Possessors and relative clauses

Legacy link recovery
Legacy section links (scroll sideways); old articles remain available.
hale-platero-relativesrelative-cprelativespossessor-bindingpossessor-gappossessor-yipossessor-commandpossessor-raisingexternal-headperson-headagent-ext

On the spine (2026-11-14 summary)

2026-11-14 overhaul (approved by Grant for the live site, 2026-11-14). This summary states the topic on the spine (V→v→T and subject raising to Spec,TP assumed; each pattern is what else moves). It supersedes the framing of the dated sections below, whose data, judgments and citations are unchanged.

  • HYPOTHESIS A relative clause is built exactly like a main clause (Merge with theta roles, V→v→T, subject to Spec,TP, and the H3 Move in the inverse). The enclitic is an overt D⁰ taking the whole CP and linked to one participant by Agree; nothing moves out (H27b). yiztał=yę́ę (NX-312); biztał=yę́ę (NX-313); the overt pronoun in bichʼįʼ yáshtiʼ=yę́ę (NX-314) shows there is no gap. Results §8.
  • HYPOTHESIS The enclitic is linked to the lowest third-person participant with no noun inside (H28′), and only to a third person (H31): haztał =yę́ę (NX-319). An unlinked D⁰ names a fact (H34): NX-372.
  • OPEN *NX-345 vs NX-346 (with léiʼ) is not understood. Possessor positions have not yet been restated on the spine; the legacy sections below keep their earlier analyses as contrast.

Hale & Platero 1974: relatives, missing nouns, and the yi-/bi- parser

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.

Working note. Kept as written (URL, data and status labels unchanged). Filed under The fourth person; the consolidated argument is the essay fourth-person. Rule IDs: rules index.
Verbs on this page 18 entries in the verb dictionary (show)

yiztał ‘kick’ · jizloh ‘rope’ · yiztsʼǫs ‘kiss’ · yishxash ‘bite’ · yiyiisxį́ ‘kill (one)’ · dabiiłtsą́ ‘see’ · yáłtiʼ ‘speak, talk (to)’ · yi-dloh ‘laugh’ · ʼ-ł-wosh ‘sleep’ · yóóʼeelwod ‘run away’ · chʼíníyá ‘go out, went out’ · deezhtlizh ‘fall, trip’ · biłʼadéłdǫǫh ‘shoot’ · néísʼah ‘butcher’ · hadoolghaazh ‘cry out, yell out’ · niiníłbą́ą́z ‘drive (a vehicle) up to a point’ · dilwoʼ ‘be fast’ · hoołtį́į́ł ‘rain (progressing)’

Restated on the unified structure (2026-11-15). Relative clauses are restated as a full CP, built like a main clause, under an overt D⁰ (=yę́ę, =ígíí) linked to one participant by Agree; nothing moves out (H27b, H28′, H31, H34). 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, relative rows. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.

The 1974 paper in which Ken Hale and Paul Platero argue that a Navajo relative clause is built by raising, not by deleting a copy of the shared noun, and that the relativizer -ę́ę/-ígíí is a determiner. Along the way they state, as surface principles of interpretation, two rules this site later borrowed from Platero and Speas: the noun next to the verb is the object under yi- and the subject under bi-, and a noun is parsed with the verb that follows it if they make a good simple sentence. They also print a speaker split on that second rule. This page reads every Navajo sentence in the paper against Grant's two phenomena, and it answers T14e: what the 1974 work says about a noun that follows a relative clause. Written 2026-10-15, 4:46am run (PT).

Ground rules (Grant's). Hale, Platero, Willie and Speas are each treated as possibly partly right. A source that does not state the feature-bundle hypothesis (FB, Not quite a word) is not evidence against it. Only data that contradict it count. Dowty's (1991) proto-roles are the baseline, and labels such as “agent” name a participant's role in the action. Grant does not speak Navajo; every Navajo form below is copied from a page image of the paper, with its example number and printed page. None is invented.

Bottom line

HYPOTHESIS Nothing in Hale & Platero 1974 contradicts FB or T-MID. Of the 17 data groups (§11), 0 cut against FB's core, 1 supports it, and 16 are compatible. The support is their own analysis of the relativizer. What is shared between the relative clause and the main clause is a determiner that can raise without its noun. The noun is optional: it stays inside the relative (the preferred form), follows it, or is absent (pp. 26–27, (57)). That is a determiner-level element bound to a slot, with the noun as an optional rider, which is what FB says.

HYPOTHESIS One result cuts against a rule FB borrowed. FB-3a(a) LINK-NEAR, taken from Platero's later NP-PRO constraint, was stated as a hard rule. Hale & Platero's own version of it, the parsing principle (44), is “employed, with varying degrees of strictness”. For some speakers it gives way when context favours another reading ((40)–(43), pp. 20–23). So LINK-NEAR has to be a default, not a categorical rule (FB-3a(a′), §9). This answers T22c for this source: H&P treat it as parsing, with speaker variation. FB's core (the bundles, rank, the other linking rules) is untouched.

HYPOTHESIS T14e, plainly. Willie's “Platero (1974)” is Platero's IJAL article, not this paper, and that article is still unopened. This paper is by the same author, from the same year, on the same question. It does print nouns after the relativizer that are read as the relative's head: (1), (15), (20), (36). But in every one of them, the noun names the same participant as the slot the whole relative fills in the main clause. So Willie's parse, in which the noun is a separate main-clause adjunct on that same slot, gives exactly the same meaning. In the two sentences where the parses could differ, (32) and (46), H&P's speakers reject the “noun after the relative is its head” reading, just as Willie's grammar would. No printed sentence shows a difference in judgments. The disagreement is about analysis, not data. There is a real speaker difference, but it is a different one. H&P's (17) has two readings (either noun can be the head), and Willie says the parallel two-reading pattern is “not true of my speech” (1991:162, on Platero 1982).

VersionEffect of Hale & Platero 1974
T-STRONG (nouns are never arguments; roles alone decide links)Still fails. Two more cases where position decides, not roles: in (27) the girl is the external participant in both conjuncts although she acts in one and is acted on in the other; in (32) the string forces ‘The man I shot butchered the deer’, against the intended meaning (p. 18).
T-MID / FB (the arguments are role-bearing φ-bundles spelled as prefixes; nouns link to them)Survives, no contradicting datum. Gains a 1974 source for FB-3a(b) ((28)), a determiner-level head for relatives (pp. 26–27), and a 1974 statement that grammatical relations of nouns are read at the surface ((50)). LINK-NEAR(a) is demoted to a default.
T-WEAK (Pronominal Argument Hypothesis)Survives. H&P write in a 1970s transformational framework with NPs in deep-structure argument positions. Their data never need those positions; their own interpretation principles do the work at the surface.
Platero/Speas configurational lineNot refuted, not forced. H&P's deep structures put nouns in argument positions, but their solution (50) explicitly moves the interpretation of grammatical relations to the surface string.

On this page: 1 Source and caveats · 2 The argument: deletion vs raising · 3 T14e: a noun after the relative · 4 yi-/bi- and the parser · 5 Enclitic raising and the determiner · 6 Rule set A vs B · 7 What each source gets right and misses · 8 Rules · 9 Derivations HP1–HP6 · 10 Datum-by-datum table · 11 Open questions T23a–g

1. Source and caveats

SOURCED Hale, Kenneth & Paul R. Platero 1974. “Aspects of Navajo Anaphora: Relativization and Pronominalization.” Diné Bizaad Nánílʼį́įh / Navajo Language Review 1(1):9–28 (printed folios 9–28). In ERIC ED101549 (the whole first issue; ed. Paul R. Platero), the same issue as Creamer 1974 (eight ranks). A slightly expanded version of a paper read at the XII Conference on American Indian Languages (AAA, New Orleans, 1973) and at the LSA winter meeting (San Diego, 1973) (pp. 9, 28).

  • Page mapping. PDF page = printed page + 3 (PDF 12 = p. 9, PDF 31 = p. 28), checked on every page. All Navajo was read on page images at 150 and 300 dpi. The OCR text layer was used only to find passages; it is unusable for Navajo (e.g. it renders bįįh as “b ilh” and Tłʼéédą́ą́ʼ as “T11 44d441”).
  • Hale's preface to the issue (p. 5) introduces the paper as “a theoretical article, but the details of the new analysis are not fully worked out”, and asks that it “be read with a severely critical eye, both in regard to the Navajo data we use and in regard to the analysis”.
  • Glosses. H&P give a word-by-word gloss in parentheses under most examples, with “:” joining the words of one gloss. We copy their glosses, writing “.” for “:” (Leipzig). They do not segment verbs, and neither do we. Where they print only a translation, we give only the translation.
  • Spelling as printed. The typewriter writes ł as a slashed l and the glottal stop as '. We write ł and ’. Kept as printed: deezhtlizh (6)–(7) and tlʼiish (38)–(39), both with a plain l (modern spelling has ł); yiyiisxį (13), hook but no tone mark; yił adeesdǫǫh with a space in (13) and (38) but yiłʼadeesdǫǫh in the tree (39); -(n)ę́e without the second hook in (20); niiníłbą́z with one vowel in (51) but niiníłbą́ą́z in (55) and the prose (p. 24); nʼiníłbą́ą́z in (52). The relativizer is printed -ę́ę, -(n)ę́ę or -ígíí in examples and “/-yę́ę/ or /-ígíí/, depending on tense” in the prose (p. 27).
  • Slips in the original. P. 13 says the optional raising rule accounts for (1) “in which the rule would not have applied” and (4) “in which the rule would have applied”. That is the reverse of the trees (10)–(11) and of p. 26 (“sentences like (1) — with raising”). We follow the trees. Kaufman's paper is “Kaufman (1973)” in the text (p. 24) but “Kaufman, Ellen (1972)” in the references.
  • Which “Platero 1974”. Willie 1991's bibliography lists Platero 1974 as “The Navajo Relative Clause”, IJAL 40:202–246, which is not this paper. This paper reports Platero's MIT master's thesis (1973), A Study of the Navajo Relative Clause, and argues against its deletion analysis. Speas 1986's (4.15) paradigm is from Platero 1978, not from here.
  • Not this paper. Hale & Platero 1996 (lexical argument structure, the ł-alternation) is on plain ł-causer.

2. The argument: from deletion to raising SOURCED

Navajo relatives come in two forms (p. 9). In one, common in verb-final languages, the head noun follows the relative clause and the shared noun inside it is “apparently simply deleted”. In the other, which is “preferred”, there is no head after the clause, and the shared noun sits “in the position corresponding to its logical function in the subordinate clause” (p. 10).

Hale & Platero 1974:9–10, (1) and (4)
(1) Tł’éédą́ą́’ yáłti’-ę́ę hastiin ałhosh.
    last.night speak-REL man sleep
    ‘The man who spoke last night is sleeping.’
(4) Tł’éédą́ą́’ hastiin yáłti’-ę́ę ałhosh.
    last.night man speak-REL sleep
    (the preferred alternative to (1))

Platero's 1973 thesis derived both from one deep structure with two copies of hastiin (their (2)): delete the lower copy and you get (1), rule (3); delete the head and you get (4), rule (5). Ordinary pronominalization in Navajo is also deletion, forward or backward:

Hale & Platero 1974:11, (6)–(7)
(6) Hastiin ch’íníyá-(a)go deezhtlizh.
    man went.out-COMP fell
    ‘When the man went out, he tripped.’
(7) Ch’íníyá-(a)go hastiin deezhtlizh.
    went.out-COMP man fell
    ‘When he went out, the man tripped.’

So Platero merged the two processes into one deletion rule (9), with backward deletion “only possible if the deletee is in a subordinate clause” (pp. 11–12). H&P then argue the merger is wrong. Relativization obeys Ross's island constraints and pronominalization does not; this is Platero's own result (p. 13; the thesis data are not reprinted). In relatives one copy must go, while elsewhere deletion is “preferred” but optional (p. 13):

Hale & Platero 1974:13, (13)
Łééchąą’í dibé yiyiisxį biniinaa shizhé’é łééchąą’í yił adeesdǫǫh.
‘Because the dog killed a sheep, my father killed the dog.’

No word gloss printed. Both instances of łééchąąʼí ‘dog’ are kept, and the sentence is acceptable.

An optional raising rule (12) would instead move the lower NP into the empty head position. The rest of the paper defends it. H&P's first step is to show that the constraints on deletion that seemed common to both processes are really principles of interpretation of the surface string (§4). Their second step is positive evidence for raising (§5). Their conclusion (p. 28): “the struggle between the two competing conceptions of Navajo relativization is won decisively by the raising hypothesis”. They leave open whether raising copies.

3. T14e: a noun after the relative

SOURCED Willie 1991:157–158 (page images), on her (12) yidloh=ę́ę ʼatʼééd biztał ‘The girl kicked the one who was laughing’: “In (11) [sic; the example meant is (12)] ʼatʼééd can not be an adjunct to the relativized verb since it follows it. This interpretation is counter to Platero (1974) who claims that a nominal following a relatived verb can be part of the relative clause; this may be a dialect difference.” Willie prints “(11)” twice in this passage. The second must be (12), because (12) is the example in which ʼatʼééd follows the relative.

SOURCED What the 1974 paper prints. Four sentences have a noun after the relativizer that H&P read as the relative's head:

Hale & Platero 1974:14–15, 19, (15), (20), (36); (1) above
(15) Ashkii yiyiiłtsą́-(n)ę́ę at’ééd yidloh.
     boy saw-REL girl laugh
     ‘The girl that saw the boy is laughing.’
(20) Ashkii biiłtsą́-(n)ę́e at’ééd yidloh.
     ‘The girl that was seen by the boy is laughing.’
(36) Bił’adéłdǫǫh-ę́ę bįįh níséł’ah.
     I.shot-REL deer I.butchered
     ‘I butchered the deer I shot.’

HYPOTHESIS Willie's parse gives the same meanings. On Willie's rules (REL-EDGE, REL-ADJ), the post-relative noun is a main-clause adjunct, and the relative is a second adjunct. In (1), (15) and (20) the main verb (‘sleep’, ‘laugh’) has one slot, so both adjuncts sit on it: ‘the one who spoke, the man, is sleeping’. In (36) the main verb has a 1st-person subject and one 3rd-person slot, so again both sit on that slot. The truth conditions are H&P's in every case. The analyses differ only on constituency (one noun phrase or two adjuncts).

SOURCED Where the parses could differ, H&P's speakers side with Willie's reading. In (32) and (46) the main verb has two open 3rd-person slots, so “the noun after the relative is its head” and “the noun after the relative is a main-clause argument” give different meanings. Only the second is available:

Hale & Platero 1974:18, 22, (32), (46)
(32) Hastiin bił’adéłdǫǫh-ę́ę bįįh néís’ah.
     man I.shot-REL deer butchered
     only: ‘The man I shot butchered the deer.’          (33)
     not:  ‘The man butchered the deer I shot.’          (from (30))
(46) Ashkii łééchąą’í bishxash-ę́ę łį́į́’ yizloh.
     boy dog bit-REL horse roped
     only: ‘The boy that was bitten by the dog roped the horse.’   (48)
     not:  ‘The boy roped the horse that was bitten by the dog.’    (49)

H&P explain the missing readings by the parser (44): the noun before the relative verb is taken as its clausemate (‘I shot the man’ (45) and ‘The boy was bitten by the dog’ (47) are good simple sentences). The post-relative noun is then left for the main verb. Willie's REL-EDGE predicts the same outcome from the other end. Neither paper prints a sentence that separates the two explanations. Willie's (12) would do it, because it has no noun before the relative. H&P never judge such a sentence.

SOURCED The real speaker difference is elsewhere.

Hale & Platero 1974:15, (17)
(17) Ashkii at’ééd yiyiiłtsą́-(n)ę́ę yidloh.
     boy girl saw-REL laugh
     (a) ‘The girl that the boy saw is laughing.’
     (b) ‘The boy that saw the girl is laughing.’

H&P: “a tolerable ambiguity which is sometimes avoided by the use of the so-called ‘fourth person’” (p. 14). Willie's (20a) ʼatʼééd ʼashkii yiztsʼǫs=yę́ę yóóʼeelwod has the same shape with only one reading, the external argument as head (her REL-EXT), and she writes that the two readings Platero (1982) reports are “not true of my speech”.

HYPOTHESIS T14e verdict. (i) The claim Willie attributes to Platero is real in this 1974 paper: a noun after the relativizer can be the relative's head ((1), (15), (20), (36)), though the internally headed form is preferred. (ii) On every printed sentence, Willie's “a noun after the relative belongs to the main clause” gives the same meaning, so these data do not show a dialect difference. The difference is in the analysis (one phrase or two adjuncts). (iii) A genuine speaker difference does exist on which noun may be the head when both are inside the relative ((17) vs Willie (20a)). Another exists on how strictly the parser (44) applies ((40)–(43)). (iv) The decisive sentence, a (12)-type string judged by Platero-type speakers, is not in either source (T23a). The IJAL article may print one (T23b).

4. yi-/bi- and the parser: missing nouns are a surface problem SOURCED

4.1 The yi-/bi- principle (28)

(28) (a) In sentences of the form (NP) NP yi-V, the NP immediately preceding
         the verb is the logical object.
     (b) In sentences of the form (NP) NP bi-V, the NP immediately preceding
         the verb is the logical subject.                           (p. 17)
(29) (a) Łį́į́’ dzaanééz yi-ztał.    horse mule yi-kicked   ‘The horse kicked the mule.’
     (b) Dzaanééz łį́į́’ bi-ztał.    mule horse bi-kicked   ‘The mule was kicked by the horse.’

H&P: “This principle exists independently of deletion”. It is “a principle which will be envoked to interpret sentences from which some noun phrase has been deleted regardless of the way in which the noun phrase was removed” (p. 17). It explains why backward deletion in (14) gives a string that can only mean (16), ‘The girl that saw the boy’ (15). It also explains why the inverted relative (20) works.

HYPOTHESIS (28) is FB-3a(b), stated in 1974 by Platero and Hale. That is earlier than Platero's NP-PRO constraint (1982:288, quoted by Speas 1986:250) and the closest-NP generalization Speas reports (1986:285). In bundle terms: the noun next to the verb links to the internal bundle, and voice (FB-4) has already decided which role the internal bundle bears, the patient under yi- or the agent under bi-.

4.2 Coordination: position after voice, not role

Hale & Platero 1974:16, (22), (27)
(22) At’ééd ch’íníyá dóó ashkii yiyiiłtsą́.
     girl went.out and boy saw
     ‘The girl went out and saw the boy.’     NOT ‘… and the boy saw her.’ (25)
(27) At’ééd ch’íníyá dóó ashkii biiłtsą́.
     ‘The girl went out and was seen by the boy.’

HYPOTHESIS In both, the lone noun ashkii links to the internal bundle of the second verb (28). The girl, carried over from the first clause, fills the external bundle. She is the one who sees in (22) and the one who is seen in (27). What carries her over is a position (external), not a role in the action. That is the same pattern as Speas's (4.54) (Speas §5.4), and it is again against T-STRONG.

4.3 The clausemate parser (44), and its speaker split

(44) In a string S of the form (NP) NP V, the subsequence (NP) NP is a
     clausemate with V if the string S exists as a well-formed simple
     sentence of Navajo.                                          (p. 21)

H&P write the string as Σ; we write S, because Σ is the site's symbol for a verb's role schema.

(44) gives (32) and (46) their only readings (§3). It also covers forward deletion across a relative (pp. 19–21):

Hale & Platero 1974:19–21, (38), (41)
(38) Ashkii tl’iish bishxash-ę́ę yił adeesdǫǫh.
     boy snake bit-REL shot
     ‘The boy shot the snake he was bitten by.’
(41) At’ééd hadoolghaazh-go ashkii łééchąą’í bishxash-ę́ę yiztał.
     (from (40): girl cry.out-COMP boy [girl dog bit-REL] dog kicked)
     some speakers:  ‘When the girl yelled out, the boy kicked the dog that bit her.’ (42)
     other speakers: only ‘… the dog that bit him.’                                 (43)

For the first group, (42) “is the preferred interpretation because it corresponds to the most likely juxtaposition of events” (p. 21). H&P: (44) is “employed, with varying degrees of strictness, by speakers of Navajo” and “can be relaxed by some speakers, provided the context is sufficiently strong” (pp. 21, 23).

HYPOTHESIS Consequence for FB. FB-3a(a) said an overt noun “may not skip an open bundle” of the nearest following verb. In (42), ashkii skips the open patient bundle of bishxash (whose agent is the dog) to link to the main verb. For the first group of speakers that is good. So FB-3a(a) as a hard rule is contradicted for those speakers, and it must become a default (FB-3a(a′), §8). For the strict speakers it holds as stated. This is the same parse Platero later called “difficult to elicit” (Speas 1986:281).

4.4 The architecture (50)

(50) (i)   The syntactic rules of Navajo … which … effect the removal of a
           noun phrase constituent … apply without constraint.
     (ii)  There exist certain principles of semantic interpretation — like
           (28) and (44) — which apply at the surface structure level of
           syntactic representation to interpret grammatical relations.
     (iii) A syntactic derivation is identified as ill formed if the surface
           structure interpretation principles assign it a semantic reading
           which is at variance with the meaning assigned by the deep
           structure.                                               (p. 23)

HYPOTHESIS This is close to FB's division of labour. Who does what to whom is fixed early (for H&P in deep structure; for FB on the bundles, by Σ and Agree inside the word). Which noun goes with which participant is decided at the surface (rule set B) by (28) and (44), and a mismatch kills the derivation. The difference is that H&P store the meaning on deep-structure NPs, and FB stores it on the bundles. No datum in the paper needs the deep NPs: every judgment is computed from the surface string plus the verb's voice.

5. Enclitic raising and the determiner SOURCED

H&P's positive evidence comes from Kaufman's “unbounded rightward movement” of spatial enclitics (Kaufman 1972/1973, not opened):

Hale & Platero 1974:24, 26, (51), (52), (55)
(52) Shizhé’é chidí aa-jį’ n’iníłbą́ą́z.
     my.father car there-up.to drove
     ‘My father drove the car up to that point.’
(51) Shizhé’é chidí niiníłbą́z-í-jį’-dę́ę́’ hoołtį́į́ł.
     my.father car drove.up.to-COMP-up.to-from rain
     ‘Rain is progressing from the point which my father drove the car up to.’
(55) Shizhé’é chidí niiníłbą́ą́z-dę́ę́’ hoołtį́į́ł.
     (synonymous with (51))

The enclitic -jįʼ ‘up to a point’ is selected by the embedded verb (‘he drove it to a terminal point’). Normally it sits on a complement before the verb (52). In (51) it “appears as a suffix to the verb of the embedded clause” (p. 24), after the complementizer -í-. The main verb selects its own enclitic -dę́ę́ʼ ‘from’. The raised enclitic lands where a relative head lands, “immediately following the embedding” (p. 25). In (55) there is neither -jįʼ nor a complementizer. H&P take this to mean that -í- is part of the enclitic phrase [EP -í- -jįʼ] (56) and that the whole phrase either raises or deletes (pp. 26–27). If it stayed behind, “the enclitic would be left dangling, unattached to any phonologically constituted constituent in that clause” (p. 26).

The determiner analysis (p. 27): -í- “is a determiner or a pronoun of some kind”, perhaps the element in shí ‘I’, bí ‘he, she, it’, nihí ‘we, you (pl.)’. The relativizer that appears “whether or not the noun phrase itself is in the raised position” is “a determiner emanating from the embedded noun phrase”. There are “two versions of raising — one in which only the determiner raises and another in which both the noun phrase and the determiner raise”. The test (57): the same elements occur on simple nouns.

Hale & Platero 1974:27, (57)
(i)  Łį́į́’-ę́ę dilwo’.    ‘The aforementioned horse is fast.’
(ii) Łį́į́’-ígíí dilwo’.   ‘The (one which is a) horse is fast.’

HYPOTHESIS Why this supports FB. On H&P's own best analysis, the one element that must be present on the relative is a determiner (pronoun-like) element. The noun may raise with it (1), stay behind (4), or be absent (36). That is Willie's later “the enclitic -- a demonstrative pronoun -- is the head of the relative” (1991:158; REL-HEAD), reached from a different theory. It is also what FB predicts: what links a clause to a slot is a φ/D-level element, not a lexical noun. OPEN One wrinkle: if noun and determiner raise together, (57) leads us to expect the order noun–determiner (łį́į́ʼ-ę́ę). In (1) the determiner comes first (yáłtiʼ-ę́ę hastiin). H&P leave “the exact formulation of the raising rule” open (p. 28). The order fits Willie's two-adjunct parse of (1) better than a single raised [N-D] (T23c).

6. Rule set A vs rule set B: the word as changeover HYPOTHESIS

Grant's phenomenon 1 says that inside the verb word ordering is left-headed (rule set A, the prefix string) and above the word it is right-headed (rule set B), with the word as the changeover (headedness). Example (51) shows both sides in one phonological word:

niiníłbą́z  |  -í-  -jį’  -dę́ę́’
[verb word: prefixes + stem, built by A]  |  [clause-level enclitics, ordered by B]
(H&P do not segment the verb word; we do not either)
H&P factSideBearing
A complement precedes its verb: aa-jįʼ nʼiníłbą́ą́z (52)BHead-final clause (lexical V takes its complement on the left)
The relative precedes its head noun (1). The relativizer, complementizer and enclitics close the clause at its right edge ((1), (4), (51), (57))BIf -ę́ę/-ígíí are determiners (p. 27), these are functional heads after their complements. That favours (i′) on the headedness page: functional heads outside the word are final (T1).
Raising moves the enclitic phrase rightward to the clause edge, “unbounded” (pp. 24–25)BThe B-level landing site is on the right
The enclitic must attach to “a phonologically constituted constituent” on its left (p. 26)A/B seamB-level heads lean on the word that A builds. The changeover is a host–clitic boundary at the right edge of the verb word.
yi-/bi- inside the verb decide how the preverbal noun is read (28)A feeds BVoice is fixed in the word (A). Nouns are linked at clause level (B).

OPEN Whether -í- really is the pronominal -í of shí, bí, nihí is H&P's “perhaps” (p. 27). If it is, then the same φ-level element would be clause-final when it closes a clause (B) and inside a free pronoun. That cannot be decided from this paper.

7. What each source gets right, and what it misses

SourceGets right (kept)Misses or leaves open
Hale & Platero 1974Two surface forms of the relative. Relativization ≠ pronominalization (islands; obligatoriness). The yi-/bi- principle (28) and the parser (44) as surface interpretation. Speaker variation in (44). The relativizer as a determiner that can raise alone.Keep NPs in deep argument positions though no datum needs them. The raising rule is not formulated, and the D–N order in (1) is unexplained. They print no (12)-type sentence.
Willie 1991 Ch. 6The enclitic is the head (the same result from a different theory). Nouns as adjuncts. A noun after the relative belongs to the main clause, which is consistent with every H&P meaning.Calls the difference with Platero a possible dialect difference without a minimal pair. On the printed data it is an analysis difference. Her single-reading head choice differs from H&P (17).
Platero 1982 / Speas 1986 (via Speas)Closest-NP and NP-PRO generalizations (the descendants of (28) and (44)).Treat them as grammar. H&P 1974 show that at least (44) varies across speakers, so it acts as a parsing default.

8. Rules

FB-3a(a′) LINK-NEAR (default)  An overt noun is preferentially linked to an
                 open bundle of the nearest following verb when the noun(s)
                 and that verb would form a well-formed simple clause (all
                 of FB-3, FB-3a(b), FB-4 satisfied). For some speakers strong
                 context can override the preference (H&P (41)–(42)).
                 Replaces the categorical FB-3a(a).
FB-3a(b)  (unchanged; now SOURCED for 1974 as H&P (28))  The noun string-closest
                 to the verb links to the INTERNAL bundle; voice (FB-4) has
                 fixed which role that is.
FB-7  SURF-LINK  Bundles, their roles and coreference are fixed in the word
                 (A: Σ, Agree, FB-4). Overt nouns are linked at clause level
                 (B) by FB-3, FB-3a(b), FB-3a(a′). A derivation whose B-linking
                 contradicts the A-values is filtered out.   (≈ H&P (50))
REL-D     The relativizer (-ę́ę / -ígíí) is a D/φ-level enclitic at the right
                 edge of K that identifies one bundle of K's verb. A noun for
                 that bundle is optional: inside K (H&P (4)), after REL (1),
                 or none (36).                 (≈ H&P p. 27; Willie REL-HEAD)
B-ENC     Clause-level enclitics (COMP -í-, spatial -jį’ / -dę́ę́’, REL, -go)
                 follow the verb word and need a host on their left. A lower
                 enclitic phrase may raise rightward to the edge of its clause
                 (H&P (51)) or delete (55).
RuleStatusBasis
FB-3a(a′)SOURCED as H&P (44) and its speaker variation; bundle form HYPOTHESIS(32), (41)–(47), pp. 18–23
FB-3a(b)SOURCED (H&P (28), p. 17; Platero 1982 via Speas); bundle form HYPOTHESIS(15), (20), (22), (27), (29)
FB-7HYPOTHESIS, modelled on H&P (50) SOURCED(14)–(17), (30)–(33)
REL-DSOURCED as H&P's determiner analysis (p. 27) and Willie's REL-HEAD; φ-level HYPOTHESIS (FB-5)(1), (4), (36), (57)
B-ENCSOURCED (pp. 24–27); the A/B reading GRANT'S CONJECTURE formalized as HYPOTHESIS(51)–(56)

9. VPISH-aware derivations

Conventions as on the Speas page and derivations: Σ = role schema step; Merge/Agree/Move at clause level; A = word-internal linearization (rule set A, the prefix string); B = clause level (rule set B: nouns, clauses, enclitics). V-P = VPISH raising of the external bundle to Spec,M/S inside the word. Proto-role labels and entailments are ours (Dowty 1991 baseline). The verb words are spelled out whole, because H&P do not segment them.

HP1. (4) and (1): one meaning, two surfaces

Inner verb (speak)
Σ      Σ(√SPEAK) = {k1[ρ₁: vol, sent]}                               HYPOTHESIS (labels)
Merge  k1 in Spec,v ; V-P: k1 → Spec,M/S                              VPISH
A      spell-out yáłti’                                                SOURCED (form)
B      REL-D: -ę́ę closes K at its right edge, identifies k1           SOURCED (p. 27, as D)
B      (4): hastiin before the verb, linked to k1 (FB-3)               SOURCED (string)
       (1): hastiin after REL, linked to the same participant          SOURCED (string)
Outer verb (sleep)
Σ      {m1[ρ₁]} ; Merge m1 Spec,v ; A spell-out ałhosh
B      [K-REL] linked to m1 (REL-ADJ); in (1) hastiin also on m1
Result: same truth conditions. H&P: (4) = determiner raises alone, (1) = noun
and determiner raise. Willie: (1) = two adjuncts on m1. FB is neutral here;
the noun is optional in both.

HP2. (15) and (17): the inner links are fixed by (28); the head is not

Inner  Σ(√SEE) = {k1[ρ experiencer], k2[ρ stimulus]} ; voice yi- (FB-4)
A      spell-out yiyiiłtsą́
B      (15): ashkii string-closest → k2 (internal = object, FB-3a(b) = (28a))
             head = k1 (the seer) → 'the girl that saw the boy'            SOURCED
B      (17): at’ééd → k2 ; ashkii → k1 (two nouns, SOV default)
             REL-D may identify k1 or k2 → readings (a) and (b)             SOURCED
Outer  yidloh {m1} ; [K-REL] → m1
Result: who saw whom is fixed in both readings; only the choice of head
varies. For Willie's speakers the head must be the external k1 (REL-EXT), one
reading only. Rank (FB-4) cannot help: boy and girl are of equal rank.

HP3. (32): where the intended reading dies

Inner  Σ(√SHOOT) = {k1[π1, ρA], k2[π3, ρP]} ; A spell-out bił’adéłdǫǫh   SOURCED ('I:shot')
       REL-D: -ę́ę identifies k2 (the only 3rd-person bundle)
Outer  Σ(√BUTCHER) = {m1[ρA], m2[ρP]} ; A spell-out néís’ah
B      bįįh string-closest to néís’ah → m2 (FB-3a(b))
B      hastiin precedes bił’adéłdǫǫh; 'Hastiin bił’adéłdǫǫh' is a good simple
       clause (45) → FB-3a(a′): hastiin → k2
B      [K-REL] → m1 (the remaining slot)
Result: 'The man I shot butchered the deer' (33), the only reading (SOURCED).
The deep meaning (30), 'The man butchered the deer I shot', would need
hastiin → m1, skipping k2. FB-3a(a′) disprefers that. H&P: the pull to
read hastiin with 'I shot' "is so strong that it precludes any other
interpretation" (p. 21). Position beats the intended roles, which is
against T-STRONG.

HP4. (27): the carried-over participant is the external bundle

C1     ch’íníyá {c1[ρ₁: vol, mov]} ← at’ééd
C2     Σ(√SEE) = {k1, k2} ; voice bi- (FB-4): the agent bundle is internal
A      spell-out biiłtsą́
B      ashkii string-closest → internal (agent here; (28b))
B      null external (the patient here) = at’ééd (continuity; H&P (18))
Result: 'The girl went out and was seen by the boy.' In (22), with yi-, the
same position is the seer. The continuity is carried by external↔external,
not by the role in the action.

HP5. (41): two kinds of speaker

Adjunct  [at’ééd hadoolghaazh-go] : at’ééd → the crier
Inner    Σ(√BITE) = {k1, k2} ; voice bi- ; łééchąą’í closest → internal = biter
Main     Σ(√KICK) = {m1, m2} ; spell-out yiztał ; [K-REL] (head = the dog) → m2
B  strict:  'ashkii łééchąą’í bishxash' is a good clause (47) → ashkii → k1
            → '… kicked the dog that bit HIM' (43)
B  lenient: FB-3a(a′) overridden by context; ashkii → m1 ; k1 null = the girl
            → '… kicked the dog that bit HER' (42)
Result: the same string, two grammars of linking. Only the default rule
differs; the bundles and voice are the same.

HP6. (51) and (55): A builds the word, B places the enclitics

Embedded  Σ(√DRIVE.TO.POINT) = {k1[ρA], k2[ρ theme: the car], EP[goal]}
          EP = [-í- -jį’], the complement selected by the verb             SOURCED (p. 25)
A         spell-out niiníłbą́z / niiníłbą́ą́z (unsegmented)
B         raise EP rightward to the edge of the embedded clause → -í-jį’  (51)
          or delete EP → no -í-, no -jį’                                    (55)
Main      hoołtį́į́ł selects -dę́ę́’ 'from'; the embedded clause is its complement
B         -dę́ę́’ follows → niiníłbą́z-í-jį’-dę́ę́’
Result: the complement is selected by the head inside the word (A) and spelled
out at the clause's right edge (B). One phonological word straddles the
changeover.

10. Datum-by-datum table

H&P datum (page)H&P conclusionFB readingVerdict for FB
(1)/(4) two relative forms (pp. 9–10)One meaning, two derivationsNoun optional; REL-D identifies the bundleCOMPATIBLE
(6)/(7) forward/backward deletion (p. 11)Pronominalization by deletionNull bundles are complete (FB-2)COMPATIBLE (not discriminating)
(13) both nouns kept (p. 13)Deletion optional outside relativesTwo nouns on bundles in two clausesCOMPATIBLE (not discriminating)
(14)–(16) backward deletion misread (p. 14)Interpretation, not a rule constraintFB-3a(b)COMPATIBLE
(17) two head readings (p. 15)Tolerable ambiguity; fourth person can avoid itREL-D free between k1/k2; rank cannot decideCOMPATIBLE; speaker difference with Willie (20a)
(19)/(20) inverted relative (p. 15)Inversion feeds deletionFB-4 + FB-3a(b)COMPATIBLE
(21)–(27) coordination (p. 16)Same interpretation principleExternal↔external continuityCOMPATIBLE; against T-STRONG
(28)/(29) yi-/bi- principle (p. 17)Independent of deletion= FB-3a(b), sourced for 1974COMPATIBLE
(30)–(33), (45) deer shot (p. 18, 22)Parser (44) forces one readingFB-3a(a′) + FB-3a(b)COMPATIBLE; against T-STRONG
(34)–(36) ‘I butchered the deer I shot’ (pp. 18–19)Backward deletion fine hereNoun after REL on the one 3rd slot; no matrix subject nounCOMPATIBLE
(37)–(39) snake (pp. 19–20)Constraint (37) needs a coreference exceptionashkii links into K; null m1 = ashkii by continuity (FB-6 PAR)COMPATIBLE
(40)–(43) girl/dog, speaker split (pp. 20–21)(44) relaxable for some speakersFB-3a(a) as a hard rule fails for them; FB-3a(a′)CUTS AGAINST FB-3a(a) as categorical; core COMPATIBLE
(44), (46)–(49) clausemate parser (pp. 21–22)Surface parsing strategyFB-3a(a′)COMPATIBLE
(50) architecture (p. 23)Rules free; surface interpretation filtersFB-7 SURF-LINKCOMPATIBLE (parallel; no datum needs deep NPs)
Islands (p. 13; Platero 1973, not reprinted)Relativization obeys Ross's constraints; pronominalization does notRaising of a D/φ element is a movement dependencyCOMPATIBLE; the decisive data are unprinted (T23d)
(51)–(56) enclitic raising (pp. 24–27)Incontrovertible evidence for rightward raisingB-ENC; A selects, B spells outCOMPATIBLE (bears on headedness, §6)
(57) and pp. 26–27: relativizer = determiner, raises alone or with NRaising hypothesis winsREL-D: φ/D-level head, noun optionalSUPPORTS

Tally over 17 rows: 0 CUTS AGAINST FB's core; 1 SUPPORTS; 16 COMPATIBLE, one of them with a demotion of the borrowed FB-3a(a) to a default (FB-3a(a′)). There are two more cases against T-STRONG, which had already failed.

11. Open questions T23a–g

#QuestionWhat would decide it
T23aThe decisive T14e sentence: does a Platero/H&P-type speaker accept Willie's (12) yidloh=ę́ę ʼatʼééd biztał as ‘The girl who was laughing kicked him’ (noun after REL = head)?Speaker judgments; Platero 1974 IJAL
T23bOpen Platero 1974, “The Navajo Relative Clause”, IJAL 40:202–246 (the source Willie actually cites). Does it print (12)-type data or island data?The article
T23cIf noun and determiner raise together, why is the order REL–N in (1) but N–REL in (57)? Is (1) one phrase or two adjuncts? 2026-10-17 (8:46pm): two working answers on relative as CP §6: the noun after REL is a separate add-on (S1), or remnant raising (S2, needs the antisymmetric spine); matching with REL as D (S3) mis-orders it.Prosody, or extraction from the post-REL noun; H&P leave the rule open (p. 28)
T23dPlatero's island data: what does relativization fail to extract out of? Out of a clause linked to a bundle (an adjunct) or out of an argument? Answered 2026-10-17 (8:46pm) from Platero 1973 Ch. II (relative as CP §3.2): relativization fails out of one conjunct of dóó and out of a relative clause; it is doubtful out of an -ígíí nominalized clause (speakers split); it succeeds through direct-discourse complements and a -go complement. The islands are clauses under a determiner, whether they are arguments or not.Platero 1973 thesis, 1974 IJAL; bears on D7 (extraction)
T23eIs the lenient reading (42) available for (32) too, given a context that favours ‘the man butchered the deer I shot’?Speaker judgments; this is T22c in its sharpest form
T23fIs -í- (COMP in (51)) the same element as the -í of shí, bí, nihí (p. 27)? If so, the clause-closing head is pronominal (φ-level)Morphology of the independent pronouns (Young & Morgan)
T23gSpellings: niiníłbą́z (51) vs niiníłbą́ą́z (55); nʼiníłbą́ą́z (52); deezhtlizh, tlʼiish with plain l; yiyiisxį without toneYoung & Morgan; another printing

Sources: Hale & Platero 1974 (ERIC ED101549, page images; box copies under [local working copy, not in the repository], not committed). Willie 1991:157–158 and 162 (page images of pp. 157–158; p. 162 from the earlier Ch. 6 reading). Cited by H&P, not opened: Platero 1973 (MIT MA thesis), Kaufman 1972/1973, Hale 1973, Ross 1967, Chomsky (p.c.). Speas 1986 and Creamer 1974 were opened in earlier runs (sources).

The relative clause as a clause under a ‘the one’ head: is the shared participant raised or matched? Platero 1973's island and enclitic-phrase data (T27m)

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. 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. 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.

Working note. Kept as written (URL, data and status labels unchanged). Filed under The fourth person; the consolidated argument is the essay fourth-person. Rule IDs: rules index.

Verbs on this page → dictionary: yiztał ‘kick’ · yishxash ‘bite’ · yáłtiʼ ‘speak, talk (to)’ · yi-dloh ‘laugh’ · ʼ-ł-wosh ‘sleep’ · yíniiʼ ‘hear’ · hózin ‘be known; with shił béé: I know it’ · yíchxǫʼ ‘be ruined; (classifier) ruin’ · dilwoʼ ‘be fast’ · sidá ‘sits; with yii’/bii’ ‘in’: sits in’

Restated on the unified structure (2026-11-15). Relative clauses are restated as a full CP, built like a main clause, under an overt D⁰ (=yę́ę, =ígíí) linked to one participant by Agree; nothing moves out (H27b, H28′, H31, H34). 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. 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, relative rows. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.

The framework audit (row F10) found that the site's relative-clause pages rest on 1970s transformations: Platero's deletion rule, Hale & Platero's raising rule, and Willie 1991's “the enclitic is the head”. Task T27m asked for a restatement in today's terms, with the relative as a clause (CP) under a determiner-like enclitic, and for a decision between raising (the shared participant starts inside the relative and moves to its edge) and matching (there is a second, outside copy of the head, and the inside one is deleted under identity). The decision was to rest on the island data and the “missing noun” data. Those island data are in Platero's 1973 MIT thesis (opened this morning for the clause spine). This run read its Chapters I–II (pp. 1–55) and notes (pp. 95–96) on page images. Written 2026-10-17, 8:46pm run (PT).

Ground rules (Grant's). No analysis (Platero, Hale & Platero, Willie, Kayne) is assumed fully right. Silence counts against nothing; only contradicting data count. Grant does not speak Navajo. Every Navajo form below is copied from a Platero 1973 page image read this run (300 dpi), or from a site page that records it with page and example number. No form, gloss, judgment or index is invented. Tests that need sentences nobody printed are given in letters only. Dowty's proto-roles are the baseline, and “role in the action” names a participant's Dowty entailments. Every rule name coined here is ours and a HYPOTHESIS.

Notation. Site convention (no conventions file has arrived from the Syntax · Semantics · Interface bot). Σ = role-schema step; A = word level (the verb word plus its grammatical postpositions, spelled out as cycle D1 = the MVS); B = clause level (nouns, clauses, enclitics; cycle D2 and up). Spine as on the clause spine §4: V < v < M/S (= T) < Qual (= D1 = MVS) < Foc < Top < C. ρ = a pronoun slot (a role-bearing φ-bundle spelled as a prefix). K = the relative or nominalized clause; M = the main clause. Opρ = a silent copy of slot ρ's φ/D-bundle (ours). Glosses: Platero's colon gloss is copied as printed. His footnote 1 (p. 95) gives the order object prefix, mode, subject prefix, stem, so “3:perf:3:kick” becomes 3.OBJ-PFV-3.SBJ-kick in the Leipzig column; “imp” is his imperfective. Verbs are not segmented beyond his glosses.

Bottom line, in plain words

SOURCED What Platero found. The tie between a relative clause and the participant it is about can reach down through any number of ‘he said’ and ‘I think’ clauses ((55), (78)). But it cannot reach into one half of an ‘X and Y’ phrase (*(80), *(83)), and it cannot reach into a relative clause inside another relative (*(90), *(91), against well-formed (88)). Reaching into a clause that has been turned into a noun with -ígíí is doubtful, and speakers disagree ((96), (100)). These limits hold whether the noun stays inside the relative or follows it (p. 41). Ordinary “the same person again” links between a noun and a later null pronoun ignore all these limits ((115), (117)). Platero concludes that the limits are a filter on the relative tie, (119), and that “no movement is involved in relative clause formation in Navajo” (p. 52).

HYPOTHESIS What that decides (T27m). (1) The relative tie is a movement-type dependency, not plain coreference. Plain coreference is exactly what (115) and (117) show ignoring islands. So “the enclitic binds a pronoun” (Willie 1991:153), if “binds” means ordinary coreference, misses the island data, though her claim that the enclitic is the head survives. (2) What moves is not the noun. In the preferred form the noun stays in the middle of the relative ((12), p. 6), and very often there is no noun at all (Willie (1a–g), (2); H&P (36)). The thing that moves is a silent copy of a pronoun slot, Opρ. Hale & Platero's “only the determiner raises” (1974:27) is the closest older statement. (3) Raising versus matching is not decided by the islands. Both involve a movement step, so both predict the islands. What separates them is whether a second copy of the head stands outside the relative. When the noun is inside or absent, no outside copy is ever heard. When a noun follows the relative (H&P (1), Platero (10)), matching with the enclitic as a determiner puts the words in the wrong order (*clause–noun–ę́ę, but attested clause–ę́ę–noun). Matching survives only if the enclitic is a complementizer, which is Platero's own label (p. 2). The evidence opened so far points to a determiner (H&P (57); the object prefix yi- for an -ígíí clause in (101)). So the site adopts raising of the slot (Opρ to the edge of the clause, under the determiner -ę́ę/-ígíí) as its working analysis, with matching kept as the runner-up. The deciding data (reconstruction: idiom chunks, bound pronouns inside an outside head) are not printed in any source opened (T34a).

HYPOTHESIS The island, restated. Platero's (104) says “A relative connection cannot penetrate into a noun phrase.” Taken literally, that wording would also rule out his own possessor relatives (51)–(53), where the tie reaches into ‘its tail’. His trees (105a, b) show what he meant, a noun phrase built on a clause. Restated: ISL-D, Op cannot leave a clause that is the complement of the determiner (-ę́ę, -ígíí). Clauses with no determiner (direct-discourse complements, the -go complement of ‘walk about saying’) are open. Platero's own diagnostic matches this. An -ígíí clause takes the verb's object prefix yi- like a noun (101), while a ‘think’ complement does not (102). A clause that is linked to a slot, as a noun is, closes; a clause that is not, stays open.

HYPOTHESIS For Grant's thesis. Everything a relative can be about is a pronoun slot: the verb's subject or object (also under the inverse), the object prefix of a postposition ((45), (111)), the possessor prefix bi- on a noun ((51)), at any depth. The one participant Platero shows cannot be relativized is a bare noun with an enclitic, kin-di ‘at the house’ (*(110)). Enclitics take no pronoun prefix (Fernald et al., recorded on postposition units), so that house has no slot. Platero's explanation is that an enclitic's object is “not a full noun phrase” (p. 51; (113)/(114)). On every printed sentence the two explanations pick out the same cases. That is support for the weak thesis (the verb's, and the postposition's, arguments are slots; nouns are optional add-ons). The relative is about a slot, the noun can be inside, after, or absent, and a participant with no slot is invisible to it. No datum contradicts the weak thesis. STRATA is consistent: relativization can apply to slots before any noun is chosen. The adjunct and inside-the-clause views are not separated by these data.

SOURCED OPEN For Grant's changeover point. The verb-plus-postpositions unit (MVS) is not an island: subject, object, postpositional object and possessor slots are all relativizable. The boundary that blocks is a determiner-closed clause, which is a clause-level (rule set B) object. So the only thing that crosses from the word into the clause is the slot's silent copy, never a noun (HYPOTHESIS). How it leaves the D1 cycle is open (T34e, joins T31d).

On this page: 1 Sources · 2 The three older analyses, labelled · 3 Data · 4 Formal rules · 5 Raising, matching, binding: the test matrix · 6 The noun after the relative (T23c) · 7 Derivations RC1–RC9 · 8 What each analysis gets right and misses · 9 Grant's thesis, Dowty, STRATA, the changeover · 10 Tasks T34a–i

1. Sources opened or re-read this run (2026-10-17, 8:46pm PT)

  • Platero, Paul R. 1973. A Study of the Navajo Relative Clause. MIT M.S. thesis (DSpace 1721.1/116490; box copy [local working copy, not in the repository], PDF = typed page + 5). Newly read: Ch. I §§1–4 (pp. 1–13), Ch. II §§1.1–1.9 (pp. 14–35), §§2.1–2.3 islands (pp. 35–47), §3 Enclitic Phrases (pp. 48–52), §4 Conclusion (pp. 52–55), notes 1–5 (pp. 95–96). Page images at 150 dpi; every Navajo line in §3 below checked on 300 dpi crops (box: [local working copy, not in the repository]). The OCR text was used only to find passages. Already read this morning: pp. 31–34, 82–96.
  • Kayne, Richard 1994. The Antisymmetry of Syntax. Re-checked through a web fetch for Chapter 8 (“Relatives and Possessives”, §8.3 “N-Final Relative Clauses”, p. 92 per the table of contents). Not reached: the fetch is cut off in Chapter 6, a direct download returns HTTP 202 with no body (bot wall), and the DocsLib copy is a preview only. Only the table of contents and the passages already on A/B as spell-out (§5.1, §5.5) are used. Nothing in §6 below is attributed to Chapter 8.
  • Used as recorded on the site: Hale & Platero 1974:9–28 incl. (1), (4), (12), (36), (51)–(57), p. 27 (H&P 1974); Willie 1991:151–181 (1a–g), (2), (12), (16), (20a), (64)–(67), pp. 153, 158, 167 (relatives); Fernald et al. postposition lexicon introduction (enclitics take no pronoun prefix; postposition units); Dowty 1991:572 (27)–(28); the clause spine and FOC-LINK (clause spine).
  • Not opened: Platero 1974 (IJAL 40:202–246; T23b); Kaufman 1972, “Navajo Spatial Enclitics” (cited by Platero fn. 5 and by H&P); Brame 1968; Ross 1967 (cited by Platero; his constraint names are used here only as Platero uses them); any modern matching or raising literature beyond Kayne's table of contents.

2. The three older analyses, labelled before testing

SourceWhat it says (page)Framework labelRestated here as
Platero 1973Deep structure [S [NP [S S REL] NP] V] with two identical nouns (3), p. 2. -ígíí/-yę́ę is “a relative complementizer. In fact, it is a generalized nominalizer appearing on any sentence dominated by the NP node”, “Chomsky-adjoined” to the clause (p. 2). “We will also assume, again without strong evidence, that there is no verb phrase constituent of the node S” (p. 2). One deletion rule (22), forward or backward; the noun left inside is “clearly within the embedded sentence” (p. 6). Islands as a filter (119), because “no movement is involved” (p. 52).SOURCED (older author's pre-VPISH analysis; contrast only) Aspects-style flat S (no VP, so no position for a subject inside the verb phrase). Not counted as evidence on where nouns attach.MATCH (two copies, deletion under identity), with REL as C
Hale & Platero 1974Optional raising rule (12); the relativizer “is a determiner or a pronoun of some kind”; “two versions of raising — one in which only the determiner raises and another in which both the noun phrase and the determiner raise” (p. 27)SOURCED (older author's pre-VPISH analysis; contrast only) transformational; NPs in deep argument positionsRAISE-NP and RAISE-D
Willie 1991“the relativizing enclitic is the head … This enclitic binds a pronoun that is incorporated into the relativized verb” (p. 153); “there is no gapping” (p. 151); nouns in relatives are adjuncts (p. 158)GB (government and binding); no VP-internal subject position usedBIND (if “binds” is coreference) or RAISE-D (if it is an operator–variable tie)

SOURCED Platero also notes, against reading his “no movement” too strongly, that transparency of direct discourse is not evidence against movement: “One might be tempted to think that this is because Navajo relativization does not involve movement. However, Ellen Kaufman has shown that there is a movement rule in Navajo which allows extraction from discourse (Kaufman, 1972).” (fn. 5, p. 96)

3. Data: Platero 1973, read on page images

Transcription. Copied from 300 dpi crops of the typescript. Platero types ł as a barred l and the glottal stop as '; we write ł and ’. He often types a space before the hyphen of the relativizer (yiztał -ę́ę); we close it, and otherwise keep his hyphens, parentheses (-(n)ę́ę, -(h)ígíí: sounds that appear only in some contexts) and spacing. The past relativizer appears as -ę́ę, -yę́ę or -ą́ą “depending on phonological processes” (p. 2). The Leipzig column is ours, built from his gloss with his fn. 1 order. His translations are copied unchanged, and stars and question marks are his.

