm PT.
Update 2026-10-22, 8:46pm (T50a, T49b–d).SOURCED data, HYPOTHESIS reading. Instrument and voice checks every third-person sentence in four NLA Postposition Lexicon lists. -ł ‘with (someone)’ alternates yił/bił (NLA 1992, 294, 2), and -ká, -chʼįʼ, -kééʼ take the y-form. The instrument -ee is bee in both 3/3 sentences found (1953; 92, whose gloss line conflicts with its translation), so P1 passes its first test, weakly: no sourced yee has been found, so a gap reading is open. IP-7 stays OPEN. A ball takes the y-form again (19 yikééʼ), so (a) holds only as worded, by postposition class. New IP-8: the postposition of a verb + P unit is the one next to W.
Ground rules (Grant's). Every Navajo form below comes from a source opened this run, or is already on this site with its source. Forms are quoted as printed, with the source's own gloss line (in Willie's own gloss style, not re-glossed beyond what she prints), and nothing is segmented beyond what the source prints. Where no sourced form exists, a test is stated with schematic forms only (capital-letter placeholders, no Navajo string). What the sources say is graded SOURCED. Rules and mappings are ours (HYPOTHESIS). Grant's own ideas are GRANT'S CONJECTURE. What needs data is OPEN. Only contradicting data count against an analysis; a source's silence does not. No analysis, the site's own included, is assumed to be fully right. Grant does not speak Navajo, and nothing here is a native-speaker judgment.
Notation (the site's own; unchanged). Σ = role-schema step (which participants the verb's stem describes, and each one's role in the action); A = word level (rule set A: the prefixal, left-headed spine, spelled out as one word W, plus the PP sub-cycle); B = clause level (rule set B: right-headed clause syntax, noun linking, particles, enclitics). W is the changeover point from A to B. MVS = Willie's “Maximal Verb-Sentence” (W&J 2000: “Maximal Verb Complex”). A ranked slot is one that the yi-/bi- voice compares with the verb's own third-person slots (§4). ρA, ρTh, ρBen, ρGoal, ρAdr (addressee), ρExp, ρCont (what is known), ρSrc, ρCom (companion), ρIns, ρPoss: role labels ours. IP-0 to IP-6: noun between P and verb §5. PC-0 to PC-7: attached vs separate P §4. CYC: A/B as spell-out §3 (now in its v2 form, §6 below).
Bottom line, in plain words
SOURCEDThree of Willie's grammatical postpositions are printed alternating with the voice, and two of those mark someone the verb on its own does not describe. ‘Lead’ with ‘for’: (85) ʼashkii ʼatʼééd łį́į́ʼ yá yoołóós ‘The boy is leading the horse for the girl’ and (86) … bá yoołóós ‘The girl is leading the horse for the boy’ (p. 95). ‘Roll’ with ‘to’: (83) ʼAwééʼ jooł yaa nímááz ‘The baby rolled to the ball’ and (84) … baa nímááz ‘The ball rolled to the baby’ (p. 95). ‘Speak’ with ‘to’: (79)/(80) (p. 94), and (12a, b) (p. 209). Willie says “the Inverse has scope over the entire MVS” (p. 100) and “over the verb word and any associated postpositonal phrases -- oblique arguments” (p. 208, her spelling), and W&J call the MVS “the domain of the voice alternation”. So these PPs are inside the MVS.
HYPOTHESISSo the previous run's rule is too narrow. You can lead a horse without leading it for anyone, and a thing can roll without rolling to anything. So the person you do it for and the place it rolls to are not participants the stem singles out (our call), yet their PPs are inside the MVS. IP-1 as worded is contradicted by sourced data under a sourced diagnostic. The revised rule IP-1′ says: a PP is inside the MVS when the voice ranks its object together with the verb's own participants; a PP whose object is never ranked (the rope you rope with) is on the clause side. The previous run's finding about bee dibé síníloh stands, because the instrument is never shown alternating.
OPENWhat decides which roles get ranked is not settled. Two candidates fit everything opened: animacy (Willie p. 100: the features of an inverse subject “are associated with referents that are high on the animacy scale”), or the kind of relation (roles that relate two participants, such as for, to, with someone, are ranked; roles that relate the action to a thing, such as the means or the place it starts from, are not). The first sourced instrument sentence with two third persons would decide a lot (T49c, T49d).
HYPOTHESISThe spell-out rule now says what the site has been doing. CYC v2 seals the verb word W at the word level, treats each PP as its own small domain, and adds C1′: a sealed unit merged inside the phrase that W spells out comes before W. Every earlier derivation gives the same string under v2. Nothing earlier on the site changes its output, and the MVS becomes a set of PP sites, not a spell-out domain.
GRANT'S CONJECTUREFor Grant's thesis: the verb complex turns out to be the domain where role slots are ranked against each other. The verb's own participants and the role slots its postpositions add are ranked there, and the nouns are not. That is a roles-first result, and nothing opened this run contradicts the weak thesis. It does not decide between Grant's view and a configurational one (§8).
The live NLA site (curl: no connection, about 12:50pm PT), so T48a, T46h, T49a could not be done
Page citations are to Willie's printed page numbers, read on the page images. The site's earlier citation “Willie p. 224” for the inverse's scope over PPs (T49d) was not re-opened; pp. 100 and 208 say the same thing and are cited here instead.
2. The list (28), verb by verb
SOURCED Willie's list (p. 41, (28); there is no (e)): a. -aa ‘to’ Dative; b. -chʼįʼ ‘to’ Goal, Allative; c. -tsʼą́ą́ʼ ‘from’ (take away from; Malefactive); d. -á ‘for’ Benefactive; f. -ee ‘with’ Instrumental; g. -ł ‘with, together’ Comitative; h. -ghá ‘through’ (penetrate through); i. -láah ‘beyond’ Comparative, Locative. “(28 a,c,d,g) occur only with animates, (28 f) occurs only with inanimates, and others occur with both.”
Columns: Stem alone? = is the P's object a participant the verb's stem singles out, judged from the English translation only (our call, HYPOTHESIS; this is what IP-1 used). Voice on P? = is a yi-/bi- pair printed on the P with the same verb (SOURCED where yes). Noun between? = is any sourced sentence printed with a noun between this PP and W.
P (28)
Verb (example)
P's object
Stem alone?
Voice on P?
Noun between?
IP-1 (T49)
IP-1′ (this page)
a -aa
nímááz ‘rolled’ (83), (84)
where it rolled to (ρGoal)
no
yes: yaa/baa
no
wrong: predicts clause side
✓ ranked → MVS
b -chʼįʼ
yáłtiʼ ‘speaking’ (26), (79), (80); (12a, b), (13a, b) p. 209
no (p. 66: the noun before yá is “not coindexed with it”)
wrong
✓ ranked → MVS
d -á
néidííłtsʼin ‘hit (with fist)’ (30)
ρBen
no
not printed
no
clause side
open (no pair)
f -ee
hózin ‘known’ (42), (43), with -ł
what is known (ρCont)
yes
not printed
no
MVS
open (no pair)
f -ee
NLA síníloh ‘rope’ (9)–(11b)
the rope (ρIns)
no
not testable (2nd-person subject)
yes: dibé
✓ clause side
✓ never ranked → clause side
f -ee
NLA hólǫ́ ‘exist’ = ‘have’ (15), Q2
the owner (ρPoss)
only as the unit ‘have’
not printed
no
MVS
open
g -ł
déʼáázh ‘dl.walk’ (29) p. 41
the companion (counted in the dual)
yes
not testable (1st person)
no
MVS
open
g -ł
nizhóní/hózhóní ‘nice’ (41); hózin (42), (43)
who likes / knows (ρExp)
yes (p. 48: “obligatory”)
not printed
no; *(41c) puts a noun first
MVS
open
h -ghá
No example found in Willie (text-layer search). OPEN
i -láah
No example found in Willie (text-layer search). OPEN
HYPOTHESISTally. Where the voice diagnostic can be applied at all (three pairs, three postpositions), all three PPs are inside the MVS, and in at least two of them the P's object is not singled out by the stem. Where a noun has been seen between a PP and W (once, the rope), the P's object is a means and no voice pair is known. No row contradicts IP-1′. Two rows contradict IP-1. Seven rows can't decide, because the subject is first or second person or no pair is printed.
3. The examples
3.1 ‘For’ with ‘lead’: three participants, and the noun next to yá is not its object
Willie 1991 p. 64, (29)
yá
yoołóós
3.for
3.3.lead [sing. +animate]
‘She is leading it for him.’
As printed. Willie: grammatical PPs “may be phonologically integrated into the MVS, or they may form the first word in the MVS as in the following Benefactive construction” (p. 64), and this MVS has “three argument positions”.
Willie 1991 p. 66, (32)
łį́į́ʼ
yá
yoołóós
horse
3.for
3.3.lead
‘She is leading the horse for him.’
Willie: “the noun preceding the inflected postposition (yá) is not coindexed with it, and cannot be governed by it”. Her principle (31), p. 65: “In an MVS with one Postpositional Phrase, the innermost nominal is coindexed with the postpositional object in the Direct Construction”, overridden here because the horse is the thing led.
Willie 1991 p. 95, (85), (86)
ʼashkii
ʼatʼééd
łį́į́ʼ
yá / bá
yoołóós
boy
girl
horse
3.for
3.3.lead
(85) yá: ‘The boy is leading the horse for the girl.’ (86) bá: ‘The girl is leading the horse for the boy.’
Willie, p. 95: “Note that the theme argument does not take part in the inversion process.” (87)/(88) p. 96 ʼazdzą́ą́ ʼatʼééd yá/bá yoołóós repeat the pattern. Willie (p. 96): “the Inverse marking functions like obviative marking”.
SOURCED So the beneficiary slot and the leader slot are what the voice compares; the theme (the horse) is left out. HYPOTHESIS In site terms: Voice ranks ρA against ρBen, not against ρTh.
3.2 ‘To’ with ‘roll’: a one-participant verb, and a goal the voice ranks
Willie 1991 p. 95, (83), (84)
ʼAwééʼ
jooł
yaa / baa
nímááz
baby
ball
3.to
3.rolled
(83) yaa: ‘The baby rolled to the ball.’ (84) baa: ‘The ball rolled to the baby.’
Glosses as printed (one person marker on the verb). Already discussed for its intransitive verb on one-place verb + P.
3.3 ‘To’ with ‘speak’
Willie 1991 p. 94, (79), (80); p. 209, (12a, b)
ʼashkii
ʼatʼééd
yichʼįʼ / bichʼįʼ
yáłtiʼ
boy
girl
3.to
3.speaking
(79) yichʼįʼ: ‘The boy is speaking to the girl.’ (80) bichʼįʼ: ‘The boy is being spoken to by the girl.’ Likewise p. 209 (12a) John Mary yichʼįʼ yáłtiʼ ‘John is talking to Mary’, (12b) John Mary bichʼįʼ yáłtiʼ ‘John is being talked to by Mary’.
3.4 ‘From’ with ‘walk’: a complex PP
Willie 1991 p. 44, (36)
dibé
baghan
bitsʼą́ą́jįʼ
yishááł
sheep
3s.home
3s.from.at
1s.walking
‘I am walking away from the corral.’
Willie treats this as a grammatical P followed by a lexical one (pp. 43–44); two grammatical or two lexical Ps cannot be stacked. Compare (33) p. 43 ʼawééʼ bichʼįʼ yishááł “baby 3.to 1.walking” ‘I am walking to the baby’ and *(34) ʼawééʼ-góó yishááł. Note that the gloss gives -jįʼ as “at” here, while the list of lexical Ps on p. 42, (30), gives -jįʼ ‘to’.
3.5 ‘With’ for an experiencer, and two PPs together
Willie 1991 p. 47, (41a); p. 48, (42); p. 49, (43)
Kee
yóóʼeelwod=ígíí
bił
béé
hózin
Kee
3.ran away=COMP
3.with
3.about
ho.known
(43) ‘He/she knows that Kee ran away.’ (42) Shił béé hózin “1.with 3.about ho-known” ‘I know it’. (41a) shił nizhóní “1sg.with 3.nice” ‘I like it’.
Willie (p. 48): the PP whose object has an Experiencer theta role “is obligatory with these ‘psych’ verbs”; (41c) *kintah shił nizhóní, but (41d) kintah-di shił hózhóní. In (43) she takes the clause as an adjunct linked to the object of béé.
SOURCED Willie p. 209, on (13b) John háísh bichʼįʼ yáłtiʼ: the wh-word “intervenes between the Inverse postpositional phrase and the nominal (“John”) that is associated with it”. That is a word between a noun and its PP, not between a PP and W. So it does not touch X1.
3.6 What Willie says the inverse covers
SOURCED p. 100 (§3.3.5): “the Inverse has scope over the entire MVS”; “with constructions containing an oblique argument, the Inverse produces a goal or benefactive subject”; the inverse subject is definite, topic and focus, and “All of these features are associated with referents that are high on the animacy scale.” p. 208 (§8.3, questions): “The Inverse voice alternation has scope over the verb word and any associated postpositonal phrases -- oblique arguments” (spelling as printed), after (11a, b) John Mary haʼátʼíísh yeiníʼą́ / beiníʼą́. p. 96: inverse marking “functions like obviative marking”.
