Authority caveat. Nobody on this site claims to know Navajo authoritatively. Every Navajo form is copied from a named published source with its example number; none is invented. Formal rules are hypotheses to be tested against Grant's texts.

SOURCED stated in a published source we opened HYPOTHESIS our formalization or inference GRANT'S CONJECTURE Grant's own claim OPEN needs data or reading

Not quite a word: is the Navajo verb's argument a feature bundle?

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.

Working note. Kept as written (URL, data and status labels unchanged). Filed under The roles-vs-words thesis; the consolidated argument is the essay roles-not-words. Rule IDs: rules index.

Verbs on this page → dictionary: yiyiisxį́ ‘kill (one)’ · yíchxǫʼ ‘be ruined; (classifier) ruin’

Restated on the unified structure (2026-11-15). The framework is restated in §1: VPISH (arguments merged in the verb’s scope, subject in Spec,vP) with theta roles, and V→v→T with the subject raised to Spec,TP as the working default in tensed clauses; each pattern is what else moves, with an LF whose scope matches c-command. Light d, used below, is withdrawn (Grant, 2026-11-14); light n is an OPEN structural question. Position-class template statements below describe the order of prefixes; they are not the derivation (§3). Flat-S and other pre-VPISH notation below is labelled contrast. Status on the spine: see unified structure, §1. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.

Grant's intuition is that the Navajo verb doesn't take an ordinary noun as its argument. It takes a theta role, a feature bundle, or a “proto-word”, something that is not quite a word. This page states that as a precise hypothesis, compares it with the literature on pronouns as feature bundles, and tests it against the configurational objections. Written 2026-10-14 (PT).

Ground rules for this page (Grant's). Nobody has fully worked out Navajo arguments. Each existing analysis (Jelinek, Willie, Hale, Speas, Baker) may be only partly right, and none of them is treated here as the judge. For each account we show what it gets right and what it misses, and the hypothesis keeps the parts that work. A source that does not state the hypothesis is not evidence against it. Only data that contradict it count against it.

Bottom line. HYPOTHESIS FB (below) survives every Navajo datum on the site. It also absorbs the strongest configurational point: Speas's own analysis of a pronominal-argument language (Warlpiri) already says that only pronominal clitics discharge the verb's argument positions and that nominals are modifiers (Speas 1990:167–171). The live dispute is narrower than “roles vs words”. It is whether a Navajo noun links to its slot inside the verb's projection (Speas) or outside it (Jelinek, Willie). FB is compatible with both on most data. Condition C, extraction and quantifier scope would decide, and the Navajo chapter of Speas 1990 (pp. 214–281), with its relative-clause argument (pp. 221–237), is still unread. So the hypothesis is not refuted, but it has not yet been tested against her actual Navajo data.

Update 2026-10-14 (8:46pm): Speas's Navajo data now checked. HYPOTHESIS Her 1986 dissertation (Ch. 2 §2.2.9, Ch. 3, Ch. 4; page images) is read datum by datum on Speas 1986 on Navajo. Result: no datum contradicts FB (0 CUTS AGAINST, 3 SUPPORTS, 15 COMPATIBLE). Her two “nouns are arguments” tests, no weak crossover (2.95) and multiple questions (2.96c), target an A-bar-adjunct version of the PAH and do not bear on FB. Her Condition C data, (4.15a) and (4.119), lean toward FB, and she concedes that Navajo nominals “behave like adjuncts insofar as they show ATB effects” (p. 323). FB has to borrow two positional rules from Platero and Speas: FB-3a LINK-NEAR and FB-6 PAR (§6 there). The 1990 revision of her chapter is still unread, so FB-a is partly done.