3.1 Every Platero 1973 sentence used on this page

#Navajo (as printed)Platero's glossLeipzig (ours)Translation (Platero's)p.Note
(10)Tłʼéédą́ą́ʼ ałhą́ą́ʼ-ą́ą ashkii yádoołtih.last:night imp:3:snore-REL boy fut:3:speaklast.night IPFV-3.SBJ-snore-REL boy FUT-3.SBJ-speak‘The boy who was snoring last night will speak.’5Same meaning as (12), noun after the relativizer (Platero's “backward deletion”).
(12)Tłʼéédą́ą́ʼ ashkii ałhą́ą́ʼ-ą́ą yádoołtih.last:night boy imp:3:snore-REL fut:3:speaklast.night boy IPFV-3.SBJ-snore-REL FUT-3.SBJ-speak‘The boy who was snoring last night will speak.’6Noun inside: “preceded and followed by material belonging to that sentence” (p. 6).
(31)Ashkii łééchąąʼí yiztał-ę́ę nahałʼin.boy dog 3:perf:3:kick-REL imp:3:barksboy dog 3.OBJ-PFV-3.SBJ-kick-REL IPFV-3.SBJ-bark‘The dog that the boy kicked is barking.’16Object slot. “Strictly speaking (31) is ambiguous” (it could be the boy who barks, p. 17).
(45)(Shí) łééchąąʼí bá hashtaał-ígíí nahałʼin.(I) dog for:it imp:1:sing-REL imp:3:bark(1SG) dog 3-for IPFV-1.SBJ-sing-REL IPFV-3.SBJ-bark‘The dog that I am singing for is barking.’22Object slot of a postposition. Reprinted on p. 49 as hastaał-ígíí (no sh).
(51)Łééchąąʼí bi-tseeʼ kʼínígizh-ę́ę nahałʼin.dog its-tail 3:perf:3:cut:off-REL imp:3:barksdog 3.POSS-tail 3.OBJ-PFV-3.SBJ-cut.off-REL IPFV-3.SBJ-bark‘The dog whose tail I cut off is barking.’25Possessor slot. The gloss says 3 subject, the translation ‘I’ [sic].
(55)Chidí dilwoʼ nisin ní-(n)ę́ę yíchxǫʼ.car imp:3:fast perf:1:think imp:3:tell-REL perf:3:ruincar IPFV-3.SBJ-fast PFV-1.SBJ-think IPFV-3.SBJ-tell-REL PFV-3.SBJ-ruin‘The car he said he thought was fast is broken down.’27Subject slot two clauses down, inside direct-discourse complements of ‘think’ and ‘say’ (prose p. 27: /ní/ ‘to say’).
(78)Łééchąąʼí shishxash dishni-(i)go naashá-(h)ígíí nahałʼin.dog 1:perf:3:bite imp:1:say-COMP imp:1:walk-REL imp:3:barkdog 1.OBJ-PFV-3.SBJ-bite IPFV-1.SBJ-say-COMP IPFV-1.SBJ-walk-REL IPFV-3.SBJ-bark‘The dog I claim bit me is barking.’35Into a -go-marked complement of naaghá ‘walk about (saying)’ = ‘claim’ (pp. 33–35). The typescript repeats the last relative word at the line break (naashá-(h)ígíí / naasha-(h)ígíí) and has dishni without the tone mark it has in (76).
*(80)* Łééchąąʼí dóó mósí ałhigą́-(n)ę́ę nahałʼin.dog and cat recip:imp:3:fight-REL imp:3:barkdog and cat RECIP-IPFV-3.SBJ-fight-REL IPFV-3.SBJ-bark*‘The dog which and the cat were fighting is barking.’36One conjunct of dóó ‘and’; head kept inside (forward deletion).
*(83)* Dóó mósí ałhigą́-(n)ę́ę łééchąąʼí nahałʼin.and cat recip:imp:3:fight-REL dog imp:3:barksand cat RECIP-IPFV-3.SBJ-fight-REL dog IPFV-3.SBJ-bark*‘The dog which and the cat were fighting is barking.’37Same, head after the relative (backward deletion).
(88)Hastiin łééchąąʼí bishxash-ę́ę beʼeldǫǫh néidiitą́-(n)ę́ę deesdǫǫh.man dog 3:perf:3:bite-REL gun 3:perf:3:pick:up-REL perf:3:fireman dog 3.OBJ-PFV-3.SBJ-bite-REL gun 3.OBJ-PFV-3.SBJ-pick.up-REL PFV-3.SBJ-fire‘The gun that the man who was bitten by the dog picked up went off (fired).’40Two stacked relatives, each linked to its own clause: well-formed.
*(90)* Hastiin łééchąąʼí bishxash-ę́ę beʼeldǫǫh néidiitą́-(n)ę́ę nahałʼin.man dog 3:perf:3:bite-REL gun 3:perf:3:pick:up-REL imp:3:barkman dog 3.OBJ-PFV-3.SBJ-bite-REL gun 3.OBJ-PFV-3.SBJ-pick.up-REL IPFV-3.SBJ-bark*‘The dog that the man who was bitten by picked up the gun is barking.’41‘dog’ is inside the inner relative; the outer relative tries to be about it. Head inside.
*(91)* Hastiin bishxash-ę́ę beʼeldǫǫh néidiitą́-(n)ę́ę łééchąąʼí nahałʼin.man 3:perf:3:bite-REL gun 3:perf:3:pick:up-REL dog imp:3:barkman 3.OBJ-PFV-3.SBJ-bite-REL gun 3.OBJ-PFV-3.SBJ-pick.up-REL dog IPFV-3.SBJ-bark(as *(90))41Same, head after. “regardless of the direction in which deletion applies” (p. 41).
(93)Łééchąąʼí iisxį-(n)ígíí shi-ł bééhózin.dog perf:3:kill (something)-NOM me-with is:knowndog PFV-3.SBJ-kill-NOM 1-with known‘I know the fact that the dog killed (something).’42A nominalized clause as the subject of ‘is known to me’; *(94) Łééchąąʼí shi-ł bééhózin ‘I know the dog’ shows ‘dog’ is not the head (pp. 42–43).
?(96)? Łééchąąʼí iisxį-(n)ígíí shi-ł bééhózin-ígíí nahałʼin.dog perf:3:kill-NOM me-with is:known-REL imp:3:barkdog PFV-3.SBJ-kill-NOM 1-with known-REL IPFV-3.SBJ-bark‘The dog that I know to have killed (something) is barking.’43“To some native speakers of Navajo (96) is acceptable. To others it is of questionable grammaticality.” (p. 43)
(97)Łééchąąʼí nishxash-ígíí yíniiʼ.dog 2:perf:3:bite-NOM imp:1:heardog 2.OBJ-PFV-3.SBJ-bite-NOM IPFV-1.SBJ-hear‘I heard about the dog biting you.’ or ‘I heard about the fact that the dog bit you.’44Nominalized clause as object.
?(100)? Łééchąąʼí nishxash-ígíí yíniiʼ-ę́ę nahałʼin.dog 2:perf:3:bite-NOM perf:1:hear-REL imp:3:barkdog 2.OBJ-PFV-3.SBJ-bite-NOM PFV-1.SBJ-hear-REL IPFV-3.SBJ-bark‘The dog I heard bit you is barking.’45“at least as questionable as (96)” (p. 45). yíniiʼ is glossed imp in (97), perf here (as printed).
(101)Hastiin łééchąąʼí nishxash-ígíí yi-yííniiʼ.man dog 2:perf:3:bite-NOM 3-perf:3:hearman dog 2.OBJ-PFV-3.SBJ-bite-NOM 3.OBJ-PFV-3.SBJ-hear‘The man heard about the fact that the dog bit you.’46The -ígíí clause takes the object prefix yi- “just as it does in any transitive sentence” (p. 46). The first hyphen of the gloss is overtyped.
(102)Hastiin łééchąąʼí nishxash ní-zin (*yinízin).(gloss line overtyped; not reproduced)—‘The man thinks the dog bit you.’46Direct-discourse complement: no yi- on ‘think’. Contrast (103) Hastiin atsįʼ yi-nízin (*nízin) ‘The man wants meat.’ (p. 47).
(108)Hastiin kin-di sidá.man house-at imp:3:sitman house-at IPFV-3.SBJ-sit‘The man is sitting at the house.’49Enclitic phrase (EP): -di is suffixed to a bare noun.
*(110)* Hastiin kin-di sidá-(h)ígíí naaʼíízhoozh.man house-at imp:3:sit-REL perf:3:collapseman house-at IPFV-3.SBJ-sit-REL PFV-3.SBJ-collapse‘The house the man is sitting in collapsed.’ (intended)50Printed kin- di with a space.
(111)Ashkii tʼiis yi-yaa-di sidá-(h)ą́ą naaʼííkę́ę́z.boy tree it-beneath-at perf:3:sit-REL perf:3:fallboy tree 3-beneath-at PFV-3.SBJ-sit-REL PFV-3.SBJ-fall‘The tree under which the boy was sitting fell over.’50The noun is the object of the postposition -yaa (with its prefix yi-), and -di follows. Well-formed.
(113)(Shí) díí tsékooh b-iiʼ sédá.((I) this canyon it-in imp:1:sit)(1SG) this canyon 3-in IPFV-1.SBJ-sit‘I am sitting in this canyon.’51A postposition's object takes the demonstrative díí.
*(114)* Shí díí tsékooh-di sédá.(I this canyon-at imp:1:sit)1SG this canyon-at IPFV-1.SBJ-sit*‘I am sitting at this canyon.’51An enclitic's object does not.
(115)Ashkii yah-íiyá dóó łééchąąʼí bishxash-ę́ę ahoʼniiłʼin.boy in-perf:3:walk and dog 3:perf:3:bite-REL imp:3:barkboy in-PFV-3.SBJ-walk and dog 3.OBJ-PFV-3.SBJ-bite-REL IPFV-3.SBJ-bark‘The boy walked in and the dog that he was bitten by started barking.’52‘boy’ in the first conjunct is the antecedent of the null ‘him’ inside a relative in the second: no island effect. Second i of yah-íiyá read as unaccented (faint).
(117)Ashkii sh-aa níyá-(a)go łééchąąʼí nabiłʼin dóó awééʼ háácha-ígíí yíniiʼ.boy me-to perf:3:come-COMP dog 3:perf:3:bark and baby imp:3:cry-NOM perf:1:hearboy 1-to PFV-3.SBJ-come-COMP dog 3.OBJ-PFV-3.SBJ-bark and baby IPFV-3.SBJ-cry-NOM PFV-1.SBJ-hear‘When the boy came to visit me, I became aware of the fact that the dog was barking at him and the baby had started to cry.’53The null ‘him’ sits inside a conjunct inside a nominalized clause: no island effect.

SOURCED Also used, by number only (forms not reproduced here): (29a–c), (32a–c), (39)–(42) (relatives on inverted clauses, pp. 15–21), (47)–(49) (pp. 23–24), (52)–(53) (pp. 25–26), (56), (63), (68) (pp. 27–31), (84)–(85) (p. 37), (94) (p. 42).

3.2 Where the tie reaches, in one table

The participant the relative is about is …Platero's dataResultIs there a pronoun slot for it?
subject or object of the relative's verb, direct or inverted(29)–(32), (31), (39)–(42), (12), (10) (pp. 15–21)✓yes: verb prefix (3rd-person subject zero, object yi-/bi-)
object of a postposition(45), (47)–(49) (pp. 22–24); (111) (p. 50)✓yes: b-á, b-aa, bi-chʼįʼ, yi-yaa
possessor inside the object(51)–(53) (pp. 25–26)✓yes: bi-tsee’, bi-tsiigha’, bi-gaan
a slot one or two direct-discourse clauses down(55), (56), (63), (68) (pp. 26–31)✓yes
a slot inside the -go complement of naaghá ‘claim’(78) (p. 35)✓yes
one conjunct of dóó ‘and’ (noun or clause)*(80), *(83), *(84), *(85) (pp. 36–37)✗yes, but inside a coordination
a slot inside a relative clause*(90), *(91) vs (88) (pp. 40–41)✗yes, but inside a D-closed clause
a slot inside an -ígíí nominalized clause?(96), ?(100) (pp. 43, 45)? (speakers split)yes, but inside a D-closed clause
the bare noun of an enclitic phrase*(110) (p. 50)✗no: enclitics take no pronoun prefix

SOURCED Rows 1–9 are Platero's judgments. HYPOTHESIS The last column is ours. Every ✓ row has a slot; the only ✗ row that is not an island is the one with no slot.

4. Formal rules (ours; all HYPOTHESIS unless marked)

REL-CP   [DP [CP Spec,C [C′ K Crel]] Drel]
         K = an ordinary clause on the site's spine (MVS = QualP = D1, then Foc, Top).
         Crel is silent. Drel = -ę́ę/-yę́ę (past) or -ígíí (non-past) (Platero p. 2),
         head-final, an enclitic leaning on the word to its left (rule set B).
         Determiner/head status: SOURCED (H&P p. 27; Willie pp. 153, 158). The CP layer is ours.

OP-ρ     Crel bears [rel, uφ]. It Agrees with exactly one pronoun slot ρ of K and
         attracts ρ's φ/D-bundle, a silent copy Opρ, to Spec,CP (A-bar Move).
         Drel takes Op's index: Drel = 'the one (who is ρ)'.
         The noun linked to ρ, if any, does NOT move.
         (≈ H&P's "only the determiner raises", p. 27; Willie's "binds a pronoun", p. 153,
          read as an operator–variable tie.)

ACC      Only a pronoun slot can be Op's source: a verb prefix slot, the object prefix
         of a postposition, or a possessor prefix on a noun. A bare noun with no slot
         (the object of an enclitic) cannot.                    (Platero (45), (51), *(110), (111))

ISL-D    Op cannot leave a CP that is the complement of a determiner (Drel, nominalizing
         -ígíí).                           (= Platero (104) as drawn in (105a,b); *(90), *(91); ?(96), ?(100))

ISL-&    Op cannot leave one conjunct of dóó.          (Platero §2.1; *(80), *(83), *(84), *(85))

OPEN-C   A CP with no determiner over it (direct-discourse complements of ní 'say',
         nízin 'think'; the -go complement of naaghá) does not block Op.  ((55), (63), (68), (78))

D-LINK   A D-closed clause is linked to a slot of its verb as a noun is (it can trigger
         the object prefix yi-). A clause with no D is not linked to a slot.
                                                            ((101) vs (102); (103))

CO-FREE  Coreference between a null slot and a noun elsewhere (Platero's Identical NP
         Deletion; the site's linking/continuity) is not movement and ignores ISL-D and
         ISL-&.                                                  ((115), (117), pp. 52–54)

REL-ABS, REL-EXT, REL-EDGE, REL-ADJ   Willie's choice of which slot, and where the [K D] phrase
         attaches in M, are kept as on relatives §8. Not re-tested here.
RuleStatusBasis
REL-CPSOURCED head/determiner; HYPOTHESIS CP layerH&P p. 27, (57); Willie pp. 151–158; Platero p. 2 (“generalized nominalizer”)
OP-ρHYPOTHESISIslands (§5) + noun in situ (12) + no noun (Willie (2), H&P (36))
ACCHYPOTHESIS from SOURCED judgmentsPlatero pp. 14–35, 48–52; Fernald et al. (enclitics take no prefix)
ISL-D, ISL-&, OPEN-CSOURCED as Platero's generalizations; the D/C formulation is oursPlatero pp. 35–47; (104) is tentative: “The results of this inquiry are by no means firm” (p. 45)
D-LINKSOURCED (101)–(103) and Platero's reading of them (pp. 46–47); slot wording oursp. 46: the nominalized sentences “appear to function as true noun phrases”
CO-FREESOURCED (Platero pp. 52–54)(115), (117)

HYPOTHESIS Why ISL-D and not “no noun phrase”. Platero's (104) as worded would forbid the tie to “penetrate into a noun phrase”. The possessor relatives (51)–(53) do penetrate the noun phrase ‘its tail’, and Platero calls them acceptable (p. 25). His (105a, b) trees put an S under the NP both times. So his intended island is a noun phrase built on a clause, which ISL-D states. On the site's terms the possessor slot bi- sits at the left edge of the possessed noun, so Op can leave from the edge (T34c).

5. Raising, matching, binding: the test matrix

Four ways to implement the relative tie. All four keep the enclitic as the head of the whole phrase.

  • RAISE-NP: the noun itself moves to Spec,CP (H&P's “noun phrase and determiner raise”).
  • RAISE-D = OP-ρ: only the slot's bundle moves; the noun stays where linking puts it (H&P's “only the determiner raises”).
  • MATCH: a second, outside head stands next to [K REL]. Inside K a copy (or operator) moves, and one of the two copies is deleted under identity (Platero 1973's two-copy structure (3), with deletion in either direction (22)).
  • BIND: no movement; REL binds the slot the way a noun or antecedent is linked to a slot (CO-FREE).
Test (data)RAISE-NPRAISE-D (OP-ρ)MATCHBIND
T1 islands: *(80)/*(83), *(90)/*(91) vs (88), both deletion directions✓✓✓ (Platero needs a separate filter (119); with an operator it follows)✗ (plain coreference ignores islands: T2)
T2 coreference ignores the same islands: (115), (117)✓✓✓ (Platero's own argument, pp. 52–54)✗ (the two ties would be the same relation)
T3 noun inside, after an adverb of K: (12) vs (10)✗ overtly (the noun is not at K's edge); needs covert raising✓✓ (outside copy deleted, inside copy kept)✓
T4 no noun at all: Willie (1a–g), (2); H&P (36)needs a silent noun to move✓ (a slot is always there)needs two silent nouns, identical✓
T5 noun after REL: H&P (1), Platero (10); vs N–REL on a bare noun, H&P (57)head-final spine: ✗ (Spec,CP is on the left); antisymmetric remnant: ✓ (§6)✓ if the noun is a separate adjunct of M (Willie REL-EDGE)with REL = D: ✗ (predicts clause–N–REL); with REL = C: ✓✓ (two adjuncts)
T6 enclitic phrase: *(110) vs (111); (113)/*(114)✓ if a bare N cannot move✓ (ACC: no slot)✓ (Platero: identity fails, p. 51)? (nothing to bind; needs ACC too)
T7 possessor (51)–(53), noun before ‘its tail’ in placeneeds left-branch movement of the noun✓ (slot at the edge)✓✓

HYPOTHESIS Verdict for T27m. BIND, taken as plain coreference, fails T1/T2. Willie's head claim is kept, but her “binds” has to be an operator tie. Overt RAISE-NP fails T3 for the form Platero calls preferred (p. 4). RAISE-D passes every test. MATCH passes every test only if REL is a complementizer (T5) and silent identical nouns are allowed (T4). The evidence opened so far favours REL as a determiner: H&P (57) Łį́į́ʼ-ę́ę dilwoʼ ‘The aforementioned horse is fast’; Platero's own “generalized nominalizer appearing on any sentence dominated by the NP node” (p. 2); and D-LINK ((101): the -ígíí clause triggers the object prefix like a noun). So RAISE-D is adopted as the working analysis and MATCH is kept as the runner-up. The islands do not choose between them; they choose movement over coreference. The decisive data would be reconstruction effects with a pronounced outside head (T34a), and none are printed in the sources opened.

HYPOTHESIS Platero 1973 vs Hale & Platero 1974, re-scored. Platero's 1973 “no movement is involved” is right about the noun (it stays put, p. 6) and, in today's terms, wrong about the tie (it behaves like movement). H&P 1974 cite the island result as one reason to prefer raising. That is right as far as it goes, but islands are equally compatible with matching plus operator movement. Their second argument, enclitic raising ((51)–(56)), is about the enclitic phrase, not the noun.

6. The noun after the relative (T23c)

SOURCED On a bare noun the enclitic follows the noun: H&P (57) Łį́į́ʼ-ę́ę dilwoʼ. In a relative with the noun after it, the enclitic precedes the noun: H&P (1) Tłʼéédą́ą́ʼ yáłtiʼ-ę́ę hastiin ałhosh; Platero (10) Tłʼéédą́ą́ʼ ałhą́ą́ʼ-ą́ą ashkii yádoołtih. With the noun inside, as in (12), nothing follows the enclitic but the main verb.

HYPOTHESIS Three structures for (10), schematically:

S1 (two adjuncts; Willie REL-EDGE)   [DP [CP Opρ [K adv V]] Drel]   N   VM
     The DP and the noun are separate add-ons, both linked to the main verb's slot.
     Order adv–V–REL–N–VM ✓. Meaning = H&P's on every printed sentence (T14e).

S2 (antisymmetric remnant raising; Kayne-style, built only from Kayne §5.1/§5.5)
     [DP [K adv V]j [D′ Drel [CP N [C′ C tj]]]]   VM
     The noun raises to Spec,CP; the rest of K then moves to Spec,DP.
     Order adv–V–REL–N ✓. One phrase. Needs the antisymmetric spine (T29b open).

S3 (matching with REL as D, one phrase)   [DP [NP [CP Op [K adv V]] N] Drel]
     Order adv–V–N–REL ✗ (unattested in the sources for this meaning).

HYPOTHESIS So T23c (“why REL–N in (1) but N–REL in (57)?”) has two answers that work. S1: the noun after the relative is not inside the relative's phrase at all. S2: it is, by remnant raising. S3 is out. Chapter 8 of Kayne 1994 (§8.3 “N-Final Relative Clauses”) was not reachable, so S2 is not attributed to Kayne's own analysis of prenominal relatives (T34g). Between S1 and S2 the deciding sentence is still T23a (Willie's (12) yidloh=ę́ę ʼatʼééd biztał judged by a Platero-type speaker).

7. VPISH-aware derivations RC1–RC9

Format as on the clause spine. Every statement about where things merge or move is ours (HYPOTHESIS). The strings, glosses and judgments are Platero's (SOURCED). Role labels follow his translations, and entailments are read Dowty-style (role in the action). V-P = VPISH raising of the external slot from Spec,v to Spec,M/S inside the word (the site's working choice). Under V-H (in situ) nothing below changes except the launching site of Op for a subject.

RC1. (12) Tłʼéédą́ą́ʼ ashkii ałhą́ą́ʼ-ą́ą yádoołtih: the noun stays inside, a slot copy moves

  1. Σ K: Σ(√SNORE) = {ρ1} (the snorer: sentience, a body-internal process; no other participant). π = 3.
  2. A Merge V (unsegmented), v; ρ1 in Spec,v (VPISH); Merge M/S; Agree M/S ⟶ ρ1 [3]; Move ρ1 → Spec,M/S (V-P). Merge Qual. Spell-out D1 = MVS = ałhą́ą́ʼ.
  3. B Merge Foc, Top. Link ashkii ⟶ ρ1 (FOC-LINK: the noun right before the MVS). Merge (adjoin) the adverb tłʼéédą́ą́ʼ above it.
  4. B Merge Crel [rel, uφ]. Agree Crel ⟶ ρ1. Move Opρ1 (silent) → Spec,CP. ACC ✓ (ρ1 is a slot). The noun ashkii is not moved, so it stays between the adverb and the verb, as on p. 6.
  5. B Merge Drel -ą́ą (past) with CP; D takes Op's index: ‘the one who snored’. Spell-out: the enclitic leans on the verb word: ałhą́ą́ʼ-ą́ą.
  6. Σ M: Σ(√SPEAK) = {m1}. A as step 2 → yádoołtih. B Link [DP K D] ⟶ m1 (REL-ADJ). Spell-out D2: Tłʼéédą́ą́ʼ ashkii ałhą́ą́ʼ-ą́ą ⌢ yádoołtih ✓.

HYPOTHESIS (10) differs only at step 3: ashkii is not linked inside K, and is linked to m1 as a separate add-on after the DP (S1 of §6). Both mean the same, as Platero says (pp. 5–6).

RC2. (45) (Shí) łééchąąʼí bá hashtaał-ígíí nahałʼin: the slot is on a postposition

  1. Σ K: Σ(√SING) = {ρA [π 1] (volition)}; the beneficiary is the object slot ρX [3] of -á ‘for’. Whether -á is grammatical or lexical is not settled by Platero (T32 tasks).
  2. A Merge P -á with ρX; Agree P ⟶ ρX → b- (a 1st-person subject, so the plain 3rd-person form: one-place verb + P). V (unsegmented), v, ρA Spec,v → Spec,M/S. Spell-out: bá ⌢ hashtaał.
  3. B Link łééchąąʼí ⟶ ρX (next to the PP). Optional shí ⟶ ρA.
  4. B Crel Agree ⟶ ρX; Move OpρX → Spec,CP (ACC ✓: the prefix b- is a slot). Merge D -ígíí (non-past): ‘the one I sing for’.
  5. B M: nahałʼin {m1}; [DP] ⟶ m1 ✓.

RC3. (51) Łééchąąʼí bi-tseeʼ kʼínígizh-ę́ę nahałʼin: the slot is a possessor

  1. Σ K: Σ(√CUT.OFF) = {ρA (volition, causation), ρP (change of state, affected)}. ρP is filled by the noun phrase [ρPoss -tsee’], whose possessor prefix is a slot of its own (ρPoss [3], bi-).
  2. A Verb word as usual → kʼínígizh (Platero's gloss 3:perf:3; translation ‘I’ [sic]).
  3. B Link bi-tsee’ ⟶ ρP; Link łééchąąʼí ⟶ ρPoss (it stands right before the possessed noun).
  4. B Crel Agree ⟶ ρPoss; Move OpρPoss from the left edge of the possessed noun phrase → Spec,CP. No determiner closes ‘its tail’ around a clause, so ISL-D is not triggered. Platero's (104), read literally, would block this; his (105) trees do not (§4).
  5. B D -ę́ę: ‘the one whose tail was cut off’; [DP] ⟶ the slot of nahałʼin ✓.

RC4. (55) Chidí dilwoʼ nisin ní-(n)ę́ę yíchxǫʼ: through two clauses with no determiner

  1. Σ K3: Σ(√FAST) = {ρTh} (π 3). K2: Σ(√THINK) = {ρExp [π 1]} + a direct-discourse clause. K1: Σ(√SAY) = {ρA [π 3]} + a direct-discourse clause.
  2. A Three verb words: dilwo’, nisin, ní. Neither ‘think’ nor ‘say’ takes the object prefix for its clause ((102); D-LINK): the clauses are not linked to slots.
  3. B Link chidí ⟶ ρTh in K3. Merge K3 as a bare CP complement in K2, and K2 in K1.
  4. B Crel on K1: Agree ⟶ ρTh of K3; Move Op through Spec of each bare CP to Spec,CPrel. OPEN-C ✓ at each step (no D above K3 or K2). Merge D -(n)ę́ę.
  5. B M: yíchxǫʼ {m1}; [DP] ⟶ m1 ✓ ‘The car he said he thought was fast is broken down.’

(78) is the same, with the ‘say’ clause inside the -go complement of naaghá: Op passes a CP headed by -go. So -go (C) does not close a clause, while -ę́ę/-ígíí (D) does. That fits the clause spine's choice of =go as C (clause spine). Whether an adjunct =go clause (Willie's chains) is an island is not tested by Platero (T34i).

RC5. *(90), *(91) vs (88): one relative inside another

  1. Σ Inner Kin: Σ(√BITE) = {ρA (the dog: volition, causation, contact), ρP (the man: affected)}, inverse bishxash (the agent slot is internal). Outer Kout: Σ(√PICK.UP) = {σA (the man), σP (the gun)}.
  2. B (88): in Kin, OpρP → Spec,CPin; D -ę́ę: ‘the one the dog bit’ = the man. [DPin] ⟶ σA. In Kout, OpσP → Spec,CPout; D -(n)ę́ę: ‘the gun that …’. [DPout] ⟶ the slot of deesdǫǫh ‘fired’ ✓. Each Op moves only inside its own clause.
  3. B *(90)/*(91): the outer relative is meant to be about the dog, which is ρA of Kin. Crel of Kout Agree ⟶ ρA(Kin); Move Op out of CPin, which is the complement of D -ę́ę. ISL-D ✗. The crash happens whether the noun łééchąąʼí is linked inside Kin (*(90)) or after the outer relative (*(91)), because the noun is never what moves. This derives Platero's “regardless of the direction in which deletion applies” (p. 41).

RC6. *(80) Łééchąąʼí dóó mósí ałhigą́-(n)ę́ę nahałʼin: one conjunct

  1. Σ K: Σ(√FIGHT.RECIP) = {ρ1 [plural or dual reading; reciprocal]}. One slot.
  2. B Link [&P łééchąąʼí dóó mósí] ⟶ ρ1. The dog is not a slot of its own, only part of the phrase linked to ρ1.
  3. B Intended: ‘the dog (which with the cat fought)’. Op would have to be the dog alone. ACC ✗ (no slot for one conjunct) and ISL-& ✗. With Op = ρ1 the relative means ‘the ones who fought’, which is (82)-type, not the intended reading (p. 37).

HYPOTHESIS On the slot view, the coordinate-structure ban partly follows from ACC: a conjunct has no slot of its own. ISL-& is still needed for (85), where the conjunct is a whole clause with its own slots (letters-only check: Op from a slot of the second conjunct clause [S1 dóó S2] ✗).

RC7. *(110) vs (111): no slot, no relative

  1. Σ (110) K: Σ(√SIT) = {ρA (the man; sentience)}. The location kin-di ‘at the house’ is an enclitic phrase: -di is suffixed to a bare noun and has no pronoun prefix.
  2. B Crel looks for a slot with the house's index. There is none. ACC ✗ → *(110). (Platero: the enclitic's object is “not a full noun phrase”, so identity with the head fails; p. 51.)
  3. Σ (111) K: Σ(√SIT) = {ρA (the boy)}; the location is [yi-yaa]-di: the postposition -yaa ‘beneath’ has an object slot ρX (yi-), and -di ‘at’ attaches outside it.
  4. A Agree P ⟶ ρX: yi-. (With a 3rd-person subject this is the grammatical-P form; whether -yaa joins the MVS here is open, T33d.)
  5. B Link tʼiis ⟶ ρX. Crel Agree ⟶ ρX; Move OpρX → Spec,CP. ACC ✓. D -(h)ą́ą: ‘the one the boy sat under’ ⟶ the slot of naaʼííkę́ę́z ✓.

HYPOTHESIS The same split may explain why H&P's enclitic raising ((51), H&P §5) works. There the relativized goal of -jįʼ is the pronoun-like -í- (H&P: “a determiner or a pronoun of some kind”, p. 27), not a bare noun. A pronominal element can be Op's source; a bare noun cannot. Whether -í- counts as a slot is OPEN (T23f, T34g).

RC8. (115): coreference crosses both islands

  1. B Conjunct 1: Ashkii yah-íiyá: ashkii ⟶ the walker's slot.
  2. B Conjunct 2: inside the relative, Σ(√BITE) = {ρA (the dog), ρP (the bitten one)}, inverse bishxash. OpρA → Spec,CP (the relative is about the dog: ‘the dog that he was bitten by’). ρP has no noun.
  3. B CO-FREE: ρP is coreferent with ashkii in conjunct 1. This tie crosses a coordination and enters a D-closed clause, and it is fine (p. 53).

HYPOTHESIS So the relative tie (Op movement) and the noun-to-slot or antecedent-to-slot tie (coreference) are two relations with different locality, as Platero argued (pp. 52–54). On the site, noun linking (FB-3, FOC-LINK) belongs with the second. This is a small point for T31d: linking a noun to a slot behaves like interpretation, not like Agree-plus-movement.

RC9. Letters only: what would separate RAISE-D from MATCH (T34a)

Needed: a relative whose head noun N is pronounced OUTSIDE K (after REL), where
        N contains something that must be interpreted INSIDE K:
  (a) an idiom chunk whose idiomatic reading needs K's verb   [Nidiom]  … Vidiom-REL  Nidiom?
  (b) a possessor pronoun in N bound by a quantifier inside K
RAISE-D: N is not in K at any point → no idiomatic reading in (a), no binding in (b)
         (S1), unless S2 (remnant raising), which predicts both.
MATCH:   an inside twin of N exists, so an inside (idiomatic / bound) reading is at
         least available in principle; which copy is read is not fixed by Platero 1973.
No such sentence is printed in Platero 1973, H&P 1974, or Willie 1991. Not constructed.

8. What each analysis gets right, and what it misses

SourceGets right (kept)Misses or leaves open
Platero 1973The full accessibility pattern (subject, object, inverted, postpositional object, possessor, any depth through direct discourse). The islands, and that they hold for both surface forms (p. 41). That plain coreference does not obey them (pp. 52–54). The enclitic-phrase contrast and the díí test ((110)–(114)). The yi- test for clauses that act as nouns ((101)–(103)). Honest marking of weak data ((96), (100); “by no means firm”, p. 45).“No movement” (p. 52): in today's terms the tie behaves like movement of a silent element, which his own fn. 5 leaves room for. A flat S with no VP (p. 2), so no place for a subject inside the verb phrase. (104) as worded overreaches his possessor data. REL called a complementizer, but also “a generalized nominalizer”, the same element as on simple nouns.
Hale & Platero 1974Raising beats deletion; REL is a determiner; “only the determiner raises” is available (p. 27). That is the analysis adopted here (RAISE-D).The islands do not by themselves favour raising over matching with an operator. The raising rule is not formulated, and the REL–N order of (1) is unexplained (T23c; §6 gives two answers).
Willie 1991The enclitic is the head; no gap is pronounced; nouns in relatives are add-ons; a noun after REL belongs to the main clause (S1 in §6 works on all printed data).“Binds a pronoun”, read as coreference, cannot tell the island-sensitive relative tie from the island-free coreference tie (T1/T2). Island data are not discussed (silence, not counted against her). Read as an operator tie, her analysis becomes RAISE-D.
Kayne 1994 (as far as opened)Final heads as complements moved to a higher specifier (§5.1, §5.5), which makes the remnant structure S2 available.Ch. 8 not opened; nothing about relatives is attributed.

9. Grant's thesis, Dowty, STRATA, and the changeover

GRANT'S CONJECTURE Weak thesis: the verb's arguments are role-bearing pronoun slots, and nouns are optional add-ons linked to them. HYPOTHESIS Verdict: supported, no contradicting datum. (i) What a relative is about is always a slot (§3.2). (ii) The noun can be inside (12), after (10), or absent (Willie (2)), and nothing in the derivation of the relative changes. (iii) A participant with no slot, the house in *(110), cannot be relativized, though it is plainly a participant in the situation. (iv) The relative tie is a different relation from noun linking (T1/T2), as the thesis expects if nouns are add-ons linked by interpretation. The refinement from one-place verb + P stands. Slots also occur outside the verb's own frame (on postpositions and possessed nouns), so the thesis reads “the arguments are slots”, not “every slot is the verb's argument”.

HYPOTHESIS Strong thesis (nouns are never arguments) is not tested by these data. Platero's own explanation for *(110), that an enclitic's object is “not a full noun phrase” (p. 51), treats full noun phrases as the things that relativize. ACC treats slots that way. On every printed sentence they coincide, because every full noun phrase in the data sits on a slot (T34f).

HYPOTHESIS Dowty baseline. Relativization is blind to the role in the action. Agents, patients, beneficiaries of -á, possessors and themes are all accessible, and the inverse changes nothing about accessibility. Dowty's proto-role entailments fix which slot is which in Σ (step 1 of every derivation). The relative then sees only slots and their internal/external position (Willie's REL-ABS/REL-EXT). That is an extension of the Dowty baseline (proto-roles build the slots; clause-level operations see the slots), not a departure from it.

GRANT'S CONJECTURE STRATA (theta roles first, event-structure roles as another layer, noun selection later). HYPOTHESIS Consistent, and it makes one sharp prediction that the data meet: no condition on relatives refers to which noun is chosen. ACC refers to slots (Layers 1–2), the islands to clause structure, and REL-EXT to whether nouns are present, not which ones. The house in (110) enters only at the noun layer, with no slot from Layer 1 or 2, so it is invisible to the relative. Three views compared: the adjunct view (Willie: nouns are add-ons) and STRATA both say “the relative is about a slot”. The inside-the-clause view (Platero, Speas: nouns are arguments) says “the relative is about a full noun phrase” (p. 51). On printed data all three make the same judgments, so they are not separated here. T34f is the test that would separate them.

GRANT'S CONJECTURE Changeover (the MVS as the point where rule set A hands over to rule set B). HYPOTHESIS The relative supports a clean division. Rule set A builds the slot inside the word (including the postposition's and the possessor's prefix). Rule set B builds Crel, moves the slot's silent copy, and spells the determiner as an enclitic on the word's right edge. The MVS is not a barrier, since every slot inside it is accessible. A D-closed clause is a barrier, and it is a rule-set-B object. The only thing that crosses from A to B is Opρ, never a noun. OPEN If D1 (the MVS) is a spell-out cycle, Op must leave from its edge. A subject slot in Spec,M/S is below Qual, and object slots are lower still. So either every slot can pass through Spec,QualP, or D1 is not a cycle for Op (T34e, joins T31d).

HYPOTHESIS Lexicalization at the phrase level. Nothing here is idiomatic, but D-LINK adds one phrase-level unit: an -ígíí clause behaves as a single noun-like unit for its verb (it gets yi-, (101)). The relative [K D] is a phrase that functions as one word-sized add-on in the main clause. That is phrase-level packaging, not lexical storage.

10. Tasks T34a–i

#QuestionWhat would decide it
T34aRAISE-D vs MATCH: reconstruction with a pronounced outside head (idiom chunk, bound pronoun) (RC9)A speaker; Platero 1974 (IJAL) if it prints such data
T34bIs -ę́ę/-ígíí a D (H&P (57); D-LINK) or a C (Platero p. 2) in relatives, or both (a D over a silent C)? Joins T14d, T23fCo-occurrence with díí ‘this’ on a relative; Young & Morgan on the enclitics
T34c(104) vs the possessor relatives (51)–(53): is the island only “CP under D”? Can a possessor be relativized out of a possessed noun that is itself inside a relative?Letters only: [[N1 bi-N2 V]-REL N3 V]-REL … with the outer relative about N1; a speaker
T34dThe speaker split on ?(96), ?(100): is the nominalized-clause island weaker than the relative-clause island? Does it differ for factive =ígíí vs =nígíí with ‘believe’ (Willie (64)–(67))?Speakers; Willie's ‘believe’ data with a relative added
T34eHow does Op leave the D1 (MVS) cycle: through Spec,QualP, or is D1 not a cycle for Op? (joins T31d, T29a)Theory-internal; a relative about a slot inside a word that contains a da’/doo-type disjunct (T29a)
T34fACC (slots) vs Platero's “full noun phrase” condition: a phrase that takes díí but has no pronoun slot, or a slot whose noun cannot take dííA search of Young & Morgan; a speaker. Separates the three views of §9.
T34gH&P's enclitic raising (51) vs Platero's *(110): is -í- a pronominal slot (so ACC is met)? Is Kayne's §8.3 analysis of prenominal relatives S2?Kaufman 1972; Kayne 1994 ch. 8 (access blocked this run)
T34hDoes Platero 1974 (IJAL) keep the 1973 island data and judgments, especially (96), (100), and the possessor case?The article (= T23b)
T34iIs an adjunct =go clause (Willie's chains) an island for the relative, unlike the -go complement of ‘claim’ in (78)?Letters only: [[N V=go] V-REL] … with the relative about a slot inside the =go clause; a speaker

Sources: Platero 1973 (MIT M.S. thesis; page images, box copies under [local working copy, not in the repository], not committed). Hale & Platero 1974, Willie 1991, Fernald et al. and Dowty 1991 as recorded on the linked pages. Kayne 1994 table of contents only for Chapter 8. Not opened: Platero 1974, Kaufman 1972, Brame 1968, Ross 1967.

The one who … : relative clauses, nominalized clauses and the fourth person

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.

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.

Working note. Kept as written (URL, data and status labels unchanged). Filed under The fourth person; the consolidated argument is the essay fourth-person. Rule IDs: rules index.
Verbs on this page 14 entries in the verb dictionary (show)

yiztał ‘kick’ · néidííłtsʼin ‘hit with the fist’ · jizloh ‘rope’ · yiztsʼǫs ‘kiss’ · dabiiłtsą́ ‘see’ · nayiisniiʼ ‘buy’ · yi-dloh ‘laugh’ · yóóʼeelwod ‘run away’ · yoodlą́ ‘believe’ · honeesná ‘win’ · yíniiʼ ‘hear’ · ʼínáályaa ‘be fixed’ · nizin ‘think; wish (with an optative clause)’ · hózin ‘be known; with shił béé: I know it’

Restated on the unified structure (2026-11-15). Relative clauses are restated as a full CP, built like a main clause, under an overt D⁰ (=yę́ę, =ígíí) linked to one participant by Agree; nothing moves out (H27b, H28′, H31, H34). 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, relative 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 6 (relative and nominalized clauses, printed pp. 151–181). Four results bear on Grant's two phenomena. (1) A Navajo relative clause is an ordinary full clause with an enclitic =yę́ę ‘the one …’ added at its right edge. Willie argues the enclitic is itself the head of the relative (pp. 153, 158). The enclitic sits at clause level (rule set B), and the pronoun it picks out is a prefix inside the verb word (rule set A). (2) Which argument the relative is “about” is decided by slot and person, not by a word. By default it is the absolutive (internal) argument (16). The inverse makes the agent internal, so yiztał=yę́ę is ‘the one he kicked’ and biztał=yę́ę is ‘the one who kicked him’ (1c, 1d). A fourth person can never be the head of a restrictive relative (p. 175). (3) Parallel processing holds with relatives: two all-third-person transitive clauses keep the same two referents, and the main verb's yi-/bi- decides which one the relative names ((27)–(31), p. 166). This answers T12f for relatives. (4) With ‘believe’ and a nominalized clause, two third persons may be the same or different people (64). So across four constructions, the plain third-person default comes out three ways (same, different, free), while the fourth person behaves the same in all of them (one fourth person = different, two = same).

On this page: 1 Question · 2 Source · 3 How a relative is built · 4 Which argument is the head · 5 Two transitive clauses · 6 The fourth person · 7 Nominalized clauses · 8 Rules · 9 The 3/3 default in four constructions · 10 Derivations R1–R6 · 11 Bearing on T1, T12a, T12f, T13d · 12 Open

1. The question

The reported-speech page showed that the 3/3 default flips between =go chains (same cast) and ‘say’ + complement (different subjects). It left T12f open for relative clauses and T13d open (is the default set by the construction?). For Grant's two phenomena, this page asks:

  • Theta roles as arguments (phenomenon 2). When a relative clause names ‘the one who …’, what picks the slot it names: a role (agent/patient), a position (internal/external), or a person value? No noun is needed (1a–g), so the choice cannot come from a lexical item.
  • Rule set A vs rule set B (phenomenon 1). The relativizer is an enclitic at the right edge of a clause, and the slot it binds is a prefix inside the verb. Where is the changeover?

2. Source opened this run (2026-10-10 afternoon PT)

  • Willie, MaryAnn 1991. Navajo Pronouns and Obviation. PhD dissertation, University of Arizona (scan on the box). Chapter 6, printed pp. 151–181 (PDF = printed + 6), read as 150 dpi page images. Every Navajo line quoted here was checked on 300 dpi crops. The OCR text layer was used only to find pages.

Transcription and glosses. Navajo is copied as printed (glottal stop as ’). Glosses are Willie's, copied exactly; her gloss order is object.subject, which she restates on p. 153: “(Recall that the object pronoun precedes the subject pronoun in the Navajo verb.)” Subscript indices are hers. Willie gives no Leipzig line, and we add none. Where print and gloss disagree we copy both and mark [sic].

Print notes. The enclitic is mostly printed =yę́ę, but =ę́ę in (11)–(15), (25), (44); with a hyphen for “=” in (18), (26), (28); without nasal hooks (=yée) in (40) and (46); with a space in (4) haztał =yę́ę. (43) prints “=nę́ę” glossed “=REl”. (66) prints =ígíí where (64)–(65) have =nígíí. Plain l where ł is expected: (30), (32) yiyiiltsą́, (33) biiltsą́. (22) prints yóóʼeelwcd. The first vowel of yóóʼ is typed with a mark that looks like a double acute; we write ó. There is no example (23) in this chapter. (5) is repeated on p. 163; (60)–(61) repeat Ch. 5 (17)–(18).

Cited by Willie but not opened by us: Willie 1990; Platero 1974, 1982; Akmajian 1977; Schauber 1979.

3. How a relative is built (§6.0–6.2) SOURCED

Willie 1991:151: the relative is formed “by enclitizing to a main clause a relativizing determiner which means ‘the one...’”; “there is no gapping”; the enclitics are =ígíí and =ę́ę; “Any argument type can be relativized”. P. 153: these relatives are not headless; “the relativizing enclitic is the head (Willie, 1990). This enclitic binds a pronoun that is incorporated into the relativized verb.”

Willie 1991:151–153, (1a–g)
(1a)łichííʼ=yę́ę
3.red=REL
‘the one that is red’ (relativized subject of neuter verb)
(1b)yicha=yę́ę
3.crying=REL
‘the one who was crying’
(1c)yiztał=yę́ę
3.3.kicked=REL
‘the one he kicked’
(1d)biztał=yę́ę
INV 3.3.kicked=REL
‘the one who kicked him’
(1e)bichʼįʼyáshtiʼ=yę́ę
3.to1.speak=REL
‘the one I spoke to’
(1f)biʼdistał=yę́ę
PASS 3.3.kicked
‘the one who was kicked’ (no REL in the printed gloss)
(1g)shiłbééhózin=ígíí
1.with3.about3.known=REL
‘that which is known to me’

(1g) is the relative of shił béé hózin ‘I know it’, a psych verb whose two participants are both on postpositions; Chs. 2 and 4 gloss hózin “ho-known” (pleonastic & psych §5.2, T17g).

Willie 1991:153–157, (2)–(10)
(2)yiztał=yę́ęyicha
3i.3j.kick=RELi3i.crying
‘The one he kicked is crying.’
(3)jiztał=yę́ę
3.4.kicked=REL
‘the one that person kicked’
(4)haztał =yę́ę
4.3.kicked=REL
‘the one who kicked that person’
(5)ní nidloh=yę́ękʼad nicha
you 2.laugh=RELnow 2.cry
‘You who were laughing are now crying.’ (teasing)
(7)yidloh=yę́ęyidootsʼǫs
3.laughing=REL3.3.will kiss
‘He will kiss the one who was laughing.’
(8)yidloh=yę́ębidootsʼǫs
3.laughing=RELINV 3.3.will kiss
‘The one who was laughing will kiss him.’
(9)ʼatʼéédyidootsʼǫs
girl3.3.will kiss
‘He will kiss the girl.’
(10)yidloh=yę́ębidiʼdootsʼǫs
3.laughing=RELPASS.3.3.will kiss
‘The who was laughing will be kissed.’ [sic]

P. 156: “the Direct/Inverse voice alternation of the main clause verb affects the theta role assigned to the external argument to which the relative is adjoined.” Willie compares (7) with (9): the relative and the noun are “both referring expressions”, adjoined the same way.

Willie 1991:157–158, (11)–(12)
(11)ʼatʼéédyidloh=ę́ębiztał
girl3.laughing=RELINV3.3.kicked
‘The girl who was laughing kicked him.’
(12)yidloh=ę́ęʼatʼéédbiztał
3-laugh=REL†girlINV 3-3-kicked
‘The girl kicked the one who was laughing.’

SOURCED † Segmentation note (2026-11-11 audit pass 4): Willie 1991 (12) prints yidloh=ę́ę with the gloss “3-laugh=REL”: the gloss names more morphemes than the form shows as separate pieces (fused or unwritten morphemes), so the two lines do not match one-to-one in the source. Kept exactly as printed.

P. 158: a noun that follows the relative cannot be its adjunct (Willie notes Platero 1974 differs: “may be a dialect difference”). 2026-10-15: checked against Hale & Platero 1974 (page). The 1974 paper does print post-relative nouns read as heads, but on every printed sentence Willie's parse gives the same meaning (T14e). “nominals are not arguments, but adjuncts in relative clauses … the enclitic -- a demonstrative pronoun -- is the head of the relative.”

4. Which argument is the head (§6.3) SOURCED

Willie 1991:159–165, (14)–(26)
(14)yizloh=ę́ę
3.3.roped=REL
‘the one that was roped’
(15)bizloh=ę́ę
INV3.3.roped=REL
‘the one who roped him’
(18)yiztsʼǫs-yę́ęyóóʼeelwod
3.3.kissed=REL3.ran away
‘The one he kissed ran away.’
(19)biztsʼǫs=yę́ęyóóʼeelwod
INV3.3.kissed=REL3-ran away
‘The one who kissed him ran away.’
(20a)ʼatʼéédʼashkiiyiztsʼǫs=yę́ęyóóʼeelwod
‘The girl who kissed the/a boy ran away.’
(20b)ʼatʼéédʼashkiibiztsʼǫs=yę́ęyóóʼeelwod
‘The girl who was kissed by the boy ran away.’
(22)ʼashkiiyiztsʼǫs=yę́ęyóóʼeelwcd [sic]
a. ‘The one who kissed the boy ran away.’ (restrictive, preferred) b. ‘The boy, who was kissed by her, ran away.’ (non-restrictive)
(24)ʼashkiibiztsʼǫs=yę́ęyóóʼeelwod
a. ‘The boy, the one who kissed her, ran away.’ *b. (no restrictive reading)

Willie's generalizations, verbatim. (16): “The heads of relative clauses are absolutives.” P. 160: “with the Inverse voice, the agent is the internal argument.” P. 161: (18) is patient-headed and (19) agent-headed “because the agent is the internal argument in the Inverse construction.” (21): “When both arguments of the relative have nominal adjuncts, the relativizing enclitic binds the external argument, which must be definite.” She adds that Platero 1982 reports two readings for some speakers, “not true of my speech.” On (24): the inverse relative is non-restrictive because the inverse requires a definite subject.

5. Two transitive clauses: parallel processing (§6.3.4) SOURCED

(27) Parallel Processing, p. 166: “In a sequence of two transitive clauses with all third person arguments, the two arguments of the first and second clause are interpreted as coreferential.” “Voice alternations in either clause … change the relations between arguments and theta roles, but the same referents must be involved. This applies to sentences with relative clauses also.”