4. Rules IP-1′, IP-5′, IP-7
IP-1′ RANK-SITE (HYPOTHESIS; replaces IP-1)
A PP is first-merged inside the verb's phrase (inside QualP, so inside the MVS) iff the role of its
object slot is one the yi-/bi- Voice can rank against W's own 3rd-person slots.
A PP whose slot is never ranked is adjoined on the clause side (above FocP), as in IP-1.
Diagnostic (Willie p. 100, p. 208; W&J 2000:270): a P printed with a yi-/bi- pair under one verb
is ranked.
IP-5′ ROLE-REL (HYPOTHESIS; revises IP-5)
Whether a slot is ranked depends on the role, not on the P's form (one -ee: owner of 'exist'
next to W, rope of 'rope' clause side) and not on the noun linked to it (p. 66: the noun before
yá is not its object).
IP-5 said "on the stem's role schema". IP-5′ says "on the ranking": the stem's own participants
PLUS the role slots its postpositions add.
IP-7 WHAT IS RANKABLE (OPEN: two candidates, both fit every row of §2)
(a) ANIMACY: a P's slot is rankable iff the P is one that takes animate objects
(Willie p. 100: the inverse subject's features go with referents high on the animacy scale).
(b) RELATION: a P's slot is rankable iff its role relates two participants
(for, to, with-someone, experiencer); means, path and source-place are not.
They come apart on (i) a goal or companion that is a thing; (ii) a beneficiary that is
inanimate; (iii) -ee for an owner (animate, but -ee is listed as inanimate-only).
HYPOTHESISPredictions. P1 (= T49d): the instrument's P never alternates yee/bee with the voice. One sourced alternating instrument would refute IP-1′ for the rope sentences, or would put the instrument back inside the MVS and make bee dibé síníloh the counterexample to Willie's (44a) after all. P2: no noun between a ranked PP and W (X1 restated: the deciding test of T49 §7 now applies to yá, yaa, yichʼįʼ). P3: a focus noun may stand between an unranked PP (a source PP like bitsʼą́ą́jįʼ) and W. Nothing opened tests P2 or P3.
OPENThe escape for IP-1. If yi-/bi- on a postposition is clause-level obviation (Willie p. 96 compares it to obviative marking) and not the MVS's voice, the diagnostic doesn't show where the PP is, and IP-1 could survive. That would mean dropping Willie's p. 100/208 scope statement and W&J's “domain of the voice alternation”. We don't do that here, but it isn't refuted.
5. Tensions in the list itself (Willie's own data)
SOURCED data, HYPOTHESIS reading. (i) -aa is listed as animate-only, but (83) has jooł ‘ball’ as its object. (ii) -tsʼą́ą́ʼ is listed as animate-only, but (36) has ‘the sheep's home’ (the corral). (iii) -ee is listed as inanimate-only, but the NLA's ‘have’ sentences use it for the owner (Q2 chidíísh bee hólǫ́; (15) Dibé nee hólǫ́). So the animacy column of (28) describes typical objects, not a constraint. That weakens IP-7(a) as a rule about the P's object, but not as a rule about the referents the voice compares (in (83) the baby is animate). Willie's “grammatical” class is a set of forms. Membership in the MVS is a matter of site, and the site depends on the verb and the role. That was the T49 reading, and the verb-by-verb data confirm it.
6. CYC v2: the rule text with D1 = W and C1′ (T49h)
The v1 text (on A/B as spell-out §3) said D1 = W, the 10-17 note moved D1 to the MVS, the 10-21 note moved it back to W, and the 10-22 4:46am note added C1′ in prose. v2 states all of that in the rule.
CYC v2 (HYPOTHESIS, ours)
For each spell-out domain D, let H(D) = the heads of D's spine and N(D) = everything else in D.
Any unit sealed in an earlier domain counts as ONE element of N(D).
C1 every x ∈ N(D) precedes every y ∈ H(D);
C1′ for sealed units x, y ∈ N(D): if x is first-merged inside a projection of a head that y
spells out (as its complement, specifier or adjunct), then x ⌢ y;
C2 otherwise, within N(D): if x asymmetrically c-commands y, then x ⌢ y;
C3 within H(D): if x asymmetrically c-commands y, then x ⌢ y.
Domains:
D1 = W: the heads Qual, M/S, v, V with the pronoun slots and preverbs merged among them; sealed.
sub = each PP (P head; its pronoun a non-head), each possessed N (N head), each relative
(-ę́ę head); each sealed before it enters D2.
D2 = the clause: C-type particles (ya’, -go, -ę́ę …) and silent Foc, Top, C as heads;
W, PPs, nouns and sealed clauses as non-heads.
The MVS is not a domain. It is the set of PP sites inside QualP (IP-1′), ordered in D2 by C1′.
COROLLARIES (all HYPOTHESIS; K1–K4 unchanged in content)
K1 Qual ⌢ M/S ⌢ v ⌢ V C3 in D1.
K2 Obj ⌢ V; pronoun ⌢ P; possessor ⌢ N C1 in D1 and in each sub-domain.
K3 [W] ⌢ ya’ / -go / -ę́ę C1 in D2.
K4 C3 never compares a head inside W with one outside it.
K5 (new) every PP sited inside QualP precedes W C1′ in D2.
K6 (new) a noun in Spec,FocP or Spec,TopP precedes every PP inside QualP and W C2 in D2.
K7 (new) a clause-side PP and a noun are ordered by height alone C2 in D2.
So: rule set A = CYC on D1 (and the sub-domains); rule set B = CYC on D2.
The changeover = the sealing of W.
HYPOTHESISWhat changed from the 4:46am C1′. The 4:46am prose said “argument of”. v2 says “inside a projection of”. That covers the beneficiary PP, which the verb does not select (Willie (29)'s third argument comes from the postposition) and which we put adjoined inside VP. Without the widening, (85) would come out unordered.
HYPOTHESISAntisymmetry. C1′ is C1 one level up: a unit inside a phrase precedes the unit that pronounces that phrase's heads. So v2 is still “Kayne's axiom within each plane, plus non-heads first” (spell-out §3), with “non-head” now also read over sealed units.
7. Derivations CV1–CV8 under CYC v2
Steps are labelled by rule set: Σ role schema, A word (W, and the PP sub-cycle), B clause. Each step names its operation: Merge (builds structure), Agree (values person/number), Move, Link (a noun linked to a slot), Seal (spell-out of a unit). VPISH: every participant slot is first-merged inside the verb's phrase. In the V-P variant the subject slot raises inside the word to Spec,M/SP; in V-H it stays. The strings are the same either way. The spine is Hale 2003's Qual > M/S > v > V, with silent Foc and Top above it (clause spine). Voice is the site's yi-/bi- head (inverse), above QualP. No flat-S tree is used. No verb is segmented beyond its source's gloss.
Σ Σ(√LEAD) = {ρA (causation, volition, movement: the leader), ρTh (moved, led: the horse)}. The beneficiary is not in the stem's Σ. Role labels ours.
AMerge V with ρTh; Merge v, ρA in Spec,v (VPISH); Merge M/S; Agree M/S ⟶ ρA [3]; Merge Qual. (Merge built the slots; Agree only copies person and number.) Seal D1 → W = yoołóós (“3.3.lead”).
A PP: Merge-á with slot ρBen [3]; Merge the PP adjoined to VP (IP-1′: ρBen is ranked, shown by (85)/(86)). The PP's pronoun is not valued yet.
BMerge Voice; Agree Voice ⟶ {ρA, ρBen} (ρTh is skipped, p. 95). Direct (85): ρA outranks → P's pronoun y-. Inverse (86): → b-. Seal the PP → yá / bá. This is the one step where the PP's sealing waits on the clause level: PC-1 already had the voice chosen in the MVS.
BMerge Foc; łį́į́ʼ in Spec,FocP; Link ↔ ρTh (horse fits ‘led’; Willie p. 66).
BMerge Top twice; ʼatʼééd in the lower Spec,TopP, ʼashkii in the higher. Link: (85) ʼashkii ↔ ρA, ʼatʼééd ↔ ρBen; (86) the reverse (Willie's translations).
BSeal D2: C2 gives ʼashkii ⌢ ʼatʼééd ⌢ łį́į́ʼ ⌢ {PP, W}; C1′ gives PP ⌢ W (the PP is inside VP, a projection of a head W spells). → ʼashkii ʼatʼééd łį́į́ʼ yá yoołóós ✓ / … bá yoołóós ✓.
Moves: 0 (V-H). HYPOTHESIS Under v1 with D1 = MVS the same string came out, but the PP had to be inside a sealed MVS, and the yes/no ending couldn't then stand between a PP and W (it does: NLA (6)). v2 gets both.
CV2. Willie (83)/(84) ʼAwééʼ jooł yaa/baa nímááz: a ranked goal with a one-participant verb
Σ Σ(√ROLL) = {ρTh (movement: the roller)}. One slot.
AMerge V, ρTh; Merge v, M/S; Agree M/S ⟶ ρTh [3] (unaccusative: the slot is first-merged low, VPISH); Merge Qual; Seal W = nímááz (“3.rolled”).
A PP: Merge-aa with ρGoal [3]; adjoined to VP (IP-1′).
BAgree Voice ⟶ {ρTh, ρGoal}: direct (83) y-, inverse (84) b-; Seal PP → yaa / baa.
BMerge Top ʼAwééʼ, Foc jooł; Link (83) baby ↔ ρTh, ball ↔ ρGoal; (84) the reverse.
Moves: 0. HYPOTHESIS The voice here ranks a stem slot against a P slot only, which is why the inverse is possible with a one-place verb (one-place verb + P).
Σ Σ of the unit [-ł … -ee … hózin] = {ρExp (sentience: the knower), ρCont (what is known)}. Both are P slots. W's own subject is the areal ho- (gloss “ho.known”; pleonastic psych).
AMerge V, v, M/S, Qual with ho-; Seal W = hózin.
A PP1: Merge-ł with ρExp [3]; PP2: Merge-ee with ρCont [3]. Both inside QualP, PP1 higher (our assignment: the experiencer outranks the content). Agree each P ⟶ its slot [3]; Seal → bił, béé. Voice: no pair printed, so the voice value of the b- is not tested (OPEN).
B The relative-type clause Kee yóóʼeelwod=ígíí is its own derivation, sealed (its =ígíí is its head, C1 → last). Merge in Spec,TopP; Link ↔ ρCont (Willie: adjunct to the object of béé).
Moves: 0. HYPOTHESIS The order of two PPs inside QualP is height (C2), not C1′, because C1′ only orders a unit before the unit that spells its host. The order bił ⌢ béé follows only if the experiencer is the higher slot. That is an assumption here, not a result.
AMerge V, v, M/S; Agree M/S ⟶ ρTh [1sg]; Seal W = yishááł (“1s.walking”).
A Sub-domain (possessed N): possessor dibé, N ghan with b- (C1: possessor ⌢ N) → dibé baghan.
A Sub-domain (PP): Merge-tsʼą́ą́ʼ with ρSrc [3] → bitsʼą́ą́ʼ; Merge lexical -jįʼ over it (Willie pp. 43–44); Seal → bitsʼą́ą́jįʼ.
B Site: IP-7(b) puts the source clause side; IP-7(a) is undecided (the P is listed animate-only, the object is a place). The 1st-person subject leaves the voice idle, so §2's diagnostic is silent. Both sites are derived: clause side adjoined above FocP, or inside VP.
BMerge the possessed N as the PP's object or as a linked noun (Willie's gloss doesn't decide); Link ↔ ρSrc. Seal D2: either C2 or C1′ gives dibé baghan ⌢ bitsʼą́ą́jįʼ ⌢ W ✓.
Moves: 0. OPEN The two sites give one string here. They would come apart if a focus noun were added (P3).
As ID1 on noun between P and verb, with two changes in the reason, not the output. (i) The site is set by IP-1′ (ρIns never ranked) instead of IP-1. (ii) In the final seal, bee is clause side, so not inside any projection W spells, and C1′ is silent. C2 then gives da’ ⌢ DP ⌢ bee ⌢ dibé ⌢ W ✓. B
B Site (a) inside VP: C1′ nee ⌢ W; dibé in Spec,FocP, C2 dibé ⌢ nee. Site (b) clause side above FocP: C2 gives nee ⌢ dibé unless dibé is in Spec,TopP. The printed order dibé nee hólǫ́ is derivable either way, so (15) doesn't decide the site (a second-person owner leaves the voice idle).
Bya’ is a D2 head; C1 → last. Output Dibé nee hólǫ́, yaʼ ✓ (the comma is prosody).
B Unchanged. N7 is three nested sub-domains, each ordered by C1. N8 has no PP, so the two nouns are ordered by C2 and both come before W, since they c-command its heads. C1′ never applies.
CV8. Spell-out N1–N5 (Hale 2003 words)
A Unchanged. They are D1-internal. C1′ orders sealed units only, and nothing inside W is sealed. The case-realizing causative P of N5 is a head inside W (PC-4), not a sealed PP.