Willie 1991:166–170, (28)–(40)
(28)yiztał-yę́ęyiyiiłtsą́
3.3.kicked=REL3.3.saw
‘He saw the one he kicked.’
(29)yiztał=yę́ębiiłtsą́
3.3.kicked=RELINV3.3.saw
‘The one he kicked saw him.’
(30)biztał=yę́ęyiyiiltsą́
‘He saw the one that kicked him.’
(31)biztał=yę́ębiiłtsą́
‘The one that kicked him saw him.’
(32)yiztał=yę́ęʼatʼéédyiyiiltsą́
‘The one he kicked saw the girl.’
(33)yiztał=yę́ęʼatʼéédbiiltsą́
‘The one he kicked was seen by the girl.’
(36)ʼatʼéédyiztał=yę́ęʼashkiinéidííłtsʼin
girl3.3.kicked=RELboy3.3.hit
‘The one who kicked the girl hit the boy.’
*(39)ʼatʼéédbiztał=yę́ęʼashkiinábidííłtsʼin
‘[the girl, the one who kicked himi, was hit by the boyj]’
(40)ʼatʼéédbiztał=yéeʼashkii léiʼnábidííłtsʼin
‘The girl, the one who kicked himi, was hit by some boyj.’

P. 167: “If yi- is present, the head of the relative is an adjunct to the patient argument of the main clause verb; if bi- is present the relative is an adjunct to the agent argument.” A noun after the relative is an adjunct of the main verb, since “the relativizing enclitic marks the rightmost boundary of the relative”, and then “(27) above does not apply, since the noun marks disjoint reference -- by specifying another referent.” On *(39) Willie writes “I find it impossible to process”: it would need disjoint internal arguments, “impossible unless the second of these is indefinite”, and (40) with léiʼ ‘some’ is acceptable.

6. The fourth person in sentences with relatives (§6.4) SOURCED

P. 170: “the function of the fourth person in marking disjoint reference is parallel to the constructions where nominals are adjoined.” P. 171: relatives with a 1st or 2nd person head must be non-restrictive (“already maximally definite”), as in (41)–(43). A 4th person head must also be non-restrictive (44); the 4 here is deference and “fixes reference uniquely”.

Willie 1991:170–176, (41)–(59)
(41a)siztsʼǫs=yę́ę
1.3.kissed=REL
‘the one who kissed me’
(41b)sétsʼǫs=yę́ę
3.1.kissed=REL
‘the one I kissed’
(44)joodloh=ę́ęyóóʼajoolwod
4.laughing=REL4.ran away
‘That one, the one who laughed, ran away.’
(45)haztał=yę́ęyiyiiłtsą́
4.3.kicked=REL3.3.saw
‘He saw the one who kicked that person.’
(46)jiztał=yéeyiyiiłtsą́
4.3.kicked=REL [sic; cf. (3) “3.4”]3.3.saw
‘He saw the one who that person kicked.’
(47)yiztał=yę́ęjiiłtsą́
3.3.kicked=REL4.3.saw [sic; cf. (49) “3.4.saw”]
‘That person saw the one who was kicked.’
(48)yiztał=yę́ęhoołtsą́
3.3.kicked=REL4.3.saw
‘The one that he kicked saw that person.’
(49)haztał=yę́ęjiiłtsą́
4.3.kicked=REL3.4.saw
‘That personi saw the one who kicked himi.’
(50)jiztał=yę́ęjiiłtsą́
3.4.kicked=REL3.4.saw
‘The onei that personj kicked, that personj saw himi.’
(51)haztał=yę́ęhoołtsą́
4.3.kicked=REL4.3.saw
‘That one who kicked that person saw that person.’
(52)jiztał=yę́ęhoołtsą́
3.4.kicked=REL4.3.saw
‘That one that person kicked saw that person.’
*(53)ʼatʼéédjoodloh=yę́ęyóóʼajoolwod
girl4-laughing=REL†4-ran away
*(54)yiztsʼǫs=yę́ęʼatʼéédhoołtsą́
(55a)jiztsʼǫs=yę́ęʼatʼéédyiyiiłtsą́
‘The one that person kissed saw the girl.’
(55b)haztsʼǫs=yę́ęʼatʼéédyiyiiłtsą́
‘The one who kissed that person saw the girl.’
(57)ʼatʼéédhaztsʼǫs=yę́ęjiiłtsą́
girl4.3.kissed=REL3.4.saw
‘That personi saw the girl that kissed himi.’ (politeness)
(59)ʼatʼéédjiztsʼǫs=yę́ęjiiłtsą́
‘That personi saw the girl hei kissed.’

SOURCED † Segmentation note (2026-11-11 audit pass 4): Willie 1991 (53), p. 174 prints joodloh=yę́ę with the gloss “4-laughing=REL”: the gloss names more morphemes than the form shows as separate pieces (fused or unwritten morphemes), so the two lines do not match one-to-one in the source. Kept exactly as printed.

Pp. 173–174, verbatim: “Where fourth person appears in just one clause, it marks disjoint reference, and where it appears in both clauses it marks coreference. This follows the fact that anaphors must match their antecedents in phi features. Fourth person in all these examples is being used for deference.” On *(53): “Thus, nominals and fourth person pronouns have a parallel syntactic function, that of blocking coreference.” *(54) is out because the main verb's 4 slot and the noun on its 3 slot leave no 3 argument for the relative to attach to. P. 175: “The fourth person cannot be the head of a restrictive relative; the head is third person, glossed ‘the one’.”

7. Factive and propositional nominalized clauses (§6.5) SOURCED

Willie 1991:178–181, (62)–(68)
(62)chidíʼínáályaa=yę́ęyáʼátʼééh
car3.fixed=REL3.good
‘The car that was fixed is good.’
(63)chidíʼínáályaa=ígííyáʼátʼééh
‘It is good that the car is fixed.’
(64)honeesná=nígííyoodlą́
3-win-NOM†3-believes
a. ‘Hei believes that hei,j won.’ b. ‘He believes the winner.’
(65)Honeesná=nígííjoodlą́
3-win-NOM†4-believes
a. ‘That personi believes that hej won.’ b. ‘That person believes the winner.’
(66)Hozhneesná=ígííjoodlą́
a. ‘Hei believes that hei won.’ *b. ‘… hej won.’ (no gloss line printed)
(67)Maryłį́į́ʼnayiisniiʼ=ígííyíniiʼ
horse3.3.buy=COMP3s.1s.heard
‘I heard that Mary bought a horse.’
??(68)Maryłį́į́ʼnayiisniiʼ=goyíniiʼ
‘I heard it, when Mary bought the a horse.’ (sic)

SOURCED † Segmentation note (2026-11-11 audit pass 4): Willie 1991 (64) prints honeesná=nígíí with the gloss “3-win-NOM”: the gloss names more morphemes than the form shows as separate pieces (fused or unwritten morphemes), so the two lines do not match one-to-one in the source. Kept exactly as printed.

SOURCED † Segmentation note (2026-11-11 audit pass 4): Willie 1991 (65) prints Honeesná=nígíí with the gloss “3-win-NOM”: the gloss names more morphemes than the form shows as separate pieces (fused or unwritten morphemes), so the two lines do not match one-to-one in the source. Kept exactly as printed.

Willie: no adjoined nominal is possible in (66), and “Parallel examples occur with other verbs of cognition.” P. 181: “=ígíí is factive, while =go occurs in temporal/conditional constructions.” §6.6 summary: relatives headed by a 1st, 2nd or 4th person must be non-restrictive, and so must inverse relatives.

8. Rules

Notation as on the reported-speech page. K = the embedded (relative or nominalized) clause, M = the main clause, REL = the enclitic.

REL-HEAD   REL (=yę́ę / =ígíí, "the one ...") is the head of the relative. It is a
           determiner/demonstrative that closes K at its right edge and binds one
           pronominal slot ρ of K. No gap; K is a full clause.            (pp. 151, 153, 158)
REL-ABS    Default: REL binds the internal (absolutive) argument of K.          (16)
           Direct (yi-): internal = ρP.  Inverse (bi-): internal = ρA.     (pp. 160–161)
REL-EXT    If both arguments of K have nominal adjuncts, REL binds the external
           argument, which must be definite.                                     (21)
REL-NR     A relative headed by π = 1, 2 or 4, or an inverse relative, is
           non-restrictive.                                   (pp. 171, 175; (24); §6.6)
REL-4-HEAD π(ρ bound by REL) ≠ 4 in a restrictive relative.               (p. 175)
REL-EDGE   REL marks the right boundary of K; a noun after REL belongs to M. (pp. 158, 167)
REL-ADJ    [K=REL] is adjoined to a slot of M, like a noun: yi- on M → to ρP(M);
           bi- on M → to ρA(M).                                         (pp. 156, 167)
REL-PP     π = 3/3 in K and M (both transitive): the two referents are the same two
           (roles may swap).                                     (27), (28)–(31)
REL-NOUN   A noun in M on the slot REL does not attach to suspends REL-PP
           (disjoint reference); *(39) unless that noun is indefinite (40).
REL-4      π = 4 in one clause only → disjoint; π = 4 in both → coreferent. (pp. 173–174)
           Nouns and π = 4 both block coreference: *(53), *(54).
BEL-3/3    'believe' + [K=nígíí], 3/3: same or different referent.             (64)
BEL-3/4    3 in K, 4 on 'believe': disjoint.                                    (65)
BEL-4/4    4 in K, 4 on 'believe': coreferent only.                             (66)
RuleStatusBasis
REL-HEAD, REL-EDGESOURCEDWillie's analysis, pp. 151, 153, 158, 167. Willie 1990 (cited) not opened. 2026-10-17 (8:46pm): restated on relative as CP. REL is the head (D); “binds a pronoun” must be an operator tie (a slot copy moving to Spec,CP), because Platero 1973 shows the relative tie obeys islands that coreference ignores.
REL-ABS, REL-EXTSOURCED(16), (21), (14)–(26). Platero 1982 reports other speakers differ (not opened).
REL-NR, REL-4-HEADSOURCED(22)–(24), (41)–(44), p. 175. 2026-10-29 (12:54pm): on person-head, (41a) shows that REL-ABS is the 3/3 case only. With one 3rd person, REL binds it even when it is the external agent.
REL-ADJ, REL-PP, REL-NOUNSOURCED(27)–(40), pp. 156, 166–170.
REL-4SOURCED(45)–(59), pp. 170–176. The [sic] glosses of (46), (47) are read from the translations (T14c).
BEL-3/3, 3/4, 4/4SOURCED for ‘believe’ only(64)–(66). Willie says other cognition verbs are parallel but prints none.
Writing ρ for the bound slot, and “internal” as a positionHYPOTHESISOur notation. Whether “internal/external” is a structural position or a case label is OPEN (T14b).

9. The 3/3 default in four constructions (T13d)

ConstructionPlain 3 / 33 / 44 / 4Source
=go chain (adjunct clause)same cast (GO-PP)differentsameclause chains, Willie Ch. 5
Relative =yę́ę (adjunct of a slot)same cast (REL-PP)differentsame(27)–(31), (45)–(52)
‘say’ + complement (no enclitic)different subjects (RS-33)differentsamereported speech, (22)–(26) Ch. 5
‘believe’ + =nígíí clauseeither (BEL-3/3)differentsame(64)–(66)
  • SOURCED In every row, one fourth person means a different person and two fourth persons mean the same person. PI-MATCH (i) and (ii) therefore hold across all four constructions. PI-MATCH (iii) (the 3/3 default is set by the construction) is confirmed with a third value: free.
  • HYPOTHESIS The two constructions where the clause is an adjunct (=go, relatives) share the cast. The two where the clause is an argument of the main verb (‘say’, ‘believe’) do not force sharing, and ‘say’ adds obligatory disjointness. Why ‘say’ and ‘believe’ differ is OPEN (T14a). One candidate: the ‘say’ complement is the only subordinate clause with no enclitic (p. 149), while ‘believe’ takes a nominalized clause.

10. Derivations R1–R6, step by step

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). Verb words are not segmented beyond Willie's glosses. V-P: the subject slot raises to Spec,M/SP inside the word; V-H: it stays in situ (Hale). REL is treated as a D head taking K as its complement (Willie's “determiner”, pp. 151, 158). The tree is ours.

R1. (2) yiztał=yę́ę yicha ‘The one he kicked is crying.’

  1. Σ K: Σ(√KICK) = {ρA, ρP}, π = 3/3, direct. Merge V + ρP, v + ρA, M/S; Agree; Move (V-P) ρA. A yiztał.
  2. Merge D = REL with K. REL-ABS: direct, so REL binds the internal slot ρP (Willie's index: 3i = object = RELi). B D is head-final: yiztał=yę́ę.
  3. Σ M: Σ(√CRY) = {ρ1}, π = 3. A yicha.
  4. REL-ADJ: [K=REL] is adjoined to ρ1(M), to its left. B yiztał=yę́ę yicha. Index: 3i.crying.
  5. HYPOTHESIS The A → B changeover is at the enclitic. Everything up to yiztał is A; =yę́ę and the adjunction are B. The slot the relative names is decided inside A (direct vs inverse), but it is read off by a B-level head.

R2. (18) vs (19): one root, two heads

  1. Σ K: Σ(√KISS) = {ρA, ρP}, π = 3/3. (18) direct yiztsʼǫs, (19) inverse biztsʼǫs. A only the yi-/bi- prefix differs.
  2. REL-ABS: (18) internal = ρP → ‘the one he kissed’. (19) inverse, internal = ρA (p. 160) → ‘the one who kissed him’.
  3. B [K=yę́ę] adjoined to ρ1 of yóóʼeelwod ‘3.ran away’.
  4. HYPOTHESIS Same root, same person values, same word order: the role named by ‘the one’ is fixed by which slot the voice prefix makes internal. This is Grant's “slots, not lexical items” point at its cleanest: the head of the relative is a position in a role schema.

R3. (29) yiztał=yę́ę biiłtsą́ ‘The one he kicked saw him.’

  1. K as in R1: REL binds ρP(K) (the kicked one, x). ρA(K) = y (the kicker).
  2. Σ M: Σ(√SEE) = {ρA, ρP}, π = 3/3, inverse. A biiłtsą́.
  3. REL-ADJ: bi- on M → [K=REL] is adjoined to ρA(M). So x = seer.
  4. REL-PP: the same two referents. The remaining slot ρP(M) = y. Result: x (kicked) saw y (kicker). Roles have swapped across the clauses; the cast has not.
  5. B yiztał=yę́ę biiłtsą́.

R4. (49) haztał=yę́ę jiiłtsą́ ‘That personi saw the one who kicked himi.’

  1. Σ K: √KICK, gloss “4.3”: ρP π = 4, ρA π = 3. A haztał. REL-4-HEAD: REL cannot bind the 4; it binds ρA (π = 3), ‘the one who kicked’. (No yi-/bi- choice with a 4: 4-NO-INV.)
  2. Σ M: √SEE, gloss “3.4”: ρP π = 3, ρA π = 4. A jiiłtsą́.
  3. REL-ADJ: [K=REL] (π = 3) goes to the π = 3 slot of M, ρP: the kicker is the one seen.
  4. REL-4: π = 4 in both clauses → coreferent. The kicked one (ρP of K) = the seer (ρA of M), as Willie's translation and index show.
  5. SOURCED via the translation, and important for T12a: the two coreferent 4 slots are patient in K and agent in M. The match goes by person value, not by role. (52) is the mirror case (agent in K, patient in M). B haztał=yę́ę jiiłtsą́.

R5. (64)–(66): ‘believe’ with a nominalized clause

  1. Σ K: √WIN, {ρ1}. (64)/(65) π = 3 honeesná; (66) π = 4 hozhneesná. A.
  2. Merge D/C = =nígíí / =ígíí. B enclitic at the right edge.
  3. Σ M: √BELIEVE = {ρExp, ρTh}. HYPOTHESIS The nominalized clause fills ρTh (reading a, ‘believes that …’), or REL binds ρ1(K) and names a person (reading b, ‘believes the winner’). π(ρExp) = 3 yoodlą́ or 4 joodlą́.
  4. Reference: (64) 3/3 free (BEL-3/3); (65) 3/4 disjoint (BEL-3/4); (66) 4/4 coreferent and reading b is out (BEL-4/4). B K precedes M.

R6. *(54) yiztsʼǫs=yę́ę ʼatʼééd hoołtsą́: no slot left

  1. K: √KISS, 3/3 direct; REL binds ρP(K). [K=REL] is π = 3 and needs a π = 3 slot of M.
  2. M: hoołtsą́ “4.3.saw”. The π = 3 slot is taken by the noun ʼatʼééd (REL-EDGE: the noun after REL belongs to M). The other slot is π = 4, which a π = 3 relative cannot attach to (REL-4).
  3. ✗ No slot of M is available for [K=REL]. SOURCED (Willie p. 174). HYPOTHESIS In schema terms: adjunction targets a slot by person value, and each slot takes at most one linked expression.

11. Bearing on T1, T12a, T12f, T13d

  • T12f (does parallel processing hold outside =go?): answered for relatives. SOURCED Yes. Willie states (27) and says it “applies to sentences with relative clauses also” (p. 166). A noun or a fourth person on the main verb suspends it (REL-NOUN, REL-4).
  • T13d (is the 3/3 default set by the construction?): yes, with three values. See §9. Same in adjunct clauses, different under ‘say’, free under ‘believe’.
  • T12a (roles or functions?): new evidence. SOURCED In relatives, cast-sharing is over referents with roles free ((29), (30)), and 4/4 matching follows the person value across a patient/agent split ((49), (52)). HYPOTHESIS If the same holds in =go chains, Willie's “coindexed according to theta role” (p. 148) describes the default pairing, not a requirement.
  • T1 (are enclitics like -ę́ę heads?). SOURCED Willie argues the relativizing enclitic is the head, a demonstrative pronoun (pp. 153, 158). HYPOTHESIS That makes it a head-final D in rule set B, binding a slot built in rule set A.
  • Grant's phenomena. GRANT'S CONJECTURE Theta roles, not lexical items, are the arguments. Chapter 6 is consistent with this: the head of a relative is chosen by position (internal/external), by voice, and by person value, and nouns are adjuncts, never arguments (p. 158). It does not test whether chemistry has a parallel; no chemical claim is made here.

12. Open questions

IDQuestionWhat would settle it
T14aWhy does ‘say’ force different subjects, ‘believe’ allow either, and adjunct clauses (=go, relatives) force the same cast?More complement-taking verbs (think, want, know) with 3/3; Willie prints only ‘believe’. Progress 2026-10-10 (evening): ‘think’ (nizin) with an optative clause links the clause's 4 subject to the thinker, 3 or 2 (Willie 1991:120–121, (18)–(19); fourth person uses §11). 3/3 with ‘think’ is still unprinted. OPEN
T14bIs “internal/external argument” a structural (VPISH) position or a case label? (20b) and (21) are the test: with two nouns, REL binds the external argument.Willie 1990; Hale's work on the inverse; speaker data on (20b). OPEN
T14c(46) prints “4.3.kicked” for jiztał (elsewhere “3.4”), and (47) prints “4.3.saw” for jiiłtsą́ (elsewhere “3.4.saw”). Misprints?The translations favour misprints; another source with these forms. OPEN
T14dHow are =yę́ę, =ę́ę, =ígíí, =nígíí, =nę́ę related, and is the factive =ígíí (63), (67) the same morpheme as the relative =ígíí (1g)?Y&M 1987; a grammar of the enclitics. OPEN
T14ePlatero (1974, 1982) reports a noun after the relative as its adjunct and two readings in (22)-type sentences for some speakers. Dialect, or analysis?Answered for Hale & Platero 1974 (2026-10-15, 4:46am): page. A noun after the relativizer can be the relative's head ((1), (15), (20), (36)), with the internally headed form preferred. In every printed case Willie's two-adjunct parse gives the same meaning, and where the parses would differ ((32), (46)) H&P's speakers give Willie's reading. So for this claim the difference is analysis, not data. A real speaker difference remains on head choice with two nouns inside the relative (H&P (17), two readings, vs Willie (20a), one). Willie's “Platero (1974)” is the IJAL article, still unopened (T23a–b). OPEN (partly answered)
T14fIs *(39) a grammar violation or a processing failure (“I find it impossible to process”), given that léiʼ rescues it (40)?Speaker judgments with other indefinites. OPEN

Added 2026-10-14 (8:46pm): Platero's relative-clause paradigm ((4.15), 14 readings, 5 good; (4.52)–(4.55)), as reported in Speas 1986:254–285, is analysed on Speas 1986 on Navajo. It bears on REL-HEAD/REL-ADJ: the matrix null argument may corefer with a noun inside the relative (no Condition C), and the noun closest to the verbs is the object of both verbs (Platero, via Speas p. 285).

Not read this run: Willie §4.2.4 (narratives), §4.2.5 (the Impersonal) (since read: fourth person uses); Willie 1990; Platero 1974, 1982. Hale & Platero 1974 (NLR) since read: page (2026-10-15).

Who can "his" point back to under bi-? A possessor test for the deep inverse (T53a)

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.

Working note. Kept as written (URL, data and status labels unchanged). Filed under Possessors and object-first orders; the consolidated argument is the essay possessors. Rule IDs: rules index.

Verbs on this page → dictionary: yiztsʼǫs ‘kiss’ · dabiiłtsą́ ‘see’ · yíchxǫʼ ‘be ruined; (classifier) ruin’

Restated on the unified structure (2026-11-15). Possessors are restated as the left specifier of the noun phrase, with the possessor prefix as agreement (as on a grammatical postposition). This page’s particular analysis has not yet been restated on the unified structure; light n stays 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. 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, possessors. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.

The deep-or-shallow page ran four of Oxford's (2023) tests on Navajo bi- and found it behaves like a "deep" inverse, one where the patient really moves up over the agent. Oxford's deciding test, binding (D2), was left open, because no Navajo binding data had been opened (T53a). This page finds some. Judith Aissen (2000) quotes Young & Morgan (1980) on sentences where "his" inside one noun phrase refers to the other person in the sentence. In plain words: when the doer owns the thing acted on ('the boy saw his own horse'), Navajo uses only the ordinary yi- form. When the person acted on owns the doer ('Mr. Long's wife is looking at him'), only the bi- form is possible. That is the same split English makes between an active and a passive with "every man ... his mother", and it is what Oxford uses to show that a forced inverse is still a deep one. But it is not decisive. Aissen explains the same four judgments without any structure, by ranking owner over owned. In every good example the owner's name also comes before "his", so plain word order explains them too.

On this page: 1 Question · 2 Sources · 3 Data · 4 Aissen's account · 5 Does this run Oxford's D2? · 6 Three rival explanations · 7 A check against Willie's possessor rule · 8 Rules · 9 Derivations · 10 Grant's ideas · 11 Bottom line · 12 Open

Notation (the site's own; no Syntax · Semantics · Interface bot conventions file has arrived, so the provisional notation is kept). Σ = role-schema step; 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). W is where A hands over to B. A possessed noun is its own small word, written N-W, with one slot π (the possessor). ρA = agent slot, ρP = patient slot (labels ours). [prx] / [obv] = proximate / obviative, Aissen's and Oxford's terms for the "main" and "other" third person. Subscript i, j mark who is who, as in the sources. RCC: deep-or-shallow §6.

Reminder: Merge builds structure; moving something already built is internal Merge (Move). Agree values features such as person, number and here [prx]/[obv] between a head and an argument; nothing moves. A shallow inverse is Agree only. A deep inverse is Agree plus one Move of the patient over the agent (Algonquian inverse §6). Binding is neither. It is a condition on the finished tree: an antecedent must c-command what it binds. That is why it can tell the two kinds of inverse apart.

1. The question

  • T53a (from deep-or-shallow §11): can the raised patient in a bi- sentence bind into, or take scope over, the agent? This is Oxford's D2. In Algonquian the third-person inverse reverses binding ((A4)): the patient binds a pronoun inside the agent.
  • Behind it: T52d (is bi- deep?) and T11b (fourth person compared), which feared a Navajo binding test could not be built at all.

2. Sources opened this run (2026-10-24, ~12:46pm PT run)

  • Aissen, Judith 2000. "YI and BI: Proximate and Obviative in Navajo." In A. Carnie, E. Jelinek & M. Willie (eds.), Papers in Honor of Ken Hale (MIT Working Papers on Endangered and Less Familiar Languages 1), pp. 129–150. The author's own RTF from her UCSC web space (people.ucsc.edu/~aissen/navajo.rtf), read whole, footnotes included (pulled from the RTF directly, because the converted text dropped them). Page numbers of the printed volume are not in the file, so this page cites example and section numbers.
  • Willie, MaryAnn 1991. Navajo Pronouns and Obviation. PhD dissertation, University of Arizona (box scan). Page images at 130 and 300 dpi of pp. 60–61 ((19), (20), (14), (21)); text layer pp. 57–61 and 92–93 re-read. Pp. 185–186 ((4)–(5)) as already transcribed on possessor raising §3.
  • Oxford, Will 2023. "A tale of two inverses." Syntax 26. Author's preprint (lingbuzz/006567), text layer; §7.3 pp. 29–30 re-read ((45), fn. 19, (46)–(47), the English parallel).

Young & Morgan 1980 (the source of the key sentences) was not opened. Its data are here secondhand, via Aissen. Aissen 1997 (Language 73) is on the box but is image-only, and no OCR tool is installed, so its Algonquian possessor data (Rhodes 1993) were not read. Rhodes 1993 was not opened. Update 2026-10-24 ~8:46pm PT: Aissen 1997 has now been read via OCR on the possessor-gap page.

3. The data SOURCED

Fidelity. Aissen's file sets Navajo in the SIL IPA93 font. We decoded it by the font's own key: ¬ = ł, 7 = nasal hook (ogonek), @ = high tone. The decoding was checked against forms printed elsewhere (yiztsʼǫs, yíchxǫʼ, biníłʼį́). This is the author's file, not page images of the printed volume. Forms are printed exactly as in the file. In (27) and (29) the file writes bi-lį́į́ without a final ’ although its (1a) has łį́į́ʼ; we do not add one. (27b) writes ashkii without its initial ’, as printed. The first gloss line is Aissen's; the second is our Leipzig-style line, and it adds nothing her prose does not say ((8): BI = proximate object pronoun, YI = obviative object pronoun; fn. 13: BI also occurs "in genitive function").

Aissen 2000 (27a, b), citing Young & Morgan 1980:171
(1a)ʼashkiibi-lį́į́yi-yiiłtsą́.
boyBI-horseYI-saw
boy3POSS.PRX-horse3OBJ.OBV-saw
‘The boyi saw hisi horse.’
(1b)*bi-lį́į́ashkiibi-yiiłtsą́.
BI-horseboyBI-saw
3POSS.PRX-horseboy3OBJ.PRX-saw
‘(Hisi horse was seen by the boyi.)’

Aissen: Young and Morgan "observe (p. 171) that while the sentence ‘the boy saw the horse’ can be expressed either in the YI form or the BI form, the sentence ‘the boy saw his horse’ (interpreted with coreference) can only be expressed in the YI form." Her fn. 14 quotes them: "If a second noun is the property of the first, subject/object inversion is blocked with respect to their interaction."

Aissen 2000 (29), citing Young & Morgan 1980:171
(2)*Bi-lį́į́ashkiiyi-yiiłtsą́.
BI-horseboyYI-saw
3POSS.PRX-horseboy3OBJ.OBV-saw
‘(His horse saw the boy.)’

Aissen: "Young and Morgan (1980, 171) note that ‘his horse saw the boy’ cannot be translated by the YI form of the verb." "Young and Morgan do not say how ‘hisi horse saw the boyi’ would be expressed in Navajo. But they do cite the example in (30), which is equivalent in the relevant ways (1980, 172)."

Aissen 2000 (30), citing Young & Morgan 1980:172
(3)Hastiin Néézbeʼesdzáánbi-níłʼį́.
man longBI-wifeBI-look.at
man long3POSS.PRX-wife3OBJ.PRX-look.at
‘Mr. Long is being looked at by his wife. (I.e. Mr. Long’s wife is looking at him.)’

The gloss line "BI-wife" is Aissen's; the possessive b- is fused into beʼesdzáán in the spelling. Her analysis (31): "Since Hastiin Nééz, ‘Mr. Long’ is proximate, so are the genitive within ‘his wife’ and the object of ‘look at’, which are coreferential with Hastiin Nééz."

Willie 1991:60–61, (19), (14), (21); page images (300 dpi)
(4a)[hastiin sání][bi-chidí]yiyííłchxǫʼ
man old3POSS.car†3.3.ruined
‘The old man ruined his car.’ (19)
(4b)[hastiin sání bi-chidí]yíchxǫʼ
man old 3POSS.car3-3-ruined
‘The old man’s car is ruined.’ (14)
(4c)[hastiin sání bi-chidí]yiyííłchxǫʼ
man old 3POSS.car3-3-ruined
‘He/she ruined the old man’s car.’ (21)

SOURCED † Segmentation note (2026-11-11 audit pass 4): Willie 1991 (19) prints [bi-chidí] with the gloss “3POSS.car”: the gloss joins with periods or colons pieces the form separates with hyphens, so the two lines do not match one-to-one in the source. Kept exactly as printed.

Glosses and brackets are Willie's, as printed. P. 60, (20): "When a transitive sentence includes two nominal adjuncts, the inner nominal is coindexed with the internal argument, and the outer nominal (by default) is coindexed with the external argument." P. 61: in (21) "this single constituent is interpreted as an adjunct to the patient argument."

Willie's (19) prints no subscripts, so the English "his" could in principle mean someone else. Unlike (1a), it does not on its own report coreference. What it adds is first-hand evidence that the string [noun] [bi-noun] yi-verb has two parses, two adjuncts (4a) or one possessive phrase (4c). So (1a) also has a second reading, ‘he saw the boy’s horse’. Aissen's and Young & Morgan's judgment is about the two-adjunct reading.

4. Aissen's account SOURCED

  • (8): "BI is the object proximate pronoun; YI is the object obviative pronoun." In a 3/3 clause one argument is proximate and the other obviative.
  • (6b), §4.4: "A genitive outranks its head." Stated as a constraint, (26) = (33): Avoid a possessum proximate.
  • Coreference: "all coreferents within an obviation span have the same obviation status" (on (29)).
  • (27a): the boy, the possessor inside 'his horse' and the subject are all proximate; 'horse' and the object are obviative → YI. (27b) would need the reverse and break (26). (29): YI makes 'boy' obviative, so the coreferent possessor is obviative and 'horse' proximate, which breaks (26). (30): the object is proximate (BI), and so is the coreferent possessor inside the subject; nothing is broken.
  • Aissen's approach "makes no appeal to the binding theory" (§4.2). Word order: "the proximate precedes the obviative" (§7; constraint (11)).
  • Cross-language: "Sentences of this type (genitive within the subject coreferential with the object) consistently require special treatment in languages with obviation systems." Tzotzil and Chamorro require a passive. "In others, e.g. Algonquian, this configuration requires a marked form of the verb, the inverse (Rhodes 1993; Aissen 1997)." (Secondhand for those languages; not opened.)

5. Does this run Oxford's D2?

SOURCED Oxford 2023 §7.3 (pp. 29–30) explains why the Algonquian inverse is forced when an obviative acts on a proximate, and why that forced inverse is still the deep kind, with (45), the Proximate C-command Constraint: "A proximate argument cannot be c-commanded by an obviative co-argument." The speaker must "use a voice construction that places the proximate in subject position: the default voice when the proximate is the agent and the inverse voice when the proximate is the patient." His English parallel: "the passive Every mani is loved by hisi mother does not have a simple active paraphrase (*Hisi mother loves every mani), and the active Every mani loves hisi mother does not have a simple passive paraphrase (*Hisi mother is loved by every mani)."

CellEnglish (Oxford)Navajo (Y&M via Aissen)Match?
Agent antecedes "his" in the patient; plain voiceEvery mani loves hisi mother ✓(1a) ʼashkii bi-lį́į́ yi-yiiłtsą́ ✓yes
Same, marked voice*Hisi mother is loved by every mani*(1b) bi-lį́į́ ashkii bi-yiiłtsą́yes
Patient antecedes "his" in the agent; plain voice*Hisi mother loves every mani*(2) Bi-lį́į́ ashkii yi-yiiłtsą́yes
Same, marked voiceEvery mani is loved by hisi mother ✓(3) Hastiin Nééz beʼesdzáán bi-níłʼį́ ✓ (a different verb and nouns; Aissen: "equivalent in the relevant ways")yes

HYPOTHESIS The four cells line up with the deep-inverse pattern. (3) is the first Navajo sentence on this site in which the patient's referent is picked up by a pronoun inside the agent under bi-, which is the reversed configuration Oxford's D2 looks for. (2) shows that the plain yi- form cannot do it. Put Aissen's (26) together with Oxford's (45): the possessor is proximate, so the noun phrase it sits in is obviative, so its co-argument (the antecedent) is proximate and must not be c-commanded by it. Then the antecedent must be the structurally higher argument: the agent in yi-, the raised patient in bi-. That derives all four cells, but only if bi- moves the patient over the agent. Under a shallow, agreement-only bi-, the agent 'his wife' in (3) would still c-command 'Mr. Long', and (3) should be as bad as (2).

HYPOTHESIS But it is coreference, not quantifier binding. Oxford's English parallel and his Oji-Cree (A4) both use a quantified antecedent ("every man", "all the women"), which can only be bound. The Navajo antecedents are plain nouns ('the boy', 'Mr. Long'). So these data show where a "his" may corefer, which is a weaker test than binding in the strict sense. That is why the result counts as a lean, not a verdict.

6. Three rival explanations

AccountWhat it needsPredicts (1a), *(1b), *(2), (3)?Needs bi- to move the patient?
(i) Deep inverse + PCC (Oxford (45) + Aissen (26))c-command; the patient raised over the agent in bi-all fouryes
(ii) Aissen's obviation tierBI/YI = [prx]/[obv] object; (26); coreferents share status. No structure.all four (§4)no
(iii) Linear orderThe antecedent noun must precede the possessed noun ("Proximate precedes Obviative", Aissen (11), plus Willie's linking (20))all four: in (1a) and (3) the antecedent comes first; in *(1b) and *(2) bi-noun comes firstno

HYPOTHESIS In these data the three accounts cannot be told apart. Noun order in Navajo follows voice (the outer noun links to the higher argument, Willie (20), reversed under bi-), so "comes first" and "is structurally higher" go together in every sentence found. Account (ii) is the one Aissen actually gives, and it uses no tree at all. So the possessor data are compatible with a deep bi- (they fit RCC and PCC), but they do not show it, and they do not count against a shallow one.

OPEN What would separate them. (a) The fourth person. Aissen §6 says the 4th person stays outside the obviation tier (Willie 1991:127, via Aissen). If the antecedent or the possessor is fourth person, (i) and (ii) predict no restriction, while a pure c-command binding account still predicts one. (b) A quantified antecedent ('every boy') under bi-, which tests binding proper. (c) A sentence where the owner's noun is left out, so word order cannot help, for example 'his horse' plus a verb with no other noun. No source opened gives any of these. None is supplied here.

HYPOTHESIS Young & Morgan's own wording fits only half. "If a second noun is the property of the first, subject/object inversion is blocked" (via Aissen fn. 14) describes (1a)/*(1b). In (3) the second noun ('his wife') is the property of the first ('Mr. Long'), and the bi- form is the only good one. Aissen's (26) covers both. This is our reading of a one-sentence quotation, and Young & Morgan's page may say more (T57b).

7. A check against Willie's possessor rule

SOURCED Willie 1991:185–186 ((5a), (5b), transcribed on possessor raising §3): ʼashkii ʼatʼééd yighan=góó naayá ‘The boy went to the girl's house’ vs ʼashkii ʼatʼééd baghan=góó naayá ‘The girl went to the boy's house’. Her rule: "the bi- pronouns signals that the preceding nominal is not coindexed with it".

HYPOTHESIS Willie's (5) fits Aissen's reading of possessive BI/YI. In (5a) the possessor is the girl, who is not the first noun (the obviative one, on Aissen (11)) → yi-. In (5b) the possessor is the boy, the first noun (proximate) → bi-. (1a) is not a counterexample to Willie's wording even though ʼashkii right before bi-lį́į́ is its possessor. Her (4) ʼatʼééd baghan ‘the girl's home’ also has the possessor right before bi-, so her rule must be about the 3/3 setting with two nouns before the possessed one, as in (5).

8. Rules

POSS-1  (= Aissen (8), fn. 13) The prefix on a possessed noun spells its possessor slot π;       SOURCED
        BI = [prx], YI = [obv], as on verbs.
POSS-2  (= Aissen (26)/(33)) If π is a 3rd-person [prx], the possessed noun phrase is [obv].        SOURCED (Aissen's)
POSS-3  Coreferent 3rd persons in one span share [prx]/[obv] (Aissen, on (29)).                  SOURCED (Aissen's)
POSS-4  (= Oxford (45), carried over) A [prx] argument slot may not be c-commanded by an           SOURCED for Algonquian;
        [obv] co-argument slot.  In Navajo this is RCC with [prx]/[obv] in place of rank.        HYPOTHESIS for Navajo
POSS-5  POSS-1–4 + deep bi- (ρP Moves over ρA) derive (1a), *(1b), *(2), (3).                   HYPOTHESIS
        POSS-1–3 alone (no c-command) derive them too (Aissen); so do the order facts.            HYPOTHESIS (confound)
POSS-6  Not decisive for D2: the antecedents are plain nouns (coreference), not quantifiers.      HYPOTHESIS

9. Derivations, step by step

Labels: Σ role schema, A word (W for the verb, N-W for a possessed noun), B clause. VPISH throughout: ρA is first-merged in Spec,v, ρP as the complement of V. The inverse is the site's INV analysis: bi- = the patient slot Moved to an outer Spec,v above ρA (the inverse as Agree and EPP). Each slot's bundle carries person, number, an index and [prx]/[obv] from the start (Σ). Aissen's trees are Jelinek-style (nouns as adjuncts to a flat clause, coindexed with pronoun slots), which is pre-VPISH notation, so they are restated here and not used as evidence about structure.

B-1. (1a) ʼashkii bi-lį́į́ yi-yiiłtsą́ ‘The boyi saw hisi horse’ (good)

  1. Σ Σ(SEE) = {ρA (seer) i, ρP (seen) j}. For the noun: Σ(HORSEposs) = {π i}.
  2. A N-W: Merge the stem 'horse' with n; Merge π in its specifier; Agree n ⟶ π: π is [prx] (POSS-2 makes the noun phrase [obv]); spelled bi- (POSS-1). Seal N-W = bi-lį́į́.
  3. A W: Merge V with ρP j; Merge v; Merge ρA i in Spec,v. POSS-3: ρA shares i with π, so ρA is [prx]; ρP is [obv].
  4. A Agree (object prefix) ⟶ ρP [obv] → yi-. No inverse: ρA [prx] c-commands ρP [obv], and POSS-4 is met. Spell-out, Seal W = yi-yiiłtsą́.
  5. B Link nouns (Willie (20)): the outer noun ʼashkii links to ρA (i), and the inner N-W bi-lį́į́ to ρP (j). The index on π inside N-W matches ʼashkii's. Right-headed clause: nouns, then W.

Ruleset A does all the feature work inside two words (π inside N-W; ρA, ρP inside W). Ruleset B only lines nouns up with slots.

B-2. *(1b) bi-lį́į́ ashkii bi-yiiłtsą́ (bad)

  1. Σ, A N-W as B-1 steps 1–3: π i [prx], so ρA i [prx] and ρP j [obv].
  2. A Inverse: Move ρP to outer Spec,v, above ρA. Agree would spell ρP as bi- only if it were [prx] (POSS-1). It is [obv], so bi- mismatches.
  3. A And even apart from the prefix, POSS-4 fails: ρP [obv] now c-commands its co-argument ρA [prx]. Crash.

Two independent failures, Aissen's (step 2, morphology only) and Oxford's (step 3, structure). This sentence cannot tell them apart.

B-3. *(2) Bi-lį́į́ ashkii yi-yiiłtsą́ ‘(Hisi horse saw the boyi)’ (bad)

  1. Σ Σ(SEE) = {ρA (seer = the horse) j, ρP (seen = the boy) i}; N-W 'his horse' with π i.
  2. A N-W: π i [prx] → bi-; the phrase 'his horse' is [obv] (POSS-2). POSS-3: ρP i is [prx]; ρA j, which 'his horse' will link to, is [obv].
  3. A Direct (no Move): Agree object prefix ⟶ ρP [prx] should give bi-, but the form has yi-: mismatch (Aissen). And ρA [obv] in Spec,v c-commands ρP [prx]: POSS-4 fails (Oxford). Crash.

B-4. (3) Hastiin Nééz beʼesdzáán bi-níłʼį́ ‘Mr. Longi's wife is looking at himi’ (good)

  1. Σ Σ(LOOK.AT) = {ρA (looker = the wife) j, ρP (looked-at = Mr. Long) i}; N-W 'his wife' with π i.
  2. A N-W: π i [prx] → b- (fused in beʼesdzáán); 'his wife' is [obv]. POSS-3: ρP i [prx], ρA j [obv].
  3. A W: Merge V + ρP, v, ρA in Spec,v. Inverse: Move ρP to outer Spec,v. Now ρP [prx] c-commands ρA [obv]: POSS-4 met (and RCC: the higher-ranked slot is on top).
  4. A Agree object prefix ⟶ ρP [prx] → bi-. Spell-out, Seal W = bi-níłʼį́.
  5. B Under bi- the linking is reversed: the outer noun Hastiin Nééz links to the raised ρP, the inner N-W beʼesdzáán to ρA. π inside N-W matches Hastiin Nééz.

Only B-4 separates the accounts in principle. If step 3's Move did not happen (a shallow bi-), POSS-4 would fail exactly as in B-3, and (3) would be out. Aissen's account still lets it in, because it never checks c-command. So (3) supports "deep" only if POSS-4 (c-command) is the right constraint for Navajo, which these data cannot show.

10. Grant's ideas

GRANT'S CONJECTURE Verbs take feature bundles, not nouns. Not contradicted, and this is the most favourable data yet. Every condition in B-1 to B-4 is stated over bundles: the possessor slot π inside the noun word and the verb's two slots, each carrying person, an index and [prx]/[obv]. What the nouns are ('horse', 'wife', 'car') plays no role, and the nouns only line up with slots at B. A pronominal-argument analysis predicts the same, so it is compatible, not proof.

GRANT'S CONJECTURE Strata (theta roles first, event roles next, noun selection last). Neutral to slightly favourable. The obviation values must be fixed on the bundles before the prefixes are spelled out at A, but they depend on who corefers with whom, which looks like noun-level information. In the derivations above this works only because each bundle carries its index from Σ, before any noun is chosen. So coreference has to be a property of bundles, not of nouns, which fits a late noun stratum. The event-role layer is not tested.

HYPOTHESIS The word as the A/B changeover. The possessor sits inside its own small word (N-W), so the possessor constraint links two A-domains (N-W and W) through one index, and B only checks that the nouns line up. If Grant's idea of word-internal Merge or Agree sites is right, π in N-W is one: a slot whose [prx] value is set by Agree inside the noun word and then constrains the verb word. Chemistry: no correlate is claimed.

HYPOTHESIS Theta roles stay put. In (1a) and (3) the seer/looker is the agent and the seen/looked-at is the patient. Only the [prx]/[obv] layer and (on the deep reading) the height of the patient change. This is the same split the site already draws between roles (fixed at Merge) and voice (Algonquian §6).

HYPOTHESIS Update 2026-10-25 (~4:46am PT): One "his", two third persons reads Horseherder 1998 first-hand. Owner yi- needs a coreferent third-person antecedent and a different, c-commanding singular third-person doer (SOURCED). Ranking predicts all her facts except that plural doers block yi-, the same ranking-predicts-✓/speakers-say-* pattern as W&J (70b), which also has a plural doer. That adds a plural-doer lean to T58a and gives a new D2 test (deep vs shallow bi-) that does not run through ranking.

HYPOTHESIS Update 2026-10-25 (~12:46pm PT): What comes first decides reads Platero 1978 first-hand. 'The boyi was seen by hisi father' is good with bi-, with the name outside 'his father', and the yi- form cannot mean it (SOURCED). His precede-and-command rule, plus bi- putting the unspoken patient first, predicts W&J (70b)'s star with no plural condition and no Move in his notation. Restated in VPISH, (70b) separates "patient above the doer's phrase" (deep bi- or a null topic) from shallow bi-, so D2 stays OPEN. His (3.18) also clashes with Young & Morgan on yi- (T60b).

11. Bottom line

HYPOTHESIS Update 2026-10-24 ~8:46pm PT. Aissen 1997, read first-hand, derives the Algonquian and Tzotzil versions of this restriction by ranking alone. So POSS-5's "fits deep" is downgraded to "fits any form where the patient outranks the agent" (GAP-4). A new lean in favor of structure comes instead from Willie & Jelinek 2000 (70b), where the owner's name sits inside the doer's phrase. See where the owner's name sits.

Navajo does have binding-type data for the inverse question: a "his" inside the doer's noun phrase can point back to the person acted on only in the bi- form, and a "his" inside the acted-on noun phrase can point back to the doer only in the yi- form (SOURCED, Young & Morgan 1980 via Aissen 2000). That is the pattern a deep inverse predicts (HYPOTHESIS). Aissen's owner-outranks-owned rule and plain word order predict exactly the same judgments without any movement, and the examples use names, not "every", so D2 moves from "no data" to "fits deep, not decisive" (OPEN).

12. Open

IDQuestionWhat would settle it
T53aBinding/scope under bi- (Oxford's D2).Partly answered 2026-10-24 ~12:46pm PT (§5–6): coreference data found and they fit "deep" (HYPOTHESIS), but obviation and word order explain them too. OPEN
T57aFourth-person test: with a 4th-person antecedent or possessor (outside the obviation tier, Aissen §6), does the possessor restriction disappear (obviation: (i), (ii)) or stay (c-command binding)?A text-layer search of Willie 1991 for ha-/ji- found no such sentence; Young & Morgan 1987; speakers. OPEN Update 2026-10-24 ~8:46pm PT: partly answered on possessor-gap §6 (W&J 2000 (77)).
T57bYoung & Morgan 1980:171–172 first-hand: the exact forms of (27), (29), (30), their full statement, and whether they give the bi- version of 'his horse saw the boy'.Young & Morgan 1980 (not on the box). OPEN
T57cA quantified antecedent ('every boy … his horse') under yi- vs bi-: binding proper, not coreference.A grammar or speakers; no opened source has one. OPEN
T57dThe Algonquian possessor constraint first-hand (Rhodes 1993; Aissen 1997): does it use the deep (3′→3) inverse, as Aissen's summary implies?Aissen 1997 needs OCR on the box; Rhodes 1993 not found. OPEN Update 2026-10-24 ~8:46pm PT: answered on possessor-gap §4: yes, the 3′→3 inverse (Cree, Ojibwe), but Aissen derives it by ranking with no movement, and Tzotzil uses a passive, so it is not a deep-inverse test (HYPOTHESIS).

Sources: Aissen 2000 (author's RTF), §§2, 4.2, 4.4, 5, 6, 7, fn. 13–14, examples (8), (11), (24)–(31); Willie 1991 pp. 57–61 (page images pp. 60–61), 92–93, 185–186; Oxford 2023 preprint pp. 29–30. Cross-links: deep or shallow inverse?, the Algonquian inverse, possessor raising, the inverse as Agree and EPP, fourth person compared, headedness, open questions, sources.

Where the owner's name sits: Aissen 1997 read first-hand, and a Navajo sentence the ranking rule gets wrong (T57d, T57a)

Working note. Kept as written (URL, data and status labels unchanged). Filed under Possessors and object-first orders; the consolidated argument is the essay possessors. Rule IDs: rules index.

Verbs on this page → dictionary: dabiiłtsą́ ‘see’

Restated on the unified structure (2026-11-15). Possessors are restated as the left specifier of the noun phrase, with the possessor prefix as agreement (as on a grammatical postposition). This page’s particular analysis has not yet been restated on the unified structure; light n stays OPEN. 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, possessors. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.

The possessor-binding page found that when the person acted on owns the doer ('Mr. Long's wife is looking at him'), Navajo allows only the bi- form. It left two things open: what the Algonquian version of this restriction looks like first-hand (T57d), and what happens with a fourth-person owner (T57a). This page reads Aissen's 1997 Language paper on obviation (OCR of the page images, every example quoted here checked on the image). In plain words: Fox, Cree and Ojibwe show the same restriction as Navajo, and Tzotzil (Mayan) shows it too, but there a passive is the fix instead of an inverse. Aissen explains all of them with one ranking rule, owner outranks owned, and no claim that anything moves. So the restriction on its own does not show that Navajo bi- moves the patient. But there is a Navajo sentence the ranking rule gets wrong. Willie & Jelinek (2000) star 'all John'si children saw himi' even with bi-, which is exactly the form the ranking rule calls for. In that sentence the owner's name sits inside the doer's phrase. In every good example found, it stands outside. A bi- that moves the patient up, together with Condition C (which Hale 1999 shows is active in Navajo), predicts this. The ranking rule alone does not. This is a lean, not a result: the starred sentence also differs in having 'all' and in coming from a different source.

Update 2026-10-25 (~8:46pm PT): Saxon 2003 (via Young & Morgan 1987) gives clauses with a thing as subject that have two bi-s. That is contradicting data for "bi- = the one proximate". The ranking conclusion on this page still stands for its animate-subject data, but it should be read as limited to animate subjects. See two-osv §6 HYPOTHESIS.

On this page: 1 Question · 2 Sources · 3 Aissen 1997: the restriction outside Navajo · 4 Her account: ranking, not movement · 5 The Navajo sentences side by side · 6 The fourth person (T57a) · 7 Three accounts, scored · 8 Rules · 9 Derivations · 10 Grant's ideas · 11 Bottom line · 12 Open questions

Notation as on the possessor-binding page. Σ = the role schema. A = the word level (prefixal, left-headed; W for the verb word, N-W for a possessed noun word with possessor slot π). B = the clause level (right-headed). ρA/ρP are the agent and patient slots; [prx]/[obv] are proximate/obviative. Merge vs Agree: Merge builds structure (it puts two pieces together, for example a noun phrase next to the verb word). Agree values features (it copies person, number, index or [prx]/[obv] from one bundle to a probe). Move is Merge again of something already in the structure.

1. Question

T57d: in the Algonquian languages, is the possessor restriction carried by the (deep, 3′→3) inverse, as Aissen 2000's summary implies? And does that make the restriction a test for a deep inverse (one where the patient moves above the agent), which is how the possessor-binding page used it (POSS-4/5)? T57a: with a fourth-person owner, does the restriction disappear?

2. Sources

SOURCED Aissen, Judith 1997. "On the syntax of obviation." Language 73(4): 705–750. Box copy, image-only. This run OCR'd all 47 pages with tesseract 5.5.0, and every example quoted below was checked on the page image: (16), (19), (21), (36), (44), (116), and the prose on pp. 712, 736, 741, 745–746. Non-Navajo data are given with Aissen's own glosses (and Rhodes's glosses for Ojibwe, as she prints them). See sources.

SOURCED Willie, MaryAnn & Eloise Jelinek 2000. "Navajo as a discourse configurational language." In Fernald & Platero (eds.), The Athabaskan Languages (OUP). Re-read on page images (120 dpi) of pp. 275–278 (Jelinek collection scan, PDF pp. 155–158): (70), (71), (73), (75)–(77) and the prose on p. 277. (70) and (77) were first transcribed on the fourth-person page §3.6. The possessor-binding run missed them.

SOURCED Willie, MaryAnn 1991. Navajo Pronouns and Obviation. Page images (130 dpi) of pp. 216–217: (28a, b), (29).

Carried over (not re-opened): Aissen 2000 (author's RTF), (27)–(31), Table 1 (from Faltz 1998:172), §6; and Hale 1999 on Condition C (Conflicting Truths page).

Not opened: Rhodes 1993 (Ojibwe data are from Aissen's citation of it), Wolfart 1973, Young & Morgan 1980 (archive.org copy is access-restricted), Platero 1982, Hale, Jeanne & Platero 1977, Thompson 1989/1994.