Status table
Item
Status
Basis
Inverse scope covers grammatical PPs
SOURCED
Willie p. 100 (“entire MVS”; goal or benefactive subjects), p. 208 (“any associated postpositonal phrases”); W&J 2000:270
yi-/bi- pairs on -á, -aa, -chʼįʼ
SOURCED
(79)–(88); (12a, b)
IP-1 (Σ decides the site), as worded
HYPOTHESIS, contradicted by (83)–(86) under Willie's diagnostic
§2; survives only by the obviation escape (§4)
IP-1′ (ranking decides the site)
HYPOTHESIS, no contradicting row
§2, §4
IP-5′ (role, not P form or noun)
HYPOTHESIS
same -ee twice; p. 66
IP-7 (what is rankable)
OPEN
(a) and (b) both fit
X1 (noun between a ranked PP and W)
OPEN
none in Willie Ch. 3, Ch. 8
Animacy column of (28) as a constraint
HYPOTHESIS: a tendency only
(83), (36), NLA Q2 (§5)
-ghá, -láah
OPEN
no examples
CYC v2 (D1 = W, C1′ widened)
HYPOTHESIS
CV1–CV8 all ✓
What each analysis gets right and misses (on this question)
Analysis
Gets right
Misses
Willie 1991
The inverse's scope over the PPs, with pairs for three Ps; “three argument positions” for the benefactive; the noun before yá not governed by it (p. 66)
The animacy column of (28) has counterexamples in her own data; no examples for -ghá, -láah; no instrument with two third persons
Willie & Jelinek 2000
MVS = the domain of the voice alternation, which is the diagnostic used here
Silent on which Ps are in that domain with which verbs
Speas 1986 (configurational)
Not tested on this question; nothing here contradicts it
No account of a PP slot that the voice ranks
This site, T49 (IP-1)
A P's site depends on the verb, not the P's form; the rope sentences
“Participant the stem describes” is too narrow: the beneficiary of ‘lead’ and the goal of ‘roll’
This site, CYC v1
The orders, every time
Its text disagreed with its own update notes; C1′ in prose didn't cover adjoined PPs
8. Grant's frame
GRANT'S CONJECTURETheta roles as a proto-role schema, not lexical items. Strengthened in one respect. The verb complex is where role slots are ranked against each other: the stem's own and those its postpositions add. Nouns are linked from outside it, and Willie p. 66 says a noun next to yá needn't be its object. So what the voice operates on is role slots, not words. That fits the weak thesis (verbs take role-bearing slots; nouns are linked add-ons). Nothing opened this run contradicts it. A configurational analysis can restate the same facts with nouns in argument positions, so this is a fit, not a separating test.
HYPOTHESISThe layers. The site's STRATA order (theta roles first, event-structure roles next, noun selection later) gets a refinement. Between the stem's theta roles and the event-structure ingredients (means, path) there is a ranking layer that includes the P-added participants (beneficiary, goal, addressee). The means and the path sit outside it, the nouns later still. This is an observation from eleven rows, not a test.
GRANT'S CONJECTUREThe word as the changeover point. Unchanged and now written into the rule (CYC v2): the changeover is the sealing of W. The MVS is not a second word-like seal. It is a set of sites ordered at the clause level.
GRANT'S CONJECTUREWord-internal Merge or Agree sites. Nothing new inside W. One Agree step happens across the seal: Voice values the pronoun of a PP that is sealed separately (CV1 step 4). That is an Agree site that spans the word boundary without being inside the word.
GRANT'S CONJECTUREPhrase-level lexicalization.bił … béé hózin ‘know’ and bee hólǫ́ ‘have’ are verb + P units whose P slots carry the core roles. yá yoołóós is not a unit (the ‘for’ is free), yet its PP is in the MVS. So “inside the MVS” doesn't mean “lexicalized together”.
9. Tasks T50a–h and statuses
Task
What it needs
T50a
A sourced yi-/bi- pair (or its absence across many examples) for -ee Instrumental, -tsʼą́ą́ʼ, -ł Comitative: decides IP-7 (a) vs (b). Sources: NLA lexicons (down), Young & Morgan
T50b
P2: a noun between yá/yaa/yichʼįʼ and W (predicted impossible)
T50c
P3: a focus noun between an unranked source PP and W (predicted possible)
T50d
Examples for -ghá ‘through’ and -láah ‘beyond’
T50e
Is yi-/bi- on a P the MVS's voice or clause-level obviation (Willie p. 96)? Decides whether IP-1 can be saved
T50f
The order of two PPs inside QualP: bił béé (43); any pair with the order reversed
T50g
Clause spine: rewrite its MVS-as-D1 text to v2 (only an update note was added this run)
T50h
The beneficiary's site: VP-adjoined (here) vs an applicative head; Willie's “three argument positions”
Statuses: T49f: done for every P of (28) Willie exemplifies (two have no examples, → T50d). T49h: done on A/B as spell-out; the clause spine has an update note only (→ T50g). T49d: sharpened as prediction P1 (→ T50a). T49a: restated as P2 (→ T50b). T49b, T49c, T49e, T49g: unchanged. T48a–h, T46h: not attempted (live NLA down).
Does the voice reach the instrument? Third-person sentences from the NLA postposition lists, checked one by one (T50a, T49b–d, with T50b, T50f)
2026-11-11 audit pass 2. Apostrophe fixes only. Navajo glottal stops typed as ' or ’ (for example in prefix names like ʼa- and ʼeʼe-, or words like Awééʼ) now print as ʼ (U+02BC), and a build step no longer turns the closing quote of an English translation inside a Navajo row into a glottal stop (‘…the girls.’, not ‘…the girls.ʼ). No form, gloss or translation wording changed.
2026-11-11 audit pass 6. The headline said the instrument bee occurs “in both 3/3 sentences found”. Only entry 1953 is 3/3 on both its gloss line and its translation. Entry 92 is 3/3 only by its translation: its gloss line “3-1-hit.P” marks a first-person subject (already flagged lower on the page). The headline now says so, and the 3/3 status of 92 is OPEN (T51a). See the pass 6 report.
Restated on the unified structure (2026-11-15). Postpositions are restated as [PP DP P⁰] with left specifiers: a grammatical P agrees with the DP in its specifier, a lexical P has no probe (H64); a selected PP is merged in VP, and Willie’s “cannot be interrupted” follows from the default, HS-1 and left specifiers. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Flat-S and other pre-VPISH notation below is labelled contrast. Status on the spine: see unified structure, §2b and §4, postposition row. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
The previous run (postpositions by verb) replaced IP-1 with IP-1′: a postpositional phrase (PP) is inside the verb complex, the MVS, when the yi-/bi- voice ranks its slot against the verb's own third-person slots. It predicted (P1) that the instrument's postposition -ee never alternates yee/bee with the voice, and it left open which roles can be ranked (IP-7). Willie 1991 has no instrument sentence with two third persons. This run takes every example sentence in four lists of the Navajo Language Academy (NLA) Postposition Lexicon (instrumental, accompaniment, benefactive, goal: 138 entries, about 120 different sentences), read from Wayback Machine captures because the live NLA site is down. It picks out the sentences where a third person does something to, for or with another third person, and records which form each postposition takes. The voice can only show up in those sentences. Two operations are kept apart. Merge builds structure: it puts an element at a position. Agree values features such as person and number: a probe copies a value from the closest matching goal. Written 2026-10-22, 8:46pm PT.
Ground rules (Grant's). Every Navajo form below comes from a source opened this run, or is already on this site with its source. Forms are quoted as the NLA displays them, with the NLA's own gloss line, not re-glossed. Nothing is segmented beyond what the gloss line shows. Where the gloss line and the translation disagree, this is flagged, and neither is corrected. Where no sourced form exists, a test is stated with schematic forms only (capital-letter placeholders, no Navajo string). What the sources say is graded SOURCED. Rules and mappings are ours (HYPOTHESIS). Grant's own ideas are GRANT'S CONJECTURE. What needs data is OPEN. Only contradicting data count against an analysis; a source's silence does not. No analysis, the site's own included, is assumed to be fully right. Grant does not speak Navajo, and nothing here is a native-speaker judgment.
Notation (the site's own; unchanged). Σ = role-schema step (which participants the verb's stem describes, and each one's role in the action); A = word level (rule set A: the prefixal, left-headed spine, spelled out as one word W, plus the PP sub-cycle); B = clause level (rule set B: right-headed clause syntax, noun linking, particles, enclitics). W is the changeover point from A to B. MVS = Willie's “Maximal Verb-Sentence”. A ranked slot is one the yi-/bi- voice compares with the verb's own third-person slots. 3/3 sentence = a sentence whose subject and whose postposition's object are both third persons, the only kind in which the voice can show on the postposition (W&J 2000 p. 267, on the site since T49). y-form / b-form = the postposition with y(i)- or b(i)- for its third-person object. On Willie's account (pp. 94–96, 100) the y-form is the direct form, used when the subject outranks the postposition's object; the b-form is used otherwise. ρA, ρTh, ρIns, ρCom (companion), ρBen, ρGoal, ρX: role labels ours. IP-1′, IP-5′, IP-7, P1–P3: postpositions by verb §4. X1–X4: noun between P and verb §7. CYC v2, C1′, C2: postpositions by verb, CYC v2.
Bottom line, in plain words
SOURCEDFour postpositions that mark a person take the y-form in 3/3 sentences, and one of them, ‘with (someone)’, is printed in both forms. ‘For’ yá (entries 13, 489, 249), ‘to’ yaa (249), ‘after, to get’ yíká (17), ‘to’ yichʼįʼ (675), ‘behind, following’ yikééʼ (19), and ‘with’ yił (1992, 294). The b-form bił appears in 3/3 sentences where the postposition's object is named first: Jáan Mary bił baghan ‘Mary lives with John’ (2), and the second clause of 294, in the same sentence as yił. So -ł ‘with (someone)’ alternates with the voice, like ‘for’ and ‘to’ in Willie.
SOURCEDThe instrument ‘with (a thing)’ is bee in both sentences found, never yee. Only 1953 is 3/3 on both lines; 92 is 3/3 by its translation, but its gloss line “3-1-hit.P” marks a first-person subject OPEN (T51a). Tsinaabąąs bee nidaʼabąąsgo … ‘when wagons were being used’ (1953; gloss “pl-3-drive”), and Shimá sání lééchą́ą́ʼí tsin bee nidiiłhaal ‘My grandma hit the dog with a stick’ (92). No yee for the instrument occurs in any of the four lists.
HYPOTHESISThat is what the site's rule predicted: the person you do something for, to or with is ranked by the voice, and the thing you do it with is not. The y-form would be expected in both sentences if the instrument were ranked: the drivers and the grandmother outrank a wagon and a stick on any scale. So prediction P1 survives its first test and the rope result (bee dibé síníloh, T49) stands. But it is weak. It rests on two sentences, and in one of them (92) the NLA's gloss line says the subject is first person (“3-1-hit.P”) while its translation says it is the grandmother. A second explanation is not ruled out either: the instrument postposition might simply have no y-form. The only claim found that it has one is a popular website (“bee (or yee – the other 3rd person form)”), with no example.
SOURCEDThe object can come before or after the instrument.… lééchą́ą́ʼí tsin bee nidiiłhaal (dog, stick, with, hit; 92) and Béésh ʼáłtsʼózí beeʼatsidí bee ʼaháníłneʼ ‘I beat the wire in two with a hammer’ (90, Young & Morgan 1987:41, via the NLA) put the object first. The rope sentences put it after. That answers T49b, but it doesn't separate the analyses: all of them can derive both orders (§5).
HYPOTHESISNothing but a small word ever stands between a postposition that marks a person and the verb. In all about 120 sentences, what stands there is nothing, a preverb (dah), an adverb (yaa ‘down’, hashtʼeʼ ‘ready’), another postposition, or once the quantifier łaʼ ‘some’, after ‘with’ whose object is bread (986). There is no noun. So prediction P2 is not contradicted. These are not judgments, though, so it isn't confirmed either.
HYPOTHESISWhen two postpositions come before the verb, the one that makes up a fixed unit with the verb is the one next to it. Examples: yá ʼataʼ halneʼ (for, between, talk) ‘is translating for’ (13), yaa bił hoozhǫǫhgo ‘happy about it’ with the unit bił hózhǫ́ ‘be happy’ (1958; 792), and shíká bił ʼoolwoł ‘is driving to get me’ with the ‘with’ of the driving pattern (15). Willie's bił béé hózin fits too, and the NLA writes that last one as a single word, bił bééhózin. There are four cases and no counterexample. It fits Grant's point that a verb and a postposition can form a unit at the phrase level.
OPENWhat makes a role rankable is still not settled. A ball gets the y-form a second time (yikééʼ ‘after it’, chased by a dog, 19; Willie (83) had jooł yaa). So a version of the animacy candidate that looks at what the postposition's object refers to is ruled out a second time (the previous run had already set it aside). What is left is the version that looks at the postposition itself (ranked if it is one of Willie's animate-object postpositions), which says nothing about -kééʼ, since that is not on her list. The other candidate, that roles relating two participants are ranked and roles relating the action to a means are not, fits everything here. Neither is decided.
GRANT'S CONJECTURE Grant's weak thesis, that verbs take role-bearing slots and nouns are optional add-ons linked to them, gets one more fit: the voice ranks role slots, and the instrument's slot sits outside the ranking. Nothing opened this run contradicts it, and nothing separates it from a configurational account (§8).
1. Sources opened this run
Source
Access
Used for
Navajo Language Academy, Navajo Postposition Lexicon (Swarthmore Talking Dictionary, version 10.2023 as displayed), semantic lists instrumental, accompaniment, benefactive, goal
Wayback Machine captures, saved to the box as HTML and parsed line by line: instrumental, captured 2026-06-16 7:54am PT; accompaniment (“50 entries”), 2026-05-21 7:53am PT; benefactive (“41 entries”), 2025-01-25 9:08am PT; goal, 2026-06-15 1:25pm PT. The malefactive and source lists have no captures.
All 138 entries: each sentence as displayed, its gloss line, translation, the postposition the database highlights, and the source (often “YM 1987:page”) and speaker. Entry numbers are the NLA's.
navajowotd.com, “bee | Navajo Word of the Day” (post of 2013-01-09)
Web page
The statement “bee (or yee – the other 3rd person form)”. A popular source, not a grammar. Used only as a pointer (§3.4).
Creamer 1974 (box scan, printed p. 34, page image)
Page image (re-read)
Only as a lead: the name for ‘elephant’ in a sentence on p. 34 contains the string yee (§3.4). Not segmented.
The live NLA site (TLS error at about 8:50pm PT; talkingdictionary.org timed out), so T48a, T46h and T49a could not be done, and single entries could not be listened to or re-checked. Young & Morgan 1987 itself was not opened: its page numbers are cited as the NLA gives them.
The NLA's displayed spacing sometimes splits words at its database link spans. Where a spacing question matters, the form is quoted as it appears in the gloss-aligned sentence line. The gloss lines are the NLA's and contain visible slips (e.g. “3-for” under bee in 92). They are flagged, not corrected.
2. The sweep: which postposition form appears in each 3/3 sentence
SOURCED The rows are every sentence in the four lists in which both the subject and the postposition's object are third persons, going by the gloss line and the translation. Sentences with a first- or second-person subject or postposition object can't show the voice (W&J p. 267) and are not listed. They are the large majority, including every other instrument sentence (89, 90, 78, 116, 565, 914, 1022).
Entry
Postposition
Form
Subject
P's object
Which is named first
Reading
13
-á ‘for’
yá
Mary
her grandmother
subject
direct
489 (YM 1987:311)
-á ‘for’
yá
Mr. Long
our cattle
subject
direct
249 (YM 1987:180)
-á ‘for’; -aa ‘to’
yá; yaa
my little brother
his puppy
subject
direct (the gloss line has no gloss for yá)
17
-ká ‘after, for’ (not in Willie's list (28))
yíká
s/he (no noun)
Mary
the P's object (only noun)
direct
675
-chʼįʼ ‘to’
yichʼįʼ
the teacher
John or Mary
subject
direct
19 (YM 1987:120)
-kééʼ ‘behind, following’ (not in Willie's list (28))
yikééʼ
he (my dog)
it (the ball)
neither named in this clause
direct; the P's object is a thing
1992
-ł ‘with’
yił
the little boy
his mother
subject
direct
294 (YM 1987:750), main clause
-ł ‘with’
yił
an old man
the children
subject
direct
294, ‘pile’ clause
-ł ‘with’
bił
the children
him (the old man)
the P's object (the old man, first word)
inverse-like (“Lit. the children are piled with him”)
2
-ł ‘with’
bił
Mary
John
the P's object
inverse-like (Willie's pattern for b-)
site S2
-ł ‘with’
bił
lightning
Monster Slayer
the P's object
inverse-like (already on the site)
1953
-ee ‘with’, instrument
bee
they (gloss “pl-3”; translation impersonal)
wagons
the P's object (only noun)
b-form where a ranked slot would give the y-form
92
-ee ‘with’, instrument
bee
my grandma (translation) / 1st person (gloss line “3-1-hit.P”)
a stick
subject (translation reading)
b-form where a ranked slot would give the y-form, if the translation is right
2029 (YM 1987:672)
-ee
bee
3 (gloss “3-sheds.light.P”)
the fire
the P's object
not decisive: two things, the P's object first
247 (YM 1987:12)
-á ‘for’
bá
the telephone poles
(the phone line; not named)
—
not decisive: two things, loose translation
552 (YM 1987:353)
-á ‘for’
bá
good jobs (exist)
the educated people
the P's object
b-form expected on any ranking (a thing subject)
775 and others
-ee owner (‘have’)
bee (vs nee ‘with you’)
a car (exists)
s/he (the owner)
—
b-form expected on any ranking (a thing subject); not decisive
52 (YM 1987:59)
-ł ‘with’
yił
unclear (my father or a man)
unclear
—
unclear: reciprocal ‘fight’, the translation is garbled
1958
-aa and -ł, two Ps
yaa bił
an areal ‘it’ of ‘be happy’
the bracelet-giving; my uncle
—
unclear (the gloss line has no gloss for either P)
HYPOTHESISTally. Seven postpositions or uses have a decisive 3/3 sentence. Six of them take the y-form at least once: ‘for’, ‘to’ (two Ps), ‘after’, ‘behind’, ‘with (someone)’. The instrument is the seventh, with two b-forms and no y-form. For -ł, both forms occur under the same pattern Willie gives for -á and -aa: y- with the subject named first, b- with the postposition's object named first. This is T50a for -ł and -ee. -tsʼą́ą́ʼ ‘away from’ has no 3/3 sentence in these lists, and the malefactive list was never captured.
3. The sentences
3.1 The instrument with two third persons (T49c)
NLA Postposition Lexicon, instrumental list, entry 1953 (speaker Ron Gene; no printed source given)
(1953)
Tsinaabąąs bee nidaʼabąąsgo shiʼdizhchį́.
wagon 3-with pl-3-drive.I-GO 1-pass-born.P
‘I was born when wagons were being used.’
The voice can show only on the first part (Tsinaabąąs bee nidaʼabąąsgo), whose subject the gloss line marks third person plural. The highlighted P is bee.
The database highlights tsin bee as the PP and files the entry under instrumental. Two slips in the gloss line: “3-for” under bee (elsewhere “3-with”), and “3-1-hit.P”, a first-person subject, against a translation with the grandmother as subject. If the gloss line is right, the sentence is not 3/3 and tells nothing about the voice. The site's rule against segmenting beyond the source means we can't settle this from the verb form.
HYPOTHESIS Both sentences have the b-form. In each, if the instrument's slot were ranked, the subject (people who drive; a grandmother) would outrank the postposition's object (wagons; a stick), which would give the direct y-form yee. Willie (83) shows that a thing as the postposition's object does not block the y-form: ʼAwééʼ jooł yaa nímááz ‘The baby rolled to the ball’. So the b-form here is not explained by the object being a thing.
man old INDEF children 3-with 3-pile.NI-COMP 3-with 3-tell.CI
‘An old man is telling stories to a pile of children. (Lit. the children are piled with him)’
One sentence, two clauses, the same postposition in both forms: bił in the clause whose subject is the children and whose ‘with’-object is the old man; yił in the clause whose subject is the old man.
The postposition's object (John) is named first and the subject (Mary) second, with bił. That is Willie's inverse pattern for (84) and (86). Caveat: ‘live with’ is symmetric, so the translation can't show which reading the speaker meant.
‘When I’m out herding with my dog I bat fly balls for him and he chases them.’ (YM 1987:120; Wallace Dale)
In (19) only the last clause is 3/3: the dog runs after the balls, yikééʼ. The NLA glosses -kééʼ here “3-after”, and its list of postpositions gives “-kééʼ behind, following”. Entry 249 (YM 1987:180, … náneeskaadí łaʼ yá yiidiyiisdláád dóó yaa yiizhjaaʼ ‘my little brother tore off pieces of tortilla and gave them … to his puppy’) is quoted only in part because its gloss line skips yá.
3.4 Is there a yee at all?
OPEN No sentence in the four lists has yee as the instrument. The string yee’ occurs in the NLA material only as an adverb, ‘very’. navajowotd.com says of bee: “bee (or yee – the other 3rd person form)”, with no example. On Creamer 1974 p. 34 (page image) the name given for the elephant, Chį́į́h yeeʼidlohii (as printed), contains the string yee. Whether that is a y-form of -ee inside a name is not something we can tell without segmenting, which we don't do.
HYPOTHESIS So the instrument data have two readings. (i) -ee has a y-form, and the voice doesn't reach the instrument (P1, IP-1′). (ii) -ee has no y-form, so its b-form tells us nothing. A single sourced yee with the meaning ‘with it, by means of it’ in a direct 3/3 sentence would refute (i) for that verb. A sourced yee anywhere, even under another verb, would rule out (ii) (T51b).
3.5 Object before the instrument (T49b)
NLA Postposition Lexicon, instrumental list, entries 90/255 (source YM 1987:41; speakers Wallace Dale, Lorene B Legah)
(90)
Béésh ʼáłtsʼózí bee ʼatsidí bee ʼaháníłneʼ.
wire hammer 3-with 3-1-beat.P
‘I beat the wire in two with a hammer.’
As displayed, with a space inside the word the gloss line treats as one (“hammer” = bee ʼatsidí, which the database also highlights). The order is object, instrument noun, bee, verb. 92 has the same order. The rope sentences, NLA (9)–(11b), have instrument noun, bee, object, verb.
3.6 What stands between a postposition and the verb (T50b)
SOURCED A script listed the word after every pronoun-bearing postposition in all about 120 sentences. Apart from the verb itself, what follows the postposition is: the preverb dah (475, 478, 525); the adverb yaa ‘down’ (71, glossed “down”); hashtʼeʼ ‘ready’ (457); daatsʼí ‘perhaps’ after a PP that ends in -go (665); another postposition (13, 15/16, 150, 1958); and once a quantifier:
NLA Postposition Lexicon, accompaniment list, entry 986 (sources YM 1987:779, 781; speaker johnn harvey, as displayed)
after.a.while-GO bread blue 3-1-eat.F and coffee 3-with some 3-1-drink.F
‘In a while, I will eat blue corn bread and drink some coffee with it.’
The postposition's object is the bread (a thing), and the subject is first person, so the voice can't show. łaʼ ‘some’ goes with the coffee, whose noun comes before bił.
HYPOTHESIS No noun stands between any ranked (y-form) PP and the verb, so P2 has no counterexample. But the sentences weren't built to test it, so silence is all it is (P2 stays a prediction). łaʼ after bił is a part of the object between a -ł PP and the verb. If -ł with a thing were ranked, which is what IP-7(a) on the P-class reading says, this would be the first sign against P2, unless łaʼ is a low quantifier like a preverb. Under IP-7(b) (a thing as companion is not ranked) it is expected. It is recorded, not counted (T51e).
3.7 Two postpositions before one verb (T50f)
NLA Postposition Lexicon: benefactive 13, benefactive 1958, accompaniment 792 (YM 1987:459), benefactive and goal 15/16; accompaniment 684 and 897; Willie 1991 (43) on the site
‘My uncle made a bracelet and cheerfully gave it to his mother.’ (Johnny Harvey; spacing and the spelling bima as displayed)
(792)
Nił hǫ́zhǫ́ǫ́sh doodaiiʼ doósh nił hózhǫ́ǫ da?
2-with-happy.NI-Q or neg-Q 2-with happy.NI neg
‘Are you happy or not happy?’ (YM 1987:459; Lorene B Legah)
(15)
Naʼnízhoozhídę́ę́ʼ shíká bił ʼoolwoł.
Gallup-from 1-for 3-with 3-drive.I
‘S/he is driving from Gallup to get me.’ (Wallace Dale)
(684)
Ni daatsʼí nił bééhózin ʼéí doodaiiʼ nicheii daatsʼí bił bééhózin?
2 perhaps 2-with 3-know.CI TOP or 2-grandfather perhaps 3-with 3-know.CI
‘Do you know or does your grandfather know?’ (Lorene B Legah)
HYPOTHESIS In each pair, the postposition next to the verb is the one that belongs to a verb + P unit. ʼataʼ + ‘talk’ = ‘translate’ (13). -ił + ‘be pleasant’ = ‘be happy’ (792, then 1958 with yaa ‘about it’ further out). -ił + ‘drive’ is the travel pattern of -ił travel pattern (15, with -ká ‘to get’ further out). -ee + ‘be known’ = ‘know’ (Willie (43) bił béé hózin, which the NLA prints as bił bééhózin, with the -ee inside the verb word). The other postposition (‘for’, ‘about’, ‘to get’, the knower) stands further out. That makes four cases with no counterexample. The fifth two-P sequence, bá ʼahá níjaaʼ (150, ‘divided equally between’), is not counted, because it isn't clear which of the two forms a unit. The gloss line of 1958 does not gloss yaa or bił separately. Reading them as ‘about it’ and the ‘be happy’ unit is ours.