3. Aissen 1997: the restriction outside Navajo SOURCED

The rule (p. 711, §3.2): "when both a possessed noun (POSSESSUM) and its possessor (GENITIVE) are third person (animate), the genitive must outrank the possessum". Stated as (18) GEN > HEAD. In Algonquian nominals "the possessed nominal is always obviative" ((2), p. 707, Cree ote-ma 'his(3) horse(3′)').

Aissen 1997:712, (16) (Fox; page image checked)

neniwa okwisani 'the man (prox)'s son (obv)'. The versions with the son proximate (*16b) or both proximate (*16c) are starred.

Aissen 1997:712, (19) (Plains Cree, from Wolfart 1973:25; page image checked)
(1)Ca·note·maki·-ma·kwamik.
John(3)his(3)dog(3′)bite (TA 3′–3)
'John's dog bit him (sc. John).'

The verb is inverse (3′ acting on 3). Aissen: the direct form is impossible on this reading (Wolfart 1973:25; Frantz 1966; Rhodes 1993). Her gloss does not show whether Ca·n and ote·ma form one phrase.

Aissen 1997:713, (21) (Ojibwe, from Rhodes 1993, glosses by Rhodes; page image checked)
(2a)Wwaabmaanwgwisan.
3-see(TA)-DIRECT-OBV3-son-OBV
'Hei sees hisi/j son.' also 'Hisi son sees himj/*i.'
(2b)Wwaabmigoonwgwisan.
3-see(TA)-INVERSE-OBV3-son-OBV
'Hei sees hisj/*i son.' also 'Hisi son sees himi/j.'

Underlying forms as printed: /o-waabam-aa-an o-gwis-an/, /o-waabam-igo-a o-gwis-an/. Fn. 11: Rhodes 1990's apparent counterexample (from Bloomfield 1958:158) has no coreference in its context.

Aissen 1997:720, (36); 722, (44) (Tzotzil; page images checked)
(3a)*Tas-saʼproiy-ajnilliManvel-ei.
ICPA3-seekA3-wifetheManuel-ENC
'Manueli's wife is looking for himi.' (active)
(3b)Tasaʼ-atyuʼuny-ajnilliManvel-e.
ICPseek-PSVbyA3-wifetheManuel-ENC
'Manueli was sought by hisi wife.' (passive)

Aissen gives the relational structure of (3a) as (37): *Ta ssa' proi [y-ajnil li Manvelei], with Manuel inside the subject. The image of (44a) prints "Manvele-ENC" in the gloss line; we write "Manuel-ENC". Tzotzil is VOS (p. 720), so the final li Manvel-e in (3b) can be the passive subject. Her translation reads it that way, but the string alone does not show the bracket. The same contrast holds for (36b)/(44b) 'the boyi's dog bit himi' / 'the boyi was bitten by hisi dog'.

4. Her account: ranking, not movement SOURCED

  • Direction, not voice. Aissen assumes the Algonquian inverse does not change grammatical relations: both direct and inverse are transitive, and the agent is the subject in both (p. 708, after Dahlstrom). Fn. 5 names the alternatives, LeSourd 1976 (inverse = passive) and Perlmutter & Rhodes 1988 / Rhodes 1994 (relations reversed). §6.1 says the framework "is compatible with relation-changing analyses".
  • (10) Co-linking: "Distinct nominals … bear the same relation in an obviation span if and only if [they] are presupposed coreferential" (p. 710). (24) Minimal span: subject, object and genitive must share one span (p. 714).
  • The gap follows from GEN > HEAD + co-linking + (14)/(15) alignment. In (2b) the son is obviative because he is possessed, and his owner is proximate. If the owner is the object ('his son sees himi'), the object is proximate and the subject obviative, which needs INVERSE ALIGN, the -igo- form. Tzotzil has no inverse, so the same ranking needs a passive (§4.2).
  • (116), pp. 740–741: inverse and passive both arise where "the patient outranks the agent in obviation status; the difference is in the mediating syntax."
  • First and second person (p. 712): "Since the possessum can only be third person, while the genitive can be any person, structures with first and second person genitives always satisfy GEN > HEAD."
  • Against binding theory (Appendix, pp. 745–746). Aissen rejects Condition C / c-command accounts of these gaps (Woolford 1991 and Trechsel 1995, for Jakaltek) on three grounds: (i) (120) *Ta s-sa' Manveli li y-ajnil-e proi 'Hisi wife is looking for Manueli', with the name outside, is still bad; (ii) the gap disappears with an inanimate owner ((75a) I-y-ich' tal proi y-ajval li ixtalal-ei 'The owner of the ring brought it here', p. 730), and "basic phrase structure" cannot be sensitive to animacy; (iii) with a quantified or wh- antecedent, (122a, b), the subject can bind a possessor in the object ('Each childi lost hisi/heri toy'). So binding by an operator follows ordinary structure, separately from the obviation gap.
  • Navajo (p. 736): "Navajo also has coreferential blockages involving object pro with genitive antecedent (Platero 1982; Hale et al. 1977)." Aissen takes this, with the animacy gaps, as a sign that obviation is active in Navajo (diagnostic (97b), p. 735). P. 743: these languages "and Navajo—are all head-marking".

HYPOTHESIS Answer to T57d. Yes, the Algonquian restriction is carried by the 3′→3 inverse (Cree TA 3′–3, Ojibwe -igo-), and the Cree and Ojibwe cells line up with Navajo (Cree (19) ↔ Navajo Hastiin Nééz beʼesdzáán bi-níłʼį́; Ojibwe (21) ↔ ʼashkii bi-lį́į́ yi-yiiłtsą́ / its starred bi- version). But Aissen's account of them uses an inverse that does not move anything, and the same gap is fixed by a passive in Tzotzil. So the restriction shows that the patient outranks the agent in [prx]/[obv]. On its own it does not show that the patient sits higher in the tree. The possessor-binding page's POSS-5 confound, account (ii), is therefore not a side issue. It is the analysis in the original paper.

5. The Navajo sentences side by side

Willie & Jelinek 2000:276, (70); 278, (77) (page images checked)
(4)tʼááʼałtsoJáanba-ʼáłchínídabiiłtsą́
justallJohn3POSS-childrenPL-3ACC-3NOM-saw
a. '[All John'si children]j saw himk.'   b. * '[All John'si children]j saw himi.'
(5)Jáantʼááʼałtsoníha-ʼáłchínídahoołtsą́
Johnjustall4POSS-childrenPL-4ACC-3NOM-saw
'Johni, just all that person'si childrenj saw that personi.' (All John'si childrenj saw himi.)

W&J's tree for (70) marks the verb "-saw INV / INVERSE", with the possessive phrase as a FOCUS adjunct and 3ACC as Topic. (71): "Coreference between possessive arguments internal to a complex NP and the pronominal arguments of a following verb-sentence is excluded." "For most speakers, coreference of this kind is impossible; for others, it is strongly dispreferred." P. 277: "use of the inverse marks the complex POSS NP as disjoint in reference with the topicalized patient of the following verb-sentence." "The fourth person in Navajo, like first and second, cannot enter into the direct/inverse alternation." W&J's tree is pre-VPISH notation (an S node with a focus adjunct and an IP holding the pronoun slots). It is used here for its judgments and glosses, not as evidence about structure.

SentenceVerbOwner's nameOwner = patient?Judgment
(30) Hastiin Nééz beʼesdzáán bi-níłʼį́ (Y&M 1980:172 via Aissen 2000)bi-outside 'his wife' (on Willie's two-adjunct parse, possessor-binding §3)yes✓
(4b) W&J (70b)bi- (W&J: INVERSE)inside 'all John's children'yes* (most speakers)
(5) W&J (77)4th object ho-, no yi-/bi-outside ('John, all that person's children')yes✓
Tzotzil (3a) / (3b)active / passiveinside (37) / subject (translation)yes* / ✓

HYPOTHESIS The ranking rule gets (4b) wrong. In (4b) the owner John is proximate (GEN > HEAD), so 'children' is obviative. John is the patient, so the object is proximate and must be bi- (Aissen 2000 (8)). That is the form used, so the tier account predicts (4b) is fine, just as it predicts (30). W&J star it. In each good example found, the owner's name stands outside the doer's phrase; in each bad one, it is inside. Aissen's (120) is the one datum against this pattern (name outside, still bad), but it is an active clause, which ranking already rules out.

OPEN Three things besides the position of the name separate (4b) from (30): (4b) contains 'all' (tʼáá ʼałtso); it comes from a different source and speakers (W&J vs Young & Morgan); and its antecedent is a name inside a focus phrase. No minimal pair is available (T58a).

Willie 1991:216–217, (28a, b), (29) (page images checked)
(6a)Johnháíshbaghan=diyiztsʼǫs?
Johnwho3POSS:home=at3.3.kissed
'Who did John kiss at his house?'
(6b)Johnháíshyighan=diyiztsʼǫs?
Johnwho3POSS:home=at3.3.kissed
'Who did John kiss at her house?'
(6c)Johnháíshhaghan=diyiztsʼǫs
Johnwho4POSS:home=at3.3.kissed
'Who did John kiss at that other person's house?'

Willie: "This contrast in Possessor marking is not one of voice, but shows the use of this alternation to mark obviation." "Speakers vary in the extent to which they use yi-/bi- in this way." "Since Navajo has fourth person forms, which are used to mark obviation, there seems to be less reliance on yi-/bi- for this purpose." (6c) is printed without a question mark. The text layer reads haish; the page image has háísh.

HYPOTHESIS Read with Aissen 2000 (8), (6a)/(6b) are a possessor-only [prx]/[obv] contrast: ba- picks the proximate (John), yi- the obviative (the person kissed). The verb is direct (yi-) in both. So the owner's [prx]/[obv] is valued in the possessed noun word without any change of voice in the verb word, which is what Aissen's tier needs and a movement account does not explain. (6c) adds a third owner, outside the pair.

6. The fourth person (T57a)

SOURCED Fourth person "cannot enter into the direct/inverse alternation" (W&J p. 277). It sits outside the obviation tier (Aissen 2000 §6). With a fourth-person subject and a third-person object only bi- occurs, as in bi-nijil-nish (BI-4sg-work) 'One is working on it' (Aissen 2000 Table 1, from Faltz 1998:172). With a fourth-person owner and a fourth-person object, coreference that (4b) bars is possible ((5) = W&J (77)).

HYPOTHESIS T57a, partly answered. (a) As T57a was framed ("does the yi-/bi- possessor restriction vanish?"), it cannot arise. If the owner is fourth person and corefers with the object, the object is fourth person too, so the verb carries ho-, not yi- or bi-. The yi-/bi- choice exists only with two third persons (Aissen 2000 §3.3). (b) The coreference reading itself does come back with the fourth person ((5)). This is what Aissen 1997's logic predicts: like 1st/2nd, a non-tier owner satisfies GEN > HEAD vacuously (p. 712). (c) But (5) also moves Jáan out in front of the possessed phrase. So it does not separate "fourth person" from "the name is outside". W&J present the fourth person as the fix (p. 277: the fourth "can be used to force coreference"); we cannot tell from (5) which change does the work.

7. Three accounts, scored

Account(30) ✓(4b) *(5) ✓Tzotzil (75): gap gone with inanimate ownerNeeds bi- to move the patient?
(i) Ranking only: Aissen's tier, GEN > HEAD + co-linking✓✗ predicts ✓✓ (4th off tier)✓ (via ANIM > INAN; she notes a complication with passives, p. 730)no
(ii) Deep bi- + Condition C: ρP Moves above ρA; a pronoun may not c-command a coreferent name (Hale 1999 for Navajo)✓ (only pronouns inside the doer)✓ (raised pronoun c-commands Jáan)✓ if Jáan is outside, but then the 4th person is not doing the worknot addressed (Aissen's objection (ii))yes
(iii) W&J's linking: a name inside a nominal adjunct cannot be the antecedent of a verb's pronoun; the inverse marks the POSS NP disjoint from the topic✓ (Mr. Long linked directly)✓✓ (4th forces coreference)not addressedno

HYPOTHESIS (4b) counts against (i) taken alone: ranking needs help from structure or linking. (ii) and (iii) both fit. They come apart on a sentence with the name outside the doer's phrase but still inside some phrase the raised object c-commands, and on the inanimate-owner test, which (i) handles and (ii) does not obviously handle (T58a, T58c). Aissen's Appendix objection to Condition C was aimed at Jakaltek and Tzotzil actives. It does not touch (4b), which is an inverse. A combined account, ranking for which verb form is used and structure for which coreference is allowed, is the most data-faithful option so far. Aissen's (122) (operators bind by structure) already points that way.

8. Rules

GAP-1  (= Aissen 1997 (18)) GEN > HEAD: a 3rd-person genitive outranks its possessum         SOURCED (Aissen's)
       on the participant/obviation hierarchy; 1st/2nd (and, by Aissen 2000 §6, 4th)
       genitives satisfy it vacuously.
GAP-2  (= (10), (24)) Presupposed coreferents in one span share [prx]/[obv]; subject,         SOURCED (Aissen's)
       object and genitive share one span.
GAP-3  The gap is repaired by whatever form puts [prx] patient with [obv] agent:               SOURCED (Aissen 1997 (116))
       inverse (Fox, Cree, Ojibwe), passive (Tzotzil), bi- (Navajo (30)).
GAP-4  GAP-1–3 do not require the patient to Move. The gap alone does not show deep bi-.          HYPOTHESIS (replaces POSS-5's
                                                                                                 "fits deep")
GAP-5  Navajo (4b): under bi-, a coreferent NAME inside the agent phrase is still out          SOURCED (W&J (70), (71));
       (W&J (71)). GAP-1–3 alone predict it good.                                            HYPOTHESIS (the prediction)
GAP-6  GAP-5 follows if ρP is Moved above ρA (deep bi-) and Condition C holds of the            HYPOTHESIS
       pronoun in ρP; it also follows from W&J's adjunct-linking condition (no Move).
GAP-7  4th-person owner + 4th object: no yi-/bi- choice exists; coreference available (77).      SOURCED (W&J (77), p. 277);
       Confounded with the name's position.                                                    HYPOTHESIS (the confound)

9. Derivations, step by step

Labels: Σ role schema, A word, B clause. VPISH throughout: ρA is first-merged in Spec,v, ρP as the complement of V. Each slot's bundle carries person, number, index and (for 3rd person) [prx]/[obv] from Σ. Nouns are Merged at B and linked to slots (nominal adjunction). "Deep bi-" = ρP Moves to an outer Spec,v above ρA (inverse as Agree and EPP). Each derivation is given twice where the accounts differ.

G-1. (30) Hastiin Nééz beʼesdzáán bi-níłʼį́ with no movement (account (i))

  1. Σ Σ = {ρA: 3, index w; ρP: 3, index m}. Inside the possessed noun: π: 3, index m (presupposed coreferent with ρP).
  2. A N-W: Merge π with the noun stem 'wife'. GAP-1: π outranks the possessum, so π = [prx] and the phrase 'his wife' = [obv]. Agree spells π [prx] as b- (POSS-1).
  3. A W: Merge V 'look at', ρP, v, ρA (no Move). GAP-2: ρP shares index m with π, so ρP = [prx], and ρA (linked to 'his wife') = [obv]. Agree: the object prefix reads [prx] → bi-.
  4. B Merge Hastiin Nééz (linked to ρP) and beʼesdzáán (linked to ρA) to the left of the verb word, right-headed. Order: proximate before obviative (Aissen 2000 (11)).
  5. Spell-out: Hastiin Nééz beʼesdzáán bi-níłʼį́ ✓. Nothing has Moved. Ranking alone gets the right form.

G-2. (4b) *'all John'si children saw himi': ranking only vs deep bi-

  1. Σ Σ = {ρA: 3pl, index c; ρP: 3, index j}. Possessed noun: π: 3, index j, filled by the name Jáan inside the phrase.
  2. A N-W: Merge π with 'children'. GAP-1 → π [prx], phrase [obv]. Agree spells π as ba-. At B, Merge Jáan as the possessor, then 'all' (tʼáá ʼałtso): one phrase, [tʼáá ʼałtso Jáan ba-ʼáłchíní].
  3. A W: Merge V 'see', ρP, v, ρA. GAP-2 → ρP [prx], ρA [obv]. Agree: bi-; da- counts the agent (number & scope). Spelled out as dabiiłtsą́.
  4. Account (i) stops here: every condition is met, so it predicts ✓. W&J: * (most speakers). Mismatch.
  5. Account (ii) continues: A Move ρP to the outer Spec,v (deep bi-). Now ρP, a null pronoun with index j, c-commands ρA, and with it the linked phrase containing Jáanj, if the phrase is interpreted at ρA's site.
  6. B Condition C: a name may not be c-commanded by a coreferent pronoun → index j on ρP is barred; only the disjoint reading (4a) remains ✓. Under a shallow bi- (no Move), ρA stays above ρP and (4b) would be fine, as in the Spanish parallel W&J give ((72)).

Step 5 relies on assumptions the Hale derivation page audits (A1–A10): mainly, that a linked noun phrase counts for Condition C at its slot's site. That assumption is not settled here.

G-3. (5) fourth-person owner and object (W&J (77))

  1. Σ Σ = {ρA: 3pl, index c; ρP: 4, index j}. Possessed noun: π: 4, index j.
  2. A N-W: Merge π with 'children'; Agree spells π = 4 as ha-. GAP-1 is vacuous: π is off the tier.
  3. A W: ρP = 4 is off the tier. Only one 3rd person (ρA) remains, so no [prx]/[obv] contrast arises. Agree: object ho-, plural da- → dahoołtsą́. No direct/inverse choice (W&J p. 277).
  4. B Merge Jáan (linked to ρP, index j) first, outside, and then [tʼáá ʼałtsoní ha-ʼáłchíní] (linked to ρA).
  5. Spell-out ✓, coreferent. On (i), the 4th person removes the ranking clash. On (ii), the name is outside the agent phrase, so Condition C is never at issue. On (iii), the 4th person "forces coreference". All three get ✓. (5) does not choose among them.

G-4. Willie (6a)–(6c): the owner's rank set inside the noun word

  1. Σ Σ = {ρA: 3 (John); ρP: 3 (who)}. Possessor of 'home': π = 3 or 4.
  2. A W: John is topic → ρA [prx], ρP [obv]. Agree: object prefix yi- → yiztsʼǫs (the same in all three).
  3. A N-W (inside the postpositional phrase): Merge π with 'home'. Agree copies [prx] from John's index → ba- (6a); [obv] from the person kissed → yi- (6b). With π = 4, it is off the tier → ha- (6c), a third person outside the pair.
  4. B Merge [π-ghan=di] as a lexical-postposition phrase to the left of the verb, right-headed (Willie: a "Lexical postposition, not a Grammatical postposition").
  5. Spell-out ✓. The [prx]/[obv] value is assigned at A inside a noun word, with no change of verb voice. This is the ranking half of the combined account.

10. Grant's ideas

GRANT'S CONJECTURE Feature bundles as arguments. Not contradicted. GEN > HEAD and co-linking are stated over person, index and [prx]/[obv]: that is, over bundles, not over noun content. (4b) shows that where a name sits matters, which is a fact about the later layer where nouns are Merged, not about the argument slots. It is not counter-evidence to slots being bundles. Aissen's animacy effect (Tzotzil (75)) does show an inherent property of the owner (animate vs inanimate) taking part. That is neutral if animacy is a bundle feature present in Σ, as on the schema page.

GRANT'S CONJECTURE Strata (theta roles first, event structure next, lexical items later). OPEN. The combined account fits a layered order. [prx]/[obv] are valued on bundles at the word level (G-1 steps 2–3, G-4 step 3). The Condition C effect needs the name, so it can only arise once nouns are Merged at B (G-2 step 6). That is consistent with lexical items entering late, but it is one example, and account (iii) gives the same ordering without Move.

HYPOTHESIS Update 2026-10-25 (~4:46am PT): One "his", two third persons reads Horseherder 1998 first-hand. Owner yi- needs a coreferent third-person antecedent and a different, c-commanding singular third-person doer (SOURCED). Ranking predicts all her facts except that plural doers block yi-, the same ranking-predicts-✓/speakers-say-* pattern as W&J (70b), which also has a plural doer. That adds a plural-doer lean to T58a and gives a new D2 test (deep vs shallow bi-) that does not run through ranking.

HYPOTHESIS Update 2026-10-25 (~12:46pm PT): What comes first decides reads Platero 1978 first-hand. 'The boyi was seen by hisi father' is good with bi-, with the name outside 'his father', and the yi- form cannot mean it (SOURCED). His precede-and-command rule, plus bi- putting the unspoken patient first, predicts W&J (70b)'s star with no plural condition and no Move in his notation. Restated in VPISH, (70b) separates "patient above the doer's phrase" (deep bi- or a null topic) from shallow bi-, so D2 stays OPEN. His (3.18) also clashes with Young & Morgan on yi- (T60b).

11. Bottom line

  • SOURCED Aissen 1997 shows the same owner-outranks-owned gap in Fox, Cree and Ojibwe (fixed by the inverse) and Tzotzil (fixed by the passive). She derives it by ranking, assumes the Algonquian inverse is not relation-changing, and argues against c-command accounts.
  • HYPOTHESIS T57d answered: the restriction alone is not a deep-inverse test. The possessor-binding page's "fits deep" (POSS-5) is downgraded to "fits any form where the patient outranks the agent" (GAP-4). D2 stays OPEN.
  • HYPOTHESIS New lean: W&J (70b) is bad under bi- although ranking predicts it good. A deep bi- plus Condition C predicts it; so does W&J's linking condition. Confounded with 'all' and source (T58a).
  • HYPOTHESIS T57a partly answered: with a fourth-person owner the yi-/bi- choice cannot arise; the coreference returns (W&J (77)), but the name also moves outside, so the fourth person is not isolated.

12. Open questions

IDQuestionWhat would settle it
T58aWhy is W&J (70b) bad when (30) is good: the name inside the doer's phrase, the quantifier 'all', or source/speaker difference?A minimal pair: (30) with 'all', or (70) without 'all', or with Jáan outside. Speakers or a source with both. OPEN
T58bThe Navajo genitive-antecedent blockages Aissen cites (Platero 1982; Hale, Jeanne & Platero 1977) first-hand: active, inverse or both?Platero 1982 (IJAL 48) and HJP 1977, not opened. OPEN
T58cInanimate-owner test (Tzotzil (75) analogue): does the Navajo restriction disappear when the owner is inanimate ('the owner of the ring brought it')?A Navajo sentence with an inanimate possessor coreferent with the object; none opened. Predictions: (i) ranking: gap may vanish; (ii) Condition C: gap stays. OPEN
T58dThompson 1989/1994: the earlier obviation proposal for Navajo Aissen cites. What does it add?Not opened. OPEN
T58eIn Cree (19) Ca·n ote·ma ki·-ma·kwamik, is Ca·n inside the possessed phrase (like Navajo (4b), yet good) or a separate phrase (like (30))? If inside, the inverse fixes what Navajo bi- does not.Wolfart 1973:25; Rhodes 1993 (still not found). OPEN
T58fOperator binding (Aissen (122), Oxford D2 = T57c): does a quantified owner under bi- pattern with structure, as in Tzotzil, or with ranking?Merged with T57c; no Navajo data opened. OPEN

Sources: Aissen 1997 pp. 705–750 (OCR + page images); Willie & Jelinek 2000 pp. 275–278 (page images); Willie 1991 pp. 216–217 (page images); Aissen 2000 (RTF, carried over). See also possessor-binding, deep or shallow inverse, the Algonquian inverse, fourth person §3.6, Hale 1999 Condition C, open questions, sources.

One "his", two third persons: Horseherder 1998 on the owner prefix yi-, and a plural doer that breaks it (T58f, T58a)

Working note. Kept as written (URL, data and status labels unchanged). Filed under Possessors and object-first orders; the consolidated argument is the essay possessors. Rule IDs: rules index.

Verbs on this page → dictionary: jizloh ‘rope’ · dabiiłtsą́ ‘see’ · nizin ‘think; wish (with an optative clause)’

Restated on the unified structure (2026-11-15). Possessors are restated as the left specifier of the noun phrase, with the possessor prefix as agreement (as on a grammatical postposition). This page’s particular analysis has not yet been restated on the unified structure; light n stays 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, §2b, possessors. 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.

The last two runs asked when Navajo lets the owner prefix on a noun ("his", "her") point back to someone in the sentence, and why W&J's 'all John'si children saw himi' is bad under bi- (T58a). This run read a source not used here before. Nicole Horseherder's 1998 UBC M.A. thesis is entirely about the owner prefix yi- on nouns, built on her own fieldwork with four Navajo speakers. In plain words: yi- 'his/her' on a noun works only when two singular third persons are around. One is the person it means, who may be named or may just be the person being talked about. The other is a different person who is the doer of the sentence. If the doer is 'I', 'you' or a plural ('the girls'), yi- fails, even though it would refer to someone else. The ranking (obviation) account used on the last two pages predicts the 'I'/'you' failures. It does not predict the plural one, and neither does the verb's own yi-, which is fine with plural doers. That matters for T58a: W&J's bad sentence also has a plural doer ('all John's children', with da-). So a plural doer is now a sourced suspect, alongside "the name is inside the doer's phrase". This is a lean, not a result.

On this page: 1 Question · 2 Source · 3 The data · 4 Her analysis, and its tree · 5 The same facts under ranking · 6 What this does to T58a · 7 A new deep-inverse test · 8 Rules · 9 Derivations · 10 Grant's ideas · 11 Bottom line · 12 Fidelity notes · 13 Open questions

Notation as on the possessor-gap page. Σ = the role schema. A = the word level (prefixal, left-headed; W = the verb word, N-W = a possessed noun word with possessor slot π). B = the clause level (right-headed). ρA/ρP = agent and patient slots; [prx]/[obv] = proximate/obviative. Merge vs Agree: Merge builds structure; it puts two pieces together, for example the slot π inside the noun word, or a noun phrase next to the verb word. Agree values features; it checks or copies person, number and index between bundles. This page is mostly about Agree: Horseherder's condition is a feature match (person and number) between the prefix and two other bundles. Move is Merge again of something already in the structure.

1. Question

T58f (merged with T57c): is the owner prefix bound like an operator (A′-binding), and does that matter for the yi-/bi- possessor facts? T58a: why is W&J (70b) bad under bi- when (30) is good? Candidates so far: the name inside the doer's phrase, 'all', or source. And for both questions: what does a first-hand, single-topic Navajo source say yi- on a noun needs?

2. Source

SOURCED Horseherder, Nicole 1998. A Binding-Theoretic Analysis of Navajo Possessor yi-. M.A. thesis, University of British Columbia. DOI 10.14288/1.0088584 (handle 2429/8149). Open-access PDF (69 pp., scanned with an OCR layer) fetched from UBC cIRcle's public bitstream. The OCR drops nasal hooks and most tone marks, so every Navajo form quoted below was read off the page images (110–600 dpi): pp. 31, 35, 38, 39, 42, 44, 45, 47, 48, 50, 54, 57, 60–61. Thesis page n = PDF page n + 6. See sources.

Her consultants (pp. 9–10) are four speakers from three generations. One, the youngest (b. 1982), "could not distinguish between yi- and bi-" (p. 7). Judgments are hers and her consultants'.

Carried over, not re-opened this run: Willie 1991:32 (9) dayiiłtsą́ (number & scope); W&J 2000 (70), (77) and Willie 1991 (28)–(29) (possessor-gap §5); Aissen 2000 (8), (26)/(33) (possessor-binding).

Not opened: the works she draws her minimal pairs and theory from, namely Young & Morgan 1987:9, Saxon 1984a/1986/1995 (Dogrib), Aoun 1985, Huang & Tang 1991. They are cited here only as she reports them.

3. The data SOURCED

Glosses are Leipzig-style restatements of hers. Her own gloss labels are given in the notes, and her translation and indices are kept. "Pf" = perfective. Her "3-3" / "3.3" on the verb = third-person object + third-person subject.

Horseherder 1998:31, (1a), (1b), (2a), (2b) (page image checked)
(1)*yi-łįį́ʼyayizloh
3POSS(yi)-horse3-for3OBJ.3SBJ.PFV.rope
'hei roped hisk horse' (bad "if out of context")
(2)Joeyi-łįį́ʼyizloh
Joe3POSS(yi)-horse3OBJ.3SBJ.PFV.rope
'hef roped Joe'sj horse'
(3a)haayitʼeegoFrankJoeyikaʼelwod
Q.in.what.wayFrankJoe3-for/after.run
'How did Frankf help Joej?'
(3b)yi-łįį́ʼyaayizloh
3POSS(yi)-horse3-for3OBJ.3SBJ.PFV.rope
'hef roped hisj horse for himj' (a possible answer to (3a))

Her gloss labels: "yi-horse", "3.for", "3-3.Pf.rope". The postposition is printed ya in (1a) and yaa in (2b), and yá on pp. 44 and 47–48; all are transcribed as printed. Fn. 2: "the interpretation [Joei roped [hisi horse]] is possible for some speakers. However this reading is hard to get." Her conclusion (p. 31): yi- "cannot stand alone: it must be co-indexed with a nominal antecedent", and (p. 31) that antecedent "can be provided by the discourse context".

(2) has one reading only (pp. 35–37, (4a)–(4c)): 'he roped Joe's horse'. It cannot mean 'Joe roped his (own) horse'. With context (3a), (3b)'s yi- can be Joe, but not Frank, the doer (p. 34, (3a)–(3d)). A third reading, 'Frank roped Kody's horse for Joe', needs Kody to be named: "Kody yi-łįį́ʼ yaa yizloh" (p. 35).

Horseherder 1998:38, (5); 39, (7a, b) (page images checked)
(4)Paulyi-łįį́ʼyiyiiłtsąnizin
Paul3POSS(yi)-horse3OBJ.3SBJ.PFV.see3.think/want
'Hei thinks Paulk saw his*i/*k/j horse.' Only: 'He thinks Paul saw someone's horse' (p. 39)
(5a)Haaʼíí biniinaDugiMelabichʼįʼbaháchiʼ
what reason.forDugiMela3(bi)-toward3.angry
'Why is Mela angry with Dugi?'
(5b)yi-łééchaaʼíyineezʼį́ʼ-igiibiniinabichʼįʼbaháchiʼ
3POSS(yi)-dog3OBJ.3SBJ.PFV.steal-RELreason.for3(bi)-toward3.angry
'She is angry because he stole her dog.' Literally: 'He stole her dog, that is why she is angry with him.'

(4): yi- "must be disjoint from both the subject of the main clause and the subject of the subordinate clause" and still needs a topic in the discourse (p. 38). She concludes that yi- is "like an R-expression: it must be free everywhere" (p. 39). (5b): here yi- (Mela) corefers with the main-clause subject 'she', and she argues the reason clause is an adjunct, not c-commanded by it (pp. 39–41). Fn. 3: "the response with yi- in (7b) was induced by the OSV (bi-) pattern in the question in (7a)". Her gloss of the relative verb: "that.3-3.stole-clitic".

Horseherder 1998:42, (9a, b) (page image checked, 600 dpi)
(6a)HayitʼéégoJoehooghan-dinádzá
howJoehome-at3SBJ.PFV.return
'How did Joe get home?'
(6b)*yiłįį́ʼhooghan-dibiłnádzá
3POSS(yi)-horsehome-at3(bi)-with3SBJ.PFV.return
'He rode home' (literally 'His horse rode him home')

Bad although there is a discourse antecedent (Joe) and yi- is on the subject, so no subject c-commands it. Hence her third condition (p. 42): "yi- is a disjoint anaphor … it must have two antecedents: an A′-antecedent with which it corefers, and crucially an A-antecedent from which it is disjoint" (following Saxon on Dogrib).

Horseherder 1998:44, (2); 45–46, (3a, b), fn. 1; 47–48, (6), (7); 48, (8) (page images checked)
(7)shitsilíshicheiyiłįį́ʼyáyizloh
1SG.POSS-younger.brother1SG.POSS-grandfather3POSS(yi)-horse3-for3OBJ.3SBJ.PFV.rope
'My younger brother roped my grandfather's horse for him' ✓ (two overt third persons)
(8a)HayitʼéégoJoebikaʼeshwod… *yiłįį́ʼ séloh
howJoe3-after.1SG.SBJ.PFV.run(=help)3POSS(yi)-horse 3OBJ.1SG.SBJ.PFV.rope
'How did I help Joe? … *I roped his horse' (Joe is available; the doer is 'I')
(9a)atʼeekéMelayiyiiłtsą́… *yiłįį́ʼ yá yizloh
girl.PLMela3OBJ.3SBJ.PFV.see3POSS(yi)-horse 3-for 3OBJ.3SBJ.PFV.rope
'The girls saw Mela … *They roped her horse for her'
(9b)MelaDugiyiyiiłtsą́…yiłįį́ʼ yá yizloh
MelaDugi3OBJ.3SBJ.PFV.see3POSS(yi)-horse 3-for 3OBJ.3SBJ.PFV.rope
'Mela saw Dugi … she roped his horse for him' ✓
(10)*shí,yiłįį́ʼséloh/ Shí, shiłįį́ʼ séloh ✓
1SG3POSS(yi)-horse3OBJ.1SG.SBJ.PFV.rope/ 1SG 1SG.POSS-horse …
'*I, I roped his horse' / 'I, I roped my horse'. Same pattern with nihí 'we' (glossed 1.du.Pl; … siidloh) and ni 'you' (… siníloh)

Her gloss labels: "1.poss.brother", "3.poss.horse", "3.3.Pf.rope", "3-1.Pf.run.after(help)", "Pl.girl", "1.S.", "1.3.Pf.rope". (4a) 'How did you help Joe' (bikaʼiinilwod) patterns with (8a); its (4b) is printed *yiłįį́ʼ siniloh (p. 46). Example numbers (1)–(10) on this page are ours; hers are in each block header. Fn. 1 (p. 45): 'Mela helped me … *yiłįį́ʼ yizloh "she roped my horse"' is also out. Her summary (p. 48): "a first, second or plural antecedent cannot license yi-". In (9a) and (9b) neither verb carries plural da-; both are printed yiyiiłtsą́ / yizloh.

The verb's yi- has no such number limit (carried over, SOURCED): Willie 1991:32 (9) dayiiłtsą́ 'They saw him/them' (number & scope §2.1) has a plural doer and an object yi-. Horseherder herself (pp. 59–61) lists how the verb's yi- and the owner's yi- differ: the verb's yi- also needs a disjoint antecedent (the subject), but its coreferent antecedent is in an A-position (the object noun itself), while the owner's must be in an A′-position.

4. Her analysis, and its tree SOURCED (hers)

  • Generalized Binding (Aoun 1985, as she uses it): yi- is an A′-anaphor (it must be bound from an A′-position, either a possessor in Spec,DP / adjoined to DP or a discourse Topic) and it must be A-free everywhere, like a name (pp. 38–41, 44).
  • Disjoint anaphor (Saxon on Dogrib ye-): it also needs an A-antecedent from which it is disjoint (p. 42). This is the doer, the "nearest c-commanding DP in an A-position" (p. 45 schema).
  • Phi-features (pp. 50–55): yi- is "inherently specified for third person singular", and "both the coreferent A′-antecedent and the disjoint A-antecedent … must be specified as third singular", formalized with Huang & Tang's (1991) separate φ- and R-indices ((vi), (13), (14)). She calls the blocking by a non-coreferent 1st/2nd/plural DP the main support for the disjoint-anaphor analysis (p. 55).
  • Nouns in A-positions (p. 38): ruling out (2)-as-'Joe roped his own horse' and *(3b) with Frank "implies that lexical DPs (e.g. 'his horse') are in A-position". That counts against the strict Pronominal Argument view, on which the noun phrases are adjuncts.
  • Dogrib (pp. 56–58): ye- also needs a disjoint antecedent but, unlike Navajo yi-, no A′-antecedent. Navajo (1b) 'Joej built a canoe for hisi/j father' with yi- on 'father' is starred without context. With Frank added it is good: (2) Frank Joe yizheeʼí tsinaʼeeł ya ayiila 'Frankf built a canoe for Joe'sj father' (p. 57, page image checked). She cites the Dogrib data from Saxon 1995.

HYPOTHESIS Framework label. Her trees (pp. 33–35, 51–52) are Government-and-Binding IP trees: pro in Spec,IP, a VP with a PP adjoined, and "agr agr rope" under V. They are not flat-S trees, but they are pre-VPISH in the sense used on the framework audit: no vP, and the subject is not first-merged in Spec,v. They are used here for her judgments and her binding conditions, not as structural evidence. §9 restates them in the vP/TP spine. Her "A-position" claim survives the restatement only if a linked noun phrase counts at its slot's site (assumption A3 on the Hale derivation page), which is not settled.

5. The same facts under ranking (Aissen) HYPOTHESIS

Aissen 2000's reading (carried over): bi-/b- spells a proximate, yi- an obviative (POSS-1, POSS-3), and coreferent third persons in one span share [prx]/[obv]. Fact by fact:

Fact (Horseherder)Binding account (hers)Ranking account (Aissen)
(1) bare yi-łįį́ʼ is out without contextno A′-antecedent✓ also: an obviative needs a proximate in the span, and with only 'he' and an unidentified owner there is no contrast to read. Weaker: ranking says nothing about needing a discourse antecedent.
(2) 'Joe roped his own horse' outA-free (Condition C-like)✓: owner = doer → same span value → [prx] → b-, not yi-
(4) yi- disjoint from both clauses' subjectsA-free everywhere✓ if the span covers the complement clause; not checked
(6b) yi- on the subject outno disjoint A-antecedent✓: the owner must outrank its possessum (GEN > HEAD), so the owner Joe is [prx] → b-, not yi-
(8a), (10) doer = 'I'/'you' blocks yi-φ-mismatch with the disjoint antecedent✓: 1st/2nd are off the third-person tier; the owner is then the only third person, so it is [prx] → b-
(9a) plural doer blocks yi-φ-mismatch (number)✗: two third persons, 'the girls' [prx], Mela [obv] → yi- predicted good. Ranking (in Aissen 1997's Algonquian too) is not sensitive to number.
verb dayiiłtsą́ plural doer fineverbal yi- is a different case (pp. 59–61)✓

HYPOTHESIS So the two accounts agree on everything except number, and number is exactly where the ranking account fails. That is the same failure pattern as W&J (70b) on the possessor-gap page: ranking predicts ✓, speakers say *.

6. What this does to T58a HYPOTHESIS

SentenceOwner prefixDoerOwner's nameRanking predictsJudgment
(30) Hastiin Nééz beʼesdzáán bi-níłʼį́ (Y&M via Aissen 2000)b(i)-singular ('his wife')outside✓✓
W&J (70b) tʼáá ʼałtso Jáan ba-ʼáłchíní dabiiłtsą́, coreferentba-plural (+ da-, + 'all')inside✓*
Horseherder (9a) here = her p. 47 (6) … yiłįį́ʼ yá yizlohyi-plural (no da-)discourse (Mela)✓*
Horseherder (9b) here = her p. 48 (7)yi-singulardiscourse (Dugi)✓✓
  • HYPOTHESIS New lean for T58a: the plural doer. Both sentences that ranking gets wrong have a plural doer and a singular owner who is a different person from the doer. In (9a) the owner's name is not inside the doer's phrase at all, so "name inside the doer's phrase" cannot explain (9a). One condition, "a possessor relation across a plural doer fails", would cover both.
  • OPEN Why it is only a lean. The prefixes differ (ba- vs yi-), and so does the relation: in (70b) the owner is coreferent with the patient, while in (9a) the owner is the beneficiary/patient's owner. The sources and speakers also differ (W&J vs Horseherder's consultants). (70a), the disjoint reading of the same string, is good, so a plural doer does not ruin the sentence, only the coreference. In (9a) the plural is not marked on the verb, so the subject may be dual. Horseherder does not discuss bi- possessors with plural antecedents.
  • HYPOTHESIS The T58a confound list is now four: name inside the doer's phrase; 'all'; plural doer; source. The cheapest minimal pair is still W&J (70) with a singular possessum ('John's son saw him', bi-) (T59a).

7. A new deep-inverse test (D2) HYPOTHESIS

Horseherder's disjointness condition is stated over c-command: the disjoint A-antecedent is "the nearest c-commanding DP in an A-position" (p. 45), which is why yi- on the subject is out ((6b)). That gives a test the possessor-gap page lacked, because it does not run through ranking:

  • Schematic only, no Navajo form claimed: [NP] [yi-NA] bi-V, i.e. under the bi- (inverse) form, a yi- owner on the doer's phrase, meaning 'the patient's/someone's X did V to the patient'.
  • Deep bi- (ρP Moves above ρA): the patient c-commands the doer's phrase, so the yi- owner on it has a disjoint A-antecedent → predicted possible (if the owner is someone other than the patient).
  • Shallow bi- (no Move): the doer's phrase is still the highest A-position, like (6b) → predicted out.
  • Ranking: under bi- the doer's phrase is [obv]. GEN > HEAD needs its owner to outrank it, i.e. be [prx], but the patient already holds [prx] and the owner is a different person. If a span allows one proximate (as in Aissen's Algonquian), the owner can be neither → predicted * (a fourth-person owner aside). So ranking sides with shallow here, and only deep bi- plus Horseherder's c-command condition predicts ✓.

OPEN No such sentence has been opened (T59b). Fn. 3, the bi- question inducing a yi- answer (5a)/(5b), is suggestive, but there the bi- is on a postposition and the yi- owner is inside a reason clause, so it does not test this.

8. Rules

YI-1  An owner yi- on a noun needs a coreferent 3rd-person antecedent in an A′-position:      SOURCED (Horseherder pp. 31, 41, 44)
      a possessor name in its own phrase, or a discourse topic. (1) vs (2), (3).
YI-2  An owner yi- is disjoint from every c-commanding A-position subject, in its own        SOURCED (pp. 34–39; (4))
      clause and above it ("free everywhere").
YI-3  An owner yi- needs a disjoint 3rd-person A-antecedent, so yi- on the subject's phrase   SOURCED (p. 42, (6b); p. 57)
      fails even with a discourse antecedent.
YI-4  Both antecedents must be 3rd person singular: a 1st/2nd or plural doer blocks yi-     SOURCED (pp. 45–55; (8a), (9a), (10))
      even though it is not the owner. (Her: yi- is inherently 3sg, and dual φ/R indices.)
YI-5  Verbal yi- has no such number limit (dayiiłtsą́ 'they saw him').                     SOURCED (Willie 1991:32 (9), carried over)
YI-6  Ranking (POSS-1–3, GAP-1–2) derives YI-2, YI-3 and the 1st/2nd half of YI-4,         HYPOTHESIS
      but not the plural half.
YI-7  T58a: W&J (70b) and Horseherder (9a) share a plural doer and a ranking-predicts-✓/    HYPOTHESIS (lean)
      speakers-say-* profile. A plural-doer condition covers both; "name inside the doer's
      phrase" covers only (70b).
YI-8  Deep vs shallow bi- (D2) predict opposite judgments for [N_P] [yi-N_A] bi-V (YI-3);   HYPOTHESIS (test); data OPEN
      ranking (one proximate per span) predicts *, like shallow.

9. Derivations, step by step

Labels: Σ role schema, A word, B clause. VPISH throughout: ρA is first-merged in Spec,v and ρP as the complement of V. The possessed noun word (N-W) is built at A with its owner slot π. Noun phrases are Merged at B and linked to slots (nominal adjunction). Horseherder's Spec,IP pro is restated as ρA (first-merged in Spec,v). Her "A′-position" for the possessor is restated as the specifier of the possessed noun phrase at B, or as a Topic position for a discourse antecedent. Each bundle carries person, number and index.

Y-1. (2) Joe yi-łįį́ʼ yizloh 'hef roped Joe'sj horse' ✓

  1. Σ Σ = {ρA: 3sg, index f; ρP: 3, index h (the horse)}. In the possessed noun: π: 3sg, index j.
  2. A N-W (left-headed, prefixal): Merge π with the stem 'horse'. Spell-out of π as yi- is held until its conditions can be checked (YI-1–4 need material outside the word).
  3. A W: Merge V 'rope' with ρP, then v, then ρA in Spec,v. Agree: the object prefix reads ρP (3), the subject marking reads ρA (3) → yizloh.
  4. B Merge Joe in the possessed phrase's specifier (her A′-position), index j, giving [Joe yi-łįį́ʼ]. Link it to ρP and place it left of the verb word (right-headed clause).
  5. B Checks. Agree (feature matching) π ↔ Joe: 3sg = 3sg, same index → YI-1 ✓. π ↔ ρA: 3sg = 3sg, indices j ≠ f, and ρA c-commands ρP's site → YI-3 ✓, YI-2 ✓.
  6. Spell-out: π → yi-; Joe yi-łįį́ʼ yizloh ✓, read 'hef roped Joe'sj horse'. If f = j ('Joe roped his own horse'), YI-2 fails → * (her (4c); some speakers marginal, fn. 2).

Y-2. *(8a) '…*I roped his horse': the doer is 1st person

  1. Σ Σ = {ρA: 1sg; ρP: 3 (horse)}; π: 3sg, index j; discourse Topic: Joe, 3sg, j.
  2. A N-W: Merge π + 'horse'. W: Merge V, ρP, v, ρA. Agree: subject 1sg → séloh.
  3. B π's A′-antecedent = Topic Joe: φ match ✓ (YI-1). π's disjoint A-antecedent = ρA: Agree fails, 1sg ≠ 3sg → YI-4 ✗. Spelling π as yi- crashes: *yiłįį́ʼ séloh.
  4. Ranking route (YI-6): ρA is off the third-person tier; the only third person is Joe/π, so π = [prx] → b-, not yi-. Same result, different reason.

Y-3. *(9a) 'The girls saw Mela … *they roped her horse for her': the doer is plural

  1. Σ Σ = {ρA: 3pl (the girls), index g; ρP: 3 (horse); beneficiary inside yá: 3sg, index m}; π: 3sg, index m; Topic: Mela, 3sg, m.
  2. A N-W: Merge π + 'horse'. W: Merge V 'rope', ρP, v, ρA; Agree: 3/3 → yizloh (no da- printed). The postposition word yá 'for her' is built at A.
  3. B YI-1: π ↔ Topic Mela, 3sg = 3sg ✓. YI-3/4: π ↔ ρA: index m ≠ g ✓, but number 3sg ≠ 3pl → Agree fails → *.
  4. Ranking route: 'the girls' [prx], Mela [obv], π = Mela → [obv] → yi- predicted ✓. Mismatch: only a number-sensitive condition gives *.
  5. Contrast with Y-4: if number were simply copied onto the verb's yi-, dayiiłtsą́ would also be out. It is fine, so the number check is specific to the owner slot π in the noun word (YI-5). HYPOTHESIS: in the verb word, number on the subject is read by a separate probe, da- (number & scope). In the noun word there is no such probe, and π's disjointness check includes number.

Y-4. dayiiłtsą́ 'They saw him' (Willie 1991:32 (9), carried over) ✓

  1. Σ Σ = {ρA: 3pl; ρP: 3}.
  2. A W: Merge V, ρP, v, ρA. Agree: object prefix reads ρP → yi-; da- reads ρA's plural. Spell-out: da-yi-iłtsą́ ✓.
  3. No owner slot, so no YI-4 check. Horseherder's own comparison (pp. 59–61): the verbal yi-'s coreferent antecedent is the object in an A-position. No A′-antecedent is needed.

Y-5. The D2 test, schematic: [NP] [yi-NA] bi-V

  1. Σ Σ = {ρA: 3sg, the doer, owned by a third person k; ρP: 3sg, index p}; π (in the doer's phrase): 3sg, k; Topic for k.
  2. A W: Merge V, ρP, v, ρA. Inverse: Agree spells bi-. Deep: Move ρP to an outer Spec,v above ρA. Shallow: no Move.
  3. B Link [yi-NA] to ρA, and NP to ρP.
  4. YI-3 for π: deep → ρP (3sg, p ≠ k) c-commands ρA's site ✓ → predicted good. Shallow → nothing c-commands ρA's site → predicted * (like (6b)). Ranking → * (§7). OPEN: no data; this relies on assumption A3 (the linked noun phrase counts at its slot's site).

10. Grant's ideas

GRANT'S CONJECTURE Feature bundles (or proto-words) as arguments. Not contradicted, and this is the most bundle-like condition found so far. YI-4 is stated entirely over person and number on two bundles, and one of the two antecedents can be a discourse topic with no noun in the sentence at all ((3b), (8a), (9b)). The condition that blocks (9a) is about a bundle that is not the owner. That is a pure feature-matching (Agree) effect, not a fact about noun content. Horseherder's claim that 'his horse' is in an A-position counts against the strict adjunct view, not against bundles. A noun linked to a slot that already holds a bundle sits at that slot's site (if A3 holds).

GRANT'S CONJECTURE Strata (theta roles first, event structure next, lexical items later). OPEN, consistent. In Y-1 to Y-3, YI-2 to YI-4 are checked on bundles (Σ/A). Only YI-1 needs anything at B, and even that can be met by a discourse topic. Nothing in Horseherder requires the noun 'horse' to be present before the owner's features are fixed.

GRANT'S CONJECTURE Merge/Agree sites inside the word (Merge/Agree in the word). π is a Merge site inside the noun word (A). Its spell-out as yi- waits on Agree with material at B (Y-1 step 2). So the word-level rule set builds the slot, while the clause level supplies the antecedents the slot is checked against. This is a word-internal Agree site whose check crosses the A/B changeover. HYPOTHESIS

HYPOTHESIS Update 2026-10-25 (~12:46pm PT): What comes first decides reads Platero 1978 first-hand. 'The boyi was seen by hisi father' is good with bi-, with the name outside 'his father', and the yi- form cannot mean it (SOURCED). His precede-and-command rule, plus bi- putting the unspoken patient first, predicts W&J (70b)'s star with no plural condition and no Move in his notation. Restated in VPISH, (70b) separates "patient above the doer's phrase" (deep bi- or a null topic) from shallow bi-, so D2 stays OPEN. His (3.18) also clashes with Young & Morgan on yi- (T60b). For this page: the plural-doer lean (YI-7) is not needed for (70b), which has a ba- owner. The plural block on owner yi- ((9a)) stands as its own sourced fact.

11. Bottom line

  • SOURCED Horseherder 1998 (first-hand, page images checked): owner yi- needs a coreferent third-person antecedent (named or topic) and a different, c-commanding third-person doer. It is out on the subject's phrase, out across clauses with the subject, and out when the doer is 'I', 'you' or plural.
  • HYPOTHESIS T58f partly answered: owner yi- is bound from an A′ (topic/possessor) position, in operator-like fashion. But no quantified antecedents are tested, so the T57c quantifier question stays OPEN.
  • HYPOTHESIS Ranking predicts everything except the plural block. Plural doers are where ranking fails in both Horseherder (9a) and W&J (70b): a new T58a lean (plural doer), alongside "name inside the doer's phrase" (T58a partly answered).
  • HYPOTHESIS A new D2 test: yi- on the doer's phrase under bi- (deep predicts ✓, shallow predicts *). Data OPEN.