4. Rules: what changes
IP-1′ RANK-SITE unchanged. New support: -ł (both forms, 1992/294/2), -ká, -ch’į’, -kéé’ (y-forms).
No contradicting row. HYPOTHESIS
P1 (= T49d) "the instrument's P never alternates yee/bee with the voice"
First test passed: bee in both 3/3 instrument sentences (1953; 92 as translated).
Weak: two sentences, one with a conflicting gloss, and a possible gap reading
(no sourced yee). HYPOTHESIS
P2 no noun between a ranked PP and W. No counterexample in about 120 sentences.
Not a judgment test. HYPOTHESIS
IP-7 WHAT IS RANKABLE still OPEN.
(a) only as worded (P-CLASS): a P's slot is rankable iff the P is one of the animate-
object Ps of Willie (28). A reading by the P-object's animacy is contradicted again
(Willie (83) jooł yaa, already noted; now NLA 19 yikéé' with a ball).
(b) RELATION: fits every row (the ball is a participant of 'chase' and 'roll to';
the wagon and the stick are means).
They come apart on -kéé' (not in Willie's list (28)) and on -ł with a thing (986, untestable: 1st person).
IP-8 UNIT-INNER (new) When two PPs precede W, the PP that forms a verb + P unit with the stem
is first-merged lower (V's complement), and the other PP is adjoined higher.
C2 (higher first) then orders the outer PP first, and C1′ puts both before W.
Rests on 13, 1958 (+792), 15, Willie (43). No counterexample. HYPOTHESIS
HYPOTHESISWhy the instrument gets the b-form under IP-1′. The clause-side PP is adjoined above FocP, and so above Voice, which sits above QualP. Voice can only Agree with slots in its own domain, so it never sees ρIns. The postposition's pronoun is valued by the P's own Agree with its slot ([3]), and with no ranking value it is spelled with the default b-. That is one mechanism for three facts: bee in a 3/3 sentence, the noun between bee and W in the rope sentences, and no y-form.
5. Object before or after the instrument (T49b): do the analyses come apart?
NTh in a Spec,TopP above the PP ✓ (X2: “possible only with Nobj as topic”)
NTh scrambled out of VP past the PP ✓
✓
HYPOTHESIS Both orders are attested, and every option derives both. X2 is now SOURCED as an order, but the topic reading it needs under O1 can't be checked: the NLA gives no context, and neither sentence has an enclitic that marks a topic. X3, the rope order with two third persons, is still not found. It would show whether the noun after bee is read by the voice (W&J p. 266).
6. Derivations IV1–IV4
Steps are labelled by rule set: Σ role schema, A word (W, and the PP sub-cycle), B clause. Each step names its operation: Merge (builds structure), Agree (values person/number), Move, Link (a noun linked to a slot), Seal (spell-out of a unit). VPISH: every participant slot is first-merged inside the verb's phrase. The spine is the site's Qual > M/S > v > V, with Voice (the site's yi-/bi- head, inverse) above QualP and silent Foc and Top above it. No flat-S tree is used. No verb is segmented beyond the NLA's gloss line. The verb words are sealed as wholes, and their internal steps are only as detailed as the gloss shows.
IV1. NLA 92 Shimá sání lééchą́ą́ʼí tsin bee nidiiłhaal, on the translation's reading: an unranked instrument
Σ Σ(√HIT) = {ρA (causation, volition: the hitter), ρTh (contacted, affected: the dog)}. The stick is entailed only as “some means”, so it is not in Σ. Role labels ours.
AMerge V with ρTh; Merge v, ρA in Spec,v (VPISH); Merge M/S; Agree M/S ⟶ ρA (Merge built the slot; Agree copies its person and number). Merge Qual. Seal D1 → W = nidiiłhaal, as printed. The gloss line's “3-1” is the open question, so the person value is left as printed.
A PP sub-cycle: Merge-ee with its slot ρIns; Agree P ⟶ ρIns [3]. No ranking value arrives, so the pronoun is spelled with the default b- form. Seal → bee.
BMerge Voice above QualP; Agree Voice ⟶ {ρA, ρTh}, the only third-person slots in its domain. The PP is not there yet.
BMerge Foc (empty). Merge the PP [tsin bee] adjoined above FocP (IP-1′: ρIns is not ranked). tsinLink ↔ ρIns inside the PP's phrase.
BMerge Top twice: lééchą́ą́ʼí in the lower Spec,TopP, Link ↔ ρTh; Shimá sání in the higher, Link ↔ ρA (translation).
Moves: 0. HYPOTHESIS The b-form follows from the PP's site: Voice's Agree can't reach a PP merged above it. Under O2, or with the PP in VP, Voice's domain contains ρIns, and with ρA outranking a stick the direct form yee is predicted ✗. That is the X4 test, and on this sentence it goes against the PP-in-VP option, as long as the translation's subject is right.
Σ Σ(√WALK.du) = {ρA (movement, volition: the walker)}. The companion's slot is added by -ł: ρCom. Whether the dual on W counts the companion is not settled here.
AMerge V, v with ρA (VPISH), M/S, Qual; Agree M/S ⟶ ρA [3, du as glossed]. Seal D1 → W = yiʼash (“3-walk.du.Prog”).
A PP: Merge-ił with ρCom [3]; Merge the PP adjoined to VP (IP-1′: ρCom is ranked, shown by 294/2). Pronoun not yet valued.
BMerge Voice; Agree Voice ⟶ {ρA, ρCom}: direct (the boy outranks, as translated) → y-. Seal the PP → yił.
BMerge Foc (empty); Merge the goal PP [kingóó] (noun + -góó, no pronoun to rank) adjoined above FocP.
BMerge Top twice: bimá (its own sealed possessed-N sub-domain; its bi- is the possessor's, outside Voice) in the lower Spec,TopP, Link ↔ ρCom; ʼashkii yázhí éí in the higher, Link ↔ ρA (éí glossed TOP).
Moves: 0. HYPOTHESIS The clause-side goal PP stands between the companion's noun and the ranked PP, never between the ranked PP and W.
IV3. NLA 294 Hastiin sání léiʼ ʼáłchíní bił yanáaʼáago yił nahalneʼ: one P, two forms
A Clause 1 (-go clause, its own CP-derivation): W1 = yanáaʼáago (“3-pile.NI-COMP”), subject slot ρTh (the piled children). PP -ił with ρCom [3] (the old man), adjoined to VP.
B Voice1: Agree ⟶ {ρTh, ρCom}. The ‘with’-object, linked to the old man (named first, in the matrix topic), outranks → inverse → b-; Seal → bił. ʼáłchíníLink ↔ ρTh.
A Clause 2: W2 = nahalne’ (“3-tell.CI”), ρA (the teller); PP -ił with ρCom [3] (the children), adjoined to VP.
B Voice2: Agree ⟶ {ρA, ρCom}. The teller outranks → direct → y-; Seal → yił. Hastiin sání léiʼ in Spec,TopP, Link ↔ ρA of clause 2. The -go clause is adjoined below Top.
HYPOTHESIS How Voice1 sees the old man as the higher-ranked one, through a link from the matrix topic into the subordinate clause, is our assumption. The NLA's literal translation supports the roles, not the mechanism. The point the sentence makes doesn't depend on it: the same -ł takes both forms with two different verbs in one sentence.
IV4. NLA 13 Mary bimá sání yá ʼataʼ halneʼ: two PPs, the unit P inside (IP-8)
Σ Σ(ʼataʼ + √TALK, the ‘translate’ unit) = {ρA (the translator), ρX (who is talked between: here unspecified)}. The beneficiary is added by -á. Role labels ours.
A PPunit: Merge-ta’ with ρX; Agree P ⟶ ρX [unspecified] → ʼa- (the NLA glosses it “indef.”). Seal → ʼataʼ. Merge it as V's complement (IP-8: the unit's own P, first-merged lowest).
AMerge v, ρA (VPISH); M/S; Agree M/S ⟶ ρA [3]; Qual. Seal D1 → W = halne’ (“3-talk.DI”).
A PPBen: Merge-á with ρBen [3], adjoined to VP, higher than the complement PP.
BMerge Voice; Agree ⟶ {ρA, ρBen} (ρX is unspecified, not third person): direct → y-; Seal → yá.
BMerge Foc: bimá sání in Spec,FocP, Link ↔ ρBen; Merge Top: Mary, Link ↔ ρA.
BSeal D2: C2 gives Mary ⌢ bimá sání ⌢ {PPBen, PPunit, W}. Inside QualP, C2 again: the adjoined yá is higher than the complement ʼataʼ. C1′ puts both before W. → Mary bimá sání yá ʼataʼ halneʼ ✓.
Moves: 0. HYPOTHESIS The reversed order, ʼataʼ yá halneʼ, would need the unit P to be higher than the beneficiary. IP-8 says that doesn't happen. No opened source prints it, and none prints it as bad either (T51f). The site's CV3 (Willie (43) bił béé hózin) gets the same string under IP-8 with -ee as the unit P, and also if -ee is inside W, as the NLA's spelling bééhózin suggests.
Status table
Item
Status
Basis
-ł ‘with (someone)’ alternates yił/bił in 3/3 sentences
SOURCED
NLA 1992, 294, 2; site S2
y-forms of -á, -aa, -ká, -chʼįʼ, -kééʼ in 3/3 sentences
SOURCED
NLA 13, 489, 249, 17, 675, 19
Instrument -ee: bee in 3/3, no yee in the lists
SOURCED
NLA 1953, 92 (gloss conflict)
P1: the instrument is never ranked
HYPOTHESIS, first test passed, weak
two sentences; gap reading open (§3.4)
X2 / T49b: object before the instrument
SOURCED as order
NLA 92, 90; does not separate O1/O2/O4
X3: rope order with two third persons
OPEN
not found
P2: no noun between a ranked PP and W
HYPOTHESIS, no counterexample
about 120 sentences; not a judgment test
IP-7 (a) read by the P-object's animacy
contradicted (Willie (83), already noted; NLA 19)
—
IP-7 (a) P-class reading vs (b) relation
OPEN
both fit; separate on -kééʼ, -ł with a thing
IP-8: the unit P is next to W
HYPOTHESIS
NLA 13, 1958 (+792), 15; Willie (43)
A y-form yee exists
OPEN
one popular-site statement; Creamer p. 34 string, unsegmented
What each analysis gets right and misses (on this question)
Analysis
Gets right
Misses
Willie 1991
The direct/inverse pattern on Ps (y- with the subject first, b- with the P's object first) holds for -ł in the NLA, as for her -á, -aa, -chʼįʼ
Lists -ee Instrumental as grammatical (inside the MVS) without an instrument voice pair; on the NLA data the instrument behaves differently from the other Ps
NLA Postposition Lexicon
Many sourced 3/3 sentences with speakers and Young & Morgan pages; tags each P use
Gloss-line slips (92, 249, 1958); spacing split at link spans; translations sometimes loose (52, 247)
Speas (configurational, as on the site)
Derives both object/instrument orders (scrambling)
With the instrument PP in VP, Voice should reach it and give yee in 92, 1953 (if yee exists): needs an extra statement that instruments are skipped
This site, IP-1′
Every row of §2; the b-form of the instrument follows from its site
Rests on two instrument sentences; doesn't yet say what makes a role rankable (IP-7)
This site, IP-7 (a)
Fits every decisive row as worded (by P class)
Silent on Ps outside (28), such as -kééʼ, -ká; a reading by the object's animacy fails on a ball twice
7. Grant's frame
GRANT'S CONJECTURETheta roles as a proto-role schema, not lexical items. One more fit. What the voice compares are role slots: the stem's own (ρA), and those a postposition adds for a person (beneficiary, goal, companion). The slot of a means (wagon, stick, rope) is outside the comparison. The nouns are linked to these slots from outside, at the clause level (IV1–IV4). This fits the weak thesis. It is not a separating test against a configurational analysis, which can say the same thing with a rule that skips instruments.
HYPOTHESISThe layers (STRATA). The refinement of the last run, a ranking layer between the stem's roles and the event-structure ingredients, now has data outside Willie: in the NLA sentences the companion is in it and the instrument isn't. The means is an ingredient of how the action is done, not a participant the voice ranks.
GRANT'S CONJECTUREPhrase-level lexicalization. IP-8 gives this a word-order signature. In the four cases, the postposition that forms a unit with the verb (ʼataʼ halneʼ ‘translate’, bił hózhǫ́ ‘be happy’, bił + ‘drive’, béé hózin ‘know’) sits next to the verb, and the freely added one sits outside. If that holds, being part of a verb + P unit is visible in the structure (first-merged as V's complement), not only in the meaning.
OPENWord-internal Merge sites. One lead. The NLA writes ‘know’ as bééhózin, one word, where Willie prints béé hózin. If the NLA's spelling reflects structure, the -ee of ‘know’ is a P merged inside W, which would be a word-internal Merge site. Spelling is not structure, so this is a question (T51g), not evidence.
GRANT'S CONJECTUREThe word as the changeover point. Unchanged. Every derivation here seals W first and orders PPs and nouns at the clause level.