12. Fidelity notes

  • The OCR text layer is unreliable for Navajo ("yilft'", "yiyiiitsa", "seloh"). All quoted forms were read off the page images. The thesis prints 'horse' as łįį́ʼ, with nasal hooks on both vowels and an acute visible only on the second, and we reproduce that. The bar on ł is not always clear on the scan.
  • Printed variants kept as printed: ya/yaa/yá 'for'; nizin and yiyiiłtsą without acutes in (4) (p. 38) but yiyiiłtsą́ on p. 47; atʼeeké (p. 47).
  • P. 57 (3b) Yichei 'adaadzodli 'Her grandfather is boastful' is printed without an asterisk, but the text says the subject-possessor examples are "ungrammatical in both languages". Not used as a judgment here.
  • Glosses are Leipzig-style restatements; her labels are given in each note. "3POSS(yi)" marks the prefix without committing to [obv].
  • Her trees are GB IP trees (pre-VPISH), labeled and restated in §9, not used as structural evidence.
  • Young & Morgan 1987:9 (her (5) minimal pairs), Saxon 1984a/1986/1995, Aoun 1985 and Huang & Tang 1991 were not opened. The open.library.ubc.ca page asks for browser verification; the PDF was fetched from cIRcle's REST bitstream, which is marked public. No paywall or access restriction was bypassed.

13. Open questions

IDQuestionWhat would settle it
T59aT58a narrowed: plural doer or name inside the doer's phrase? Is 'John'si son saw himi' (singular, bi-, name inside) good?A singular version of W&J (70b), or (9a) with a singular doer and the name inside the doer's phrase. OPEN
T59bD2 test (YI-8): is [NP] [yi-NA] bi-V possible with an owner distinct from both?Any sourced Navajo sentence with yi- on the doer's noun under bi-. OPEN
T59cIs Horseherder's plural block about plural or about dual (no da- in (9a))? Does it hold with da-?(9a) with da- on the verb; Horseherder's consultants or another source. OPEN
T59dDoes the plural block also hold for yi- on postpositions (her Ch. 6 leaves verbs and postpositions open)?Postposition yi- with a plural doer; the instrument-voice lists may contain one. OPEN
T59eYoung & Morgan 1987:9 minimal pairs (her (5), including an inanimate possessum) first-hand.Y&M 1987, not opened. OPEN
T59fDogrib ye- first-hand (Saxon 1986, 1995): no A′-antecedent needed. Does Dogrib show the plural block?Saxon 1986 thesis / 1995; not opened. OPEN

Sources: Horseherder 1998 (UBC M.A. thesis; page images pp. 31–61); carried over: Willie 1991:32 (9), W&J 2000 (70), Aissen 2000. See also possessor-gap, possessor-binding, deep or shallow inverse, number & scope, Hale derivation (A3), open questions, sources.

What comes first decides: Platero 1978 on who "his" can mean, and a no-movement route to W&J (70b) (T59a)

Working note. Kept as written (URL, data and status labels unchanged). Filed under Possessors and object-first orders; the consolidated argument is the essay possessors. Rule IDs: rules index.

Verbs on this page → dictionary: jizloh ‘rope’ · dabiiłtsą́ ‘see’

Restated on the unified structure (2026-11-15). Possessors are restated as the left specifier of the noun phrase, with the possessor prefix as agreement (as on a grammatical postposition). This page’s particular analysis has not yet been restated on the unified structure; light n stays 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. 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, possessors. 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.

The last three runs asked why W&J's 'all John'si children saw himi' is bad under bi- (T58a, T59a). The candidates were the name inside the doer's phrase, 'all', a plural doer and a deep (patient-raising) bi-. This run opened a source cited on this site so far only through Hale 1999: Paul Platero's 1978 MIT dissertation Missing Noun Phrases in Navajo, written by a Navajo-speaking linguist. In plain words, Platero shows three things. First, Navajo decides who "his" or a dropped "him" can mean partly by what comes first: a name cannot be the same person as an unspoken "he/him" that comes before it and outranks it in the sentence. Second, the bi- form puts the unspoken person acted on first. Third, 'The boy was seen by his father' is good with bi- when the boy's name stands on its own, outside 'his father'. Put the first two together and they predict W&J's bad sentence from word order alone. You don't need the plural, and you don't need the patient to move inside the verb, as long as the unspoken patient sits in front of the doer's phrase. 'All' matters only because it forces the name inside the doer's phrase. That narrows T59a, and it moves the deep-inverse question (D2) to a sharper point. One Platero datum also clashes with Young & Morgan as Aissen reports them (§7).

Update 2026-10-25 (~8:46pm PT): Option (iii) vs (ii) in §6 (T60c) is taken up on two ways to put the object first. Saxon 2003 treats only Focus Movement (yi- kept) as A-bar movement. Her bi- landing site, Spec,DiscP, behaves like an A-position, so standard Condition C gives (70b) under option (ii) HYPOTHESIS. There is no binding datum yet (T61a). D2 OPEN.

On this page: 1 Question · 2 Source · 3 The data · 4 His analysis (pre-VPISH) · 5 What this does to T59a / T58a · 6 Restated in VPISH: what (70b) now tests · 7 A clash with Young & Morgan · 8 Rules · 9 Derivations · 10 Grant's ideas · 11 Bottom line · 12 Fidelity notes · 13 Open questions

Notation as on the possessor-yi and possessor-gap pages. Σ = the role schema. A = the word level (prefixal, left-headed; W = the verb word, N-W = a possessed noun word with owner slot π). B = the clause level (right-headed). ρA/ρP = agent and patient slots. Merge vs Agree: Merge builds structure. It puts two pieces together, e.g. the owner slot π into the noun word, or a noun phrase next to the verb word, and Move is Merge again of something already built. Agree values features: it copies or checks person, number and index between bundles. This page is mostly about Merge: where things end up (what precedes and c-commands what) decides coreference. Platero's PRO = an unpronounced noun phrase in an argument slot (his term). Below it is restated as a slot bundle.

1. Question

T59a: does coreference survive in a singular version of W&J (70b) ('John'si son saw himi', bi-, name inside)? This separates the plural doer from 'all' and from the name being inside. T58a, behind it: why is (70b) bad when ranking (Aissen) predicts it good? And for D2 (deep vs shallow bi-): does the answer need the patient to Move?

2. Source

SOURCED Platero, Paul Randolph 1978. Missing Noun Phrases in Navajo. PhD dissertation, MIT (supervisor Kenneth Hale). MIT DSpace handle 1721.1/16328. 174-page scanned PDF with a text layer, fetched from DSpace's public bitstream. PDF page = printed page. The text layer is lossy for Navajo (e.g. "yiyiilts$", "bizh6'6"), so every Navajo form below was read off the page images (130–400 dpi): pp. 34–37, 80–82, 84, 86–87, 120, 127, 135. Platero writes ashkii and atʼééd without the initial glottal stop, and we keep that. "≠" in his examples means "cannot mean". See sources.

Platero is a Navajo speaker. The judgments are his, and some he attributes to earlier work (Hale & Platero 1974; HJP 1976). This site has earlier opened Platero 1973/74 (relative as CP) and cited Platero 1978 via Hale 1999 (conflicting truths).

Carried over, not re-opened this run: W&J 2000 (70), (71) (possessor-gap §5); Aissen 2000 (27), (29), (30), citing Young & Morgan 1980:171–172 (possessor-binding §3); Horseherder 1998 (9a) (possessor-yi §3).

3. The data SOURCED

The first gloss line in each block is Platero's parenthetic gloss as printed. The second is our Leipzig-style line, and it adds nothing his prose does not say: he calls the verb's yi-/bi- "the object prefix" (p. 135), and bi- on a noun its possessor. "3OBJ(yi)" and "3OBJ(bi)" mark which prefix appears without committing to an analysis.

Platero 1978:34, (1.57); 36, (1.62), (1.63); 37, (1.66) (page images checked, 130 and 400 dpi)
(1)Ashkiibizhéʼéyiyiiłtsą́.
boyhis:fathersee
boy3POSS-father3OBJ(yi).3SBJ.PFV.see
(a) 'The boy saw his father.'   (b) ≠ 'The boy's father saw him.'
(2)Ashkiibizhéʼébiiłtsą́.
boyhis:fatherbi:see
boy3POSS-father3OBJ(bi).3SBJ.PFV.see
(a) 'The boy was seen by his father.'   (b) ≠ 'The boy's father was seen by him.'
(3)Atʼééd yahʼííyá dóóashkiibizhéʼéyiyiiłtsą́.
girl enter andboyhis:fathersee
girl 3SBJ.PFV.enter andboy3POSS-father3OBJ(yi).3SBJ.PFV.see
(a) 'The girl entered and saw the boy's father.'   (b) ≠ 'The girl entered and the boy's father saw her.'
(4)Atʼééd yahʼííyá dóóashkiibizhéʼébiiłtsą́.
girl enter andboyhis:fatherbi:see
girl 3SBJ.PFV.enter andboy3POSS-father3OBJ(bi).3SBJ.PFV.see
(a) 'The girl entered and was seen by the boy's father.'   (b) ≠ 'The girl entered and the boy's father was seen by her.'

P. 36: "It is, of course, possible for Navajo to express the (b) meanings cited under (1.57) and (1.62) by resorting to the inverted—i.e., O S bi-V—forms of transitive sentences." So (2a) is how Navajo says 'the boy'si father saw himi'. He parses (2a) as (1.64): ashkii [ashkii bi-zhéʼé] biiłtsą́, with the boy's name standing on its own and the second ashkii (inside 'his father') left unpronounced. He excludes the parse with the name inside 'his father' and an overt patient after it, (1.65) [ashkii bi-zhéʼé] ashkii biiłtsą́ (p. 37). In (2), (4) the hyphenation bi-zhéʼé appears in his trees; the sentences print bizhéʼé.

Platero 1978:80, (3.8); 82, (3.10); 127, (4.23), (4.24) (page images checked; (4.24) at 400 dpi)
(5)PROashkiibizhéʼéyiyiiłtsą́
PROboy3POSS-father3OBJ(yi).3SBJ.PFV.see
'He/shei saw the boyj's father.' PRO cannot be the boy: it "both precedes and commands" ashkii (p. 81).
(6)ashkiiPRObizhéʼéyiyiiłtsą́
boyPRO3POSS-father3OBJ(yi).3SBJ.PFV.see
'The boyi saw hisi father.' Also (4.23) 'The boyi saw his/herj father': the owner may be someone outside the sentence.
(7)Atʼééd yahʼííyáagoashkiibizhéʼéyiyiiłtsą́.
girl enter:COMPboyhis:her:fathersee
girl 3SBJ.PFV.enter-COMPboy3POSS-father3OBJ(yi).3SBJ.PFV.see
(a) 'When the girl entered, the boyi saw hisi father.'   (b) 'When the girli entered, the boy saw heri father.'

(5) and (6) are his phrase-marker strings (trees (3.8), (3.10)) with PRO written in. (3.12) (p. 82) notes that the pronounced string ashkii bizhéʼé yiyiiłtsą́ is ambiguous between them. With the name repeated in place of PRO, (3.8′) ashkii [ashkii bizhéʼé] yiyiiłtsą́ (p. 84) also bars coreference of the two ashkii.

Platero 1978:86, (3.18), (3.19); 87, (3.20) (page images checked; (3.20) at 400 dpi)
(8)[ashkiibizhéʼé]ashkiiyiyiiłtsą́
boyhis:fatherboysee
boy3POSS-fatherboy3OBJ(yi).3SBJ.PFV.see
coreference of the two ashkii "is available" ('the boyi's father saw the boyi')
(9)[ashkiibííghahgi]ashkiitłʼiishyiyiiłtsą́
boybesideboysnakesee
boy3-besideboysnake3OBJ(yi).3SBJ.PFV.see
coreference "is possible" ('beside the boyi, the boyi saw a snake')
(10)Ashkii[ashkii bííghahgi]tłʼiishyiyiiłtsą́.
boyboy 3-besidesnake3OBJ(yi).3SBJ.PFV.see
coreference "is impossible"

English renderings of (8) and (9) are ours. Platero gives only the coreference judgment and word glosses. His glosses: "boy his:father boy see"; "boy beside boy snake see". Brackets in (8), (9) follow his trees (an NP or PP node above the first ashkii), and the brackets in (10) are his.

Platero 1978:120, (4.7); 135, (4.33), (4.34) (page images checked)
(11a)atʼéédashkiibiztsʼǫsʼThe girl was kissed by the boy.ʼ
(11b)PROashkiibiztsʼǫsʼHe/she was kissed by the boy.ʼ
girl / PROboy3OBJ(bi).3SBJ.PFV.kiss(his gloss: "bi:kiss")

(4.34) "Interpretation of Grammatical Relations" (IGR): an NP is Object right before yi-V and Subject right before bi-V. Of two NPs, the first is Subject before "NP yi-V" and Object before "NP bi-V". In plain words: "the NP nearest the verb is object if the object marker is /yi-/, but it is subject if [it is /bi-/]" (p. 135). His "subject"/"object" are the seer/kisser and the seen/kissed: (11a) 'the girl was kissed by the boy'. Revised as (4.37) (p. 137) for strings "(NPb) NPa V". It "applies to overt strings" and is "oblivious to PROs" (pp. 138, 142). PRO gets the "complementary" relation "by some sort of general convention" (p. 143), and its position is deduced: with bi- and one overt NP, PRO comes first, (4.39) PRO atʼééd biztsʼǫs (p. 139).

4. His analysis SOURCED (his), labeled pre-VPISH

  • Every slot is filled (pp. 116, 122): the phrase marker "matches the subcategorizational properties of the nucleus". A missing noun phrase is an NP over PRO, in verbs, in possessed nouns (bi-zhéʼé "is said to require an NP argument functioning as possessor", p. 120) and in postpositions. He finds "no particular evidence to assume that there is a verb phrase" (p. 117).
  • The Non-Coreference Rule (3.15), after Lasnik 1976: "If NP1 precedes and commands NP2, and NP2 is not a pronoun [or PRO, in the Navajo case], NP1 and NP2 are non-coreferential." Coreference is otherwise free, limited only by things like number consistency ((3.22), p. 88). There is no rule assigning coreference (§3.3).
  • Command: NP and PP nodes count as boundaries ((3.18), (3.19)), so he adopts Reinhart's c-command (3.21) (p. 87).
  • The IGR (4.34)/(4.37): grammatical relations are read off the overt string by order and the yi-/bi- prefix. It replaces his earlier NP-PRO Constraint (4.1) and the Second Noun Phrase Constraint (1.56) (pp. 139–142). It alone excludes the parse (4.17), with a gap right before the verb, that would give (1b).

HYPOTHESIS Framework label. His trees are flat S trees (S → NP NP V, no VP). They are pre-VPISH notation in the sense of the framework audit, and they are used here for his judgments and for the shape of his rules, not as evidence for a flat clause. §6 and §9 restate them in the vP/TP spine. One thing survives the restatement unchanged: the Non-Coreference Rule is stated over precedence plus c-command, and both still exist in a vP/TP tree.

5. What this does to T59a / T58a HYPOTHESIS

Run W&J (70b) through Platero's two rules. The string is tʼáá ʼałtso Jáan ba-ʼáłchíní dabiiłtsą́ ('just all John's children', PL-3ACC-3NOM-saw; W&J's glosses, carried over).

  1. 'All' modifies 'children', so [tʼáá ʼałtso Jáan ba-ʼáłchíní] is one noun phrase. A parse with Jáan on its own, like (1.64), would split 'all' from 'children'. HYPOTHESIS (constituency inferred from W&J's bracket and translation).
  2. IGR: one overt NP right before a bi-V is the Subject (the seer). So 'all John's children' are the seers, and the seen is PRO.
  3. Position: with bi- and one overt NP, PRO comes first (as in (11b) and his (4.39)). Flat S: PRO precedes the subject phrase and c-commands everything in it, Jáan included.
  4. Non-Coreference Rule: PRO precedes and commands Jáan, and Jáan is a name, not a pronoun → PRO ≠ John. Only (70a) 'saw himk' survives. That is W&J's judgment.

HYPOTHESIS Result. Platero's rules predict W&J's star with no number condition, no ranking failure and no Move inside the verb word. It is exactly his (5)/(3.8) configuration, with bi- putting the unspoken patient first. 'All' plays a role only as a bracket-forcer: it removes the (1.64) parse, under which the name stands outside and coreference is fine, as in (2a).

  • HYPOTHESIS T59a partly answered. The singular string 'the boy his-father bi-saw' is good with coreference ((2a), SOURCED). But Platero assigns it the name-outside parse (1.64), so it does not show that coreference survives with the name inside. His rules predict that the name-inside reading of the same string is out (the same steps 2–4 as above). The predicted reading is 'hej was seen by the boyi's father', j ≠ i. He prints no judgment for it, so this remains a prediction. A string judgment alone cannot separate the parses. The test needs a singular possessed phrase that is forced to be one constituent (T60a).
  • HYPOTHESIS The plural-doer lean (YI-7) is not needed for (70b). Platero derives the star with a singular or plural doer alike. The plural block remains a separate, sourced fact about owner yi- (Horseherder (9a)), where no name is inside the doer's phrase. (70b) has a ba-/bi- owner, not yi-. So the two facts may have two causes, and the T58a confound list shrinks to three: name inside (now with a mechanism), 'all' (now explained as the bracket-forcer) and source.
  • SOURCED Platero's 1978 judgments for (1), (2) independently confirm Young & Morgan 1980 (as reported by Aissen 2000): with a coreferent owner, 'X's father saw X' needs bi-, not yi-.

6. Restated in VPISH: what (70b) now tests HYPOTHESIS

In the vP/TP spine, ρA is first-merged in Spec,v and ρP as the complement of V, so in the base ρA c-commands ρP. Platero's flat-S "PRO first" has to be translated. Step 3 needs the null patient to sit somewhere that precedes and c-commands the doer's phrase. There are three ways to get that, and each has a source or a page behind it:

OptionWhere the null patient is(70b) coreferenceBehind it
(i) shallow bi-, no topicρP stays below ρA✓ predicted (nothing c-commands Jáan) ✗ wrongdeep or shallow
(ii) deep bi-ρP Moves to an outer Spec,v above ρA (A-movement)* (Condition C / Non-Coreference) ✓Oxford 2023, via GAP-6
(iii) null topic at Ba B-level Topic linked to ρP, left of the doer's phrase (A′-position)* ✓W&J's own tree: "3ACC" as Topic, the possessive phrase as Focus (possessor-gap §5)

HYPOTHESIS So (70b) does not decide between deep bi- and shallow bi-. It decides between "the patient ends up in front of and above the doer's phrase" ((ii) or (iii)) and "it doesn't" ((i)). That weakens possessor-gap's lean that "deep bi- + Condition C" is specially needed (GAP-6 already listed W&J's adjunct condition as a rival). Platero's rule is a third rival, with no Move at all in his notation. D2 stays OPEN. What would separate (ii) from (iii) is a test that tells an A-position from an A′-position, for example whether the fronted patient can bind a reciprocal in the doer's phrase, or shows weak crossover (T60c).

HYPOTHESIS Platero's PP and NP boundaries ((8), (9) vs (10)) fit a right-headed clause level B where each noun phrase is its own c-command domain. The owner's name inside [ashkii bizhéʼé] commands nothing outside its phrase ((8) ✓). Only a whole phrase or a slot placed to the left and above commands. That is ruleset B behavior. It holds whatever goes on inside the noun word at A.

7. A clash with Young & Morgan (as Aissen reports them)

SentenceVerbOwner inside the doer's phraseCoreferent with the patientJudgment
(8) Platero (3.18) [ashkii bizhéʼé] ashkii yiyiiłtsą́yi-the name ashkiiyes (patient = the name repeated)✓ "available"
Y&M 1980:171 via Aissen 2000 (29) *Bi-lį́į́ ashkii yi-yiiłtsą́yi-a null owner (bi-)yes ('his horse saw the boy')*

SOURCED Both are printed judgments on the same shape: [owner's X]doer + patient + yi-V, with the owner coreferent with the patient. Platero says yes; Young & Morgan (via Aissen) say no.

HYPOTHESIS Ranking (Aissen's co-linking: coreferents share [prx]/[obv]) predicts * for both: the boy as owner is [prx], so as patient he is [prx], which needs bi-. So (8) is contradicting data for strict ranking, if the judgment holds. Differences that could matter: (8) repeats the full name where Y&M has a null owner; (8) was offered as a structural test of command, not of voice; the sources and speakers differ (and Willie 1991:217, carried over, says speakers vary in using yi-/bi- for obviation). OPEN (T60b). Platero's (6)/(4.23) also lets a bi- owner be someone other than a third-person doer, where Horseherder's speakers use owner yi- (her (2) Joe yi-łįį́ʼ yizloh). That could be the same speaker variation (T60d).

8. Rules

PC-1  Non-Coreference: if NP1 precedes and c-commands NP2 and NP2 is not a pronoun/PRO,      SOURCED (Platero (3.15), after Lasnik 1976;
      they are non-coreferential. Otherwise coreference is free (number permitting).               (3.8), (3.8′), (3.22))
PC-2  NP and PP nodes bound command: a name inside a possessive phrase or PP commands        SOURCED (Platero (3.18)–(3.21))
      nothing outside it.
PC-3  IGR: the NP nearest yi-V is object, nearest bi-V is subject (seer); of two NPs the     SOURCED (Platero (4.34), (4.37))
      first is subject with yi-V, object with bi-V. Reads overt strings only; PRO gets the
      complementary relation, and its position is deduced (bi- + one NP → PRO first).
PC-4  'X's father saw X' (coreferent) is bi- (2a), name outside 'his father'; the yi- form   SOURCED (Platero (1.57), (1.63), (1.64))
      cannot mean it (1b).
PC-5  PC-1 + PC-3 predict W&J (70b)'s star: bi- + one overt NP puts the null patient      HYPOTHESIS (derivation P-3)
      first, above the name in the doer's phrase. No number condition, no Move in the flat
      notation; 'all' only forces the name-inside bracket.
PC-6  In VPISH, PC-5 needs the null patient above and before the doer's phrase: deep bi-   HYPOTHESIS; D2 OPEN
      (A-Move) or a B-level Topic (A′). Shallow bi- with no topic predicts ✓ (wrong).
PC-7  Platero (3.18) yi- with a coreferent repeated name inside the doer's phrase ✓ vs      SOURCED (both judgments); conflict OPEN
      Y&M (29) *: contradicting data for strict ranking, if it holds.
PC-8  T59a: the singular bi- string is good with coreference only on the name-outside parse  HYPOTHESIS (partly answered)
      (2a); the name-inside reading is predicted out but unprinted.

9. Derivations, step by step

Labels: Σ role schema, A word, B clause. VPISH throughout. Platero's PRO is restated as a slot bundle (person, number, index). An overt noun phrase is Merged at B and linked to a slot (nominal adjunction); where his account needs the noun in the slot, that is noted. Precedence is read at B (right-headed: phrases left of the verb word).

P-1. (2a) Ashkii bizhéʼé biiłtsą́ 'The boyi was seen by hisi father' ✓

  1. Σ Σ = {ρA: 3sg, index f (the father); ρP: 3sg, index b (the boy)}. In the possessed noun: π: 3sg, index b.
  2. A N-W (left-headed): Merge π with the stem 'father' → bi-zhéʼé. Agree: π 3sg is spelled bi-.
  3. A W: Merge V 'see' + ρP, then v, then ρA in Spec,v. Agree: the object prefix is spelled bi- (inverse: the patient outranks; POSS-1) → biiłtsą́. Whether ρP also Moves (deep) does not matter here.
  4. B Merge ashkii (linked to ρP) leftmost, then [bizhéʼé] (linked to ρA), then the verb word (right-headed). PC-3: of two NPs before bi-V, the first is object → ashkii = seen, 'his father' = seer ✓.
  5. B PC-1: ashkii precedes π, but π is a pronoun bundle → not blocked; π = b allowed. Spell-out ✓ (2a).

P-2. (1b) *Ashkii bizhéʼé yiyiiłtsą́ as 'the boy'si father saw himi'

  1. Σ as in P-1. A W with yi- (direct) → yiyiiłtsą́.
  2. B To mean 'the boy's father saw him', [ashkii bizhéʼé] must be linked to ρA with a null patient. PC-3: one overt NP before yi-V is object, so [ashkii bizhéʼé] is read as the seen → reading lost. (Platero: this also bars the gap right before the verb, (4.17).)
  3. Also out under ranking (Aissen): the coreferent owner and patient are [prx], so the verb needs bi-. Two independent reasons, one judgment.

P-3. W&J (70b) *'all John'si children saw himi': three restatements

  1. Σ Σ = {ρA: 3pl, index c; ρP: 3sg, index j}; π (in the doer's phrase): 3sg, j.
  2. A N-W: Merge π + 'children' → ba-ʼáłchíní. W: Merge V, ρP, v, ρA; Agree: bi-, and da- counts the agent → dabiiłtsą́.
  3. B Merge Jáan as owner, then 'all' over the whole: [tʼáá ʼałtso Jáan ba-ʼáłchíní], one phrase, linked to ρA (PC-3: the one NP before bi-V is subject).
  4. (i) shallow, no topic: ρA's site c-commands ρP, and nothing c-commands Jáan → PC-1 does not apply → ✓ predicted. Mismatch.
  5. (ii) deep: A Move ρP to an outer Spec,v. At B, ρP (index j) precedes and c-commands the doer's phrase (if the phrase counts at ρA's site, assumption A3 on the Hale derivation page) → PC-1 bars j = John → *. ✓
  6. (iii) topic: B Merge a null Topic linked to ρP at the left edge, above the Focus phrase. It precedes and c-commands Jáan → PC-1 → *. ✓ No Move inside W.
  7. Platero's flat notation is (iii)-like: PRO simply comes first. Ranking alone: ✓ (wrong), as on possessor-gap G-2.

P-4. (8) [ashkii bizhéʼé] ashkii yiyiiłtsą́, coreferent ✓ (Platero)

  1. Σ Σ = {ρA: 3sg, f; ρP: 3sg, b}; π: 3sg, b.
  2. A N-W: π + 'father' → bizhéʼé. W: Merge V, ρP, v, ρA; Agree spells the object yi- → yiyiiłtsą́.
  3. B Merge [ashkii bizhéʼé] (→ ρA) and ashkii (→ ρP), in that order. PC-3: two NPs + yi-V → first subject, second object ✓.
  4. B PC-2: the first ashkii is inside an NP, so it does not c-command the second → PC-1 silent → coreference available (Platero ✓).
  5. Ranking: owner = patient → both [prx] → object must be bi- → * predicted. Clash (§7). OPEN

10. Grant's ideas

GRANT'S CONJECTURE Feature bundles (or proto-words) as arguments. Not contradicted. Platero's system is close in one respect: every argument slot exists whether or not a noun is pronounced, PRO "matches the subcategorizational properties of the nucleus" (p. 122), and PRO has person and number but no noun content (its number is read off the verb, (3.22)). The overt nouns are then matched to slots by an interpretive rule over the string (the IGR, "oblivious to PROs"). In this site's terms that is slot bundles first and nouns linked later. But Platero himself puts the nouns in the argument positions (lexical insertion, p. 116), so his work is consistent with bundles, not evidence for them. No datum here contradicts the idea.

GRANT'S CONJECTURE Strata (theta roles first, event structure next, lexical items later). OPEN, consistent. In P-1 to P-4 the slots and their prefixes are fixed at Σ/A. PC-1 needs to know whether NP2 is a name or a pronoun, so it can only apply once nouns are in (B). That is the same ordering the possessor-gap page found for Condition C.

GRANT'S CONJECTURE Merge/Agree sites inside the word. Platero's possessed noun "requires an NP argument functioning as possessor" (p. 120). That is a slot built inside the noun word, the owner slot π, a word-internal Merge site. PC-2 shows that what fills π, or the name next to it, commands nothing outside its phrase. So the noun word and its phrase are a closed domain for coreference, and the clause level sees only the phrase as a whole. HYPOTHESIS: the A/B changeover is also a command boundary.

HYPOTHESIS Theta roles stay put. In (1)/(2), (3)/(4) and (11) the seer and the seen keep their roles. Only the prefix and the order change, and with them who comes first and so who can corefer with whom.

Chemistry: SPECULATIVE, flag only. "What comes first and above blocks sameness" is an ordering-plus-domain constraint. No chemical correlate is identified or claimed.

11. Bottom line

  • SOURCED Platero 1978 (first-hand, page images checked): 'the boyi's father saw himi' is said with bi-, (2a) Ashkii bizhéʼé biiłtsą́, and the yi- form cannot mean it, (1b). Navajo coreference obeys a precede-and-c-command rule, with NP and PP as boundaries ((5), (8)–(10)). Bi- with one overt NP puts the unspoken patient first ((11b), (4.39)).
  • HYPOTHESIS These rules predict W&J (70b)'s star with no plural condition and no Move in his notation. 'All' is a bracket-forcer. T59a partly answered: singular + coreference is fine only on the name-outside parse.
  • HYPOTHESIS In VPISH, (70b) separates "patient ends up above and before the doer's phrase" (deep bi- or a null topic) from shallow bi- with no topic. It does not separate deep from topic. D2 OPEN.
  • SOURCED + OPEN Platero (3.18) vs Young & Morgan (29) conflict on a coreferent owner inside the doer's phrase under yi-.

12. Fidelity notes

  • The text layer is unreliable ("yiyiilts$", "bizh6'6", "biighahgi", "yah'iiyAago"). Every quoted form was read off page images: yiyiiłtsą́, biiłtsą́, bizhéʼé, atʼééd, yahʼííyá, yahʼííyáago, bííghahgi (both vowels with acutes, checked at 400 dpi), tłʼiish, biztsʼǫs.
  • Kept as printed: no initial glottal stop on ashkii/atʼééd; capitalization only sentence-initially; Platero's glosses in the first gloss line.
  • (5), (6) are his phrase-marker strings with PRO, not pronounced sentences. Their pronounced form is (1). English renderings of (8), (9) are ours; he gives coreference judgments only.
  • The prediction for the name-inside reading of (2) (§5) is ours, from his rules. He prints no judgment for it.
  • His flat S trees are pre-VPISH. They are labeled as such and restated in §6, §9, and are used for judgments only.
  • W&J (70), Aissen 2000 / Y&M 1980, and Horseherder (9a) are carried over from earlier pages, not re-opened.
  • Read but not used: p. 133 (4.32a) and Platero's remark that HJP 1976:403 wrongly called such relatives ambiguous. Only the text layer was read there, and the forms are not checked.
  • The PDF came from MIT DSpace's public REST bitstream (the plain download link returned an empty file). No access restriction was bypassed.

13. Open questions

IDQuestionWhat would settle it
T60aA bracket-forcing singular test: under bi-, is coreference out when the owner's name is forced inside a singular doer's phrase (PC-8)?Any sourced bi- sentence where [name bi-X] must be one constituent (a modifier, a relative, a clitic on the phrase). OPEN
T60bPlatero (3.18) ✓ vs Y&M (29) *: repeated name vs null owner, or speakers?Y&M 1980:171 first-hand; Platero 1982; HJP 1977; any source with [N bi-X] N yi-V. OPEN
T60cIs the first-position null patient under bi- in an A-position (deep) or a topic (A′)? Can it bind a reciprocal in the doer's phrase, or show weak crossover?A sourced bi- sentence with a reciprocal or a bound pronoun in the doer's phrase. OPEN
T60dOwner bi- disjoint from a third-person doer (Platero (4.23), (4.24b)) vs Horseherder's owner yi- in the same spot: speaker variation (Willie 1991:217)?One speaker's judgments on both; Horseherder's consultants by age. OPEN
T60eDo PC-2's PP boundaries hold under bi-? (Schematic only, no form claimed: a PP containing the patient's name, before the doer, with bi-V.)A sourced bi- analogue of (9)/(10). OPEN
T60fHJP 1976:403 vs Platero p. 133 on relatives on inverted equal-rank clauses: check the page image and HJP.Platero p. 133 image; HJP 1976. OPEN

Sources: Platero 1978 (MIT diss.; page images pp. 34–37, 80–87, 120, 127, 135); carried over: W&J 2000 (70), Aissen 2000 / Young & Morgan 1980, Horseherder 1998, Willie 1991:217. See also possessor-yi, possessor-gap, possessor-binding, deep or shallow inverse, conflicting truths, open questions, sources.

Whose neck? Possessor raising: the noun goes into the verb, the person becomes the object

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 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.

Working note. Kept as written (URL, data and status labels unchanged). Filed under Possessors and object-first orders; the consolidated argument is the essay possessors. Rule IDs: rules index.

Verbs on this page → dictionary: yiztał ‘kick’ · yiztsʼǫs ‘kiss’ · dabiiłtsą́ ‘see’ · yáłtiʼ ‘speak, talk (to)’ · yi-kʼos:kʼíiníłneʼ ‘cut off (someone's) head; “neck:sever”’ · náshidííltsʼin ‘punch’ · shinákʼeedzííníltsʼin ‘punch (someone) in the eye; “eye-punch”’ · yí-laʼ:yiztsʼǫs ‘kiss (someone's) hand’ · yitsą́:dzíítáál ‘kick (someone) in the stomach; “belly-kick”’ · ba-ghandi:níyá ‘go to (someone's) home, visit; “home:went”’ · nineez ‘be long’ · kééhatʼíí ‘live, reside (plural)’

Restated on the unified structure (2026-11-15). Possessors are restated as the left specifier of the noun phrase, with the possessor prefix as agreement (as on a grammatical postposition). This page’s particular analysis has not yet been restated on the unified structure; light n stays 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. 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, possessors. 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 7, “Possessor Raising” (pp. 182–198), which this site had not opened before. In these sentences a body-part noun (‘neck’, ‘eye’, ‘hand’, ‘stomach’) goes inside the verb word, and the body part's owner is marked as the verb's object. Willie's key statement fits Grant's second phenomenon closely: the incorporated body part “is not an argument of the transitive verb, but serves to derive a complex verb” (p. 187). The verb still has two argument slots. The noun fills neither. The person who is “affected” (p. 197) fills the object slot. Read with this site's rules, the chapter gives four results. (1) Incorporating a noun need not use up an argument slot. Here it changes the verb and leaves the slot count at two (§5). (2) The same string can be one word or two, with different meanings, and nothing in the sound tells them apart (11), pp. 191–192. So the changeover from rule set A to rule set B is a structural boundary, not a phonological one (§6). (3) The yi-/bi- choice on a possessor obeys the same “preceding nominal” rule as on a verb (5), p. 185. Applied to Ch. 8 (28), that rule answers part of T11d (§7). (4) Willie's own explanation of (10)–(11) makes the non-argument ‘neck’ count on the animacy scale (p. 190). That pulls against a ranking computed over slots alone (§8, T16c).

On this page: 1 Question · 2 Source · 3 Possessives · 4 Possessor raising · 5 A slot, not a noun · 6 One word or two · 7 T11d · 8 Animacy · 9 Rules · 10 Derivations Q1–Q6 · 11 VPISH · 12 Open

1. The question

This site models a Navajo verb's arguments as role slots, Σ(√) = {ρA, ρP, …} (proto-role schema), filled by pronominal prefixes, with nouns only linked to those slots at clause level (rule set B). Possessor raising tests that in two ways:

  • Theta roles as arguments (phenomenon 2). When a noun is put inside the verb, does it fill a slot, as a lexical argument would? Or does the slot stay and get a different filler?
  • Rule set A vs rule set B (phenomenon 1). The same body-part noun can sit inside the verb word (A) or head a noun phrase outside it (B). What decides which, and can a hearer tell?

2. Source opened this run (2026-10-11, ~4:50am PT)

  • Willie, MaryAnn 1991. Navajo Pronouns and Obviation. PhD dissertation, University of Arizona (the scan on the box). Fresh 150 dpi page images of all of Chapter 7, printed pp. 182–198 (PDF 188–204; PDF = printed + 6). Every Navajo line of (1), (3)–(25) was checked on 300 dpi crops. The (2) paradigm was read at 150 dpi, with (2g–i) checked at 300. No OCR used for any form.

Transcription and glosses. Navajo as printed, with the glottal stop as ’ and the nasal hook as ą/į. Willie's gloss lines are copied exactly, misprints included and marked [sic]. Her hyphens, colons and spaces inside the verb are kept as printed. The colon marks her analysis of a body part incorporated into the verb (“neck:sever”). Her gloss order for transitive verbs is object.subject. Where she gives no gloss line ((4), (10), (11), (18)), we add none.

Cited by Willie in this chapter, not opened: Baker 1988 (“classificatory” incorporation; p. 187 says “Baker's framework”); Mithun 1986; Saxon 1989 and Rice 1989 (argument incorporation in other Athabaskan languages, pp. 196–197); Young & Morgan 1987 (the source of (21)–(22)); the Muskogean comparison on p. 186 has no citation.

3. Possessives and the yi-/bi- alternation inside a noun phrase (§7.0–7.1, pp. 182–186) SOURCED

P. 182: in possessor raising “a possessor argument is "raised" or "advanced" to the status of a direct object. This relignment of theta roles and grammatical relations is very much like what is called "Dative Movement" across languages”. In Navajo, “postpositional goal arguments come under the scope of the Direct/Inverse voice alternation, where they are treated like direct objects”:

Willie 1991:182–183, (1)
(1a)ʼashkiiyichʼįʼyáłtiʼ
boy3.to3.speak
‘He is speaking to the boy.’
(1b)ʼashkiibichʼįʼyáłtiʼ
boyINV3.to3.speak
‘The boy is speaking to him.’
Willie 1991:183–184, (2)–(3)
(2a)shi-má1s-mother(2f)da-nihi-máPL-1-mother
(2b)ni-má2s-mother(2g)da-nihi-máPL-2-mother
(2c)bi-má3s-mother(2h)da-bi-máPL-3-mother
(2d)nihi-má1-dual-mother(2i)ha-má4s-mother
(2e)nihi-má2-dual-mother(2j)da-ha-máPL-4-mother
(3)Da-nihi-mádookóókééhatʼíída.
PL-1-motherNEGherePL-3-liveNEG
‘Our mother does not live here.’ [*Our mothers do not live here.]

P. 184: “The plural marker takes scope only over the possessor.” Then: “We saw in Chapter 3 that the distributive da- when it appears on the verb has scope over a verbal argument, which may of course be a complex Possessor NP.”

HYPOTHESIS da- picks a slot, not a noun, inside the noun word too. On the verb, da- follows the agent slot (number & scope). Inside the possessed noun, (3) shows it counting the possessor, not ‘mother’. On this site's terms the possessor is the noun's one argument slot, and the noun stem is not a slot. So in both domains da- counts an argument slot, never the lexical stem it attaches to. Willie states the scope fact. The unification is ours.

Willie 1991:185, (4)–(5)
(4)ʼatʼéédbaghan
‘the girl's home’
(5a)ʼashkiiʼatʼéédyighan=góónaayá
boygirl3POSS-home=to†3-go
‘The boy went to the girl's house.’
(5b)ʼashkiiʼatʼéédbaghan=góónaayá
boygirlINV 3POSS-home=to3-go
‘The girl went to the boy's house.’

SOURCED † Segmentation note (2026-11-11 audit pass 4): Willie 1991 (5a), p. 186 prints yighan=góó with the gloss “3POSS-home=to”: the gloss names more morphemes than the form shows as separate pieces (fused or unwritten morphemes), so the two lines do not match one-to-one in the source. Kept exactly as printed.

P. 185: “the third person possessive pronoun is bi-. In Possessor Raising, we see yi- third person possessor pronouns also, and we see an alternation between these pronouns that is an extension of the yi-/bi- alternation with pronominal arguments of the verb or postposition.” Pp. 185–186: “In Ex. (5b), the bi- pronouns signals that the preceding nominal is not coindexed with it, just as a bi- pronoun on a verb signals that the preceding nominal is not coindexed with the object of the verb.” P. 186: in Navajo, as in the Muskogean languages, “there is homophony between third person Possessor and object forms.”

4. Possessor raising (§7.2–7.5, pp. 186–196) SOURCED

P. 186: “The possessor pronoun that is raised to direct object status is marked in an object prefix on the verb, while the element that refers to the body part is incorporated into a complex derived verb, and loses its status as the head of a noun phrase.”

4.1 ‘You punched me’ vs ‘you eye-punched me’

Willie 1991:187, (6)
(6a)shinááʼ
1.eye
‘my eye’
(6b)náshidííltsʼin
1.2.punched
‘You punched me.’
(6c)shinákʼeedzííníltsʼin
1.2.eye-punched
‘You punched me in the eye.’ (You "eye-punched" me.)

P. 187: “In (6c) the body part term has been incorporated into the verb, and the first person pronoun is not a possessor, but a direct object. The incorporated body part term is not an argument of the transitive verb, but serves to derive a complex verb, to "eye-punch"; in Baker's framework, this is 'classificatory' incorporation, of the kind seen in the Algonquian languages. The first and second person subject and object prefixes on the verb in (6) are the verbal arguments.”

We do not segment (6c). Willie's gloss is whole-word, and the string shinákʼee- is not the same as (6a) shinááʼ. How ‘eye’ is spelled inside the verb is not stated (T16g).

4.2 The chicken, the woman and the neck: direct, inverse, and an ambiguity

Willie 1991:188–192, (7)–(11′)
(7)naaʼahóóhaibikʼosnineez
[chicken3-neck]3.long
‘The chicken's neck is long.’
(8)ʼasdzą́ą́naaʼahóóhaiyi-kʼos:kʼíiníłneʼ
[woman][chicken]3.3.neck:sever†
‘The woman cut the chicken's head off.’
(9)ʼasdzą́ą́naaʼahóóhaibi-kʼos:kʼíiníłneʼ
[woman][chicken]3.3.neck:sever†
‘The chicken cut the woman's head off.’
(10)naaʼahóóhaiʼasdzą́ą́yi-kʼos kʼíiníłneʼ
‘The chicken cut off the woman's head.’
(11)naaʼahóóhaiʼasdzą́ą́bi-kʼos kʼíiníłneʼ
a. ‘The woman cut the chicken's head off.’ b. ‘The chicken cut the woman's head off.’
(10′)naaʼahóóhaiʼasdzą́ą́yikʼos:kʼíiníłneʼ
[chicken][woman]3.3.neck:sever
‘The chicken cut off the woman's head.’ DIRECT
(11.a′)naaʼahóóhaiʼasdzą́ą́bikʼos:kʼíiníłneʼ
[chicken][woman]3.3.neck:sever
‘The woman cut off the chicken's head.’ INVERSE
(11.b′)naaʼahóóhaiʼasdzą́ą́ bikʼoskʼíiníłne [sic]
[chicken][woman 3POSS-neck]3.3.sever
‘The chicken cut off the woman's head.’ DIRECT

SOURCED † Segmentation note (2026-11-11 audit pass 4): Willie 1991 (8), p. 189 prints yi-kʼos:kʼíiníłneʼ with the gloss “3.3.neck:sever”: the gloss joins with periods or colons pieces the form separates with hyphens, so the two lines do not match one-to-one in the source. Kept exactly as printed.

SOURCED † Segmentation note (2026-11-11 audit pass 4): Willie 1991 (9), p. 189 prints bi-kʼos:kʼíiníłneʼ with the gloss “3.3.neck:sever”: the gloss joins with periods or colons pieces the form separates with hyphens, so the two lines do not match one-to-one in the source. Kept exactly as printed.

P. 189: “The two third person arguments of the verb refer to the woman and the chicken; neither one refers to the body part. In (8), the Direct construction, the noun immediately preceding the complex verb is coindexed with the patient argument. In the corresponding Inverse construction, (9), 'the chicken' is interpreted as the agent.” P. 190: “(10) has only one reading, (11) is ambiguous.” P. 191: “in (11.b.) … The possessed noun bik’os is not incorporated, and there is no Possessor Raising; the complex NP is the transitive object.” Pp. 191–192: “These two constructions (11.a′) and (11.b′) have the same surface structure, since the incorporation produces no phonological effects. They differ in that in (11.a′) the bi- prefix marks the construction as Inverse, and in (11.b′) the bi- prefix marks third person possessor in a complex NP.”

4.3 Hand-kissing and belly-kicking; the alienable *goat

Willie 1991:192–194, (12)–(18)
(12)Johnʼatʼéédyí-laʼ:yiztsʼǫs
[girl]3.3.hand:kissed†
‘John kissed the girl's hand.’
(13)Johnʼatʼéédbí-laʼ:yiztsʼǫs
[girl]INV 3.3.hand-kissed
‘The girl kissed John's hand.’
(14)hastiintłʼízíyitsą́:dzíítáál
[man][goat]3.3.stomach:kicked
‘The man kicked the goat in the stomach.’ DIRECT
(15)hastiintłʼízíbitsą́:dzíítáál
[man][goat]3.3.stomach:kicked [sic]
‘The goat kicked the man in the stomach.’ INVERSE
(16)hastiinJohn bitłʼízíyitsą́:dzíítáál
[man][John 3POSS-goat]3.3.belly:kicked
‘The man belly-kicked John's goat.’ DIRECT
(17)hastiinJohn bitłʼízíbitsą́:dzíítáál
[man][John 3POSS-goat]INV3.3.belly:kicked
‘John's goat belly-kicked the man.’ INVERSE
(18)*hastiinJohn yitłʼízíyitsą́:dzíítáál

SOURCED † Segmentation note (2026-11-11 audit pass 4): Willie 1991 (12) prints yí-laʼ:yiztsʼǫs with the gloss “3.3.hand:kissed”: the gloss joins with periods or colons pieces the form separates with hyphens, so the two lines do not match one-to-one in the source. Kept exactly as printed.

P. 193 (§7.4): “In my speech, this process applies primarily to body part terms, a few kin terms (bimá "mother", bimá sání "grandmother", and bighan, "home". It cannot be used for things that are alienably possessed.” P. 194: “Ex. (18) shows that the noun tłʼízí cannot subjected to Possessor Rasing, since it cannot be inalienably possessed.”

4.4 ‘Home’ and kin terms; speakers vary

Willie 1991:194–195, (19)–(22)
(19)shádíshimá sáníba-ghandi:níyá
[1.sister][1.grandmother]3.3.home:went [sic?]
‘My grandmother went to my elder sister's home.’ (My grandmother visited my elder sister.)
(20)shádíshimá sáníyi-ghandi:níyá
[1.sister][1.grandmother]INV3.3.home:went [sic?]
‘My sister went to my grandmother's home.’ (My elder sister visited my grandmother.)
(21)shínaaíshił naaʼaashbi-másání:yiyiiłtsą́
[my o.brother][my cousin]INV 3.3.grandmother:saw
‘My cousin saw my older brother's grandmother.’
(22)shínaaíshił naaʼaashyi-másání:yiyiiłtsą́
[my brother][my cousin]3-3-grandmother:saw†
‘My brother saw my cousin's grandmother.’

SOURCED † Segmentation note (2026-11-11 audit pass 4): Willie 1991 (22) prints yi-másání:yiyiiłtsą́ with the gloss “3-3-grandmother:saw”: the gloss names more morphemes than the form shows as separate pieces (fused or unwritten morphemes), so the two lines do not match one-to-one in the source. Kept exactly as printed.

P. 194: possessor raising with ghan “home” shows “the close connection between this inalienably possessed item and the possessor -- almost like a body part.” P. 195: (21)–(22) are “Examples showing Possessor Raising with kin terms that appear in Young and Morgan (1987)”. “Some Navajo speakers do not used Possessor Raising in this way, and do not permit yi- on kin terms.”

Gloss check, not a correction. In (19) the b-form is glossed plain “3.3” and in (20) the y-form is glossed “INV3.3”. Everywhere else in the chapter it is the other way round ((8)/(9), (12)/(13), (16)/(17), (21)/(22)). The translations of (19)–(20) fit the usual pattern: b- means the first noun is not the agent, y- means it is. So we mark both glosses [sic?] and change nothing (T16d).

4.5 Body part in a postpositional phrase: the alternation stays off the verb

Willie 1991:195–196, (23)–(25)
(23)hastiinłį́į́ʼbitsą́=góneʼyiztał
[man][horse]3.3.belly=in:kicked
‘The horse kicked the man in the stomach.’
(24)??? hastiinłį́į́ʼbitsą́=góneʼbiztał
[man][horse][3POSS.belly=in]INV 3.3.kicked
(25)* hastiinłį́į́ʼyitsą́=goneʼbiztał
[man][horse 3POSS.belly=in]INV3.3.kicked

Pp. 195–196 (§7.5): “In a construction where a lexical postpositional phrase is employed as a paraphrase, the yi-/bi- alternation can extend to the lexical postposition; and when this occurs, the alternation cannot appear on the verb itself.” P. 196: “In Possessor Raising it is not the element that refers to the body that is the object; it is the Possessor that is "raised" to argumental status”.

(23)'s translation has the horse as kicker, although the verb is yiztał and ‘man’ comes first. With no postpositional phrase, hastiin łį́į́ʼ yiztał would ordinarily mean the man kicked the horse (inverse §2.1). Willie does not comment. We copy it as printed and leave it open (T16e).

4.6 Willie's conclusion (§7.6, pp. 196–198)

Pp. 196–197: “Other Athabaskan languages permit the incorporation of objects and unaccusative subjects (Saxon 1989; Rice 1989). With true argument incorporation, the head noun is incorporated but there is no "raising" of a pronoun to argument status. Simple noun incorporation produces a derived intransitive. In contrast, Possessor Raising examples, like all classificatory incorporation structures, are necessarily transitive. The Navajo examples tell us where the person was affected by including the word for the body part in the verb itself, producing a derived transitive verb. In contrast, the incorporation of object and subject nouns does not occur in Navajo because the verbal arguments are pronominal prefixes that are bound in the discourse by nouns. The Navajo verb is always inflected for these pronominal arguments; therefore, nouns are not incorporated, only adjoined to the sentence.” P. 197: “Since this alternation occurs only with the third person, first, second and fourth person forms never appear in Possessor Raising constructions.” P. 198: the third-person alternation occurs in “three functional complexes: 1) the transitive verb and its arguments; 2) in grammatical postpositional phrases; and 3), in complex NP with possessor arguments”.

5. Phenomenon 2: the noun goes in, the slot count stays at two

SOURCED Willie's two contrasts (pp. 187, 196–197):

What goes into the verbDoes it fill a slot?Result
“True argument incorporation” (other Athabaskan; Saxon 1989, Rice 1989, cited, not opened)the object or unaccusative subject nounyes“a derived intransitive”
Navajo possessor raising / “classificatory” incorporationa body-part (or ‘home’, some kin) noun“not an argument”“a derived transitive verb”; the possessor fills the object slot

HYPOTHESIS PR-SLOT. Incorporating a body part Merges √BODY with the verb root inside rule set A and builds a new root √BODY:V. Its schema has the same two slots as before, Σ(√NECK:SEVER) = {ρA, ρP}. The body part adds no slot and fills none. It narrows the event to a part of the ρP participant. The person, not the body part, gets the Proto-Patient entailment “causally affected by another participant” (Dowty 1991:572, (28c)). That is Willie's “tell us where the person was affected” (p. 197), restated as an entailment. For Grant's thesis this is a clean case. A lexical item enters the verb, and the argument structure is still decided by role slots: two before, two after, with the same proto-roles.

HYPOTHESIS “Raising” without movement. Willie's term is “raised”/“advanced” (p. 182), but she also says the body part “serves to derive a complex verb” (p. 187). On this site's schema, (b) below fits her description better. We have not shown that (a) is wrong.

  1. (a) Movement: the possessor starts in Spec of the possessed noun inside the object and moves out to the object slot of the verb.
  2. (b) New schema: √NECK:SEVER is built first. Its ρP is a whole-person patient from the start, and no possessor position is ever projected.

Under (b) nothing moves, so possessor raising says nothing about VPISH (§11). A test is named in T16a.

6. Phenomenon 1: one word or two, and nothing in the sound tells you

SOURCED (11) naaʼahóóhai ʼasdzą́ą́ bi-kʼos kʼíiníłneʼ has two readings, and Willie gives each one its own structure (11.a′, 11.b′). The incorporation “produces no phonological effects” (pp. 191–192).