8. Tasks T51a–g and statuses
Task
What it needs
T51a
NLA 92: is the subject the grandmother (translation) or the speaker (gloss “3-1-hit.P”)? Needs the live entry (audio, other speakers) or another printing
T51b
A sourced yee meaning ‘with it, by means of it’: rules out the gap reading of §3.4. In a direct 3/3 sentence it would refute P1 for that verb. Sources: Young & Morgan 1987 (an -ee paradigm, if it has one), live NLA
T51c
-tsʼą́ą́ʼ ‘away from’ in a 3/3 sentence (the rest of T50a); the NLA malefactive and source lists
T51d
X3: NA NIns bee NTh W with two third persons and the voice on the verb
T51e
-ł with a thing in a 3/3 sentence (yił or bił?) separates IP-7 (a) P-class from (b); also the status of łaʼ in 986
T51f
IP-8: more two-P sequences; any with the unit P outside, e.g. a printed ʼataʼ yá halneʼ
T51g
Word division of bééhózin vs béé hózin: what the NLA and Young & Morgan say about it; and Creamer p. 34's ‘elephant’ word in Young & Morgan
A search-engine result also pointed to NLA entry 323, a sentence translated ‘The little boy hurt himself with the gun’ that uses -kʼee ‘on account of’. The entry could not be opened (the live site is down), so its form is not quoted here. If its P has the y-form, it would be a means-like role that is ranked, which bears on IP-7(b). Statuses: T50a: done for -ł (alternates) and -ee (b-form only in two 3/3 sentences); open for -tsʼą́ą́ʼ (→ T51c). T49c: done (1953; 92 with a caveat). T49d / P1: first test passed, weak (→ T51a, T51b). T49b: done (both orders sourced; not separating). T50b: no counterexample, still a prediction. T50f: new rule IP-8 (→ T51f). T50c, T50d, T50e, T50g, T50h: unchanged. T48a, T46h, T49a: not attempted (live NLA down).
Why 'I make the baby laugh' keeps bi-: the causee's prefix belongs to the postposition (T64f)
2026-11-11 audit pass 2. Apostrophe fixes only. Navajo glottal stops typed as ' or ’ (for example in prefix names like ʼa- and ʼeʼe-, or words like Awééʼ) now print as ʼ (U+02BC), and a build step no longer turns the closing quote of an English translation inside a Navajo row into a glottal stop (‘…the girls.’, not ‘…the girls.ʼ). No form, gloss or translation wording changed.
Restated on the unified structure (2026-11-15). Voice, causatives and events are restated with every theta role assigned inside vP: the classifier spells a value of v⁰ (H59, distribution NOT UNDERSTOOD), a causer is an external argument in Spec,vP, and the passive keeps a silent indefinite agent (H10b). This page’s causative analysis has not yet been restated on the unified structure (OPEN). Light d, used below, is withdrawn (Grant, 2026-11-14); light n is an OPEN structural question. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Flat-S and other pre-VPISH notation below is labelled contrast. Status on the spine: see unified structure, §4, causatives and passive rows. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
2026-11-11 audit pass 3. The A/B headedness changeover is a hypothesis and is not established; sentences that stated it as fact now say so. See the pass 3 report.
With an 'I' or 'you' doer, an ordinary third-person object gets no prefix at all (Ø). Hale's causative (50a) biʼiiss²á 'I am feeding the boy' still has b-. The last page (hale-general-b §7) read this as a postposition's agreement keeping its own spell-out rule after it moves into the verb word (HG-10). It added "Hale does not comment on this. The reading is ours." T64f asked whether this is always so. This run collects every causative with an 'I'/'we' doer in three sources already on the box. It sets them beside plain transitives with the same doer. And it finds that Hale & Platero 1996 state the generalization themselves.
In plain words: When "I" make someone or something do something, the one being made to act still gets the bi- "him/her/it" prefix. With an ordinary object and an "I" doer, that prefix is simply left off. This holds in all ten sourced causative words found, and none goes the other way SOURCED. Hale and Platero say it is "as if" the prefix were attached to a hidden postposition SOURCED. What decides the prefix is the slot it comes from, not whether the thing is alive: an overturned car gets bi- as the one made to act, while a dish as a plain object gets nothing HYPOTHESIS on SOURCED data. That goes against one source's "animate only" remark.
Notation as on hale-general-b. Σ = the role schema (Dowty's proto-roles as the baseline). A = the word level (prefixal, left-headed). B = the clause level (right-headed). ρA = the causer (agent) slot, ρcs = the causee slot (the inner verb's own doer), ρP = an ordinary patient slot. H = Hale's rule: y- only if the 3rd-person object is c-commanded by a clausemate "true" 3rd-person subject, otherwise b- (hale023 PDF p. 16). AGRo = the object agreement of E (Hale's "edge" head). AGRc = the agreement of the causative postposition -y. Merge builds structure (it joins two pieces). Agree values features such as person and number between pieces already built. Spell-out picks the sound for a valued feature bundle.
1. Question
T64f (from hale-general-b §14): "Is a causative or other incorporated P's agreement under a 1st/2nd-person subject always b- and never Ø? Is the Ø of fn. 9 limited to E's own AGRo?" Hale's fn. 9 (PDF p. 16) says a 3rd-person object under a 1st, 2nd or 4th-person subject is Ø if the object agreement is in E, and b- otherwise. A causee's agreement ends up in E (Hale p. 53: the causative P "actually incorporates into E"). So fn. 9, read literally, predicts Ø for it.
2. Sources
SOURCEDHale, Ken 1997. "Remarks on the syntax of the Navajo verb", Parts I–III (hale023.pdf). New parts read this run: PDF pp. 47–48 ((22)–(25)), 50 ((31)–(33)), 52–54 ((40)–(43), fn. 27). Page images were checked for pp. 47, 50, 53 and 54, and p. 58 is carried over from the last run.
SOURCEDHale, Ken & Paul Platero 1996. "Navajo reflections of a general theory of lexical argument structure." In Jelinek, Midgette, Rice & Saxon (eds.), Athabaskan Language Studies, pp. 1–14. Box copy: the vdoc.pub text (no page numbers; see sources). Read this run: (6)–(10) and the paragraph after (10).
SOURCEDSaxon, Leslie 2003. WECOL 2001 vol. 13, pp. 340–359. Page images at 300 dpi: PDF p. 362 (= printed p. 355; examples (41)–(42)) and PDF p. 364 (= p. 357; notes 12–14). The printed page = PDF page − 7, from (39) on PDF p. 361 = p. 354.
Not opened: Young & Morgan 1980 / 1987 (the sources the causative forms are cited from); Willie 1991 p. 30 (cited via Saxon fn. 12 (i)).
3. Rules first
CB-1 The causee's marker is not the ordinary object marker. With a 1st/2nd-person SOURCED (H&P 1996, text after (10);
subject, a 3rd-person causee is overt bi-, "as if attached to an incorporated Saxon fn. 12, p. 357)
postposition".
CB-2 An ordinary 3rd-person object agreement in E under a 1st/2nd/4th subject is Ø. SOURCED (hale023 fn. 9, p. 16;
Saxon fn. 12 (i), via Willie 1991:30)
CB-3 The causee Agrees with the causative P inside the PP. The P then incorporates SOURCED (hale023 pp. 52–53, tree (40))
into E.
CB-4 Hale's own parse of hypothetical (49a) under shí 'I': b(AGRc) next to Ø(AGRo), SOURCED (p. 58; the star comes from
i.e. two different spell-outs of 3rd person in one word, one subject. HG-9, not from the spell-out)
CB-5 Spell-out is keyed to the head that did the Agree: HYPOTHESIS (= HG-10, now with
[3, on P] → y- if H holds; otherwise b-. Never Ø. 10/10 sourced causee forms, 0 against)
[3, on E/AGRo] → y- if H holds; Ø if the subject is 1/2/4; otherwise b-.
CB-6 Animacy does not decide CB-5: an inanimate causee (car) is b- under 'I', HYPOTHESIS on SOURCED data
an inanimate plain object (dish, snow) is Ø under 'I'. (contradicts Saxon fn. 12 "animate")
CB-7 The causee b- may precede or follow iterative ná- (fn. 27); it always SOURCED (p. 54, (42a), (43a,b), fn. 27)
precedes the indefinite ’-.
CB-8 The B→A changeover keeps the head's origin: the value is fixed at B inside HYPOTHESIS (refines HG-11)
the PP, and the P-label survives Head Movement into E, where spell-out (A)
reads it.
Answer to T64f: yes, for every sourced causative found (CB-5, HYPOTHESIS on 10 SOURCED forms; CB-1 SOURCED as a generalization by Hale & Platero). "Always" stays OPEN: Saxon writes that bi- "can be used" here, which may mean it is optional, and no causative paradigm has been opened (T65a). The fn. 9 Ø is limited to E's own AGRo in all data checked (HYPOTHESIS).
4. The data: causatives with an 'I' or 'we' doer
All forms are printed as in the source. In hale023, ¬ (= ł) and ² (before a nasal vowel) are font substitutions in the PDF, which are visible on the page images too, and are not decoded here. The morpheme breakdowns in angle brackets are Hale's. The Leipzig lines are ours, built from the source's segmentation and glosses. CAUS = the causative postposition -y, CLF = classifier, INDF = indefinite object ’-.
hale023 PDF p. 50 (33); p. 53 (41b); H&P 1996 (8b), (9b)
(1)
(Shí)
awééʼ
biyeeshdloh.
(1SG)
baby
b-y-gh-sh-¬-dloh
(1SG)
baby
3.CS-CAUS-PROG-1SG.SBJ-CLF-laugh
'I make the baby laugh.' (33). Hale's (32) *(Shí) awééʼ yishdloh, with no b- and no causative -y, is starred.
'I righted your (overturned) car for you.' (43b) (Y&M87D:267)
(8)
Shí
ashkii
biʼiiss²á. (< b-ʼ-y-Ø-sh-¬-y²á)
1SG
boy
3.CS-INDF-CAUS-M-1SG.SBJ-CLF-eat
'I am feeding the boy.' (50a), carried over from hale-general-b.
Glosses for 'with', 'for', 'morning', 'sheep little' are ours, read from Hale's English translations; the ITER/REV label for ná# follows Hale's words "the iterative or reversative ná-" (p. 54). "…" marks a part of the stem complex we do not gloss.
Saxon 2003 (41), (42), PDF p. 362 = p. 355 (page image, 300 dpi)
(9) Saxon (41): the verb word nabiishłá, glossed by her "b-walk [it] around", in 'I'm walking my uncle's racehorse for him.' (Y&M 1987:525). It is the same verb as (3), with a racehorse as the causee. The sentence also has a separate word glossed "b-for" ('for him'), which has b- under 'I' too. The other words are not reproduced, because their diacritics are faint on the image.
(10) Saxon (42): 'We showed him how to drive a truck.' (Y&M 1987:223). The verb is glossed by her "b-b-l-show how": two b-s under a 1st-person plural subject. Which b- is the causee and which is another object is not stated (T65d). The word is not reproduced, for the same reason.
Saxon p. 355: "Bi- occurs in other environments: … and when the subject is not third person. The two sentences below exemplify this last type". Fn. 12, p. 357: "There is also a zero form for third person objects: the usual inflection for third person objects when the subject is first or second person is zero … (Exceptionally bi- can be used in this context for animate direct objects in the derived causative construction (Hale and Platero 1996:4f); my (41)-(42) involve this type of verb.)"
Count: ten causative verb words with a 1st-person subject ((1)–(10); (2) appears in both Hale sources and is counted once). All ten have b- for a 3rd-person causee. None has Ø. SOURCED as data. H&P state the pattern as a property of the causee markers: "they differ from ordinary position IV object markers in that the third person is overt (bi-) in the presence of a first or second person subject, as if attached to an incorporated postposition" SOURCED.
Correction to hale-general-b §7. That page said of HG-10: "Hale does not comment on this. The reading is ours." Hale & Platero 1996 do comment, in the sentence just quoted. The postposition-origin reading is theirs ("as if attached to an incorporated postposition"). What is ours is the formal statement CB-5, which says that spell-out is keyed to the head that did the Agree. A note has been added there.
5. Same doer, plain object: Ø SOURCED
Hale's simple (non-causative) ł-transitives with an 'I' doer have no overt object prefix. He glosses the object as part of a fused "3:" label with the mode.
hale023 PDF p. 47 (22b), (23b) (page image); H&P 1996 (6b)
(11)
Òeetsʼaaʼ
sé-¬-tsʼil.
dish
3:SPF:1s-¬-shatter:PERF (Hale's gloss)
dish
(Ø.OBJ)PFV.1SG.SBJ-CLF-shatter.PFV
'I shattered the dish.' (22b). (Ò is the font's Ł.)
(12)
Yas
yí-¬-h²í²íʼ.
snow
3:YPF:1s-¬-melt:PERF (Hale's gloss)
'I melted the snow.' (23b)
(13)
Kǫʼ
n-é-ł-tsiz.
fire
3:n-SPF:1s-ł-extinguish:PERF (H&P's gloss)
'I put the fire out.' H&P (6b)
Hale's (25b) *ʼAwééʼ d-é-saʼ 'I burped the baby' (p. 48) is the Ø version of (2), and it is starred. It is no clean test of the prefix alone, because it also lacks the causative -y. Hale stars it because 'belch' does not have a simple transitive.