(11.a′) INVERSE(11.b′) DIRECT
Where ‘neck’ isinside the verb word: rule set Ahead of a possessed noun outside the verb: rule set B
What bi- isthe verb's object prefix (inverse)the possessor prefix of the noun
Slots of the verbρA = woman, ρP = chickenρA = chicken, ρP = [woman's neck]
Reading‘The woman cut off the chicken's head.’‘The chicken cut off the woman's head.’

HYPOTHESIS WORD-EDGE. The A→B changeover (the right edge of the domain where prefixes are left of the stem) is set by where the verb's schema Σ is closed. It is not set by a phonological word boundary. In (11) the same sounds can be cut at two places. Before bi-: everything right of it is one verb word in A, with ‘neck’ inside. Or after bikʼos: the noun word is closed and the verb starts at kʼíiníłne. Both cuts are well-formed, with different meanings. Willie's colon (11.a′) vs space (11.b′) records her analysis, not a pause. So this site's definition of the word as the changeover between rule sets has to be a syntactic one (end of Σ / spell-out of the verb), and it cannot be read off sound. Revised 2026-10-11 (afternoon): psych verbs put required roles on postpositions outside the verb word, so Σ closes at the MVS, not the verb word (WORD-EDGE′, pleonastic & psych §9). Note that both parses keep bi- left of a stem, inside some word. That fits rule set A in either case.

SOURCED + HYPOTHESIS The alternation is made once per domain. In (23)–(25) the body part is inside a lexical postpositional phrase (B). Then yi-/bi- appears on the possessor inside that PP, and putting bi- on the verb as well is ??? (24) or * (25) (pp. 195–196). Our reading: in (23) the person is not an argument of the verb at all. It is the possessor inside an adjunct PP, so the voice alternation of the verb has nothing to rank it against. The ranking happens inside the PP's own domain. Willie states only the distribution.

7. T11d, partly answered: whose house in Ch. 8 (28)?

T11d (fourth person compared) asked whether the b-/y- on the possessor in Willie's Ch. 8 (28a) John háísh baghan=di yiztsʼǫs? ‘Who did John kiss at his house?’ / (28b) … yighan=di … ‘… at her house?’ is fixed by c-command (RANK-OBV) or by the participant it refers to.

  • SOURCED Willie's Ch. 7 rule for possessors is linear, the same one she uses for verbs: “the bi- pronouns signals that the preceding nominal is not coindexed with it” (p. 185), and the yi- possessor is coindexed with it ((5a) ‘the girl's house’).
  • HYPOTHESIS Applied to (28), the noun just before baghan=di/yighan=di is háísh ‘who’. So (28a) b- = not the one asked about, which leaves John (‘his’), and (28b) y- = the one asked about (‘her’). This matches Willie's English pronouns, but she does not state it. The application is ours.
  • OPEN This does not separate “linear” from “c-command”. In these strings the preceding nominal could also be the one that is structurally higher. A pair where the two come apart is still needed (T11d stays open in that form).

8. A non-argument on the animacy scale? SOURCED (Willie's proposal)

P. 190: “(10 - 11) appear to violate the animacy hierarchy; the first noun in the sentence is lower in the animacy scale than the second.” Her proposal: ‘chicken’ can come first, “although it is lower in animacy than ʼasdzą́ą́ 'woman', because the third noun, -k’os neck' is interpreted as participating in the animacy hierarchy, and as occupying the lowest position on the animacy scale in the sentence. This makes the relative ranking of the two other nouns irrelevant”. P. 192: this holds “For speakers who make this distinction”.

HYPOTHESIS Tension with a slot-only ranking. On the theta schema page the yi-/bi- ranking compares the two slots' fillers. By Willie's own analysis (p. 187), ‘neck’ is not a slot. If it still counts for the animacy hierarchy, the ranking sees a lexical item that is not an argument, which is a counterexample to “arguments are role slots, not lexical items” as applied to ranking. Two ways out, neither tested: (i) the ranking is over the nouns present in the clause, not over slots (Willie's view as stated); (ii) in (10)/(11) the body part is in the B parse (11.b′ type), so it is inside an argument, and the order is licensed by that argument (the woman's neck) being low. (ii) would predict that (10) also has a (11.b′)-type parse, but (10) is unambiguous and direct. We leave it at T16c.

9. Rules

Notation as on the fourth person uses page. Σ(√) is a verb's slot set, π(ρ) a slot's person value. “Incorporate N” means Merge of a noun root with the verb root inside the verb word (rule set A).

PR-OBJ     In possessor raising the possessor is marked by the verb's object prefix;
           the body-part noun is incorporated into "a complex derived verb" and
           "loses its status as the head of a noun phrase".             (p. 186; (6c), (8))
PR-NOARG   The incorporated body part is not an argument of the verb; the subject
           and object prefixes are the arguments.                        (p. 187; (6))
PR-TRANS   Possessor raising verbs are "necessarily transitive"; simple noun
           incorporation (other Athabaskan) gives "a derived intransitive". (pp. 196–197)
PR-INAL    Possessor raising applies to body parts, "a few kin terms" and 'home';
           not to alienably possessed nouns.                    (pp. 193–194; *(18))
PR-VAR     Kin terms: speakers vary; some "do not permit yi- on kin terms". (pp. 195, 197)
PR-AMB     Incorporation "produces no phonological effects", so bi-k’os + verb can
           be inverse possessor raising or a direct verb with a possessed object.
                                                                   (pp. 190–192; (11))
POSS-YB    On a 3rd-person possessor, bi- = the preceding nominal is not coindexed
           with it; yi- = it is. An "extension" of the verb's alternation.
                                                                  (pp. 185–186; (5))
POSS-DA    da- on a possessed noun "takes scope only over the possessor".  (p. 184; (3))
LEX-PP     With a lexical PP paraphrase, yi-/bi- appears on the postposition and
           "cannot appear on the verb itself".                    (pp. 195–196; (23)–(25))
PR-ANIM    [Willie's proposal] the body part counts as lowest on the animacy scale,
           which licenses a lower-animacy noun first.                    (p. 190; (10)–(11))
PR-3ONLY   [Willie's statement] "first, second and fourth person forms never appear
           in Possessor Raising constructions."                          (p. 197; cf. (6c))
RuleStatusBasis
PR-OBJ, PR-NOARG, PR-TRANS, PR-INAL, PR-VAR, PR-AMB, POSS-YB, POSS-DA, LEX-PP, PR-ANIM, PR-3ONLYSOURCEDWillie's statements and examples as cited. Rule names are ours.
PR-SLOT (§5): the derived root keeps Σ = {ρA, ρP}, with ρP = the whole person, causally affectedHYPOTHESISFrom PR-NOARG + PR-TRANS + p. 197 “where the person was affected”, with Dowty (28c).
WORD-EDGE (§6): the A→B changeover is the close of Σ, not a phonological boundaryHYPOTHESISFrom PR-AMB.
da- counts slots in nouns as in verbs (§3)HYPOTHESISPOSS-DA + the da- results on the number & scope page.
PR-3ONLY vs (6c) shinákʼeedzííníltsʼin, a 1st-person objectOPENT16b.

10. Derivations Q1–Q6, step by step

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). Verbs are not segmented beyond Willie's glosses and her own hyphens and colons. 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).

Q1. (6b) náshidííltsʼin ‘You punched me’ vs (6c) shinákʼeedzííníltsʼin ‘You eye-punched me’

  1. Σ (6b): Σ(√PUNCH) = {ρA, ρP}. π(ρA) = 2, π(ρP) = 1 (gloss “1.2”).
  2. Σ (6c): first Merge √EYE with √PUNCH, inside A. This gives the root √EYE:PUNCH. PR-NOARG: √EYE takes no slot. PR-SLOT (HYPOTHESIS): Σ(√EYE:PUNCH) = {ρA, ρP}, and ρP is causally affected at the eye. π values as in (6b), gloss “1.2”.
  3. Merge v; ρP in Comp,V, ρA in Spec,v. Agree: object and subject prefixes are valued 1 and 2. (V-P) Move ρA to Spec,M/SP. (V-H) no move. Either way the same string results.
  4. A Spell-out: the prefixes go left of the (complex) stem, giving one word each, náshidííltsʼin / shinákʼeedzííníltsʼin. Not segmented further (T16g). The word closes here: A→B.
  5. B No nouns. The clause is the word. Same slot count in both: 2. The only difference is inside the root.

Q2. (8) ʼasdzą́ą́ naaʼahóóhai yi-kʼos:kʼíiníłneʼ vs (9) … bi-kʼos:kʼíiníłneʼ

  1. Σ Merge √NECK with √SEVER in A, giving √NECK:SEVER, Σ = {ρA, ρP}, both π = 3.
  2. Merge v and the voice head. (8) direct: ρA outranks ρP, so the object prefix is yi-. (9) inverse: ρP is made external, so the prefix is bi- (as on the inverse page; nothing new here).
  3. A Spell-out: yi-/bi- left of the complex stem, giving yi-kʼos:kʼíiníłneʼ / bi-kʼos:kʼíiníłneʼ. A→B.
  4. B Adjoin ʼasdzą́ą́, naaʼahóóhai to the left. Linking (p. 189): in the direct, “the noun immediately preceding the complex verb is coindexed with the patient”, so chicken = ρP, woman = ρA (8). In the inverse, chicken = ρA, woman = ρP (9). ‘Neck’ links to nothing. “neither one refers to the body part” (p. 189).

Q3. (11) naaʼahóóhai ʼasdzą́ą́ bi-kʼos kʼíiníłneʼ: two derivations of one string

  1. Parse a = (11.a′). Σ As Q2, inverse. A bikʼos:kʼíiníłneʼ is one word. B The noun right before the verb, ‘woman’, is not the patient (inverse). So woman = ρA and chicken = ρP: ‘The woman cut off the chicken's head.’
  2. Parse b = (11.b′). Σ Σ(√SEVER) = {ρA, ρP}, with no incorporation. A separate noun word: Σ(√NECK) = {ρPoss}, π = 3. A Noun word bikʼos (possessor prefix left of the stem). Verb word kʼíiníłne [sic], glossed “3.3.sever”. Direct, but no yi- is printed on it; T16f. B [ʼasdzą́ą́ bik’os] is a complex NP (possessor before noun, head-final), linked to ρP. ‘chicken’ is linked to ρA. Reading: ‘The chicken cut off the woman's head.’
  3. HYPOTHESIS WORD-EDGE: the two parses differ only in where Σ of the verb closes. In a, the close is after the whole string from bi-; in b, it is after kʼíiníłne, with bikʼos already closed as a noun. The phonology is the same (p. 192).
  4. Why (10) yi-kʼos kʼíiníłneʼ is not ambiguous. A yi- possessor would have to be coindexed with the preceding nominal ‘woman’ (POSS-YB), and the reading ‘the chicken cut off the woman's head’ comes out either way. So the two parses give one meaning. HYPOTHESIS: Willie gives only the one reading and the direct structure (10′). She does not discuss a (10)-type b-parse.

Q4. *(18) hastiin John yitłʼízí yitsą́:dzíítáál vs (16)–(17)

  1. Σ √STOMACH:KICK (PR-SLOT), {ρA, ρP}. ρP is the one whose stomach is kicked.
  2. (16)/(17): ρP = the goat (16) or the man (17). B [John bitłʼízí] is an ordinary possessed NP, with bi- as the default possessor. It is linked to a slot of the verb as a whole noun phrase.
  3. *(18): yitłʼízí would need ‘goat’ to undergo possessor raising itself (its possessor marked by yi- in the alternation). PR-INAL: ‘goat’ is alienable, so it cannot (p. 194). HYPOTHESIS: on this site, only a noun that can be incorporated into Σ (A) can host the alternating yi- possessor. That links POSS-YB to PR-INAL, a link Willie implies by her section heading but does not state.

Q5. (23) hastiin łį́į́ʼ bitsą́=góneʼ yiztał, ???(24), *(25)

  1. Σ Σ(√KICK) = {ρA, ρP}. No incorporation. A separate PP: [bitsą́]=góneʼ ‘in (his) belly’. Its possessor π = 3.
  2. A Noun word bitsą́. Verb word yiztał (direct). B Merge =góneʼ after its complement (head-final), giving an adjunct PP.
  3. LEX-PP: the yi-/bi- choice is made on the possessor inside the PP. The verb's own prefix must not also be bi-: ???(24), *(25).
  4. OPEN Linking ‘man’ and ‘horse’ to the slots: Willie translates (23) with the horse as kicker, which the plain direct rule does not give (T16e). We stop here and do not assign the slots.

Q6. Ch. 8 (28a) John háísh baghan=di yiztsʼǫs? with POSS-YB

  1. Σ Σ(√KISS) = {ρA, ρP}; PP [Σ(√HOME) = {ρPoss}]=di.
  2. B Order: John, háísh, [baghan=di], yiztsʼǫs. The nominal before baghan is háísh.
  3. POSS-YB: b- means ρPoss ≠ the referent of háísh. With only two people in the sentence, ρPoss = John: ‘his house’. In (28b), y- means ρPoss = the referent of háísh: ‘her house’. HYPOTHESIS (§7).

11. VPISH

HYPOTHESIS Nothing in Chapter 7 decides between V-P and V-H. The raised possessor is an object (PR-OBJ), and the subject slot behaves as in any transitive (Q1–Q2). Under PR-SLOT (b) there is no movement at all, so possessor raising is not a raising to subject and cannot test VPISH. One part of it could become a VPISH test. If T16a found evidence for movement (a), that would be A-movement out of a nominal into Comp,V, inside the verb word. That is a shorter cousin of the subject movement VPISH is about, and the same locality question would apply.

12. Open questions

IDQuestionWhat would settle it
T16aIs possessor raising movement out of a possessed object (a) or a new two-slot schema with no possessor position (b)? Test: can da- on a possessor-raising verb count the raised person the way POSS-DA counts a possessor? Can a possessor-raising verb take a 4 object (see T16b)?Possessor-raising verbs with da- and with fourth-person objects; Young & Morgan 1987. OPEN
T16bWillie says “first, second and fourth person forms never appear in Possessor Raising constructions” (p. 197), but (6c) shinákʼeedzííníltsʼin ‘you eye-punched me’ has a 1st-person object and is her model case (p. 187). Does p. 197 mean only the yi-/bi- alternation (3rd person only), or is (6c) a different construction?Willie's intent; more 1st/2nd/4th-person body-part verbs. OPEN
T16cDoes the animacy ranking see the non-argument body part (PR-ANIM, Willie p. 190), or only slots (this site)? See §8 (i) vs (ii).Speakers “who make this distinction” (p. 192); (10)/(11)-type pairs with non-body-part incorporates. OPEN
T16d(19) ba-ghandi:níyá is glossed “3.3” and (20) yi-ghandi:níyá “INV3.3”, the reverse of every other pair in the chapter. Gloss swap, or a real difference for ‘home’?Another source for ‘visit (someone's home)’. OPEN
T16e(23) hastiin łį́į́ʼ bitsą́=góneʼ yiztał is translated ‘The horse kicked the man in the stomach’ with a direct verb and ‘man’ first. Does the PP change the linking of the two nouns, or is the translation a slip?A speaker; the same frame with the nouns swapped. OPEN
T16f(11.b′) kʼíiníłne has no final ’ and no visible yi-, though it is glossed “3.3.sever” and has a full NP object. Misprint, or does a possessed NP object leave the verb without the yi- seen in (10)?A ‘sever’ paradigm (Y&M). OPEN
T16gHow are the incorporated forms related to the free nouns: shinááʼ ‘my eye’ vs shinákʼee- in (6c); -tsą́ ‘stomach/belly’; -la’ ‘hand’; -kʼos ‘neck’ translated ‘head’?Y&M 1987 (not opened). We do not segment until then. OPEN
T16hCross-linguistic (for the backlog): Willie compares Muskogean possessor raising (p. 186, no citation) and Athabaskan argument incorporation (Saxon 1989, Rice 1989). Do these share PR-SLOT, or do they only share the label “possessor raising”/“incorporation”? Do not merge them until a source is opened.Saxon 1989; Rice 1989; a Muskogean grammar. OPEN

Fidelity notes. The chapter's spellings are kept as printed. Body-part boundary marks vary: yi-kʼos:kʼíiníłneʼ (8), yi-kʼos kʼíiníłneʼ (10), yikʼos:kʼíiníłneʼ (10′). Glosses vary: “hand:kissed” (12) vs “hand-kissed” (13); “stomach” (14)–(15) vs “belly” (16)–(17), with the same -tsą́:dzíítáál. (15) has no INV in the gloss although marked INVERSE [sic]. (22) has “3-3-”. (23)–(24) =góneʼ vs (25) =gone’. In (25) ‘horse’ is printed with an odd first glyph and is read as łį́į́ʼ. The translations of (8)–(11) say ‘head’ while the gloss says ‘neck’. Willie's “relignment”, “suchas”, “cannot subjected”, “Possessor Rasing”, “heirarchy”, “Possesor”, “do not used”, and the TOC's “135” for §7.5 (printed p. 195) are all as printed. The p. 184 statement about the duals (“the second dual possessive encodes hearer + hearer”) is copied as printed.

"The girl who was kissed by the boy": an inverse relative can be headed by its external patient, so Willie's absolutive rule (16) covers only relatives without nouns (T71b; T71e in part)

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.

Working note. Kept as written (URL, data and status labels unchanged). Filed under The fourth person; the consolidated argument is the essay fourth-person. Rule IDs: rules index.

Verbs on this page → dictionary: jizloh ‘rope’

Restated on the unified structure (2026-11-15). Relative clauses are restated as a full CP, built like a main clause, under an overt D⁰ (=yę́ę, =ígíí) linked to one participant by Agree; nothing moves out (H27b, H28′, H31, H34). 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, relative rows. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.

T71b asked whether any inverse relative is headed by the external patient. It is. Willie 1991 p. 162 (20b) ʼatʼééd ʼashkii biztsʼǫs=yę́ę yóóʼeelwod 'The girl who was kissed by the boy ran away', and she restates her rule there: "When both arguments of the relative have nominal adjuncts, the relativizing enclitic binds the external argument, which must be definite" ((21)) SOURCED. The data were already on the site (relatives REL-EXT, deep-shallow D4′). Third-voice asked T71b because it overlooked them, and this page corrects that. The consequence: the absolutive rule (16) is Willie's default for relatives with no noun inside, not a general head condition HYPOTHESIS. The /third-voice/ argument against a transitive-inverse analysis of biʼdistał=yę́ę survives, because that relative has no noun. But it now shows a difference between the two surviving analyses. Hale's (EI-9) gets the one attested reading from (16) alone. The intransitive-inverse analysis (EI-8′) needs Willie's ban on nouns for the indefinite agent as an extra premise HYPOTHESIS. Which argument heads the relative is decided by position, nouns and definiteness, not by role. That tells against the strong roles-not-words thesis, which stays OPEN and not established.

On this page: 1 Question · 2 Sources · 3 Rules first · 4 The data · 5 Answer to T71b · 6 The head grid · 7 What it does to /third-voice/ · 8 T71e: is the inverse agent internal? · 9 Rulesets A and B · 10 Derivations · 11 The role schema · 12 Fidelity · 13 Open questions

Notation is the same as on third-voice (no Syntax · Semantics · Interface conventions file has arrived). A = the word level (prefixal, left-headed). B = the clause level (right-headed). Σ = the role schema, with Dowty's proto-roles as the baseline. Merge builds structure by putting two pieces together. Agree values features such as person and number on a probe from a goal. Move re-merges a piece that is already in the structure in a higher position. EXT/INT = external/internal argument in Willie's sense. Under VPISH every agent is first merged in Spec,vP, and "external" names where an argument ends up. REL = the enclitic =yę́ę. N = the number of arguments of the relative that have a noun adjoined (0, 1 or 2). ABSb = S or INT (VC-11 on /third-voice/).

1. Question

/third-voice/ used Willie's (16), "The heads of relative clauses are absolutives" (p. 160), as a test of the agentive passive. It left T71b open: is there an inverse relative headed by the external patient? (16), read as S or INT, predicts none. If one exists, (16) is not general, and the uses of (16) on /third-voice/ (VC-14, the §7 argument against EI-8) must be checked again. T71e asked the related terminological question: is the inverse agent "internal" (p. 72) or "the transitive subject of the relative" (p. 160)?

2. Sources

SOURCED Willie, MaryAnn 1991. Navajo Pronouns and Obviation. PhD dissertation, University of Arizona. Page images read this run: printed pp. 159–164 (PDF = printed + 6), with a 300 dpi crop of p. 162 ((20b), (21), (22)). The p. 152 (1f) crop from the last run was rechecked. Text layer only, as a finding aid: pp. 165–170 ((25)–(40)). Ch. 3 p. 72 is cited from the image reading recorded on ergative-indefinite (EI-2). It was not reopened this run.

Site pages that already held these data: relatives (REL-ABS, REL-EXT, REL-NR) and deep-shallow (D4′).

Not opened: Platero 1982 (cited by Willie p. 162 for speakers who accept two readings of (20)) and Young & Morgan 1987.

3. Rules first

EH-1 to EH-6 are Willie's statements, quoted from the page images. EH-7 to EH-13 are ours.

EH-1  "(16) The heads of relative clauses are absolutives ... either the            SOURCED (Willie p. 160, image)
      intransitive subject or an internal argument of the transitive verb --
      and recall that with the Inverse voice, the agent is the internal
      argument, as in (15)." Stated over "the relative clauses we have seen
      so far", all of which have no noun inside.
EH-2  (17): "the argument structure of the verb determines the interpretation       SOURCED (Willie p. 160, image)
      of an adjoined nominal": ʼashkii yiztsʼǫs 'She kissed the boy'
      (*'The boy kissed her'); ʼashkii biztsʼǫs 'She was kissed by the boy'
      (*'She kissed the boy').
EH-3  On (18)/(19): (19) is "agent-headed ... because the agent is the internal     SOURCED (Willie p. 161, image)
      argument in the Inverse construction."
EH-4  "We must revise the generalization given in (17) above as follows.            SOURCED (Willie p. 162, image)
      (21) When both arguments of the relative have nominal adjuncts, the
      relativizing enclitic binds the external argument, which must be
      definite." For some speakers (Platero 1982) (20) has two readings;
      "this is not true of my speech."
EH-5  With one noun, "the Direct relative apparently has two readings": on the     SOURCED (Willie pp. 162–163, images)
      restrictive reading "the head of the relative is the external argument,
      the agent that has no adjoined nominal"; on the non-restrictive one the
      noun and the relative are both adjuncts to the main-clause subject.
      Commas in her English gloss mark the non-restrictive reading.
EH-6  "The Inverse relative with one nominal has only one reading" (24a); the      SOURCED (Willie p. 164, image)
      restrictive reading (24b) is starred, "since the Inverse voice requires
      a definite subject, which cannot be subjected to further definiteness
      marking by means of a restrictive relative."

EH-7  (T71b) Yes. (20b) is an inverse relative whose head is the EXT patient        SOURCED for the reading (EH-4, (20b));
      (the girl, who was kissed). (16) read as "S or INT" is therefore not a      HYPOTHESIS for the scope of (16)
      general head condition. Willie's own word "revise" (EH-4) restricts it
      to relatives without nouns (N = 0).
EH-8  Head grid (our tabulation of (18)–(24); see §6):                              HYPOTHESIS (compilation; cells SOURCED)
        N = 0: REL binds INT (Direct: patient; Inverse: agent)
        N = 2: REL binds EXT (Direct: agent; Inverse: patient), definite
        N = 1, Direct: REL binds the argument without a noun (EXT agent),
               restrictive; or the relative is non-restrictive
        N = 1, Inverse: non-restrictive only; the restrictive reading on
               the noun-less EXT patient is * (24b)
EH-9  Restatement: REL binds a position, chosen by N and by the         HYPOTHESIS
      definiteness of EXT. It does not bind a case. "Absolutive" in (16)
      describes the N = 0 cells only.
EH-10 /third-voice/ VC-14 and §7 survive: (12) biʼdistał=yę́ę has N = 0          HYPOTHESIS
      (no noun, and none is possible for INDF, VC-3), so (16) applies to it.
      But "heads a relative" is no longer by itself a diagnostic of ABSb.
      Only "heads a relative with N = 0" is.
EH-11 Under (16), with N = 0, the two surviving analyses of biʼdistał=yę́ę        HYPOTHESIS
      (12) differ:
        EI-9 (Hale): patient = INT (ABSb), INDF = transitive subject (not
              ABSb). (16) alone picks the patient: one reading. ✓
        EI-8′ (intransitive inverse): patient = S (ABSb), INDF = INT. If the
              INT agent is also ABSb, as EH-1 says of the inverse agent, (16)
              allows both, and only VC-3 (no noun may be adjoined to INDF;
              relatives are adjuncts, p. 158) removes the agent reading.
      EI-8′ needs one premise more than EI-9. That is a cost, not a refutation.
EH-12 (T71e, partly) In ch. 6 Willie calls the inverse agent the "internal        SOURCED for the wording;
      argument" twice (p. 160 under (16); p. 161, EH-3), the same as p. 72.    HYPOTHESIS for the reading
      She uses "external" for the inverse patient (EH-4 with (20b)).
      "The transitive subject of the relative" (p. 160, on (15)) is best read
      as the agent of a transitive relative, not as EXT.
EH-13 (Focus 2) In (18)/(19) and (20a)/(20b) the same verb and the same roles      HYPOTHESIS
      give opposite heads. The head changes with voice (position), N and
      definiteness, all of which are properties of items and clause
      structure, not of roles.

4. The data

All forms are printed in Willie pp. 160–164 and were read on page images this run. (20b) was also read in a 300 dpi crop. Willie glosses only (17)–(19). For (20)–(24) she gives translations only, so our Leipzig lines reuse the glosses she gives for the same words in (18)/(19). "source" lines give her glosses. PFV follows the convention of earlier pages for her past-tense English glosses.

Willie 1991 p. 160 (17) (page image): a noun is read off the voice
(1)ʼashkiiyiztsʼǫs
boy3O.3S.kiss.PFV
'She kissed the boy.' / *'The boy kissed her.' (17a)
(2)ʼashkiibiztsʼǫs
boyINV.3O.3S.kiss.PFV
'She was kissed by the boy.' / *'She kissed the boy.' (17b)
Willie 1991 p. 161 (18)–(19) (page image): no noun (N = 0)
(3)yiztsʼǫs-yę́ęyóóʼeelwod
source: 3.3.kissed=RELsource: 3.ran away
3O.3S.kiss.PFV=REL3S.run.away.PFV
'The one he kissed ran away.' (18), patient-headed
(4)biztsʼǫs=yę́ęyóóʼeelwod
source: INV3.3.kissed=RELsource: 3-ran away
INV.3O.3S.kiss.PFV=REL3S.run.away.PFV
'The one who kissed him ran away.' (19), agent-headed
Willie 1991 pp. 161–162 (20a)–(20b) (page images; (20b) at 300 dpi): two nouns (N = 2)
(5)ʼatʼéédʼashkiiyiztsʼǫs=yę́ęyóóʼeelwod
girlboy3O.3S.kiss.PFV=REL3S.run.away.PFV
'The girl who kissed the/a boy ran away.' (20a): head = EXT agent
(6)ʼatʼéédʼashkiibiztsʼǫs=yę́ęyóóʼeelwod
girlboyINV.3O.3S.kiss.PFV=REL3S.run.away.PFV
'The girl who was kissed by the boy ran away.' (20b): head = EXT patient
Willie 1991 p. 162 (22) and p. 164 (24) (page images; (22) at 300 dpi): one noun (N = 1)
(7)ʼashkiiyiztsʼǫs=yę́ęyóóʼeelwod
boy3O.3S.kiss.PFV=REL3S.run.away.PFV
a. 'The one who kissed the boy ran away.' (restrictive, preferred) b. 'The boy, who was kissed by her, ran away.' (non-restrictive) (22)
(8)ʼashkiibiztsʼǫs=yę́ęyóóʼeelwod
boyINV.3O.3S.kiss.PFV=REL3S.run.away.PFV
a. 'The boy, the one who kissed her, ran away.' b. *'The one who was kissed by the boy ran away.' (24)

Willie prints (18) with a hyphen (yiztsʼǫs-yę́ę) and the others with "="; we copy that. (22) is printed "yóóʼeelwcd" on the page image, and the text layer reads the same. We give the spelling of (18)–(20). The bracketing and the "head =" labels in our translation lines come from her prose on pp. 161–164.

5. Answer to T71b

Yes. In (6) = (20b) the girl is the patient of 'kiss' (by (2) = (17b), the noun next to biztsʼǫs is the agent, so the outer noun ʼatʼééd is the patient). The relative is headed by her: "The girl who was kissed by the boy ran away". Willie's rule (21) says this is no accident. With two nouns, REL "is forced to bind the external argument", and in the Inverse the external argument is the patient SOURCED.

What changes. In (16) the inverse EXT patient is not ABSb (it is the EXT of a transitive), so (16) taken as a general rule predicts that (6) is impossible. Willie avoids the contradiction herself. She introduces (21) with "We must revise the generalization". Her cross-reference reads "(17)", but (17) is data, and the only generalization before it is (16). So (16) is the N = 0 case of a larger pattern (EH-7) HYPOTHESIS. Our reading of "(17)" as a slip for (16) is also HYPOTHESIS.

Correction. /third-voice/ said that "in every inverse relative" of the ch. 6 text layer the enclitic binds the agent. That was wrong for (20b), which is on p. 162, before the pp. 166–167 range searched there. The site already had (20b) and (21) on relatives (REL-EXT) and deep-shallow (D4′). The fault was a missed cross-reference, not missing data.

6. The head grid

Nouns in the relative (N)Direct (yi-)Inverse (bi-)Status
0INT = patient (3) (18)INT = agent (4) (19)SOURCED Willie (16), p. 161
1EXT agent, restrictive (7a); or non-restrictive (7b)non-restrictive only (8a); *restrictive on EXT patient (8b)SOURCED pp. 162–164
2EXT = agent (5) (20a)EXT = patient (6) (20b)SOURCED (21), p. 162

Read by rows, the grid says that REL binds INT by default and binds EXT when the clause supplies nouns for the other argument or for both HYPOTHESIS (EH-8, EH-9). Read by columns, the same role (patient, agent) ends up in different cells depending on voice. This is the "positions, not roles" result that /deep-shallow/ D4′ drew from the same data.

A tension we cannot settle (T72a). Willie explains *(8b) by saying the inverse's definite subject "cannot be subjected to further definiteness marking by means of a restrictive relative" (EH-6). But (6) = (20b) binds that same definite EXT patient, and its English gloss has no commas, which by her own convention (EH-5) means restrictive. Either (20b) is non-restrictive and printed without commas, or the definiteness clash is lifted when the EXT has its own noun (ʼatʼééd). Willie does not say which OPEN.

7. What it does to /third-voice/

VC-14 and the §7 argument survive HYPOTHESIS. Both rest on (12) biʼdistał=yę́ę 'the one who was kicked' (p. 152 (1f)), and that relative has no noun. None is possible either, because no noun may be adjoined to the indefinite agent (VC-3, p. 36), and a noun adjoined to the patient would give N = 1. So (12) is an N = 0 relative, and (16) is the right rule for it. The transitive-inverse analysis (EI-8) still predicts the wrong head: with N = 0, REL binds INT, and under EI-8 the INT argument is the agent, which would give 'the one who kicked him'. The weaker formulation that /third-voice/ used is now wrong: "heads a relative" proves ABSb only for N = 0 relatives (EH-10).

A new difference between EI-8′ and EI-9 (EH-11) HYPOTHESIS. With N = 0, (16) lets REL bind any ABSb argument.

  • Under EI-9 (Hale: ʼ-d- is the transitive subject), only the patient is ABSb. (16) yields exactly the attested reading.
  • Under EI-8′ (an intransitive inverse), the patient is S and the agent is INT. Willie's own gloss on (16) treats the inverse INT agent as a possible head ((4) = (19)). So (16) allows both readings, and the agent reading is removed only by adding VC-3 together with "relatives are adjuncts" (p. 158). This is the T71a reasoning.

So EI-9 needs fewer premises. That is a cost for EI-8′, not a refutation. The deciding datum is still T71a: an agentive-passive relative that binds the agent, or Willie's statement that none exists.

New prediction (T72b) OPEN. With one noun on the patient, [N biʼdistał=yę́ę] (schematic, not attested) can only be non-restrictive under any analysis, because the one argument without a noun is the INDF agent, which VC-3 bars from heading. A restrictive reading, if a source showed one, would be a problem for VC-3. No such form is in any source opened.

8. T71e: is the inverse agent internal?

In ch. 6 Willie calls the inverse agent "the internal argument" twice: under (16) (p. 160) and in her comment on (19) (p. 161). She calls the inverse patient "external" in (21) as applied to (20b) SOURCED. This agrees with p. 72 ("internal ... not the subject", EI-2). The one dissenting phrase, "the transitive subject of the relative" (p. 160, of (15) bizloh=ę́ę), stands two lines above (16) and its gloss, which calls the same argument internal. We read "transitive subject" there as "the agent of the transitive relative", not as EXT (EH-12) HYPOTHESIS. T71e is only partly answered, because ch. 3 §3.3.5 was not reread together with ch. 6 on page images this run.

9. Rulesets A and B (Focus 1)

StepLevelRulesetWhat happens
1vP, clauseBMerge the pronouns and the verb. Choose the voice (EI-4). Move the higher-ranked 3rd person to EXT in the Inverse.
2WordAAgree and spell-out give yiztsʼǫs or biztsʼǫs. The prefix order (patient outside agent) is the same in both voices (VC-15).
3ClauseBAdjoin 0, 1 or 2 nouns to the left of the word. Each is linked to a position by the voice (EH-2): the noun next to the verb goes with INT, the outer one with EXT.
4Word edgeA→BMerge =yę́ę at the right edge of the clause (on the verb word). It binds INT if N = 0 and EXT if N = 2 (EH-8).
5ClauseBThe whole relative is adjoined to a main-clause argument (p. 158), and its definiteness is checked against EXT (EH-6).

This sharpens the changeover point that /third-voice/ found HYPOTHESIS. The enclitic is a word-level (A) element phonologically, but its binding target is set by step 3, by how many nouns stand outside the word. A cannot see those nouns. So REL's choice must be made in B after the word has been spelled out, or REL must be a clause-level head (as on relative-cp) that A only spells out. The word is where the two rulesets hand over, and REL sits exactly on that boundary.

10. Derivations

Each step names its operation and ruleset. Merge builds structure, Agree copies person and number onto a probe, and Move re-merges a piece higher up. The forms are Willie's. The structure is ours HYPOTHESIS.

D1  yiztsʼǫs-yę́ę yóóʼeelwod 'The one he kissed ran away' (18), Direct, N = 0
    1  Merge  [VP pro(3,P) √tsʼǫs]; Merge pro(3,A) in Spec,vP            B   VPISH
    2  Move   pro(3,A) → EXT; pro(3,P) stays INT                         B   Direct (EI-4)
    3  Agree  object slot ← [3], Direct → yi-; subject slot ← [3] → Ø    A
    4  Spell  yiztsʼǫs                                                   A
    5  Merge  REL at the right edge; N = 0 → binds INT = patient         A→B EH-8 row 0
    6  Adjoin [yiztsʼǫs-yę́ę] to the subject of yóóʼeelwod               B   p. 158

D2  biztsʼǫs=yę́ę yóóʼeelwod 'The one who kissed him ran away' (19), Inverse, N = 0
    1  Merge  as D1.1                                                    B
    2  Move   pro(3,P) → EXT; pro(3,A) stays INT                         B   Inverse (EI-4)
    3  Agree  object slot ← [3], Inverse → bi-                           A
    4  Spell  biztsʼǫs                                                   A
    5  Merge  REL; N = 0 → binds INT = agent                             A→B EH-3
    6  Adjoin to the subject of yóóʼeelwod                               B

D3  ʼatʼééd ʼashkii yiztsʼǫs=yę́ę yóóʼeelwod (20a), Direct, N = 2
    1–4 as D1                                                            B, A
    5  Adjoin ʼashkii (next to verb) → INT = patient; ʼatʼééd → EXT      B   EH-2
    6  Merge  REL; N = 2 → binds EXT = agent (the girl); EXT definite    A→B EH-4

D4  ʼatʼééd ʼashkii biztsʼǫs=yę́ę yóóʼeelwod (20b), Inverse, N = 2       ← answers T71b
    1–4 as D2                                                            B, A
    5  Adjoin ʼashkii (next to verb) → INT = agent; ʼatʼééd → EXT = patient   B   EH-2
    6  Merge  REL; N = 2 → binds EXT = patient (the girl)                A→B EH-4
    7  Check  (16) as a general rule: EXT of a transitive is not ABSb → ✗
              (16) as the N = 0 default (EH-7): not applicable → ✓

D5  ʼashkii biztsʼǫs=yę́ę yóóʼeelwod (24), Inverse, N = 1
    1–4 as D2                                                            B, A
    5  Adjoin ʼashkii → INT = agent                                      B   EH-2
    6a Merge  REL non-restrictive, linked to the agent = ʼashkii         A→B (24a) ✓
    6b Merge  REL restrictive on the noun-less EXT patient               A→B *(24b): EXT is definite (EH-6)

D6  biʼdistał=yę́ę 'the one who was kicked' (p. 152 (1f)), N = 0, rechecked
    1–4 as /third-voice/ D4: biʼdistał                                   B, A
    5  Merge  REL; N = 0 → (16) applies (EH-10)                          A→B
       EI-8  (transitive inverse): INT = INDF agent → 'the one who kicked him'  ✗ (not the attested reading)
       EI-8′ (intransitive inverse): patient = S, INDF = INT; both ABSb → needs VC-3 to exclude INDF   ✓ with one extra premise
       EI-9  (Hale): patient = INT, INDF = subject → only the patient   ✓

11. The role schema (Focus 2)

Does a relative pick its head by role? The grid says no HYPOTHESIS. The same verb -ztsʼǫs 'kiss', with the same two roles, is headed by the patient in (3) and by the agent in (4) (voice), and by the agent in (5) and the patient in (6) (voice with N = 2). What decides is the position an argument reaches (EXT/INT), how many nouns the clause has, and whether EXT can be definite. Those are facts about the clause and about the particular items, not about Σ (EH-13). Willie herself frames it this way: "the argument structure of the verb determines the interpretation of an adjoined nominal" (p. 160) SOURCED, where "argument structure" means EXT/INT as set by voice.

This fits the weak schema thesis (roles fix the word-level slot order, VC-15/16) and adds nothing to the strong thesis that Navajo verbs take roles rather than lexical items as arguments. The strong thesis stays OPEN and is not established. No chemistry analogy is drawn here.

12. Fidelity

  • All eight numbered forms were read on Willie's page images this run. (6) = (20b) and (7) = (22) were also read in a 300 dpi crop. Willie gives gloss lines only for (17)–(19). The Leipzig lines for (20)–(24) reuse her glosses for the same words and add nothing else.
  • The "head =" labels are drawn from her prose: (21) for (20), p. 163 for (22), p. 164 for (24). For (20a) and (20b) individually she states only (21) and the translations.
  • "(17)" in "revise the generalization given in (17)" is printed as such. Reading it as (16) is ours.
  • (25)–(40) (pp. 165–170) were used only as a text-layer finding aid. Nothing on this page depends on them.
  • Platero 1982, the source for the two-reading speakers, was not opened. The judgments here are Willie's own speech ("this is not true of my speech").
  • T71b was posed on /third-voice/ as if the data were unknown, but they were on /relatives/ and /deep-shallow/. This page records the correction. It does not edit the earlier page's argument, only its open-questions row.

13. Open questions

IDQuestionWhat would settle it
T71bIs there an inverse relative headed by the EXT patient?Answered: yes, (20b) with (21) SOURCED. (16) is the N = 0 default (EH-7) HYPOTHESIS.
T71eInverse agent: internal or "transitive subject"?Partly answered: internal in ch. 6 too; "transitive subject" is loose (EH-12) HYPOTHESIS. Ch. 3 §3.3.5 still to be reread on images.
T71aCan INDF head a relative?Still OPEN. Now also the test that removes EI-8′'s extra premise (EH-11).
T72a(20b) binds the definite inverse EXT with no commas in the gloss, yet *(24b) is excluded because a definite EXT "cannot be subjected to further definiteness marking by ... a restrictive relative". Is (20b) non-restrictive, or does the EXT's own noun lift the clash?Willie's or Platero 1982's discussion of (20); a restrictive/non-restrictive diagnostic for Navajo relatives. OPEN
T72bAgentive passive with one noun on the patient, [N biʼdistał=yę́ę] (schematic): non-restrictive only, as VC-3 predicts?A sourced example (Young & Morgan 1987; Willie ch. 6 has none in the pages read). OPEN
T72cPlatero 1982's speakers who accept two readings of (20): do they also allow the N = 0 default to bind EXT? This bears on whether (16) is grammar or preference.Platero 1982 (not opened; same source as T53f). OPEN
T72dWith a 1st/2nd-person argument (no voice alternation, p. 86), does the N-dependence hold? Willie pp. 170–171 ((41)) has relatives with 1st/2nd arguments that are not the head.Willie pp. 170–175 on page images. OPEN Update 2026-10-29 ~12:54pm PT: answered on person-head: no. The grid (EH-8) holds only when two third persons compete.

Sources: Willie 1991 pp. 152 (crop), 159–164 (page images), pp. 165–170 (text layer). See also third-voice, relatives, deep-shallow, relative-cp, ergative-indefinite, open questions, sources.

"The one who kissed me": with a first, second or fourth person in the relative, the head is the third-person argument whatever its role, so the noun-count grid holds only for two third persons (T72d)

Working note. Kept as written (URL, data and status labels unchanged). Filed under The fourth person; the consolidated argument is the essay fourth-person. Rule IDs: rules index.

Verbs on this page → dictionary: yiztał ‘kick’ · yishxash ‘bite’ · dabiiłtsą́ ‘see’

Restated on the unified structure (2026-11-15). Relative clauses are restated as a full CP, built like a main clause, under an overt D⁰ (=yę́ę, =ígíí) linked to one participant by Agree; nothing moves out (H27b, H28′, H31, H34). 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, relative rows. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.

T72d asked whether the noun-count grid of external-head (EH-8: with no noun the relative binds the internal argument, with two nouns the external one) also holds when one argument of the relative is first or second person. It does not. Willie 1991 (41a) siztsʼǫs=yę́ę 'the one who kissed me' has no noun, yet it is headed by the agent, and Willie p. 123 says that a third-person agent acting on a first-, second- or fourth-person patient is "the external Nominative argument" SOURCED. (41b) sétsʼǫs=yę́ę 'the one I kissed' is headed by the patient. Ten relatives with one fourth and one third person on pp. 172–175 show the same split SOURCED. In each, the head is the one third-person argument, whether it is agent or patient, external or internal. Willie's statements that first, second and fourth persons cannot head a restrictive relative (pp. 171, 175) explain why. Our restatement: the voice/noun grid and the absolutive rule (16) apply only when two third persons compete, the only case with a voice alternation (p. 86). With one third person, person alone picks the head HYPOTHESIS. Person is a word-level feature, so in this case the choice needs no clause-level information. The head again follows item features, not roles, and the strong roles-not-words thesis stays OPEN, not established.

On this page: 1 Question · 2 Sources · 3 Rules first · 4 The data · 5 Answer to T72d · 6 The grid, extended · 7 One reason for three bans? · 8 A dependency in /third-voice/ · 9 Rulesets A and B · 10 Derivations · 11 The role schema · 12 Algonquian · 13 Fidelity · 14 Open questions

Notation is the same as on external-head and third-voice. No Syntax · Semantics · Interface conventions file is in the repo or the context folder. A = the word level (prefixal, left-headed). B = the clause level (right-headed). Σ = the role schema, with Dowty's proto-roles as the baseline. Merge builds structure by putting two pieces together. Agree values features such as person and number on a probe from a goal. Move re-merges a piece already in the structure in a higher position. EXT/INT = Willie's external/internal argument. Under VPISH every agent is first merged in Spec,vP, and EXT names where an argument ends up. REL = the relativizing enclitic (=yę́ę, =ígíí, =nę́ę). N = the number of the relative's arguments that have a noun adjoined. π = person value (1, 2, 3, 4). T = the set of the relative's arguments that are on the animacy hierarchy, that is, third persons, including the indefinite agent by EI-10. Willie's number glosses give the object first: "1.3.kissed" = 1st-person object, 3rd-person subject.

1. Question

/external-head/ found that which argument heads a Navajo relative depends on voice and on how many nouns the relative has (EH-8, EH-9). All its data had two third-person arguments. T72d asked whether the same grid holds when one argument is 1st or 2nd person. That is a sharp test, because "the voice alternation can only be used if all arguments are third person" (p. 86). Willie notes on p. 170 that such relatives exist ((41)) but does not discuss which argument heads them.

2. Sources

SOURCED Willie, MaryAnn 1991. Navajo Pronouns and Obviation. PhD dissertation, University of Arizona. Page images read this run (PDF = printed + 6): pp. 170, 171, 172, 173, 175. Crops at the scan's full resolution: (41) on p. 170 and (42)–(43) on p. 171. Rendered from the PDF this run: pp. 122 and 123 (150/200 dpi), for the animacy and EXT statements. Text layer only, as a finding aid: pp. 53, 86, 174, 176.

Site pages that already held some of these forms: relatives §6 (forms of (41), (44)–(59), REL-NR, REL-4-HEAD). That page lists them but does not ask which argument is the head. (42) and (43) are new to the site.

Not opened: Platero 1982, Young & Morgan 1987.

3. Rules first

HP-1 to HP-6 are Willie's statements, quoted from page images unless marked. HP-7 to HP-14 are ours.

HP-1  "Relative clauses may have first or second person arguments that are not     SOURCED (Willie p. 170, image)
      the head of the relative." (41a) sizts'ǫs=yę́ę 'the one who kissed me';
      (41b) séts'ǫs=yę́ę 'the one I kissed'.
HP-2  "When the head of the relative is first or second person heads, the          SOURCED (Willie p. 171, image)
      relative must be non-restrictive, since the head is already maximally
      definite." (42), (43).
HP-3  "Relatives with a fourth person head must also be non-restrictive" (44).     SOURCED (Willie pp. 171–172, image)
      "Fourth person, like first and second person, fixes reference uniquely
      in the discourse and the relative does not pick out a referent."
HP-4  "The fourth person cannot be the head of a restrictive relative; the head    SOURCED (Willie p. 175, image)
      is third person, glossed 'the one'." On (56): "Here the third person
      head of the relative has a noun adjunct."
HP-5  "With first, second, and fourth person patients, it is possible to have a    SOURCED (Willie p. 123, image)
      lower-ranked agent (third person) acting upon a human patient." On (21)
      shishxash, (22) hashxash: "the lower ranked third person agent is the
      subject, the external Nominative argument, and not an internal
      (Ergative) agent, as we see in the Inverse."
HP-6  "Fourth person, like first and second, is definite, human, and not          SOURCED (Willie p. 122, image; p. 86, text)
      involved in the animacy hierarchy" (p. 122). "The voice alternation can
      only be used if all arguments are third person ... first and second
      person arguments are always definite" (p. 86). No noun can be adjoined
      to a 4th person (p. 174) or, for a single 1st/2nd argument, any but a
      free pronoun of the same person and number (p. 53, text layer).

HP-7  (T72d) No. In (41a) N = 0 and REL binds the agent. By HP-5 a 3rd agent on    SOURCED for the reading;
      a 1st patient is EXT. EH-8 row 0 ("N = 0: REL binds INT") and (16) read   HYPOTHESIS that HP-5 (stated for
      as "S or INT" both predict the 1st-person patient, which HP-2 bars from    main clauses with 'bite') carries
      a restrictive reading. The same holds for the 4.3 relatives (45), (49),    over to 'kiss'/'kick' relatives
      (51), (55b), (57), and for (56) with its one noun on the head.
HP-8  Head rule by T (restatement):                                                HYPOTHESIS
        |T| = 0  no restrictive reading; REL binds a 1/2/4 argument
                 non-restrictively ((42), (43), (44))
        |T| = 1  REL binds the member of T, whatever its role or position
                 ((41a), (41b), (45)–(52) where one π = 4, (55), (56), (57))
        |T| = 2  the voice/noun grid EH-8 decides ((18)–(24); EH-8)
HP-9  (16) "heads are absolutives" and EH-8 are therefore the |T| = 2 case. That   HYPOTHESIS
      is exactly the case with a voice alternation (HP-6), so the dependence on
      N and voice belongs to 3/3 clauses only.
HP-10 With |T| = 1 the only argument that can take a noun is the member of T      SOURCED for (56) (HP-4);
      (HP-6), so N ≤ 1 and a noun, if present, is on the head: (56) 'at'ééd     HYPOTHESIS for the generalization
      hazts'ǫs=yę́ę. In the 3/3 Direct relative with one noun the restrictive
      head is the argument without the noun ((22a), EH-5). Whether the noun sits
      on the head or on the other argument also depends on |T|.
HP-11 FIX: a restrictive REL may not bind an argument whose reference is already   HYPOTHESIS
      fixed. Willie gives three reasons that have the same form: 1/2 "maximally
      definite" (HP-2), 4 "fixes reference uniquely" (HP-3), the inverse EXT
      "definite subject, which cannot be subjected to further definiteness
      marking" (p. 164, EH-6). FIX is our merger of the three.
HP-12 (Focus 1) For |T| ≤ 1 the head can be read off the person values of the     HYPOTHESIS
      word's own prefix slots (A). Only |T| = 2 needs B information (N, voice
      choice). Person is valued by Agree inside the word. Nouns are adjoined
      outside it.
HP-13 (Focus 2) Same verb, same two roles, opposite heads: (41a) agent, (41b)     HYPOTHESIS
      patient, and likewise (45) vs (46). What decides is the person value of
      the other argument, a property of the pronominal item, not of Σ.
HP-14 The /third-voice/ case against EI-8 rests on (16) applying to (12)         HYPOTHESIS
      bi'distał=yę́ę. By HP-8 that needs |T| = 2, so it needs the INDF agent
      to be on the hierarchy (EI-10). If EI-10 failed, |T| = 1 and any
      analysis would pick the patient, which would leave EI-8 standing.

4. The data

All forms are printed in Willie pp. 170–175 and were read on page images this run. (41)–(43) were also read in full-resolution crops. "source" lines give Willie's glosses. Our Leipzig lines give her number glosses as O(bject)/S(ubject) and add nothing else. PFV follows the site convention for her past-tense English glosses. Where she prints a translation only, the Leipzig line reuses her gloss of the same word elsewhere, and the note says so.

Willie 1991 p. 170 (41) (page image, crop): one 1st-person argument, N = 0, |T| = 1
(1)siztsʼǫs=yę́ę
source: 1.3.kissed=REL
1O.3S.kiss.PFV=REL
'the one who kissed me' (41a): head = 3rd-person agent
(2)sétsʼǫs=yę́ę
source: 3.1.kissed=REL
3O.1S.kiss.PFV=REL
'the one I kissed' (41b): head = 3rd-person patient
Willie 1991 p. 171 (42)–(43) (page image, crop): 1st/2nd-person head, |T| = 0
(3)shíʼayóíʼánííníshʼníʼígíínaaʼáháshyaadoo
source: I2.1.love=REL2.1.take care ofFUT
1SG(not glossed)2O.1S.love=REL2O.1S.take.care.ofFUT
'I, who love you, will take care of you.' (42)
(4)Ní,ʼayóíʼáshííníʼníʼ=nę́ęshaayisínínah
source: you,1.2.loved =REl1.about2.forgot
2SG(not glossed)1O.2S.love.PFV=REL1.about2S.forget.PFV
'You, who loved me, forgot about me.' (43)

Willie prints no gloss under ʼayóí, and we add none. In (42) she prints the enclitic joined to the verb with no "=" (-ígíí). In (43) she prints "=nę́ę" with a space before "=" in the gloss line and "REl" for REL. We copy the forms and normalize only the gloss.

Willie 1991 pp. 172–173 (45)–(52) (page images): one π = 4 and one π = 3 in the relative, |T| = 1
(5)haztał=yę́ęyiyiiłtsą́
source: 4.3.kicked=RELsource: 3.3.saw
4O.3S.kick.PFV=REL3O.3S.see.PFV
'He saw the one who kicked that person.' (45): head = 3rd-person agent
(6)jiztał=yéeyiyiiłtsą́
source: 4.3.kicked=REL [sic]source: 3.3.saw
3O.4S.kick.PFV=REL3O.3S.see.PFV
'He saw the one who that person kicked.' (46): head = 3rd-person patient
(7)haztał=yę́ęjiiłtsą́
source: 4.3.kicked=RELsource: 3.4.saw
4O.3S.kick.PFV=REL3O.4S.see.PFV
'That personi saw the one who kicked himi.' (49): head = 3rd-person agent
(8)jiztał=yę́ęjiiłtsą́
source: 3.4.kicked=RELsource: 3.4.saw
3O.4S.kick.PFV=REL3O.4S.see.PFV
'The onei that personj kicked, that personj saw himi.' (50): head = 3rd-person patient

(51) haztał=yę́ę hoołtsą́ 'That one who kicked that person saw that person' and (52) jiztał=yę́ę hoołtsą́ 'That one that person kicked saw that person' repeat the split with a 4O.3S main verb. (46) prints "4.3.kicked" for jiztał, which is glossed "3.4" in (50) and (52), and its enclitic without nasal hooks. Both are already flagged on /relatives/ (T14c). We follow the translation.

Willie 1991 p. 175 (55)–(57) (page image): |T| = 1 with a noun
(9)jiztsʼǫs=yę́ęʼatʼéédyiyiiłtsą́
3O.4S.kiss.PFV=RELgirl3O.3S.see.PFV
'The one that person kissed saw the girl.' (55a): head = 3rd-person patient; the noun belongs to the main clause
(10)haztsʼǫs=yę́ęʼatʼéédyiyiiłtsą́
4O.3S.kiss.PFV=RELgirl3O.3S.see.PFV
'The one who kissed that person saw the girl.' (55b): head = 3rd-person agent
(11)ʼatʼéédhaztsʼǫs=yę́ęyiyiiłtsą́
girl4O.3S.kiss.PFV=REL3O.3S.see.PFV
'He saw the girl who kissed that person.' (56): head = 3rd-person agent, which has the noun
(12)ʼatʼéédhaztsʼǫs=yę́ęjiiłtsą́
source: girlsource: 4.3.kissed=RELsource: 3.4.saw
girl4O.3S.kiss.PFV=REL3O.4S.see.PFV
'That personi saw the girl that kissed himi.' (57)

Willie glosses only (57) on this page. The Leipzig lines of (55) and (56) reuse her gloss of haztsʼǫs=yę́ę in (57), of yiyiiłtsą́ in (45), and "3.4" for ji- subject forms as in (50). "Head =" labels come from her translations and her prose on p. 175 (HP-4).

5. Answer to T72d

No: the noun-count grid does not carry over. In (1) = (41a) there is no noun (N = 0) and no voice choice, and the head is the agent: 'the one who kissed me'. Willie says the third-person agent in a 3-on-1 clause is "the external Nominative argument, and not an internal (Ergative) agent" (p. 123, HP-5) SOURCED. So an N = 0 relative binds EXT. The N = 0 row of EH-8 ("REL binds INT") predicts the 1st-person patient, and so does (16) read as "S or INT". But a 1st-person head forces a non-restrictive reading (HP-2), and (41a) has the restrictive 'the one who'. (2) = (41b) has the same verb, the same N and the patient as head.

The 4.3 relatives in §4 ((45), (49), (51), (55b), (56), (57)) give the same result for fourth person, and p. 123 states the EXT status for the 4th-person patient case explicitly ((22) hashxash). The 3.4 relatives ((46), (50), (52), (55a)) are headed by the 3rd-person patient. Willie's own summary for them is "the head is third person" (p. 175). So the head is the third-person argument whatever its role and its EXT/INT position (HP-7, HP-8) HYPOTHESIS. The hedge: p. 123 states the EXT status for main clauses with 'bite'. Carrying it over to relatives with 'kiss' and 'kick' is ours, though nothing in ch. 6 suggests otherwise.

What survives. EH-8 is not refuted. It is limited. All its data are 3/3, and 3/3 is the only configuration with a voice alternation (p. 86). So the noun count and voice decide the head only when two third persons compete (HP-9) HYPOTHESIS. Willie's (16) was stated over "the relative clauses we have seen so far" (p. 160), all 3/3, so she did not claim more.

6. The grid, extended

|T|ConfigurationWhat heads the restrictive relativeExamplesStatus
01/2 or 4 only (incl. intransitive 4)no restrictive reading; the 1/2/4 argument heads a non-restrictive one(42), (43), (44)SOURCED pp. 171–172
11, 2 or 4 + one 3rd personthe 3rd person, agent or patient, EXT or INT; a noun can go only on it(41a), (41b), (45)–(52), (55)–(57)readings SOURCED; rule HYPOTHESIS
23/3, Direct or InverseEH-8: N = 0 → INT; N = 2 → EXT; N = 1 Direct → the argument without a noun; N = 1 Inverse → non-restrictive only(18)–(24)cells SOURCED; grid HYPOTHESIS

The |T| = 1 row has no voice column because there is no voice choice. It has no N column either, because N can only be 0 or 1 (on the head itself, (56)). Compare (56) ʼatʼééd haztsʼǫs=yę́ę, where the one noun is on the head, with the 3/3 restrictive reading of (22a) ʼashkii yiztsʼǫs=yę́ę 'the one who kissed the boy', where it is on the other argument (HP-10).

7. One reason for three bans? (FIX)

Willie gives three separate explanations for why something cannot head a restrictive relative. 1st and 2nd persons are "already maximally definite" (p. 171). The 4th person "fixes reference uniquely" (p. 171). The definite inverse subject "cannot be subjected to further definiteness marking by means of a restrictive relative" (p. 164) SOURCED. HP-11 reads these as one condition: a restrictive relative picks out a referent, so it cannot bind an argument whose referent is already fixed HYPOTHESIS. FIX predicts the |T| = 0 row, the bar on the 1/2/4 argument in the |T| = 1 row, and *(24b).

FIX sharpens T72a but does not settle it. FIX says (20b) ʼatʼééd ʼashkii biztsʼǫs=yę́ę, which binds the definite inverse EXT, should be non-restrictive, while its gloss has no commas. Either the commas are missing, or a noun of its own lifts FIX for EXT. Willie does not say which OPEN.

8. A dependency in /third-voice/

The argument on third-voice and external-head §7 against EI-8 (agentive passive = transitive inverse) uses (16) on (12) biʼdistał=yę́ę 'the one who was kicked' (p. 152 (1f)). By HP-8, (16) applies only when |T| = 2. The patient is 3rd person, and the INDF agent counts only if it is on the animacy hierarchy, which is EI-10 on ergative-indefinite (HYPOTHESIS, derived, never stated by Willie). If EI-10 is right, |T| = 2 and the argument stands as before. If EI-10 is wrong, |T| = 1, every analysis picks the patient, and the relative no longer counts against EI-8 (HP-14) HYPOTHESIS. The EI-8 case therefore leans on EI-10 more heavily than the earlier pages showed. It is now T73d.

9. Rulesets A and B (Focus 1)

StepLevelRulesetWhat happens
1vPBMerge patient and verb, then the agent in Spec,vP (VPISH).
2clauseBIf both are π = 3, choose the voice by the animacy ranking (p. 86 (70)) and Move the right one to EXT. If one is 1/2/4, there is no choice: the agent goes to EXT (HP-5).
3wordAAgree values the object and subject slots for person and number. Spell-out gives the verb word (siztsʼǫs, haztał, …).
4word edgeA→BREL is merged at the right edge. If |T| ≤ 1 its target can be computed from the slot values already set in step 3. If |T| = 2 it needs step 5's noun count (EH-8).
5clauseBNouns are adjoined to the left, only to π = 3 slots (HP-6). The relative is adjoined to a main-clause argument (p. 158).

This refines the changeover point that /external-head/ placed at REL HYPOTHESIS. REL's target depends on clause-level (B) information only when two third persons compete. In the other configurations everything REL needs is inside the word (A). It does not decide whether REL is a word-level item or a clause-level head spelled out in A (relative-cp). It does show the dependence on B is limited to one configuration, the 3/3 one, which is also the only one where the B-level voice choice exists.

10. Derivations

Each step names its operation and ruleset. Merge builds structure, Agree copies person and number onto a probe, and Move re-merges a piece higher up. The forms are Willie's. The structure is ours HYPOTHESIS. We do not segment the verb words. Spell-out gives Willie's printed form.

D1  sizts'ǫs=yę́ę 'the one who kissed me' (41a), |T| = 1, N = 0        ← answers T72d
    1  Merge  [VP pro(1,P) √ts'ǫs]; Merge pro(3,A) in Spec,vP            B   VPISH
    2  Move   pro(3,A) → EXT (no voice choice: π(P) = 1)                 B   HP-5, HP-6
    3  Agree  object slot ← [1SG]; subject slot ← [3]                    A
    4  Spell  sizts'ǫs                                                   A
    5  Merge  REL; T = {pro(3,A)} → binds EXT agent                     A→B HP-8 row 1
    6  Check  EH-8 row 0 (N = 0 → INT) would bind pro(1,P): ✗ by HP-2
              (restrictive 'the one who' needs a non-fixed head)

D2  séts'ǫs=yę́ę 'the one I kissed' (41b), |T| = 1, N = 0
    1  Merge  [VP pro(3,P) √ts'ǫs]; Merge pro(1,A) in Spec,vP            B
    2  Move   pro(1,A) → EXT                                             B   HP-6 (no voice)
    3  Agree  object slot ← [3]; subject slot ← [1SG]                    A
    4  Spell  séts'ǫs                                                    A
    5  Merge  REL; T = {pro(3,P)} → binds INT patient                   A→B HP-8 row 1
              (here EH-8 row 0 gives the same answer, by coincidence)

D3  haztał=yę́ę yiyiiłtsą́ 'He saw the one who kicked that person' (45)
    1  Merge  [VP pro(4,P) √tał]; Merge pro(3,A) in Spec,vP              B
    2  Move   pro(3,A) → EXT (π(P) = 4: off the hierarchy)               B   HP-5, HP-6
    3  Agree  object slot ← [4]; subject slot ← [3]                      A
    4  Spell  haztał                                                     A
    5  Merge  REL; T = {pro(3,A)} → binds EXT agent                     A→B HP-4, HP-8
    6  Adjoin [haztał=yę́ę] to a 3rd-person slot of yiyiiłtsą́            B   p. 158; p. 175
    7  π = 4 in the relative only → disjoint from the main-clause args   B   REL-4 (pp. 173–174)

D4  'at'ééd hazts'ǫs=yę́ę yiyiiłtsą́ 'He saw the girl who kissed that person' (56)
    1–4 as D3 with √ts'ǫs: hazts'ǫs                                     B, A
    5  Adjoin 'at'ééd: only π = 3 slots take nouns → the agent           B   HP-6
    6  Merge  REL; T = {pro(3,A)} → binds the agent, which has the noun  A→B HP-4, HP-10
    7  Adjoin the complex relative to the patient of yiyiiłtsą́           B   Willie p. 175

D5  shí 'ayóí 'áníínísh'ní'ígíí naa'áháshyaa doo (42), |T| = 0
    1  Merge  [VP pro(2,P) ...]; Merge pro(1,A) in Spec,vP               B
    2  Move   pro(1,A) → EXT                                             B
    3  Agree  object ← [2SG], subject ← [1SG]; Spell the verb word       A
    4  Merge  REL; T = ∅ → no restrictive reading; REL binds pro(1)     A→B HP-2, HP-11
              non-restrictively
    5  Adjoin [shí] (free pronoun, same φ) and the relative to the       B   HP-6 (p. 53, text)
       1st-person subject of naa'áháshyaa doo

11. The role schema (Focus 2)

/external-head/ showed that the relative's head changes with voice and with the number of nouns. This page adds a third factor: the person of the other argument. With the same verb, (41a) is headed by the agent and (41b) by the patient. The only difference is which slot is 1st person. (45) and (46) give the same contrast with 4th person (HP-13) HYPOTHESIS. Person belongs to the pronominal items, the arguments themselves, not to the proto-role schema. Nothing in Σ (agent-like vs patient-like) chooses the head here.

This is evidence against using relativization to argue that Navajo verbs take roles rather than lexical items as arguments. It is not evidence against the weak schema thesis, that roles fix slot order inside the word (VC-15): the 1st-person patient in (41a) and the 1st-person agent in (41b) still sit in the object and subject slots respectively. The strong thesis stays OPEN and not established. No chemistry analogy is drawn.

12. Algonquian

Algonquian is the obvious comparison, because there too 1st/2nd persons stand outside the 3/3 direct–inverse contrast (Oxford, Routledge handbook chapter, text layer opened on ergative-indefinite: the Ojibwe theme-sign hierarchy 1/2 > X > 3 > 3′ > 0). Whether Algonquian relatives or participles also take their head from the only third person when a 1st/2nd argument is present was not checked this run. No Algonquian source was opened, so this is an OPEN question (T73e), not a parallel.

13. Fidelity

  • All twelve numbered forms in §4, and (51), (52), were read on Willie's page images this run. (41)–(43) were also read in full-resolution crops. (42) and (43) are new to the site.
  • Willie glosses (41), (42), (43), (45)–(52), (57). The Leipzig lines for (55) and (56) reuse her glosses of the same words. ʼayóí is left unglossed as she leaves it.
  • The "head =" labels are drawn from her translations ('the one who kissed me', 'the one I kissed', …) and from p. 175 ("the head is third person").
  • HP-5's "external Nominative" is printed for (21) shishxash and (22) hashxash 'bite', in main clauses. Its use for the relatives is our extension (HP-7).
  • Willie never compares (41a) with (16) or (21). The conflict between them, and the |T| restatement, are ours.
  • (46)'s "4.3" and its =yée are printed that way. The translation is followed, as on /relatives/ (T14c).
  • Pp. 53, 86, 174 and 176 were used as text layer only. p. 86 was read on images in an earlier run (sources, 2026-10-28 12:46pm).

14. Open questions

IDQuestionWhat would settle it
T72dDoes the noun-count dependence hold with 1st/2nd-person arguments?Answered: no. With one third person the head is that third person, whatever its role or position ((41a), (41b), (45)–(57)) SOURCED readings. The grid holds for |T| = 2 only (HP-8, HP-9) HYPOTHESIS.
T72a(20b): restrictive or not?Still OPEN. FIX (HP-11) predicts non-restrictive.
T73aIs the 3rd agent of a 3-on-1 relative EXT, as HP-5 says for main clauses? Any ch. 3 or ch. 6 statement, or a diagnostic (definiteness, topic) applied to (41a).Willie ch. 3 §3.3 on images; Hale 1997. OPEN Update 2026-10-29 ~8:46pm PT: answered as HYPOTHESIS on agent-ext (XA-8).
T73bA 1st/2nd-person relative with a noun on the third person, [N siztsʼǫs=REL] (schematic, not attested): does the noun go on the head, as (56) does with 4th person?A sourced example (Young & Morgan 1987; Platero 1974/1982). OPEN
T73cIs FIX (HP-11) one condition or three? Test: a 3/3 relative whose head is fixed by a demonstrative or a proper name.A sourced example with judgments. OPEN
T73dThe case against EI-8 now rests on EI-10 (INDF on the hierarchy) (HP-14). Any direct evidence that INDF counts for the animacy ranking?Willie pp. 36–38, 152–157 on images; Young & Morgan 1987. OPEN
T73eAlgonquian: with a 1st/2nd argument, does a relative or participle take its head from the only third person?Oxford's chapter on page images, or a reference grammar of Algonquian participles (not yet identified). OPEN

Sources: Willie 1991 pp. 122, 123, 170–173, 175 (page images); pp. 53, 86, 174, 176 (text layer). See also external-head, relatives, third-voice, ergative-indefinite, relative-cp, open questions, sources.

"The one who kissed me" keeps an external agent: a Navajo relative is an independent clause plus an enclitic, and Willie's commas are no test of restrictiveness (T73a)

Working note. Kept as written (URL, data and status labels unchanged). Filed under The fourth person; the consolidated argument is the essay fourth-person. Rule IDs: rules index.

Verbs on this page → dictionary: hashshish ‘sting’ · yi-dloh ‘laugh’

Restated on the unified structure (2026-11-15). Relative clauses are restated as a full CP, built like a main clause, under an overt D⁰ (=yę́ę, =ígíí) linked to one participant by Agree; nothing moves out (H27b, H28′, H31, H34). 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, relative rows. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.

T73a asked whether the third-person agent of a relative like (41a) siztsʼǫs=yę́ę 'the one who kissed me' is the external argument (EXT), as Willie says it is in a 3-on-1 main clause (p. 123). Willie never says so for relatives. Her own construction rule does carry it over: "the relative clause in Navajo is constructed by enclitizing to an indepedent clause a relativizing determiner" (p. 181). If the relative is an ordinary clause plus REL, its agent is EXT just as in the main clause HYPOTHESIS. Nothing on the definiteness side blocks this, because a Direct EXT agent can be indefinite (p. 126, (26)). Restrictiveness cannot decide the question either way, and on p. 163 Willie prints a relative she calls non-restrictive without the commas that are supposed to mark one. That takes away the comma half of the T72a tension.

On this page: 1 Question · 2 Sources · 3 Rules first · 4 The data · 5 Answer to T73a · 6 Commas are not a test (T72a) · 7 Three statements of the inverse ban · 8 Side result for T73d · 9 Rulesets A and B · 10 Derivations · 11 The role schema · 12 Fidelity notes · 13 Open questions

Notation is the same as on person-head and external-head. No Syntax · Semantics · Interface conventions file is in the repo or the context folder. A = the word level (prefixal, left-headed). B = the clause level (right-headed). Σ = the role schema, with Dowty's proto-roles as the baseline. EXT/INT = Willie's external/internal argument. Merge builds structure: it puts two pieces together into a bigger one. Agree builds nothing. It copies feature values such as person and number from one place to another. Move is Merge again of a piece that is already in the structure.

1. Question

/person-head/ showed that with a 1st, 2nd or 4th person in the relative, the head is the only third-person argument, whatever its role (HP-8). In (41a) that is the agent. HP-5 quotes Willie p. 123: in a 3-on-1/2/4 main clause the 3rd-person agent is "the external Nominative argument". T73a asked whether that also holds inside the relative. It matters because /external-head/ found that, when a relative has no noun, REL binds the internal argument (EH-8 row 0). If the agent of (41a) were EXT, (41a) would be a relative with no noun whose REL binds EXT, which looks like a problem for EH-8 unless person decides first (HP-8). If it were INT, as in an inverse, then (41a) would have to be some kind of inverse, and Willie says the voice alternation needs all arguments to be third person (p. 86).

2. Sources

SOURCED Willie, MaryAnn 1991. Navajo Pronouns and Obviation. PhD dissertation, University of Arizona. Page images rendered from the PDF and read this run: pp. 50–51 (PDF 55–56; offset is printed + 5 here), 125–126 (with a 300-dpi crop of (26)), 156–157, 160–161, 163, 181, 226–227 (PDF = printed + 6). The p. 170 crop of (41) was re-read this run. P. 123 (HP-5) is quoted from the page image read in the 12:54pm run.

Already on the site and not repeated as new: (26) séígoʼ hashshish with an F3 derivation (fourth-person); (19) and (20b) (external-head); (5), (7), (8), (10) (relatives); the pp. 50–51 passage, quoted on pleonastic-psych but not tied there to EI-10. New to the site: the p. 181 construction rule and its line "Inverse relative clauses are also non-restrictive"; the p. 227 restatement; the p. 125–126 linking passage; the p. 163 comma convention set beside (5).

Not opened: Platero 1982; Young & Morgan 1987; Willie 1990 (cited on p. 226).

3. Rules first

XA-1 to XA-7 are Willie's statements, quoted from page images. XA-8 to XA-13 are ours.

XA-1  "the relative clause in Navajo is constructed by enclitizing to an        SOURCED (Willie p. 181, image)
      indepedent [sic] clause a relativizing determiner which binds an
      argument of the relativized verb." P. 226: relatives are "'internally'
      headed, which means that there is no gapping".
XA-2  (= HP-5) With a 1/2/4 patient, the 3rd-person agent is "the external       SOURCED (Willie p. 123, image, earlier run)
      Nominative argument". "An external agent is not as likely to be new
      information as an internal agent, and can be a discourse topic; it can
      be presuppositional."
XA-3  "the animacy hierarchy is closely tied to the fact that external           SOURCED (Willie pp. 126, 163, images + crop)
      arguments in Inverse constructions must be definite. This limitation
      does not apply to any Direct construction, including fourth person
      constructions. Therefore, it is possible to have an indefinite, lower
      ranked agent acting upon a human patient in a Direct construction by
      using the fourth person." Example (26).
      Inside a relative: for a Direct 3/3 relative with one noun, "On the first
      reading, the preferred one, the relative is restrictive. The head of the
      relative is the external argument, the agent that has no adjoined
      nominal" (p. 163, image; = EH-5).
XA-4  "When a transitive sentence contains a first, second or fourth person      SOURCED (Willie pp. 125–126, images)
      argument and a third person argument, a nominal may be adjoined only to
      the third person argument. This means that no principles of
      interpretation involving word order and voice alternations are
      necessary when a 1, 2 or 4 person argument is present. There is no
      question about the theta role of the third person argument, since the
      theta role of the 1, 2 or 4 argument is clearly marked in the pronoun
      incorporated into the verb. This means that the single nominal present
      in the utterance can be linked to either an external or an internal
      argument; this is not the case with third person constructions, where a
      single nominal is always linked to the internal argument."
XA-5  "I indicate this difference in the type of relative by placing commas      SOURCED (Willie p. 163, image)
      around the non-restrictive relative in the English gloss". The same
      page then repeats (5) 'You who were laughing are now crying.' with no
      commas, and says of it: "we saw both a free pronoun and a
      non-restrictive relative serving as adjuncts to the main clause pronoun."
XA-6  Three statements of the inverse restriction:                               SOURCED (Willie pp. 164, 181, 227, images)
      (a) p. 164 (= EH-6): "The Inverse relative with one nominal has only
          one reading" (*(24b));
      (b) p. 181: "Inverse relative clauses are also non-restrictive."
      (c) p. 227: "Third person arguments can have either restrictive or
          non-restrictive relatives -- except for the subjects of Inverse
          constructions, which can have only non-restrictive relatives, in
          keeping with their definite status."
XA-7  Two classes of argument prefixes: "the prefixes that mark definite          SOURCED (Willie pp. 50–51, images)
      referential arguments: first, second, third and fourth person; also
      included here are the anaphors (reflexive and reciprocal)", and "the
      prefixes that mark indefinite arguments: the 'adi that appears in the
      'Agentive Passive', the Impersonal/indefinite -'a, and the pleonastic
      ho-." "the definite pronouns permit nominal adjuncts (including free
      pronouns), while the indefinite pronouns exclude adjoined nominals."

XA-8  ANSWER to T73a. The 3rd-person agent of a 3-on-1/2/4 relative such as       HYPOTHESIS (from XA-1 + XA-2)
      (41a) is EXT. By XA-1 the relative is an independent clause with REL
      added. Nothing in XA-1 changes the clause, so XA-2 applies inside it.
XA-9  No clash with definiteness. A restrictive head needs a reference that       HYPOTHESIS (from XA-3, HP-11)
      the relative still has to pick out (FIX, HP-11). A Direct EXT agent need
      not be definite (XA-3), and Willie herself lets a Direct EXT agent head
      a restrictive relative (p. 163). So an EXT agent can head the
      restrictive 'the one who' of (41a).
XA-10 The diagnostic is blind. An inverse INT agent can also be indefinite        HYPOTHESIS
      (EI-3) and head (19) 'The one who kissed him'. So "can head a
      restrictive relative" holds of an EXT agent (XA-9) and of an INT agent
      alike, and cannot tell which (41a) has. XA-8 rests on XA-1, not on a
      test.
XA-11 Commas in Willie's English glosses are a sufficient sign of a               HYPOTHESIS (from XA-5)
      non-restrictive relative, not a necessary one. A gloss without commas
      is not evidence of a restrictive reading.
XA-12 Of XA-6 (a)–(c), the narrowest statement that fits every example read      HYPOTHESIS (OPEN which reading
      is: a restrictive REL may not bind an inverse EXT. (a) and (c) are           Willie intends)
      instances of it. (b) also fits if read the same way, and the broad
      reading of (b) ("every inverse relative") also survives, given XA-11.
XA-13 Willie's word "external" is not used uniformly. On p. 156 she calls the     SOURCED (Willie p. 156, image) for the
      argument a relative is adjoined to "the external argument" for both (7)     wording; HYPOTHESIS that it is loose
      and (8), but in (7) the relative is the patient of a Direct verb, which     usage
      p. 123 would make internal.

4. The data

All forms were read on page images this run. "source" lines give Willie's glosses. Our Leipzig lines turn her number glosses into O(bject)/S(ubject) and add nothing else. PFV follows the site convention for her past-tense English glosses. We do not segment the verb words.

Willie 1991 p. 170 (41a) (page image, crop): 3rd agent on a 1st patient, N = 0
(1)siztsʼǫs=yę́ę
source: 1.3.kissed=REL
1O.3S.kiss.PFV=REL
'the one who kissed me' (41a)
Willie 1991 p. 126 (26) (page image, 300-dpi crop): indefinite Direct agent on a 4th-person patient, main clause
(2)séígoʼhashshish
source: scorpion4-3-sting
scorpion4O.3S.sting.PFV
'A scorpion stung her.' (26)
Willie 1991 p. 161 (19) (page image): inverse relative headed by its INT agent, N = 0
(3)biztsʼǫs=yę́ęyóóʼeelwod
source: INV3.3.kissed=RELsource: 3-ran away
INV.3O.3S.kiss.PFV=REL3S.run.away.PFV
'The one who kissed him ran away.' (19); Willie: "this is because the agent is the internal argument in the Inverse construction."
Willie 1991 p. 163 (5) (page image): a relative Willie calls non-restrictive, printed without commas
(4)nínidloh=yę́ękʼadnicha
source: you2.laugh=RELnow2.cry
2SG2S.laugh=RELnow2S.cry
'You who were laughing are now crying.' (5) ("Go on, laugh some more.")
Willie 1991 p. 156 (7)–(8) (page image): a relative adjoined to the main-clause patient vs agent
(5)yidloh=yę́ęyidootsʼǫs
source: 3.laughing=REL3.3.will kiss
3S.laugh=REL3O.3S.kiss.FUT
'He will kiss the one who was laughing.' (7)
(6)yidloh=yę́ębidootsʼǫs
source: 3.laughing=RELINV 3.3.will kiss
3S.laugh=RELINV.3O.3S.kiss.FUT
'The one who was laughing will kiss him.' (8)

FUT renders Willie's "will". The gloss of (5)–(6) uses "3.laughing", which we give as 3S.laugh with no aspect label, since her English has "was laughing".

5. Answer to T73a

  1. Willie builds a relative by adding an enclitic to an independent clause, with no gap (XA-1) SOURCED. In an independent 3-on-1 clause the 3rd agent is EXT (XA-2) SOURCED.
  2. So the agent of (41a) is EXT HYPOTHESIS (XA-8). This uses one premise beyond Willie's words: that adding REL does not re-sort the arguments of the clause it attaches to. Nothing read so far says that it does.
  3. The obvious objection is definiteness. Inverse EXTs must be definite and cannot head a restrictive relative (EH-6, XA-6c), and (41a) is glossed as restrictive ('the one who'). The objection fails. The definiteness requirement is tied to the Inverse, and "does not apply to any Direct construction" (XA-3) SOURCED. (26) 'A scorpion stung her' shows an indefinite EXT agent over a 4th-person patient in a main clause. Better still, Willie herself says that in a Direct 3/3 relative with one noun, on the preferred restrictive reading, "The head of the relative is the external argument, the agent" (p. 163) SOURCED. So a restrictive relative headed by a Direct EXT agent is already in her analysis, and (41a) only adds a 1st-person patient (XA-9) HYPOTHESIS. The same passage shows she keeps the EXT/INT structure inside relatives, which is the premise XA-8 needs.
  4. The test that would have decided it directly is not available. Whether the head is EXT or INT, the restrictive reading is possible (XA-10). An inverse INT agent heads (19) too. The answer to T73a is therefore architectural (XA-1), not a diagnostic result.
  5. Consequence for EH-8: (41a) is a noun-less relative whose REL binds EXT. That is not a counterexample to EH-8 row 0, because EH-8 only applies once person has failed to decide (HP-8: |T| = 2). This is the same conclusion as /person-head/, now with the head's position fixed as well as its role HYPOTHESIS.

XA-4 gives an independent reason why no inverse reading is needed in (41a). With a 1/2/4 argument "no principles of interpretation involving word order and voice alternations are necessary", and a single nominal "can be linked to either an external or an internal argument" SOURCED. With two third persons, by contrast, "a single nominal is always linked to the internal argument". That is EH-8 row 1 for nouns in main clauses. So Willie already states, for nouns, the person-first logic that HP-8 states for REL HYPOTHESIS that the two are one rule.

6. Commas are not a test (T72a)

T72a (external-head) was the tension that (20b) ʼatʼééd ʼashkii biztsʼǫs=yę́ę yóóʼeelwod binds the definite inverse EXT, which by EH-6 and FIX should be non-restrictive, yet its gloss has no commas. On p. 163 Willie announces the convention, "placing commas around the non-restrictive relative in the English gloss", and on the same page repeats (5) ní nidloh=yę́ę kʼad nicha 'You who were laughing are now crying.' with no commas. She then calls it "a non-restrictive relative" SOURCED. The head is 2nd person, so by HP-2 it cannot be restrictive anyway.

So a gloss without commas is not evidence of a restrictive reading (XA-11) HYPOTHESIS. The comma half of T72a falls away. (20b) can be non-restrictive, as FIX predicts, without contradicting the print. T72a is partly answered. Willie still never states which reading (20b) has, so a positive answer is still OPEN.

The same caution applies to (19) and (41a). Their glosses say 'the one who', and Willie uses 'the one' for a restrictive third-person head (HP-4, p. 175). That is the evidence that they are restrictive, not the lack of commas.

7. Three statements of the inverse ban

WhereWordingScope, read literallyFits (19) as restrictive?
p. 164 (EH-6)"The Inverse relative with one nominal has only one reading"inverse relatives with N = 1yes (N = 0)
p. 181"Inverse relative clauses are also non-restrictive."every inverse relativeno
p. 227"except for the subjects of Inverse constructions, which can have only non-restrictive relatives"a relative on the inverse EXTyes (head = INT agent)

Read literally, p. 181 rules out a restrictive (19), and (19) is glossed 'The one who', Willie's restrictive form (HP-4). Read like p. 227, p. 181 says only that a restrictive REL may not bind an inverse EXT. That narrower reading covers *(24b) (EH-6) and (20b) (now non-restrictive by §6), and it leaves (19) alone. It is also what FIX (HP-11) predicts, because only the inverse EXT has its reference fixed (XA-12) HYPOTHESIS. Which reading Willie intends is OPEN (T74b).

Two further readings of p. 227 are possible. "Subjects of Inverse constructions" could mean the EXT inside the relative (as in (20b)) or the main-clause argument the relative is adjoined to (as the 'him' of (8)). The examples read do not separate them. P. 156 does not help, because there "external" covers the patient of the Direct (7) too (XA-13).

8. Side result for T73d

T73d asks for direct evidence that INDF, the ʼadi- agent of the agentive passive, counts for the animacy ranking. EI-10 says it does, and gives two supports. One is Willie's gloss "PASS 3.3" (p. 152, and "PASS.3.3" on p. 157 (10)), which files the agent as a 3. The other is EI-4, "only third person arguments can be indefinite".

Pp. 50–51 (XA-7) weaken the second support SOURCED for the text. Willie lists "first, second, third and fourth person" as prefixes for definite referential arguments, and lists ʼadi separately, among prefixes for indefinite arguments that "exclude adjoined nominals". On this taxonomy the third person that "can be indefinite" (EI-4) is a third-person pronoun whose adjoined nominal is indefinite, like séígoʼ in (26), which p. 126 calls "an indefinite, lower ranked agent". It is not ʼadi. So EI-4 does not by itself put ʼadi on the ranking HYPOTHESIS.

The first support stands: her own gloss of (10) bidiʼdootsʼǫs is "PASS.3.3". The two pages pull opposite ways. T73d stays OPEN, with one leg of EI-10 weaker. EI-10 keeps its HYPOTHESIS label; it is not withdrawn.

9. Rulesets A and B (Focus 1)

StepLevelRulesetWhat happens in (41a)
1vPBMerge patient and verb, then the agent in Spec,vP (VPISH).
2clauseBπ(P) = 1, so there is no voice choice (p. 86). The agent goes to EXT (XA-2).
3wordAAgree values the object slot [1SG] and the subject slot [3]. Spell-out gives siztsʼǫs. Its form is identical to that of the main-clause verb, because the relative is the independent clause (XA-1).
4word edgeA→BREL is merged at the right edge and binds the only π = 3 argument (HP-8). Its position (EXT) was already fixed in step 2.

For Focus 1 this narrows the changeover again HYPOTHESIS. Every B-level decision about the relative's own arguments (EXT/INT, voice) is made before the word is spelled out in A, and REL only reads the result. What REL adds is the B-level outer relation: adjunction to a main-clause argument, and the restrictive/non-restrictive choice. XA-4 says the same for nouns: with a 1/2/4 argument the word (A) settles every role, and the B-level principles of "word order and voice alternations" are idle.

10. Derivations

Each step names its operation and ruleset. Merge builds structure, Agree copies person and number onto a probe, and Move re-merges a piece higher up. The forms are Willie's. The structure is ours HYPOTHESIS.

D1  sizts'ǫs=yę́ę 'the one who kissed me' (41a)                          ← answers T73a
    1  Merge  [VP pro(1,P) √ts'ǫs]; Merge pro(3,A) in Spec,vP            B   VPISH
    2  Move   pro(3,A) → EXT; no voice choice (π(P) = 1)                 B   XA-2, p. 86
    3  Agree  object slot ← [1SG]; subject slot ← [3]                    A
    4  Spell  sizts'ǫs  (= the independent clause)                       A   XA-1
    5  Merge  REL; T = {pro(3,A)} → binds the EXT agent                 A→B HP-8, XA-8
    6  Check  restrictive OK: Direct EXT need not be definite            B   XA-3, XA-9

D2  séígo' hashshish 'A scorpion stung her.' (26), main clause
    1  Merge  [VP pro(4,P) √...]; Merge pro(3,A) in Spec,vP              B
    2  Move   pro(3,A) → EXT; no voice choice (π(P) = 4)                 B   XA-2
    3  Agree  object slot ← [4]; subject slot ← [3]; Spell hashshish     A
    4  Adjoin séígo' to the only π = 3 slot, the EXT agent               B   XA-4
    5  séígo' is indefinite on an EXT: allowed in a Direct              B   XA-3

D3  bizts'ǫs=yę́ę yóó'eelwod (19), for comparison
    1  Merge  [VP pro(3,P) √ts'ǫs]; Merge pro(3,A) in Spec,vP            B
    2  Move   pro(3,P) → EXT (Inverse); agent stays INT                  B   p. 161
    3  Agree  both slots ← [3]; INV marks the object; Spell bizts'ǫs     A
    4  Merge  REL; |T| = 2, N = 0 → binds INT = the agent               A→B EH-8 row 0
    5  Restrictive OK: INT agent is not fixed                            B   XA-10, XA-12
    6  Adjoin to the subject of yóó'eelwod                               B

D4  schematic only (no attested form): [N  1O.3S.V=REL]                   ← bears on T73b
    1–5 as D1
    6  Adjoin N: only the π = 3 slot takes a nominal → N on the head     B   XA-4 (main clauses)

D4 uses XA-4, which Willie states for main clauses, inside a relative via XA-1. No Navajo form with a noun in a 1st/2nd-person relative has been read. T73b stays OPEN.

11. The role schema (Focus 2)

The question here is whether verb arguments are better described as positions in a role schema Σ (proto-agent / proto-patient) than as lexical items. Two observations from this run bear on it.

  • In XA-4 Willie says the role of the third-person argument is "no question" because "the theta role of the 1, 2 or 4 argument is clearly marked in the pronoun incorporated into the verb" SOURCED. The noun gets its role by elimination from the slots on the verb word. Roles live in slots, and nouns are attached to slots, not selected as items. That is consistent with Σ, but it is equally consistent with an ordinary lexical verb whose two argument slots are pronominal HYPOTHESIS.
  • What decides the head of (41a) and its position is person, a feature of the item filling a slot, not the role (HP-8, XA-8). Σ alone does not predict which slot REL binds.

So the strong thesis (theta roles, not lexical items, are the verb's arguments) gets no new support and no refutation here. It stays OPEN. Grant's own revised view, that the strong version does not hold cleanly, is GRANT'S CONJECTURE. Nothing on this page settles it.

12. Fidelity notes

  • All Navajo is copied from Willie's pages as printed, including séígoʼ and hashshish from a 300-dpi crop. No form or judgment is ours. D4 is schematic.
  • Willie never says, for a relative with a 1st/2nd-person argument, that the agent of (41a) is EXT; she says it only for the Direct 3/3 relative of p. 163. XA-8 is our inference from XA-1 and XA-2.
  • "indepedent" on p. 181 is Willie's typo, kept with [sic].
  • The p. 163 comma point uses Willie's own description of (5) as non-restrictive. That (20b) is non-restrictive is still our inference (FIX).
  • P. 157 glosses (10) "The who was laughing will be kissed." We do not repair the gloss and do not use (10) beyond its "PASS.3.3" line.
  • Leipzig lines add O/S and PFV/FUT only. Willie's own glosses stay on the "source" lines.
  • XA-8 to XA-13 and D1–D4 are ours. XA-1 to XA-7 are quotations.

13. Open questions

IDQuestionWhat would settle it
T73aIs the 3rd agent of a 3-on-1 relative EXT?Answered as HYPOTHESIS (XA-8), from Willie pp. 181 + 123. Not stated by Willie for relatives.
T74aIs there any diagnostic that tells EXT from INT for a relative's head when the head is 3rd person and the other argument is 1/2/4? Restrictiveness cannot (XA-10).A form where EXT vs INT changes something visible: a following noun's linking, scope, or a free-pronoun adjunct. Willie ch. 3–4 on images. OPEN
T74bDoes p. 181 "Inverse relative clauses are also non-restrictive" mean every inverse relative (against (19)), or only those binding the inverse EXT (as on p. 227)?Willie 1990 (cited p. 226), or a consultant judgment on (19) with a restrictive context. OPEN
T74cDoes p. 227's "subjects of Inverse constructions" mean the EXT inside the relative, or the main-clause argument the relative is adjoined to?A sourced inverse main clause with a relative adjoined to its EXT patient, glossed by Willie. OPEN
T74dT73d: pp. 50–51 class ʼadi apart from the persons, while the gloss "PASS.3.3" (pp. 152, 157) files it as a 3. Which one does the ranking use?Willie pp. 36–38 on images; Young & Morgan 1987 on an agentive passive with a non-human patient (T70a). OPEN
T74eAre XA-4 (nouns) and HP-8 (REL) one rule ("person decides first, EH-8 only for 3/3")?An attested noun inside a 1st/2nd-person relative (T73b). OPEN

Still open from earlier: T73b, T73c, T73e. Sources: Willie 1991 pp. 50–51, 125–126, 156–157, 160–161, 163, 170, 181, 226–227 (page images, this run); p. 123 (image, earlier run). See also person-head, external-head, relatives, fourth-person, ergative-indefinite, pleonastic-psych, open questions, sources.

2026-11-13 morning · First-principles review 9: Willie’s relative clauses (ch. 6, §§6.0–6.3.3) — which participant “the one who …” picks out, and why the answer changes with the number of nouns inside

2026-11-14: the “On the spine” summary at the top of this page supersedes this section’s framing. Its data, judgments and citations are unchanged.

2026-11-13 audit pass 7. Grant’s instruction (2026-11-13): VPISH is not a word-order account; it frames the verb’s arguments so that syntax and semantics line up. In ordinary tensed clauses, V→v→T and subject raising Spec,vP→Spec,TP (on to Spec,CP when needed) are the working default, not GRANT’S CONJECTURE to be tested against a no-movement rival. This review was posted before that instruction. Its derivations, labels and verdicts on verb movement are rewritten here to use the default: the former no-movement rival is kept only as labelled contrast, each derivation says what else has to move, and examples that seemed to resist the default are kept as tests of it (OPEN). No Navajo form, gloss, judgment or citation changed. See the 2026-11-13 note and the pass 7 report.

HYPOTHESIS Plain-words verdict. A Navajo relative clause is a whole clause with no gap, and an enclitic =yę́ę (or =ígíí) at its right edge picks out one participant. Which participant it picks out is not fixed by grammatical function: Willie’s own rule that the head is always the “absolutive” fails on her own examples. What does fit every example on these pages is a ranking we already needed for single nouns. With no noun inside, the enclitic picks out the lower of the two participants (the patient with yi-, the agent with bi-). With one noun inside, that noun takes the lower participant and the enclitic picks out the upper one. With two nouns inside, the enclitic goes with the first. The overt pronouns left inside postposition phrases show that nothing is moved out of the clause. Light d is not needed. Telicity is OPEN. These pages have no question or negation; the two passives are declaratives.

Starting point

Assumed, by instruction: VPISH (the subject argument first merges in Spec,vP) and theta roles; since audit pass 7 (2026-11-13), also the V→v→T working default stated below. HYPOTHESIS Everything else is tested: H1–H26 from reviews 1–8, light d, the categories D⁰ and C⁰ we give the enclitic and the clause, Agree. Since audit pass 7 (2026-11-13), V→v→T and subject raising Spec,vP→Spec,TP (on to Spec,CP when needed) are the working default in tensed clauses, assumed rather than tested: VPISH frames where the arguments start and get their roles; it is not a word-order account. An example that seems to resist the default is kept and flagged OPEN. Every derivation ends in a logical form (LF) whose scope order matches c-command; a rival fails if its LF contradicts a reading the source reports. Tense and aspect are left out of the LFs (see H30). Willie’s own analysis (the enclitic as a demonstrative head binding an incorporated pronoun, nominals as adjuncts, “internal” and “external” arguments) appears only as labeled contrast.

The source’s claims, as its own

  • SOURCED Shape. The relative is built “by enclitizing to a main clause a relativizing determiner which means ‘the one …’ ”; “there is no gapping”; “Any argument type can be relativized” (p. 151). The two common enclitics are =ígíí and =ę́ę, with “phonologically determined allomorphs”, and =ę́ę is “a perfective form” (p. 151).
  • SOURCED Head. These relatives are “not headless; the relativizing enclitic is the head”, which “binds a pronoun that is incorporated into the relativized verb” (p. 153, citing Willie 1990).
  • SOURCED Generalization (16). “The heads of relative clauses are absolutives”: the intransitive subject, or the internal argument of a transitive, which in the inverse is the agent (p. 160). Willie states it for “the relative clauses we have seen so far”.
  • SOURCED Generalization (21). “When both arguments of the relative have nominal adjuncts, the relativizing enclitic binds the external argument, which must be definite.” For some speakers (Platero 1982), (20) has both readings; “this is not true of my speech” (p. 162).
  • SOURCED One noun. With a single noun, the direct relative “apparently has two readings for most, if not all speakers” (22). The preferred, restrictive one has as head “the agent that has no adjoined nominal”; the other is non-restrictive (pp. 162–163). The inverse relative with one noun (24) has only the non-restrictive reading, because “the Inverse voice requires a definite subject, which cannot be subjected to further definiteness marking by means of a restrictive relative” (p. 164).
  • SOURCED Relatives as nominals. The main verb’s yi-/bi- “determines the interpretation of adjoined nominals, whether they are simple nouns or nominalized relative clauses” (p. 165). A noun after the relativized verb is not part of the relative, “counter to Platero (1974) … this may be a dialect difference” (p. 158).
  • SOURCED Person. “Ordinarily, the heads of relative clauses are third person”; a second-person head occurs in “a few fixed construction types” (p. 154).

The checked data

SOURCED Exact forms, with example numbers and printed pages. Source glosses, printing slips and crops are at the linked index entries.

  • No noun inside (one argument type per example): NX-310 (1a), p. 151: łichííʼ=yę́ę, “the one that is red” · NX-311 (1b), p. 152: yicha=yę́ę, “the one who was crying” · NX-312 (1c): yiztał=yę́ę, “the one he kicked” · NX-313 (1d): biztał=yę́ę, “the one who kicked him” · NX-314 (1e): bichʼįʼ yáshtiʼ=yę́ę, “the one I spoke to” · NX-315 (1f): biʼdistał=yę́ę, “the one who was kicked” (source label: “subject of agentive passive”) · NX-316 (1g), p. 153: shił béé hózin=ígíí, “that which is known to me” · NX-328 (13), p. 159: yidloh=ę́ę, “the one who was laughing” · NX-329 (14): yizloh=ę́ę, “the one that was roped” · NX-330 (15): bizloh=ę́ę, “the one who roped him”.
  • Fourth person: NX-318 (3), p. 154: jiztał=yę́ę, “the one that person kicked” · NX-319 (4): haztał =yę́ę, “the one who kicked that person”.
  • One or two nouns inside: NX-331 (22), p. 162: ʼashkii yiztsʼǫs=yę́ę yóóʼeelwcd, (a) “The one who kissed the boy ran away.”, (b) “The boy, who was kissed by her, ran away.” · cited NX-013 (24), p. 164: ʼashkii biztsʼǫs=yę́ę yóóʼeelwod, (a) “The boy, the one who kissed her, ran away.”, *(b) “The one who was kissed by the boy ran away.” · cited NX-011 (20a), p. 161: ʼatʼééd ʼashkii yiztsʼǫs=yę́ę yóóʼeelwod, “The girl who kissed the/a boy ran away.” · cited NX-012 (20b), p. 162: ʼatʼééd ʼashkii biztsʼǫs=yę́ę yóóʼeelwod, “The girl who was kissed by the boy ran away.”
  • The relative inside a main clause: NX-317 (2), p. 153: yiztał=yę́ę yicha, “The one he kicked is crying.” · NX-321 (6), p. 155: yicha=yę́ę yóóʼeelwod, “The one who was crying ran away” · NX-322 (7), p. 156: yidloh=yę́ę yidootsʼǫs, “He will kiss the one who was laughing.” · NX-323 (8): yidloh=yę́ę bidootsʼǫs, “The one who was laughing will kiss him.” · NX-324 (9): ʼatʼééd yidootsʼǫs, “He will kiss the girl.” · NX-332 (25), p. 165: yidloh=ę́ę yiztsʼǫs, “He kissed the one who was laughing.” · NX-333 (26): yidloh-ę́ę biztsʼǫs, “The one who was laughing kissed him.” · NX-326 (11), p. 157: ʼatʼééd yidloh=ę́ę biztał, “The girl who was laughing kicked him.” · NX-327 (12): yidloh=ę́ę ʼatʼééd biztał, “The girl kicked the one who was laughing.”
  • Passive: NX-315 (above) and NX-325 (10), p. 157: yidloh=yę́ę bidiʼdootsʼǫs, translated “The who was laughing will be kissed.” (sic), gloss “3.laughing=REL PASS.3.3.will kiss”.
  • Non-restrictive, second person: NX-320 (5), p. 154: ní nidloh=yę́ę kʼad nicha, “You who were laughing are now crying.”

Our working glosses (word by word, not segmented; the source’s own glosses are on the index pages): NX-312 kick.3>3=REL · NX-313 INV.kick.3>3=REL · NX-314 to.3 speak.1SG=REL · NX-331 boy kiss.3>3=REL run.away.3 · NX-325 laugh.3=REL PASS.kiss.FUT.3.

OPEN Not indexed or not used. (16) and (21) are generalizations, not examples. Printing slips are kept, not repaired: NX-331 prints yóóʼeelwcd; NX-325’s translation lacks “one”; NX-315’s gloss lacks =REL; NX-333 prints a hyphen before ę́ę; p. 158 says “(11)” where (12) seems meant. The same host yidloh takes =yę́ę in NX-322, NX-323 and NX-325 but =ę́ę in NX-326, NX-327, NX-328 and NX-332, so on these pages the choice of allomorph is not visibly “phonologically determined”. Nothing below depends on it.

Hypotheses tested, revised or added

  • H27a / H27b · how the enclitic reaches its participant (new). Both treat the enclitic as an overt head (we write D⁰) that takes the whole clause (CP) as its complement. H27a: a silent operator moves from the relativized position to Spec,CP, leaving a gap. H27b: nothing moves; D⁰ is linked to one participant inside the clause by Agree, and that participant keeps whatever form it has there (a silent pronoun, or an overt pronoun prefix).
  • H28 · which participant (new). In a clause with two ordered participants (H3: in the inverse the patient has moved above the agent), the enclitic is linked to the lowest participant that has no noun inside the relative. If every participant has a noun inside, it is linked to the upper one (H28-full, a stipulation that restates Willie’s (21)). H28 is review 2’s one-noun default H4′, extended: the enclitic counts as one more nominal to be linked.
  • H29 / H29′ · the fourth-person pair (new). H29: a fourth-person object prefix ha- marks a patient that has moved above the agent, as in the inverse. H29′: the enclitic can only be linked to a third-person participant (Willie’s “ordinarily third person”, made strict).
  • H30 · tense on the enclitic (new). =yę́ę contributes a past or perfective meaning outside the verb word.
  • Retested: H3, H4′, H6 (the inverse-moved patient must be definite) and H10a/H10b (passive with no agent vs. a silent indefinite agent) from review 2; Willie’s (16) and (21). The V→v→T working default is assumed, not retested (audit pass 7).

Derivations, step by step

HYPOTHESIS Each derivation gives Merge with theta assignment, Agree (or “none”), Move, which copy is pronounced, what is inside the verb word versus outside it, and finally the LF. Extra assumptions are marked. Brackets put specifiers on the left. Heads are final. The inflected verb is not segmented beyond the source’s hyphens and clitic signs. In each LF, ι binds the participant the enclitic picks out.