Put (2) beside (11): same doer ('I'), same ł classifier, same perfective, a 3rd-person non-subject in both. The only structural difference is the causative postposition. And the only morphological difference in the object slot is b- vs Ø. That is CB-5.
6. "Animate direct objects" only? HYPOTHESIS on SOURCED data
Saxon's fn. 12 limits the exception to "animate direct objects". Her own (41)/(42) and most of Hale's causees are animate (baby, boy, lamb, racehorse). But (7) Nichidí ná nábiiss²iʼ 'I righted your (overturned) car for you' has a causee that is a car, and it is b- under 'I'. Hale's breakdown puts this b- right after the causative-attached ná# and before the causative -y (ná#b-y-…). On his analysis this is the causee's AGRc. The 'for you' is a separate word, ná, which has 2nd-person n-. So an inanimate causee takes b-, and inanimate plain objects ((11) dish, (12) snow) take Ø. Animacy does not decide it. The causative P does. This is contradicting data for the "animate" condition as stated (one form, from one source citing Y&M87D:267; T65c).
Two things are not contradicted. Saxon's "can be used" is about optionality, and nothing here bears on that (T65a). Her "direct objects" is a label, not a datum. H&P call the same markers different "from ordinary position IV object markers", so the two sources classify them differently. That is not counted either way.
7. Where the b- sits in the word SOURCED
Hale p. 53: in (2) the agreement "is left in initial position, preceding the overt inceptive E, d-". But (43a,b) show that it can also follow iterative/reversative ná-, and fn. 27 reports both orders for one verb: "both of the following are recorded for 'make him drink something (iterative)': bináʼiishdl²í²íh, nábiʼiishdl²í²íh (Cf.., Y&M87D:215)". The causee agreement "always precedes the indefinite object prefix ’-". Hale ties the variation to "properties of ná- itself". So the b- is not simply stranded in the outer PP position, and it is not fixed in the AGRo position either (his two "reasons", p. 54). For CB-5 this does not matter. Its form is b- in both positions, so the form is set by origin, not by landing site (HYPOTHESIS).
8. Rulesets A and B HYPOTHESIS
B (clause, right-headed): Hale's tree (40) (p. 53) is head-final throughout. It has [DPi Pcaus] with P on the right, [PP VP] V on the right, and the subject DP shí inside the top VP (so VPISH-compatible; not a flat-S tree). Agree between the causee and Pcaus happens here: "DPi does not need to raise to a position locally c-governed by E, since it finds its agreement partner within PPi" (p. 53). H (is the causee c-commanded by a true 3rd-person subject?) is also computed here.
A (word, left-headed prefixal): Head Movement brings V2 → V1 → M → E, and Pcaus "incorporates into E". The agreement shows up as a prefix to the left of the stem. Spell-out chooses y- / b- / Ø.
Changeover (CB-8): at Head Movement into E, as in HG-11. What is new is what crosses over. Not only the value [3, ±H] crosses, but also which head did the Agree (P vs E). Spell-out at A uses both. One word therefore contains two vocabulary tables, one for E's AGRo (with Ø) and one for a P's AGR (without Ø). If the word is the changeover point (a hypothesis, not established), it is not a blank slate: it keeps a record of the clause-level heads it was built from.
9. Derivations, step by step
Labels: Σ role schema, B clause, A word. Hale's trees already have the subject inside VP. Only his notation (Case-binding, government) is older, and it is restated here as Merge/Agree. Nouns are linked at B (nominal adjunction).
CD-1. (2) ʼAwééʼ bidiyésaʼ 'I burped the baby'
Σ
Inner event 'belch': one Proto-Agent-like slot (the belcher) = ρcs, 3sg. Outer causing event: ρA = 1sg. 'belch' is a non-alternating (12a)-type verb (Hale (25)/(26)), so ρcs cannot become a plain object (*(25b)).
B
Merge V2 'belch' (with its conflated N). Merge PP [ρcs Pcaus] (Case-bound oblique, Hale (40)). Merge V1 (ł), then ρA in Spec,VP; then M, E. (Merge builds the structure.)
B
Agree ρcs ↔ Pcaus inside the PP: value [3sg] on AGRP. (Agree values person/number; nothing new is built.) H: is ρcs c-commanded by a "true" 3rd-person subject? The subject is 1sg, so no: [−H].
B
Agree ρA ↔ M/E (subject agreement 1sg). E's AGRo has no referential object to agree with. 'belch' has none (it is not counted on the HG-9 slot).
A
Head Movement V2 → V1 → M → E. Pcaus incorporates into E (CB-3), carrying AGRP[3, −H, origin P].
A
Spell-out by CB-5: [3, on P, −H] → b- (P table: no Ø). E's own AGRo is empty, so nothing is spelled out for it. Linear order: b-d-y-é-saʼ (Hale's segmentation) ✓.
B
Link the noun ʼawééʼ → ρcs ✓.
CD-2. (11) Łeetsʼaaʼ séłtsʼil 'I shattered the dish' (form printed as Òeets’aa’ sé-¬-ts’il in hale023)
Σ
'shatter' is alternating ((22)): the patient ρP 3sg is the inner verb's specifier. The causer ρA = 1sg.
B
Merge V2 with ρP in its Spec. V2 conflates with V1 (ł); merge ρA; M; E. No PP.
B
Internal Merge ρP → Spec,MP; Agree with E's AGRo: [3sg]. H: the subject is 1sg, so [−H].
A
Head Movement V → M → E. Spell-out by CB-5: [3, on E, subject 1sg] → Ø (fn. 9). Output sé-ł-tsʼil ✓.
CD-3. (7) Nichidí ná nábiiss²iʼ 'I righted your car for you'
Σ
Inner event: the car comes to stand upright (ρcs, 3sg, inanimate). Outer: ρA = 1sg. Beneficiary ρfor = 2sg, in a separate P-word.
B
Merge as in CD-1, with the PP [ρcs Pcaus]. A separate PP [ρfor -á] is merged as an adjunct.
B
Agree ρcs ↔ Pcaus: [3sg]. Agree ρfor ↔ 'for': [2sg]. H for ρcs: [−H] (1sg subject).
A
The word 'for' is spelled out as its own word: [2sg] → n-, giving ná ✓. In the verb word, Pcaus incorporates into E. ná# (reversative) precedes the b- here (fn. 27 variation). [3, on P, −H] → b- ✓. Inanimacy plays no role.
B
Link nouns: chidí (with 2sg possessor ni-) → ρcs ✓.
Under a 3rd-person causer, the same derivation gives [+H] and y-, as in Hale (50b) Hastiin ashkii yiʼiis²á (carried over). That is the H half of CB-5, already on hale-general-b HD-2.
10. Grant's ideas
GRANT'S CONJECTUREFeature bundles / theta roles as arguments. Consistent. CB-5 reads only the slot's bundle ([3]), the subject's bundle (1/2/4 vs true 3rd), and the head that did the Agree. It never needs to know which noun fills the slot: a baby, a racehorse and a car all give b-. A rule written over normal nouns would have to say "animate" (Saxon), and (7) contradicts that (HYPOTHESIS). No contradicting data.
GRANT'S CONJECTUREStrata. Consistent. The causee is first a theta slot of the inner event (the belcher, the laugher). The causing event layer (V1 with ł) then adds the causer and demotes that slot into a postpositional phrase. Agreement is valued after that, and the noun is linked last. The prefix depends on the event layer (was this slot demoted by CAUS?), not on the noun. That is the order the strata idea predicts (HYPOTHESIS).
GRANT'S CONJECTUREMerge/Agree sites inside the word. Supported, as a HYPOTHESIS. One verb word has at least two Agree sites with different spell-out tables: the causative P's AGR (never Ø) and the subject agreement. In Hale's parse of hypothetical (49a), a third site (AGRo, Ø) sits beside them, even though that word is starred for another reason (HG-9). The P's Agree was completed in a smaller unit (the PP) and then merged into the word. That is an intra-word Merge site (P into E) carrying an already-valued Agree site.
Chemistry: SPECULATIVE, flag only. A label whose form is fixed by the kind of site a unit came from, and which keeps that form after the unit is moved into a larger host, independent of what occupies it. No chemical correlate is identified or claimed.
11. Bottom line
SOURCED Under a 1st-person subject, all ten causative verb words found in Hale 1997, Hale & Platero 1996 and Saxon 2003 mark a 3rd-person causee with b-. Plain ł-transitives with the same subject show Ø. Hale & Platero state the generalization ("as if attached to an incorporated postposition"). Saxon fn. 12 calls it an exception for "animate direct objects" that "can be used".
HYPOTHESIS T64f answered for causatives: the Ø of fn. 9 belongs to E's own AGRo, and a P's agreement keeps its own table after incorporation (CB-5 = HG-10). Animacy is not the condition ((7) car vs (11) dish), which contradicts Saxon's "animate". On the changeover hypothesis (not established), the step to the word level carries over the head's origin (CB-8).
OPEN Whether b- is obligatory (T65a); 4th-person causers (T65b); more inanimate causees (T65c); the two b-s of Saxon (42) (T65d); non-causative incorporated Ps (T65e).
12. Fidelity notes
hale023 forms were checked on page images (pp. 47, 50, 53, 54; p. 58 last run) and are printed as printed, including the font substitutions ¬, ² and Ò. CD-2's heading gives the restored spelling Łeetsʼaaʼ séłtsʼil only as a reading aid. The restoration uses the mapping already documented on sources (¬ = ł, Ò = Ł).
H&P 1996 forms (3), (13) and the (2) duplicate come from the vdoc.pub text, which has no page images. (8b) and (9b) were checked earlier against Gessner 2003's page-image reprints (see sources). (6b) was not.
Saxon (41)/(42) were read at 300 dpi. Only the verb nabiishłá is printed here. The other words, and the (42) verb, are described rather than copied, because the ogoneks and accents are faint. Saxon's printed page numbers are inferred (PDF − 7).
Young & Morgan, the original source of most forms, was not opened. The archive.org scan of the 1980 edition is not available for reading. The Y&M page references are the citing authors'.
The Leipzig lines, the labels 3.CS / ITER / REV, and CB-5 to CB-8 are this site's. Hale's own labels are AGRc / AGRo.
The "ten forms" count includes (9) and (10), whose verbs are glossed b- by Saxon. (10) has two b-s, and which is the causee is not stated.
13. Open questions
ID
Question
What would settle it
T65a
Is causee b- under 'I'/'you' obligatory (H&P "is overt") or optional (Saxon "can be used")? Is there any sourced Ø-causee causative with a 1st/2nd-person subject?
Young & Morgan 1987 causative paradigms; a native-speaker judgment. OPEN
T65b
4th-person causer: CB-5 predicts b- (the P table has no Ø, and a 4th person is not a "true" 3rd for H). Does a sourced form show it?
A causative with a ji- subject (Willie 1991 ch. 4; Y&M). OPEN
T65c
More inanimate causees under 'I' (besides (7) the car), to confirm that Saxon's "animate" is wrong rather than a tendency.
Y&M87D:267 first-hand; other causatives of inanimate inner subjects. OPEN
T65d
Saxon (42) 'We showed him how to drive a truck', glossed b-b-: which b- is the causee, and does a second P-object inside one word strain HG-9 (one referential object agreement per word)?
Y&M 1987:223 first-hand; the theme of 'teach / show how'. OPEN
T65e
Non-causative incorporated Ps: does every P that incorporates into the verb keep b- (never Ø) under 'I'/'you', as CB-5 predicts? (A free P word already does: Saxon (41) 'for him' under 'I'.)
Sourced verbs with incorporated postpositions and 1st/2nd-person subjects (Y&M; Willie 1991 ch. 3; Kari). Update 2026-10-27 ~12:47pm PT: answered on p-object-b. Hale says ordinary Ps do not incorporate (they are proclitics), and every sourced P object with an 'I' participant is b- (9/9, SOURCED); no Ø in the P paradigm. SOURCED / "always" OPEN (T67a)
Sources: Hale 1997 workshop notes (hale023) PDF pp. 16, 47–48, 50, 52–54, 58 (page images pp. 47, 50, 53, 54; p. 58 earlier); Hale & Platero 1996 (6)–(10) and the paragraph after (10) (vdoc.pub text); Saxon 2003 pp. 355, 357 (page images, PDF pp. 362, 364). See also hale-general-b, plain ł-causer, open questions, sources.
Restated on the unified structure (2026-11-15). Postpositions are restated as [PP DP P⁰] with left specifiers: a grammatical P agrees with the DP in its specifier, a lexical P has no probe (H64); a selected PP is merged in VP, and Willie’s “cannot be interrupted” follows from the default, HS-1 and left specifiers. Light d, used below, is withdrawn (Grant, 2026-11-14); light n is an OPEN structural question. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. The “roles not words” theory, where it appears below, is dropped. Status on the spine: see unified structure, §2b and §4, postposition row. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
2026-11-11 audit pass 2. Wording that used VPISH as the name of a spine, tree, skeleton or set of slots named a structure after the hypothesis. VPISH is a hypothesis name, never a node; the structure is a vP/TP spine with the subject first-merged in left Spec,vP. Only the wording changed.