  1. NX-312 versus NX-313, NX-329 versus NX-330 (no noun inside; yi- versus bi-). Merge V “kick” with its patient, a silent third-person pronoun, as complement (theta: Patient). Merge v, and the silent agent in Spec,vP (theta: Agent, VPISH). NX-313 only: Move the patient above the agent (H3), and the verb word shows bi-. Merge T and a silent C, giving the clause CP. Merge D⁰ =yę́ę with CP. Agree: D⁰ is linked to one participant (H27b). Spell-out: the verb word, then the enclitic, which is outside the verb word (the source writes =). LF NX-312: ιx ∃e[kick(e) ∧ Agent(e,he) ∧ Patient(e,x)]. LF NX-313: ιx ∃e[kick(e) ∧ Agent(e,x) ∧ Patient(e,him)]. Result: the linked participant is the lower one in both: the patient in the direct, the agent after H3’s Move in the inverse. That is H28 with no noun inside. NX-329/330 (“rope”) and the cited NX-009/010 (“kiss”) give the same split. Nothing in VPISH alone predicts it. The agent in Spec,vP is the participant closest to the clause edge, so if the enclitic simply took the closest participant (or a silent operator moved the shortest distance, under H27a), the direct form would give “the one who kicked him”. The source reports the opposite. Willie’s (16) and H28 both fit so far.
  2. NX-314 and NX-316 (the participant is inside a postposition phrase). NX-314: merge P -chʼįʼ “to” with a third-person pronoun, pronounced as the prefix bi-. The PP merges in VP with V “speak” (theta: Goal, assigned through the PP). The first-person agent is silent in Spec,vP. Merge T, C, then D⁰ =yę́ę. LF: ιx ∃e[speak(e) ∧ Agent(e,speaker) ∧ Goal(e,x)]. NX-316 is the same with two PPs, shił “with me” and béé “about it”, and the stative verb “known”, with the enclitic linked to the object of béé. LF: ιx ∃s[known(s) ∧ Exp(s,speaker) ∧ Theme(s,x)]. Result: the relativized participant is still pronounced inside the PP: as the pronoun prefix bi- in NX-314, and in NX-316 inside béé, which the source glosses “3.about” (we do not segment it). H27b derives this directly, with no Move. H27a needs one extra assumption: the operator moves but leaves a pronounced pronoun behind (resumption). So these examples favor H27b. Willie’s (16) is contradicted as a general statement: a postposition object is neither an intransitive subject nor a verb’s internal argument, and the source itself says “Any argument type can be relativized”. H28 does not apply here (only one participant can be the target), so it is not tested.
  3. NX-318 versus NX-319 (fourth person). NX-318: patient (third person) as complement of “kick”, agent (fourth person, ji-) in Spec,vP. The enclitic is linked to the patient: ιx ∃e[kick(e) ∧ Agent(e,that_person) ∧ Patient(e,x)]. NX-319: patient (fourth person, ha-) as complement, agent (third person) in Spec,vP, and the enclitic is linked to the agent: ιx ∃e[kick(e) ∧ Agent(e,x) ∧ Patient(e,that_person)]. Result: NX-319 relativizes the agent of a transitive verb with no bi-, so Willie’s (16) is contradicted on her own p. 154. H28 survives only with H29 (ha- = patient moved above the agent, as in the inverse), which makes the agent the lower participant. H29′ (the enclitic needs a third-person participant) derives both NX-318 and NX-319 without any assumption about order, and it agrees with the older page’s report that a fourth person “can never be the head of a restrictive relative” (p. 175, not reopened this run). OPEN H29 versus H29′. H29 predicts that with two overt nouns, a ha- clause puts the patient noun first.
  4. NX-331 (22) versus cited NX-013 (24): one noun inside. NX-331(a): merge V “kiss” with ʼashkii as complement (Patient). The agent is silent in Spec,vP (Agent). Direct, so no H3 Move. By H4′, a single noun goes with the lower participant, here the patient, which matches the source. Merge T, C, D⁰ =yę́ę. By H28, the lower participant already has a noun, so the enclitic is linked to the upper one, the silent agent. The relative DP then merges as the single argument of the main verb “run away”, in Spec,vP (Agent). LF: ∃e2[run_away(e2) ∧ Agent(e2,r)], r = ιx ∃e1[kiss(e1) ∧ Agent(e1,x) ∧ Patient(e1,boy)]. That is the source’s preferred, restrictive reading, with the head “the agent that has no adjoined nominal”. NX-013(b), starred: the same steps for the inverse. H3 moves the patient above the agent, ʼashkii goes with the lower participant (now the agent, matching the source), and H28 links the enclitic to the upper participant, the moved patient. So H28 predicts the starred reading “the one who was kissed by the boy”. It is ruled out only if H6 is added: the inverse-moved patient must be definite. Here that patient is the variable the relative restricts, and a bound variable is not a definite. With H6, the restrictive reading of (24) fails and only the non-restrictive (24a) remains, as the source reports. NX-011/NX-012 below show H6 is satisfied when the moved patient has its own definite noun. Result: H28 + H6 derives (22a) and *(24b). Willie’s (16) is contradicted by (22a), which the source itself calls restrictive and preferred, with the agent of a direct transitive as head. Her explanation of *(24b), “cannot be subjected to further definiteness marking”, becomes, in our terms, H6 applied to a variable. The non-restrictive readings (22b), (24a) need a separate derivation: the noun and the relative are each linked to the main clause’s argument, and the relative adds a second proposition (as in derivation 8). OPEN what the enclitic is linked to inside a non-restrictive relative.
  5. Cited NX-011 and NX-012 (two nouns inside). NX-011: ʼashkii as Patient complement, ʼatʼééd in Spec,vP as Agent, direct, so the nouns are spelled out agent first. NX-012: the same nouns and roles reversed, with H3 moving the patient ʼatʼééd above the agent ʼashkii, and the verb shows bi-. In both, the first noun is the upper participant, and the source’s head is the first noun. LF NX-011: ∃e2[run_away(e2) ∧ Agent(e2,r)], r = ιx[girl(x) ∧ ∃e1[kiss(e1) ∧ Agent(e1,x) ∧ Patient(e1,boy)]]. NX-012: r = ιx[girl(x) ∧ ∃e1[kiss(e1) ∧ Agent(e1,boy) ∧ Patient(e1,x)]]. Result: with every participant spelled out, H28’s first clause has nothing to apply to. H28-full (go with the upper participant) is needed, and it is a stipulation. That the choice can be free for other speakers (Platero 1982, as Willie reports it) fits it being a separate setting, not a consequence of H28. These relatives are internally headed: the noun that names the head is inside the relative, in its own theta position. In NX-012 the moved patient has a definite noun, so H6 is met, unlike in *(24b).
  6. The relative as a main-clause nominal: NX-322/323 (and NX-324), NX-332/333, NX-317, NX-321, NX-326/327. NX-322: the relative DP yidloh=yę́ę (built as in derivation 1, with one participant, so the enclitic takes it) is the only overt nominal of the main clause. Merge V “kiss” with a silent patient and a silent agent. By H4′, the single overt nominal goes with the lower participant: the patient in the direct (NX-322: “He will kiss the one who was laughing”), the agent after H3’s Move in the inverse (NX-323). NX-324 with the plain noun ʼatʼééd is the same. LF NX-323: ∃e2[kiss(e2) ∧ Agent(e2,r) ∧ Patient(e2,him)], r = ιx ∃e1[laugh(e1) ∧ Agent(e1,x)] (future omitted). NX-332/333 repeat the split with past “kissed”. NX-326: ʼatʼééd merges in Spec,vP of “laugh” inside the relative (Agent), so this relative is internally headed. The whole relative is then the single nominal of the inverse main clause and goes with the agent. LF: ∃e2[kick(e2) ∧ Agent(e2,r) ∧ Patient(e2,him)], r = ιx[girl(x) ∧ ∃e1[laugh(e1) ∧ Agent(e1,x)]]. NX-327: two nominals in the main clause. H3 moves the patient (the relative) above the agent (ʼatʼééd): ∃e2[kick(e2) ∧ Agent(e2,girl) ∧ Patient(e2,r)], r = ιx ∃e1[laugh(e1) ∧ Agent(e1,x)]. Result: H3 and H4′ pass out of sample with relatives in place of nouns, as Willie says. Labeled contrast: Willie and Platero (1974) disagree on whether a noun after the relativized verb can belong to the relative. Under our derivations, a noun inside the relative is pronounced before the relative’s verb, because heads are final. A Platero-type parse of NX-327, with ʼatʼééd inside the relative, would need a rightward Move inside the relative (an extra assumption). Willie’s speech, as printed, does not need it. HYPOTHESIS What else moves under the working default in NX-327: the agent ʼatʼééd raises to Spec,TP, so the patient relative has to move higher still (adjoined to TP or to Spec,CP; which is OPEN) to come first. That is one Move beyond H3 (labelled contrast: a no-movement analysis gets the order from H3 alone). The cost is the same for any inverse clause with two nominals and is not specific to relatives.
  7. NX-315 and NX-325 (passive). NX-325: merge V “kiss” with its patient as complement. The source’s gloss “PASS.3.3” marks two third persons, and its label is “agentive passive”. H10b: a silent indefinite agent merges in Spec,vP (Agent), and it cannot be linked to any nominal. H10a: no agent merges, which needs a lexical rule taking Agent away from “kiss”. The relative DP yidloh=yę́ę goes with the patient, the only participant a nominal can be linked to. LF (H10b): ∃e2∃y[kiss(e2) ∧ Agent(e2,y) ∧ Patient(e2,r)], r = ιx ∃e1[laugh(e1) ∧ Agent(e1,x)] (future omitted). NX-315: the enclitic goes with the patient inside a passive: ιx ∃e∃y[kick(e) ∧ Agent(e,y) ∧ Patient(e,x)]. Result: both fit H10b, as review 2 concluded, and neither contradicts H10a. OPEN H10b predicts that the enclitic can never pick out the passive’s silent agent (“the one by whom X was kicked”). Not printed here.
  8. NX-320 (non-restrictive, second person). Merge “laugh” with a second-person agent in Spec,vP, then =yę́ę. The main clause “cry” has a silent second-person agent in Spec,vP; ní “you” and the relative are both linked to it (the source calls both adjuncts). kʼad “now” sits before the main verb. LF, with the relative as a second assertion about the addressee: ∃e1[cry(e1) ∧ Agent(e1,you) ∧ now(e1)] ∧ ∃e2[laugh(e2) ∧ Agent(e2,you)]. Result: a non-restrictive relative adds a proposition rather than restricting a set, so ι is not used. This is the same structure we gave (22b) and (24a). HYPOTHESIS Under the working default, kʼad before the verb fits both an adverb carried along with a remnant VP and an adverb left in place (OPEN which).

Analysis A: Working default: V→v→T in both clauses, each subject Spec,vP→Spec,TP Hypothesis

Analysis A: Working default: V→v→T in both clauses, each subject Spec,vP→Spec,TPIn each clause V⁰ head-moves through v⁰ to a final T⁰, and the subject raises from left Spec,vP to left Spec,TP. V→v→T and subject Spec,vP→Spec,TP are the working default in tensed clauses (audit pass 7, 2026-11-13). With T⁰ final (HYPOTHESIS), the raised verb is pronounced last, so nothing else has to move to give the printed verb-final order. If T⁰ were initial instead, the material left in vP (object, PPs, adverbs) would have to move above T, for example as a remnant VP (HYPOTHESIS); which is right is OPEN. Inside the relative the raised subject is silent. Restrictive reading (22a). Inside the relative, the single noun ʼashkii goes with the lower participant, the patient (H4′); the enclitic is linked to the upper participant, the silent agent (H28). D⁰ =yę́ę c-commands everything in the relative, so ι scopes over the relative’s event. LF: ∃e2[run_away(e2) ∧ Agent(e2,r)], r = ιx ∃e1[kiss(e1) ∧ Agent(e1,x) ∧ Patient(e1,boy)] Full structure in the list below the figure.CPC′TPC⁰DPT′vPT⁰DPv′D′VPv⁰CPD⁰=yę́ęV′C′V⁰yóóʼeelwcdTPC⁰DPagent: silent; the participant the enclitic is linked to (H28)T′vPT⁰DPagent: silent; the participant the enclitic is linked to (H28)v′VPv⁰V′DPʼashkiiV⁰yiztsʼǫsverb wordverb wordhead-movehead-movehead-movehead-moveAgree link (H27b)MoveMove
(1) A relative with one noun inside: ʼashkii yiztsʼǫs=yę́ę yóóʼeelwcd, reading (a) (NX-331) Hypothesis NX-331, Willie 1991 ch. 6 (22), printed p. 162: ʼashkii yiztsʼǫs=yę́ę yóóʼeelwcd, (a) “The one who kissed the boy ran away.” (b) “The boy, who was kissed by her, ran away.” (yóóʼeelwcd as printed). No gloss line is printed. Leaves are printed tokens; splitting =yę́ę off as a separate head is our analysis (the source writes it as an enclitic). D⁰, C⁰ and the silent agent are our assignments. Only reading (a) is drawn.
Bracket notation (copy)

A: [CP [C′ [TP DP [T′ [vP ⟨DP⟩ [v′ [VP [V′ [V⁰ yóóʼeelwcd]]] v⁰]] T⁰]] C⁰]]

Structure as a list

Analysis A

  • CP
    • C′
      • TP
        • DP (id S-high; landing site of Move from subject)
        • T′
          • vP
            • DP (id subject; struck copy; moved to S-high)
              • D′
                • CP
                  • C′
                    • TP
                      • DP (id rS-high; landing site of Move from ragent): agent: silent; the participant the enclitic is linked to (H28)
                      • T′
                        • vP
                          • DP (id ragent; struck copy; moved to rS-high; goal of Agree link (H27b)): agent: silent; the participant the enclitic is linked to (H28)
                          • v′
                            • VP
                              • V′
                                • DP (id boy): ʼashkii
                                • V⁰ (head; id Vr; head-moves to vr): yiztsʼǫs
                            • v⁰ (head; id vr; head-moves to Tr)
                        • T⁰ (head; id Tr; hosts the “verb word” (Vr + vr + Tr))
                    • C⁰ (head; id Cr)
                • D⁰ (head; id Dr; probe for Agree link (H27b) with ragent): =yę́ę
            • v′
              • VP
                • V′
                  • V⁰ (head; id V; head-moves to v): yóóʼeelwcd
              • v⁰ (head; id v; head-moves to T)
          • T⁰ (head; id T; hosts the “verb word” (V + v + T))
      • C⁰ (head; id C)

Does light d earn its keep?

HYPOTHESIS Fair test. A light-d analysis would put a silent D⁰ on the bare nouns here (ʼashkii, ʼatʼééd), all translated as definites, and might use it to state H6 and Willie’s “must be definite”. Result: light d is not needed, and nothing here counts against it. (a) The one determiner-like head on these pages is overt: the enclitic. The ι in every LF comes from it. (b) H6 needs a way to tell a definite nominal from a bound variable (derivations 4–5). A definite bare noun can supply that by ordinary meaning (a unique individual in context) without a silent head. (c) No contrast on these pages turns on whether a bare noun has a silent D. Light d stays idle.

Telicity test

OPEN Prediction tested: whether the relativizer’s form, or which participant it picks out, depends on the telicity of the relativized verb. The enclitic =yę́ę appears with a state (NX-310 “red”, which the source calls “neuter”), activities (NX-311 “cry”, NX-320 and NX-322 “laugh”) and “kick”, whose telicity is not clear from meaning alone. =ígíí appears once, with a state (NX-316). The one variable that changes with the verb held fixed, =yę́ę versus =ę́ę on the same atelic yidloh, cannot be conditioned by telicity. The participant choice follows yi-/bi- and the nouns present, with “kick”, “rope” and “kiss” alike. These pages have no pair that differs only in telicity, so a telicity effect is neither supported nor contradicted: OPEN. OPEN H30: NX-310 is translated with a present state (“the one that is red”) although the source calls =yę́ę “a perfective form”, so H30, as a strict past meaning, is not supported by these pages.

Question × polarity × voice grid (cumulative after review 9)

OPEN No cell changes. A text-layer search of printed pp. 151–165 for doo, -ísh/-sh, daʼ, háí and haʼátʼíí, checked against the page renders, found no question and no negation. The two passives (NX-315, NX-325) are affirmative declaratives, so they do not fill a question cell.

Question cellActive / directInverse / bi-Passive / impersonal (source gloss)
Polar, affirmativeNX-119–134, 142, 203, 204, 291, 307NX-053; *NX-052OPEN (NX-217 candidate only)
Polar, negativeNX-153–161, 184, 205–209, 308 (answer NX-309)OPENOPEN (declaratives NX-216, NX-236 only)
Wh, affirmativeReview 3 set; NX-212, 220, 221Review 3 setNX-213, 215
Wh, negativeNX-138, 210, 211; NX-247 (by reading)OPENOPEN

HYPOTHESIS What this batch adds for the grid: a sourced passive with an overt nominal linked to its patient (NX-325) and a relative built on a passive (NX-315). If a passive question turns up later, H10b predicts that a wh-phrase or -ísh host can go with the patient but never with the silent agent, exactly as with the relativizer. OPEN

Where we part from the source’s analyses (labeled contrast)

  • HYPOTHESIS Not adopted: (16) “heads are absolutives”. It fails on NX-319, NX-331(a), NX-314 and NX-316, all from the same chapter. In our terms, the choice is made by the ranking H28 (the same ranking that links single nouns), plus person (H29 or H29′).
  • HYPOTHESIS Agrees with our derivations: there is no gap (overt pronouns stay inside PPs, NX-314, NX-316), and the enclitic is the head of the relative. We give it the category D⁰ and link it by Agree (H27b), not by binding an “incorporated pronoun”.
  • HYPOTHESIS Restated, not adopted as stated: Willie’s account of *(24b) (“further definiteness marking”) becomes H6 applied to a variable. Her (21) is kept only as the stipulation H28-full.

What this changes

  • HYPOTHESIS H4′ is extended to relatives (H28) and passes on NX-312/313, 329/330, 331(a), 322/323, 332/333, 326/327, with H6 needed for *(24b) and a stipulation (H28-full) for two-noun relatives.
  • HYPOTHESIS H3 and H6 pass out of sample. H6 now also does work inside relatives (*(24b) versus NX-012).
  • HYPOTHESIS The older page’s “absolutive default” (Relatives and the fourth person) is contradicted as a default by NX-319 and NX-331(a). Its person result fits H29′.
  • Working default (V→v→T, subject to Spec,TP): used in every clause, including inside relatives, where the subject is always silent. What else has to move: in two-nominal inverse clauses (NX-327) the patient moves above Spec,TP (HYPOTHESIS).
  • HYPOTHESIS Light d is still idle, and telicity is still OPEN.

OPEN predictions

  • OPEN H27a versus H27b: H27a (operator Move) predicts island effects for relativization out of an embedded clause; H27b predicts none. Willie ch. 6 §6.3.4–6.6 (pp. 165–181: transitive relatives inside transitive clauses, the fourth person, nominalized clauses) is the next place to look.
  • OPEN H29 versus H29′: a two-noun ha- clause with the order of its nouns stated.
  • OPEN H10b: no relative or question should pick out a passive’s silent agent.
  • OPEN Non-restrictive relatives: what the enclitic is linked to inside (22b), (24a) and NX-320.
  • OPEN Still missing from the grid: a polar passive question, and negative inverse and negative passive questions.

2026-11-13 afternoon · First-principles review 10: Willie’s relatives inside transitive clauses, the fourth person, and clauses that name a fact (ch. 6, §§6.3.4–6.6)

2026-11-14: the “On the spine” summary at the top of this page supersedes this section’s framing. Its data, judgments and citations are unchanged.

2026-11-13 audit pass 7. Grant’s instruction (2026-11-13): VPISH is not a word-order account; it frames the verb’s arguments so that syntax and semantics line up. In ordinary tensed clauses, V→v→T and subject raising Spec,vP→Spec,TP (on to Spec,CP when needed) are the working default, not GRANT’S CONJECTURE to be tested against a no-movement rival. This review was posted before that instruction. Its derivations, labels and verdicts on verb movement are rewritten here to use the default: the former no-movement rival is kept only as labelled contrast, each derivation says what else has to move, and examples that seemed to resist the default are kept as tests of it (OPEN). No Navajo form, gloss, judgment or citation changed. See the 2026-11-13 note and the pass 7 report.

HYPOTHESIS Plain-words verdict. One rule does most of the work on these pages. A noun, a relative clause, or the relativizing enclitic can only be tied to a third-person participant. Fourth-person (and first- and second-person) participants never take one. With that rule, review 9’s ranking still picks the right participant every time, and the extra “fourth person moves up” idea (H29) has no job left. Who is the same person as whom across the two clauses is decided by matching person, as Willie says. A clause ending in =ígíí can name a fact instead of a person when the enclitic is not tied to anyone, and that gives the two readings of “he believes (that he won / the winner)” with no extra machinery. One starred sentence, *(39), is not explained by anything we have. Light d is not needed. A telicity condition on =yę́ę is contradicted; telicity in general stays OPEN. No question, negation or passive is on these pages, so the grid does not change.

Starting point

Assumed, by instruction: VPISH (the subject argument first merges in Spec,vP) and theta roles; since audit pass 7 (2026-11-13), also the V→v→T working default stated below. HYPOTHESIS Everything else is tested: H1–H30 from reviews 1–9 (here mainly H3, H4′, H6, H27a/b, H28, H29/H29′), light d, the categories D⁰ and C⁰, Agree. Since audit pass 7 (2026-11-13), V→v→T and subject raising Spec,vP→Spec,TP (on to Spec,CP when needed) are the working default in tensed clauses, assumed rather than tested: VPISH frames where the arguments start and get their roles; it is not a word-order account. An example that seems to resist the default is kept and flagged OPEN. Every derivation ends in a logical form (LF) whose scope order matches c-command. Tense is left out of the LFs. Willie’s own analysis (“Parallel Processing”, nominals as adjuncts, internal versus external arguments, processing difficulty) appears only as labeled contrast.

The source’s claims, as its own

  • SOURCED (27) Parallel Processing. “In a sequence of two transitive clauses with all third person arguments, the two arguments of the first and second clause are interpreted as coreferential” (p. 166). It “does not apply” when a noun after the relative “marks disjoint reference” (p. 167).
  • SOURCED Main-clause voice. “If yi- is present, the head of the relative is an adjunct to the patient argument of the main clause verb; if bi- is present the relative is an adjunct to the agent argument” (p. 167).
  • SOURCED Both clauses inverse. *(39) “is excluded; I find it impossible to process … the internal arguments in the two clauses are dsjoint [sic] in reference, and this is impossible unless the second of these is indefinite” (p. 169); léiʼ “makes the construction acceptable” (p. 170).
  • SOURCED Person of the head. First- and second-person heads force a non-restrictive relative, “since the head is already maximally definite” (p. 171); so do fourth-person heads (p. 171). “The fourth person cannot be the head of a restrictive relative; the head is third person” (p. 175).
  • SOURCED Co- and disjoint reference. “Where fourth person appears in just one clause, it marks disjoint reference, and where it appears in both clauses it marks coreference. This follows the fact that anaphors must match their antecedents in phi features” (pp. 173–174). Nominals “are not adjoined to fourth person arguments, since the nominal is third person syntactically” (p. 174).
  • SOURCED Nominalized clauses. =ígíí also forms “nominalized propositional clauses”, which “do not” have a head and “refer to propositions” (p. 177); “The complementizer =ígíí is factive, while =go occurs in temporal/conditional constructions” (p. 181).
  • SOURCED Summary. “Inverse relative clauses are also non-restrictive” (p. 181).

The checked data

SOURCED Exact forms, with example numbers and printed pages. Source glosses, printing slips and crops are at the linked index entries.

  • Relative + transitive main verb, no noun (pp. 166–167): NX-334 (28): yiztał-yę́ę yiyiiłtsą́, “He saw the one he kicked.” · NX-335 (29): yiztał=yę́ę biiłtsą́, “The one he kicked saw him.” · NX-336 (30): biztał=yę́ę yiyiiltsą́, “He saw the one that kicked him.” · NX-337 (31): biztał=yę́ę biiłtsą́, “The one that kicked him saw him.”
  • Relative + noun + main verb (p. 168): NX-338 (32): yiztał=yę́ę ʼatʼééd yiyiiltsą́, “The one he kicked saw the girl.” · NX-339 (33): yiztał=yę́ę ʼatʼééd biiltsą́, “The one he kicked was seen by the girl.” · NX-340 (34): biztał=yę́ę ʼatʼééd yiyiiłtsą́, “The one that kicked him saw the girl.” · NX-341 (35): biztał=yę́ę ʼatʼééd biiłtsą́, “The girl saw the one who kicked him.”
  • Three nominals (pp. 168–170): NX-342 (36): ʼatʼééd yiztał=yę́ę ʼashkii néidííłtsʼin, “The one who kicked the girl hit the boy.” · NX-343 (37): ʼatʼééd biztał=yę́ę ʼashkii léiʼ néidííłtsʼin, “The girl who kicked him hit some boy.” · NX-344 (38): ʼatʼééd yiztał=yę́ę ʼashkii nábidííłtsʼin, “The one who kicked the girl was hit by the boy.” · NX-345 *(39): ʼatʼééd biztał=yę́ę ʼashkii nábidííłtsʼin, intended “[the girl, the one who kicked himᵢ, was hit by the boyⱼ]” · NX-346 (40): ʼatʼééd biztał=yée ʼashkii léiʼ nábidííłtsʼin, “The girl, the one who kicked himᵢ, was hit by some boyⱼ.”
  • First, second and fourth person heads (pp. 170–171): NX-347 (41a): siztsʼǫs=yę́ę, “the one who kissed me” · NX-348 (41b): sétsʼǫs=yę́ę, “the one I kissed” · cited NX-109 (42): shí ʼayóí ʼánííníshʼníʼígíí naaʼáháshyaa doo, “I, who love you, will take care of you.” (read from the p. 171 render this run; no = before ígíí in print) · NX-349 (43): Ní, ʼayóí ʼáshííníʼníʼ=nę́ę shaa yisínínah, “You, who loved me, forgot about me.” · NX-350 (44): joodloh=ę́ę yóóʼajoolwod, “That one, the one who laughed, ran away.”
  • Fourth person in one or both clauses (pp. 172–173): NX-351 (45): haztał=yę́ę yiyiiłtsą́, “He saw the one who kicked that person.” · NX-352 (46): jiztał=yée yiyiiłtsą́, “He saw the one who that person kicked.” · NX-353 (47): yiztał=yę́ę jiiłtsą́, “That person saw the one who was kicked.” · NX-354 (48): yiztał=yę́ę hoołtsą́, “The one that he kicked saw that person.” · NX-355 (49): haztał=yę́ę jiiłtsą́, “That personᵢ saw the one who kicked himᵢ.” · NX-356 (50): jiztał=yę́ę jiiłtsą́, “The oneᵢ that personⱼ kicked, that personⱼ saw himᵢ.” · NX-357 (51): haztał=yę́ę hoołtsą́, “That one who kicked that person saw that person.” · NX-358 (52): jiztał=yę́ę hoołtsą́, “That one that person kicked saw that person.”
  • Nouns and the fourth person (pp. 174–176): NX-359 *(53): ʼatʼééd joodloh=yę́ę yóóʼajoolwod · NX-360 *(54): yiztsʼǫs=yę́ę ʼatʼééd hoołtsą́ · NX-361 (55a): jiztsʼǫs=yę́ę ʼatʼééd yiyiiłtsą́, “The one that person kissed saw the girl.” · NX-362 (55b): haztsʼǫs=yę́ę ʼatʼééd yiyiiłtsą́, “The one who kissed that person saw the girl.” · NX-363 (56): ʼatʼééd haztsʼǫs=yę́ę yiyiiłtsą́, “He saw the girl who kissed that person.” · NX-364 (57): ʼatʼééd haztsʼǫs=yę́ę jiiłtsą́, “That personᵢ saw the girl that kissed himᵢ.” · NX-365 (58): ʼatʼééd jiztsʼǫs=yę́ę yiyiiłtsą́, “He saw the girl that person kissed.” · NX-366 (59): ʼatʼééd jiztsʼǫs=yę́ę jiiłtsą́, “That personᵢ saw the girl heᵢ kissed.”
  • Clauses that name a fact (pp. 178–180): NX-367 (62): chidí ʼínáályaa=yę́ę yáʼátʼééh, “The car that was fixed is good.” · NX-368 (63): chidí ʼínáályaa=ígíí yáʼátʼééh, “It is good that the car is fixed.” · NX-369 (64): honeesná=nígíí yoodlá, (a) “Heᵢ believes that heᵢ,ⱼ won.” (b) “He believes the winner.” · NX-370 (65): Honeesná=nígíí joodlá, (a) “That personᵢ believes that heⱼ won.” (b) “That person believes the winner.” · NX-371 (66): Hozhneesná=ígíí joodlá, (a) “Heᵢ believes that heᵢ won.” *(b) “Heᵢ believes that heⱼ won.” · NX-372 (67): Mary łį́į́ʼ nayiisniiʼ=ígíí yíniiʼ, “I heard that Mary bought a horse.” · NX-373 ??(68): Mary łį́į́ʼ nayiisniiʼ=go yíniiʼ, “I heard it, when Mary bought the a horse.” (sic)

Our working glosses (word by word, not segmented; the source’s own glosses are on the index pages): NX-334 kick.3>3=REL see.3>3 · NX-342 girl kick.3>3=REL boy hit.3>3 · NX-363 girl kiss.3>4=REL see.3>3. For NX-367/368 we keep the source’s “3.fixed” and do not guess the voice.

OPEN Printing slips kept, not repaired. NX-334 prints a hyphen before yę́ę; NX-336, NX-338 and NX-339 print the main verb with plain l (yiyiiltsą́, biiltsą́) where the other examples have ł; NX-346 and NX-352 print =yée without nasal hooks; NX-349’s gloss prints “=REl”. The glosses label both ji- and ha- relatives “4.3” in (45)/(46) and both jiiłtsą́ and hoołtsą́ “4.3” in (47)/(48); below we read who is fourth person from the printed translations, which are consistent with review 9’s NX-318/319 (ji- = fourth-person agent, ha-/ho- = fourth-person patient).

Hypotheses tested, revised or added

  • H31 · person-matched linking (new; generalizes review 9’s H29′). A noun, a relative DP and the enclitic D⁰ are third person. Each can be linked (by Agree) only to a third-person participant. First-, second- and fourth-person participants are spelled out only by the verb’s prefixes.
  • H28′ (H28 revised). The enclitic is linked to the lowest third-person participant that has no noun inside the relative; if every third-person participant has a noun, to the upper one (H28-full). H4′ is revised the same way: a single nominal goes with the lowest third-person participant (H4″).
  • H32 · reference of silent pronouns across clauses (new). H32a: a silent pronoun can share reference only with one of the same person (Willie’s phi-matching, restated). H32b: two fourth-person pronouns in one sentence refer to the same person. H32c: Willie’s (27): with all third persons in two transitive clauses, the silent pronouns pair up across the clauses.
  • H33 · one restrictive nominal per participant (new). A participant is linked to at most one nominal, unless the second one is non-restrictive.
  • H34 · the enclitic need not be linked (new). If D⁰ is linked to a participant, the DP names that individual (ι over the individual). If it is not linked, the DP names the proposition or fact that the clause expresses. On these pages the unlinked use is found with =ígíí/=nígíí, never with =yę́ę.
  • H35 · non-restrictive relatives (new; answers review 9’s OPEN question tentatively). A non-restrictive relative is H34’s unlinked D⁰ used as an appositive: it adds the proposition, and its silent pronoun takes its reference from the main-clause participant by H32.
  • Retested: H3, H4′, H6, H27a/H27b, H28, H29 versus H29′. The V→v→T working default is assumed, not retested (audit pass 7).

Derivations, step by step

HYPOTHESIS Each derivation gives Merge with theta assignment, Agree, Move, which copy is pronounced, what is inside or outside the verb word, and the LF. Extra assumptions beyond VPISH and theta roles are named; all are HYPOTHESIS. Specifiers are on the left. Heads are final. In each LF, ι binds the participant the enclitic picks out; subscript ₄ marks a fourth-person referent.

  1. NX-334–337 (28)–(31): a relative with no noun as the only nominal of a transitive main clause. Relative: merge V “kick” with a silent patient (theta: Patient), v, a silent agent in Spec,vP (theta: Agent, VPISH). In (30)/(31) only, Move the patient above the agent (H3; the verb shows bi-). Merge T, C, then D⁰ =yę́ę. Agree links D⁰ to the lower participant (H28′): the patient in (28)/(29), the agent in (30)/(31). Main clause: merge V “see” with a silent patient, v, a silent agent; in (29)/(31) H3 moves the patient up. The relative DP is the only overt nominal, so by H4′ it goes with the lower participant: the patient with yi- (28)/(30), the agent with bi- (29)/(31). That is exactly Willie’s p. 167 statement, derived rather than stated. LF (28): ∃e2[see(e2) ∧ Agent(e2,y) ∧ Patient(e2,r)], r = ιx ∃e1[kick(e1) ∧ Agent(e1,y) ∧ Patient(e1,x)]. LF (29): ∃e2[see(e2) ∧ Agent(e2,r) ∧ Patient(e2,y)], r = ιx ∃e1[kick(e1) ∧ Agent(e1,y) ∧ Patient(e1,x)]. Result: the order and roles follow from H3 + H4′ + H28′. The same y in both clauses (the kicker is the seer in (28)) is not derived: it is H32c, a restatement of Willie’s (27). Without H32c, y in the main clause is a free pronoun and could be anyone. OPEN whether (27) is grammar or a strong preference; Willie calls it a “processing convention”. The theta axiom does not decide it.
  2. NX-338–341 (32)–(35): relative + noun + verb. Both main-clause participants now have a nominal. Direct (32)/(34): no Move; the relative (first) is the agent, ʼatʼééd the patient. Inverse (33)/(35): H3 moves the patient, so the relative (first) is the patient and ʼatʼééd the agent. LF (35): ∃e2[see(e2) ∧ Agent(e2,girl) ∧ Patient(e2,r)], r = ιx ∃e1[kick(e1) ∧ Agent(e1,x) ∧ Patient(e1,z)]. Result: H3 passes again. H32c has no silent main-clause pronoun to pair with, so the relative’s silent pronoun (z) is free, which matches Willie’s statement that (27) “does not apply” here and the English “him”, not “her”, in (35).
  3. NX-342 (36), NX-344 (38): one noun inside a direct relative. Inside the relative, ʼatʼééd goes with the lower participant, the patient (H4′), and the enclitic with the upper, the silent agent (H28′), exactly as in review 9’s NX-331(a). Main clause: (36) direct, so relative = agent, ʼashkii = patient; (38) inverse, so H3 moves the relative (patient) above ʼashkii (agent). LF (38): ∃e2[hit(e2) ∧ Agent(e2,boy) ∧ Patient(e2,r)], r = ιx ∃e1[kick(e1) ∧ Agent(e1,x) ∧ Patient(e1,girl)]. Result: H4′ + H28′ + H3 pass out of sample. Only the restrictive reading is printed for (36) (review 9’s (22) had two).
  4. NX-343 (37), NX-346 (40): one noun inside an inverse relative. H3 moves the relative’s patient (silent “him”) above its agent; ʼatʼééd goes with the lower participant, the agent. H28′ would link the enclitic to the upper participant, the moved patient, and so predicts “the one the girl kicked”. That reading is not printed. As for review 9’s *(24b) (NX-013), H6 rules it out: the inverse-moved patient must be definite, and a variable bound by ι is not. What is left links nothing restrictively: the relative is the appositive of H35, about the girl. LF (37): ∃e2∃b[boy(b) ∧ hit(e2) ∧ Agent(e2,girl) ∧ Patient(e2,b)] ∧ ∃e1[kick(e1) ∧ Agent(e1,girl) ∧ Patient(e1,z)]. Result: H6 passes out of sample, and it matches Willie’s summary “Inverse relative clauses are also non-restrictive”. (37) is printed without commas, so we rely on her summary for the non-restrictive label.
  5. NX-345 *(39) versus NX-346 (40): NOT UNDERSTOOD. The two strings differ only in léiʼ. Our derivation for both is the same: relative as in derivation 4; main clause inverse, so H3 moves the relative (patient) above ʼashkii (agent); H6 is met because the moved patient (“the girl …”) is definite. Nothing in H3, H4′, H6, H28′, H31 or H33 separates them. H32c cannot be the reason: in (35) the relative’s silent pronoun and a definite main-clause noun are already allowed to differ. OPEN Willie’s account (labeled contrast) is processing: two inverse clauses with “disjoint internal arguments” need the second to be indefinite. The test that would separate “both clauses inverse” from “an inverse relative with its own noun, followed by a definite noun” is (37) without léiʼ, i.e. a direct main clause after the same relative. It is not printed.
  6. NX-347/348 (41) and NX-363 (56): why H29 drops out. (41a) siztsʼǫs=yę́ę: patient first person (silent), agent third person (silent), direct. Plain H28 would link the enclitic to the lower participant, “me”, which is wrong (“the one who kissed me”). H31 allows only the third-person agent, which is right. (56) ʼatʼééd haztsʼǫs=yę́ę: merge V “kiss” with a fourth-person patient (ha-), and ʼatʼééd in Spec,vP as agent. Under H29 (the ha- patient has moved above the agent), the lower participant is the agent, so the noun goes there (H4′), and plain H28 then links the enclitic to the lowest participant without a noun, the fourth-person patient. That predicts “the one the girl kissed / that person”, and the source says “the girl who kissed that person”. So H29 + H28 is contradicted. Under H31, the only third-person participant is the agent, so both the noun and the enclitic are linked to it (an internally headed relative, as in review 9’s NX-326), whether or not anything moved. The relative is then the patient of “see” (H4′, direct). LF: ∃e2[see(e2) ∧ Agent(e2,y) ∧ Patient(e2,r)], r = ιx[girl(x) ∧ ∃e1[kiss(e1) ∧ Agent(e1,x) ∧ Patient(e1,z₄)]], y ≠ z₄ (y ≠ z₄ by H32a). Result: H31 is needed (by (41a), (56), (57), and Willie’s p. 175 statement). Once it is in, H29 does no work anywhere on these pages, and review 9’s H29 test (“a two-noun ha- clause puts the patient noun first”) can never be run, because a noun cannot be linked to the fourth-person patient at all (derivation 7). H29 is dropped as idle; H29′ is kept, as part of H31. See the tree below.
  7. NX-359 *(53) and NX-360 *(54). *(53): every participant in both clauses is fourth person, so ʼatʼééd has nothing to be linked to (H31); it gets no theta role, against the theta axiom. *(54): the main verb hoołtsą́ has one fourth-person patient and one third-person agent. Both the relative and ʼatʼééd need the single third-person participant; H33 allows only one restrictive nominal per participant. Result: both stars derived (H31; H31 + H33), matching Willie’s description. OPEN why the relative in *(54) cannot be an appositive (H35) of ʼatʼééd; one guess is that an appositive must follow its noun, as ní precedes the relative in review 9’s NX-320, but this is not tested.
  8. NX-351–358 (45)–(52), NX-361–366 (55)–(59): who is the same person as whom. Each relative is built as in derivation 1 or 6, the enclitic linked by H31/H28′. In the main clause the relative goes with the main verb’s single third-person participant (H31): the patient in (45)/(46)/(47), the agent in (48), where hoo- is a fourth-person patient. The remaining silent pronouns are then matched by person: (45) main agent (third) vs. the relative’s fourth-person patient: different, so disjoint (H32a), “He saw the one who kicked that person.” (49) main fourth-person agent and the relative’s fourth-person patient: same person (H32b), “That personᵢ saw the one who kicked himᵢ.” LF (49): ∃e2[see(e2) ∧ Agent(e2,z₄) ∧ Patient(e2,r)], r = ιx ∃e1[kick(e1) ∧ Agent(e1,x) ∧ Patient(e1,z₄)]. (50)–(52) and (57)/(59) are fourth person in both clauses, so coreferent. (46)–(48), (55), (56) and (58) have it in one, so disjoint. (55a/b): two main-clause nominals, direct, so the relative is the agent and ʼatʼééd the patient (H3). Result: all 14 translations fit H31 + H32a/b. Note that in (47) the translation leaves the relative’s agent unnamed (“the one who was kicked”). H32a then says only that the agent is not the fourth-person seer.
  9. NX-349 (43), NX-350 (44), cited NX-109 (42): non-third heads. In (43) the relative’s participants are a second-person agent and a first-person patient, so H31 gives the enclitic nothing to link to. By H34/H35, the D⁰ =nę́ę is unlinked and the relative is an appositive, here on the free pronoun Ní. LF: ∃e1[forget(e1) ∧ Agent(e1,you) ∧ Theme(e1,me)] ∧ ∃s[love(s) ∧ Exp(s,you) ∧ Theme(s,me)] (roles for “forget about” and “love” are our assumption). (44): the same, with a fourth-person agent. Result: Willie’s “first, second and fourth person heads must be non-restrictive” follows from H31 + H35, with no appeal to “maximal definiteness”.
  10. NX-367/368 (62)/(63) and NX-369–371 (64)–(66): a clause that names a fact. (62): merge V “fix” with the patient (the source glosses one argument, “3.fixed”; we do not decide whether there is a silent agent). chidí goes with it; D⁰ =yę́ę is linked to it. The DP is the theme of “good”: ∃s[good(s) ∧ Theme(s,r)], r = ιx[car(x) ∧ ∃e[fix(e) ∧ Patient(e,x)]]. (63): the same clause with D⁰ =ígíí unlinked (H34), so it names the fact: ∃s[good(s) ∧ Theme(s,p)], p = ιp[p = ∃e[fix(e) ∧ Patient(e,car)]]. (64) honeesná=nígíí yoodlá: one string, two derivations. If D⁰ is unlinked, you get (a): ∃s[believe(s) ∧ Exp(s,y) ∧ Theme(s, ∃e[win(e) ∧ Agent(e,w)])] with w = y or not (no H32c: only one clause is transitive). If D⁰ is linked to the winner, you get (b): Theme(s, ιx ∃e[win(e) ∧ Agent(e,x)]). (65): a fourth-person believer and a third-person winner are disjoint (H32a), as printed. (66): fourth person in both, so they are the same (H32b), and the disjoint (b) is starred, as printed. H31 also predicts that (66) has no “believes the winner” reading (the winner is fourth person and cannot be linked); none is printed, which is consistent but weak. Result: H34 derives the (64a)/(64b) ambiguity and the (62)/(63) contrast with no extra Move. Under H27a (operator Move) the fact reading simply lacks the operator, so this does not choose between H27a and H27b.
  11. NX-372/373 (67)/(68): fact versus time. (67): inside the complement, Mary is the agent in Spec,vP and łį́į́ʼ the patient (direct, so agent first, H3). D⁰ =ígíí is unlinked (H34) and names the fact, which is the theme of “hear”: ∃e[hear(e) ∧ Exp(e,speaker) ∧ Theme(e, ιp[p = ∃h[horse(h) ∧ ∃e1[buy(e1) ∧ Agent(e1,Mary) ∧ Theme(e1,h)]]])]. The indefinite is inside the fact, below D⁰, so its scope matches c-command. (68) =go gives a time adjunct (“when”), so “hear” needs a heard sound as its theme, which is why the source calls it pragmatically odd. Result: fits H34. =go is not a D⁰ on this evidence.

Analysis A: Working default: V→v→T in both clauses, each subject Spec,vP→Spec,TP Hypothesis

Analysis A: Working default: V→v→T in both clauses, each subject Spec,vP→Spec,TPIn each clause V⁰ head-moves through v⁰ to a final T⁰, and the subject raises from left Spec,vP to left Spec,TP. V→v→T and subject Spec,vP→Spec,TP are the working default in tensed clauses (audit pass 7, 2026-11-13). With T⁰ final (HYPOTHESIS), the raised verb is pronounced last, so nothing else has to move to give the printed verb-final order. If T⁰ were initial instead, the material left in vP (object, PPs, adverbs) would have to move above T, for example as a remnant VP (HYPOTHESIS); which is right is OPEN. The main subject is silent. Inside the relative the only third-person participant is the agent, so both the noun ʼatʼééd and the enclitic are linked to it (H31; H28 restricted to third-person participants). The fourth-person patient cannot host a noun or the enclitic. The relative DP is the patient of “see” (H4′, direct verb). The main agent and the fourth-person patient do not match in person, so they are disjoint (H32a). LF: ∃e2[see(e2) ∧ Agent(e2,y) ∧ Patient(e2,r)], r = ιx[girl(x) ∧ ∃e1[kiss(e1) ∧ Agent(e1,x) ∧ Patient(e1,z₄)]], y ≠ z₄ Full structure in the list below the figure.CPC′TPC⁰DPagent: silent third person “he”T′vPT⁰DPagent: silent third person “he”v′VPv⁰V′DPV⁰yiyiiłtsą́D′CPD⁰=yę́ęC′TPC⁰DPʼatʼéédT′vPT⁰DPʼatʼéédv′VPv⁰V′DPpatient: fourth person (ha-), silentV⁰haztsʼǫsverb wordverb wordhead-movehead-movehead-movehead-moveAgree link, 3rd person (H31)MoveMove
(1) A relative with a fourth-person object: ʼatʼééd haztsʼǫs=yę́ę yiyiiłtsą́ (NX-363) Hypothesis NX-363, Willie 1991 ch. 6 (56), printed p. 175: ʼatʼééd haztsʼǫs=yę́ę yiyiiłtsą́, “He saw the girl who kissed that person.” No gloss line is printed. Leaves are printed tokens; splitting =yę́ę off as a separate head is our analysis (the source writes it as an enclitic). D⁰, C⁰, the silent agent and the silent fourth-person patient are our assignments.
Bracket notation (copy)

A: [CP [C′ [TP [DP agent: silent third person “he”] [T′ [vP ⟨DP⟩ [v′ [VP [V′ [DP [D′ [CP [C′ [TP [DP ʼatʼééd] [T′ [vP ⟨DP⟩ [v′ [VP [V′ [DP patient: fourth person (ha-), silent] [V⁰ haztsʼǫs]]] v⁰]] T⁰]] C⁰]] [D⁰ =yę́ę]]] [V⁰ yiyiiłtsą́]]] v⁰]] T⁰]] C⁰]]

Structure as a list

Analysis A

  • CP
    • C′
      • TP
        • DP (id S-high; landing site of Move from agent): agent: silent third person “he”
        • T′
          • vP
            • DP (id agent; struck copy; moved to S-high): agent: silent third person “he”
            • v′
              • VP
                • V′
                  • DP (id object)
                    • D′
                      • CP
                        • C′
                          • TP
                            • DP (id girl-high; landing site of Move from girl): ʼatʼééd
                            • T′
                              • vP
                                • DP (id girl; struck copy; moved to girl-high; goal of Agree link, 3rd person (H31)): ʼatʼééd
                                • v′
                                  • VP
                                    • V′
                                      • DP (id p4): patient: fourth person (ha-), silent
                                      • V⁰ (head; id Vr; head-moves to vr): haztsʼǫs
                                  • v⁰ (head; id vr; head-moves to Tr)
                              • T⁰ (head; id Tr; hosts the “verb word” (Vr + vr + Tr))
                          • C⁰ (head; id Cr)
                      • D⁰ (head; id Dr; probe for Agree link, 3rd person (H31) with girl): =yę́ę
                  • V⁰ (head; id V; head-moves to v): yiyiiłtsą́
              • v⁰ (head; id v; head-moves to T)
          • T⁰ (head; id T; hosts the “verb word” (V + v + T))
      • C⁰ (head; id C)

Does light d earn its keep?

HYPOTHESIS Fair test. These pages have the clearest definiteness contrast so far: *(39) versus (40), where ʼashkii “the boy” becomes ʼashkii léiʼ “some boy”. A light-d analysis would give bare ʼashkii a silent definite D⁰ and let léiʼ be an overt indefinite D⁰. Result: light d is not needed, and one version of it counts against it. (a) The only determiners that do work here are overt: the enclitics (D⁰ under H27b/H34) and léiʼ. (b) We cannot derive *(39) with or without a silent d (derivation 5), so light d does not explain it either. (c) Bare nouns are not always definite: in (67), bare łį́į́ʼ is translated “a horse”. So a silent d that is always definite is contradicted by NX-372. A silent d that can be either definite or indefinite makes no testable prediction on these pages.

Telicity test

SOURCED-contradicted (narrow claim): “=yę́ę needs a telic (or an atelic) predicate.” On pages already read, =yę́ę attaches to the state “red” (NX-310) and the activities “cry” and “laugh” (NX-311, NX-322). Here it attaches to “fix” (NX-367), which is telic by its meaning (our classification, not the source’s). So =yę́ę is not restricted in either direction. In (62)/(63) the verb, and so its telicity, is held constant, and the choice between =yę́ę and =ígíí follows individual versus fact. OPEN Whether telicity matters anywhere else in this grammar is still open. No pair on these pages differs only in telicity. The source’s “=ę́ę is a perfective form” is not tested by these data.

Question × polarity × voice grid (cumulative after review 10)

OPEN No cell changes. A text-layer search of printed pp. 165–181 for doo, -ísh/-sh, daʼ, háí, haʼátʼíí and “?”, checked against the renders, found no question and no negation. The one doo (NX-109, (42)) is glossed FUT, not negation. The English passives in (33), (38), (39) and (40) translate inverse verbs (gloss INV), not source-labeled passives, so they fill no passive cell.

Question cellActive / directInverse / bi-Passive / impersonal (source gloss)
Polar, affirmativeNX-119–134, 142, 203, 204, 291, 307NX-053; *NX-052OPEN (NX-217 candidate only)
Polar, negativeNX-153–161, 184, 205–209, 308 (answer NX-309)OPENOPEN (declaratives NX-216, NX-236 only)
Wh, affirmativeReview 3 set; NX-212, 220, 221Review 3 setNX-213, 215
Wh, negativeNX-138, 210, 211; NX-247 (by reading)OPENOPEN

Where we part from the source’s analyses (labeled contrast)

  • HYPOTHESIS Kept, restated: phi-matching for reference (H32a/b) and “nominals are third person” (H31). Willie states them; here they also choose the participant the enclitic picks out.
  • HYPOTHESIS Kept as a stipulation only: Parallel Processing (27) = H32c. It is not derived from VPISH, theta roles or Agree.
  • HYPOTHESIS Not adopted: “maximally definite” heads as the reason for non-restrictive 1st/2nd/4th-person relatives. H31 + H35 give the same result. Willie’s processing account of *(39) is not adopted, and we have no replacement (OPEN).

What this changes

  • HYPOTHESIS H31 added; H28 → H28′ and H4′ → H4″ (restricted to third-person participants). They fit every new entry they apply to; *(39) is derived by nothing we have.
  • HYPOTHESIS H29 dropped as idle; H29′ kept inside H31. H29 + H28 is contradicted by NX-363 (56).
  • HYPOTHESIS H3 and H6 pass out of sample (NX-334–346).
  • HYPOTHESIS H34/H35: the enclitic can be left unlinked. This derives fact readings (NX-368, 369a, 372) and non-restrictive relatives (NX-349, 350, 343, 346). It gives review 9’s OPEN question about non-restrictives a tentative answer.
  • Working default (V→v→T, subject to Spec,TP): used in every clause; relative-internal subjects are silent or sentence-initial. What else has to move: in two-nominal inverse main clauses (NX-339, 341, 344, 346) the patient moves above Spec,TP, as in NX-327 in review 9 (HYPOTHESIS). In NX-109 the future doo follows the verb, as the default gives.
  • HYPOTHESIS Light d is still not needed, and an always-definite silent d is contradicted by NX-372. Telicity: a telicity restriction on =yę́ę is contradicted; everything else is OPEN.

OPEN predictions

  • OPEN *(39) versus (40): the same inverse relative before a direct main clause with a definite noun (not printed).
  • OPEN H34: an unlinked =yę́ę naming a fact. Not found on these pages.
  • OPEN H32c: whether a disjoint reading of (28)–(31) is impossible or just dispreferred.
  • OPEN H27a versus H27b: still no island test. Relativizing out of an embedded =ígíí clause would be one.
  • OPEN H10b: no relative or question should pick out a passive’s silent agent. There is no passive on these pages.
  • OPEN Grid: still no polar passive question, and no negative inverse or negative passive question.