With an 'I' or 'you' doer, the verb's own third-person object prefix is left off (Ø). A postposition's third-person object never is. In every sourced clause found with a 1st- or 2nd-person participant, the postposition's 'him/her/it' is bi-: 9 tokens in 8 clauses, with no Ø and no yi-SOURCED. Willie's postposition-object paradigm has no Ø cell at all SOURCED. So the silent form belongs to the verb's own object slot, not to "third person" in general HYPOTHESIS. One correction to the question itself: Hale says no ordinary postposition actually incorporates into the verb; the ones written together with it are proclitics that lean on the finished word SOURCED.
Notation as on causee-bi and hale-general-b. A = the word level (prefixal, left-headed); B = the clause level (right-headed). Σ = the role schema (Dowty's proto-roles as the baseline). H = Hale's condition for yi-: a true 3rd-person subject c-commands the 3rd-person object. Merge builds structure (puts two pieces together into a phrase); Agree copies feature values such as person and number from one piece to a probe that needs them. The Syntax · Semantics · Interface conventions have not arrived; this is the site's own notation.
1. Question
T65e (from causee-bi §13): "Non-causative incorporated Ps: does every P that incorporates into the verb keep b- (never Ø) under 'I'/'you', as CB-5 predicts?" CB-5 says spell-out depends on the head that did the Agree: on E's object slot (AGRo), a 3rd person under a 1st/2nd/4th subject is Ø; on a P it is b- unless H holds.
2. Sources
SOURCEDWillie, MaryAnn 1991.Navajo Pronouns and Obviation. PhD dissertation, University of Arizona. Box scan; page images read this run (PDF = printed + 5): p. 28 (pronoun table (1) and the remark on cognate bi-), p. 40 (table (27)), p. 41 ((28), (29)), pp. 98–99 ((90)–(96)), p. 223 (§9.2, "ZERO allomorph").
SOURCEDHale, Ken 1997. "Remarks on the syntax of the Navajo verb" (hale023), PDF p. 53 (the paragraph before (41) and the one after (42)), re-read from the text layer; page image of p. 53 checked last run.
SOURCEDNavajo Language Academy, Postposition Lexicon (v10.2023, talkingdictionary.swarthmore.edu/navajo_postposition/). Sentences as recorded from live loads and archived copies on 2026-10-21 (notes kept on the box). The live site returned 504 this run, so they were not re-checked today. Where the NLA credits Young & Morgan 1987, the page is given; Y&M itself was not opened.
Not opened: Young & Morgan 1987; Kari 1976; McDonough 1990 (all cited by Willie p. 223).
3. The premise, corrected
T65e assumed there are non-causative Ps that incorporate. Hale (p. 53) says otherwise: "when the PP complement of a verb is phonologically dependent, i.e., a proclitic, it merely attaches to the fully derived verb to its right, binding with it rather loosely, like any other disjunct proclitic element. The causative P … exhibits a different behavior morphologically. It actually incorporates into E" SOURCED. He then hedges even that ("there are two reasons why this might be the wrong conclusion", p. 53). So the testable question becomes: does a postposition's 3rd-person object ever go silent under 'I'/'you', whether the P is a separate word or a proclitic written on the verb?
4. Rules first
PB-1 Willie's P-object paradigm (27): 1 shi-, 2 ni-, 3 yi/bi-, 4 ha (sg); no Ø. SOURCED (Willie p. 40, image)
PB-2 The verb's 3rd object: "two third person object forms are generally SOURCED (Willie p. 223, image;
recognized ... yi- and bi-, the first of which has a ZERO allomorph." citing Y&M 1987)
PB-3 Inverse bi- "is clearly cognate with the third person bi- pronouns that SOURCED (Willie p. 28, image)
appear as possessors and postpositional objects".
PB-4 "Inverse cannot apply when there is a first or second person argument in SOURCED (Willie p. 98, image)
the clause."
PB-5 Ordinary Ps do not incorporate: a proclitic PP "merely attaches to the fully SOURCED (Hale 1997 p. 53)
derived verb". Only the causative P incorporates into E (hedged).
PB-6 Spell-out of [3] by host: HYPOTHESIS (CB-5 extended to all Ps;
[3] Agreed by P (free or proclitic) → yi- if H holds; else bi-. Never Ø. 9/9 sourced P tokens, 0 against)
[3] Agreed by E's AGRo → yi- if H holds; Ø if subject 1/2/4;
else bi-.
PB-7 Because a proclitic P joins only after the word is finished (PB-5), its HYPOTHESIS (follows from PB-5 + PB-6)
agreement is valued and spelled out in the PP, before the word's A-level
rules can see it; the A-level Ø rule therefore never applies to it.
Answer to T65e: yes, in all sourced data, and for a reason that is partly sourced: there is no Ø in the P paradigm (PB-1), and the Ø is stated only for the verb's object (PB-2). "Never, for every P and every speaker" stays OPEN: the sample has 6 different Ps and no paradigm from Young & Morgan.
5. The data: a 3rd-person P object with a 1st/2nd-person participant
Forms as printed. Lines marked "source" are the source's own gloss. The Leipzig lines are ours, built only from those glosses and the translations. The verb words are not segmented beyond what the sources give.
Willie 1991 p. 41 (29); p. 98 (92) (page images)
(1)
shimá
bił
déʼáázh
source: 1.mother 3.with dl.walk
1SG.POSS-mother
3-with
DU.walk
'I am going with my mother.' (29)
(2)
shimá bił
chidí biiʼ
séké
source: my mother 3.with car 3.in DUAL.sitting
1SG.POSS-mother 3-with
car 3-in
DU.sit
'I sit with my mother in the car.' (92). Two Ps, both b-.
NLA Postposition Lexicon (recorded 2026-10-21); Y&M pages as credited by the NLA
(3)
Shí hanii tʼéiyá Naashtʼézhí bizaad shił bééhózin.
source: I contrary only English 3-language 1-with 3-know.P
'I'm not the only one who knows Zuni (despite appearances to the contrary).' (YM 1987:415). The P b-ee is written on the verb (proclitic). The verb's subject is the areal ho- 'there is knowledge' (NLA's verb selector); 'I' is the -ił experiencer.
(4)
Naatʼáanii bichʼįʼ yáshtiʼ.
source: leader 3-to 1-talk.I
'I'm talking to the leader.'
(5)
Hakʼaz ʼayóo bichʼįʼ sidziil.
source: cold really 3-towards 1-strength
'I can withstand a lot of cold.' (YM 1987:185). Inanimate P object, still b-.
'Coming back from Phoenix by Greyhound bus …' (spacing "b ee" as displayed in the archived page)
Count: 8 clauses, 9 P tokens (-ił ×3, -ii’, -ee ×2, -chʼįʼ ×2, -din), 6 different Ps. All b-; 0 Ø; 0 yi-.
6. Minimal pairs: the same P, three kinds of subject
Subject
Form (Willie, page images)
P object
Why (by PB-6)
'I' (1st person)
(92) shimá bił chidí biiʼ séké 'I sit with my mother in the car.'
b-, b-
H fails (no 3rd subject); P has no Ø, so b-.
3rd person
(93) bimá yił chidí yiiʼ siké 'He sits with his mother in the car.' (p. 99)
y-, y-
H holds.
3rd person, animate
(90) ʼasdzą́ą́ chidí yiiʼ sidá 'The woman is sitting in the car.' (p. 98)
y-
H holds.
Impersonal
(94) tsin baah haʼasʼnaʼ 'The tree was climbed.'; (95) *tsin yaah haʼasʼnaʼ (p. 99)
b-; *y-
No true 3rd subject, so H fails; b-, and not Ø.
3rd person, transitive
(96) tsin yaah haasʼnaʼ 'He climbed the tree.' (p. 99)
y-
H holds.
Willie's own account of (92) vs (93) is PB-4: the inverse "cannot apply" with a 1st/2nd-person argument, "when all arguments are third person, there is no problem" (pp. 98–99) SOURCED. Our point is narrower: where the verb's object would be Ø, the P's object is b-. The table needs no verb-object form to show this; the contrast with the verb's Ø comes from Hale's (22b)/(23b) on causee-bi §5.
7. Rulesets A and B
Step
Where
Ruleset
What happens
1
PP
B (head-final: [DP P])
Merge P with its object (a pronoun or a noun-linked pro).
2
PP
B
Agree: P's person probe is valued [3] by its object.
3
PP edge
B (HYPOTHESIS)
Spell-out of P's agreement: H is checked; yi- or bi-. The A-level Ø rule is not in this cycle.
4
Word
A (prefixal)
The verb word is built; E's AGRo, if [3] under a 1/2/4 subject, is spelled Ø.
5
Word edge
A→B changeover
A proclitic PP leans on the finished word (Hale p. 53). It brings its b- already spelled. A free PP stays a separate word.
The word is the changeover point: Ø is a word-internal (A) spell-out, and a P's agreement is decided before or outside that cycle HYPOTHESIS. The causative P is the test case that crosses the line, since it does incorporate into E, and it still keeps b- (causee-bi CB-8): the value is fixed in the PP and carried along.
8. Derivations
D1. (92) shimá bił chidí biiʼ séké 'I sit with my mother in the car.' (vP/TP spine; structure of the verb word as on derivations.)
1 Merge [vP 1SG.SBJ [VP ... sit ]] subject starts in vP (VPISH)
2 Merge [PP1 pro(3)=chidí -ii’ ] ; Agree P1 ← [3] B; head-final PP
3 Spell P1 [3]: H? subject is 1SG → H fails → b- ⇒ bii’ PB-6 (P never Ø)
4 Merge [PP2 pro(3)=shimá -ił ] ; Agree P2 ← [3] B
5 Spell P2 [3]: H fails → b- ⇒ bił PB-6
6 Merge T/E; Agree T ← 1SG (DU per Willie's gloss) ⇒ séké A; word built, no AGRo here
7 Move subject 1SG to Spec,TP (null pro) B
8 Order PP2 PP1 V (head-final clause) B
D2. (3) … bizaad shił bééhózin 'I know the language.'
1 Merge [vP ho- (areal subject) [VP ... know ]] subject = ho- (NLA verb selector)
2 Merge [PP pro(3)=bizaad -ee] ; Agree P ← [3] B
3 Spell P [3]: H? ho- is not a true 3rd person → H fails PB-6
→ b- ⇒ bee (written béé-)
4 Merge [PP 1SG -ił] ⇒ shił experiencer
5 Word ho-...-zin assembled A
6 Clitic béé= attaches to the finished word ⇒ bééhózin A→B edge (PB-5)
D3 (contrast, schematic). A plain transitive with a 1SG subject and a 3rd-person object: step 2 is Agree by E's AGRo, step 3 spells [3] as Ø (PB-2, Hale fn. 9). The difference between D1/D2 and D3 is only which head did the Agree; person and animacy are the same HYPOTHESIS.
9. What it means for the role schema (Focus 2)
The verb's own object slot (Σ's proto-patient in E) can be fully silent: the slot is there, its filler is identified by a noun or context, and nothing is pronounced for it. A role brought in by a postposition (comitative, goal, instrument, experiencer, 'without') always carries an overt person marker. This fits the idea that the verb's core role slots are schema positions that need no word of their own, while P-introduced roles come with their own pronoun HYPOTHESIS. It also gives Grant's doubt about "theta roles, not words, as arguments" a sharper shape: the claim may hold for E's slots and fail for P's slots HYPOTHESIS; see the verb guide.
10. Fidelity
Willie forms were read on page images this run. (1) is déʼáázh with acute accents on the page; the site's earlier copy on intransitive-pp had "de’aazh" and is corrected this run.
The NLA sentences were not re-checked today (the site returned 504). They are copied from notes made when the pages were opened on 2026-10-21, as already used on instrument-voice and transportation-pattern.
(1) and (2) are glossed "dl." by Willie with single-person translations; the dual reading is T33g, unchanged.
H and PB-6 are this site's restatement of Hale's and Willie's statements. PB-7 is ours.
11. Open questions
ID
Question
What would settle it
T67a
Is there any sourced Navajo form with a Ø postposition object (a bare P stem) under 'I'/'you'?
Y&M 1987 postposition paradigms; Kari 1976. OPEN
T67b
4th-person subject (ji-) with a 3rd-person P object: PB-6 predicts b-, and the verb's object Ø.
Willie ch. 4–5 examples. OPEN
T67c
Are proclitic Ps like béé- in bééhózin phonologically inside the word (stress, vowel length) or only written together? Bears on PB-7 and the A/B edge.
McDonough 1990; Hale p. 53 context. OPEN
T67d
Does the 2nd-person case match the 1st? All 8 clauses here have 'I'.
A sourced 'you' clause with a 3rd P object. OPEN
T67e
Does the P/E split in silence show up in Algonquian (oblique vs object agreement)?