Update 2026-10-15 (4:46am): Hale & Platero 1974 checked. HYPOTHESIS Their relative-clause and anaphora paper is read on page images on Hale & Platero 1974. No datum contradicts FB's core (0 CUTS AGAINST, 1 SUPPORTS, 16 COMPATIBLE). One add-on rule does change: the positional linking rule FB-3a(a) (LINK-NEAR) is a default, not a hard rule. Their parsing principle (44) is applied “with varying degrees of strictness”, and some speakers let a noun skip the nearest verb in context ((41)–(42)). SOURCED Their (28) already states FB-3a(b), that the noun next to the verb is the object under yi- and the subject under bi- (p. 17). Their determiner analysis of the relativizer (p. 27) fits FB: what a relative must contain is a pronoun-like element, and its noun is optional (§5). New on that page: FB-7 SURF-LINK (linking checked on the surface string, as in their architecture (50)).

Update 2026-10-15 (12:46pm, corrected 1:07pm): Hale 1999 “Conflicting Truths” checked (page; step-by-step derivation and assumption audit). HYPOTHESIS A conditional tension, not a refutation. Hale's (10) and (11a) report that a null subject cannot corefer with a noun inside its theme phrase or inside a clause under ‘about it’, with two people and so no rank confound (pp. 172–173). FB-5 said a bundle never triggers Condition C on a linked noun. That is in tension with (10) and (11a) if ten assumptions hold (A1–A10: Condition C is what excludes the reading, the judgments are stable, an adjunct is out of the subject's reach, and so on). None is independently established for Navajo, and Speas's (4.119) has the opposite judgment on a sentence of the same shape. The 12:46pm wording (“that is wrong”, “the answer is now inside”) was too strong. FB-1 to FB-4 are untouched. FB-3c LINK-IN, FB-5′ and FB-6′ are conditional additions, needed only if the assumptions hold. D6 below is compatible, undecided.

On this page: 1 The hypothesis · 2 Feature bundles in the literature · 3 Each account: right and missing · 4 The capacity feature (Creamer) · 5 The configurational objection, steelmanned · 6 The diagnostics under FB · 7 Derivations · 8 Verdict · 9 Next steps

1. The hypothesis

FB-1  An argument of V (or of a grammatical P) is a bundle
        β = [π person, # number, κ capacity/animacy rank, ρ role]
      realized as a pronominal prefix (or ∅ for 3rd person).
FB-2  ρ is fixed by Σ(√) (the verb's proto-role schema); π, #, κ are
      the referent's features; β is complete without any noun.
FB-3  A noun N carries features [π=3, #?, κ(N)] and no ρ. N links to
      exactly one β whose features it matches (π, #, κ, and the stem's
      sortal class on ρ); position and topicality choose among matches.
FB-4  The voice (yi-/bi-) compares κ of the two bundles: the higher-κ
      bundle is external; bi- marks external ≠ ρA.
FB-5  β is a φ-level object (variable-like), not a D-level one (referential
      R-expression); N may be D-level.
ClauseStatusBasis
FB-1: the argument is a prefix carrying person and numberSOURCEDHale's Pronominal Argument Theory: “Person-Number morphology in the Navajo verb word has the status of full argument” (handout (16), p. 7); Willie 1991:51–52
FB-1: the bundle also carries κ and ρGRANT'S CONJECTURE formalized as HYPOTHESISGrant's “schema of theta roles … prior to realization as a lexical item” (project chat, message 1)
FB-3: nouns link by matching featuresSOURCED in partWillie's ADJ-PHI: “A adjoined [sic] nominal must agree in ‘phi’ features with its pronoun” (p. 53). Sortal match: (22) “round objects”, (24) “plural objects” (pp. 61–62). Position defaults: (17), (20) (pp. 59–60)
FB-4: voice compares rankSOURCED as a description; κ as a feature of the bundle is HYPOTHESISWillie 1991:86 (70); Creamer 1974:29–37 (eight ranks)
FB-5: φ-level, not D-levelHYPOTHESISDéchaine & Wiltschko 2002 typology (§2)

HYPOTHESIS What “not quite a word” means here. In D&W's terms (§2), a φP is “neither D syntax nor N syntax”. It “simply spell[s] out φ-features”, and it is a variable rather than a referring expression (2002:410–411). That gives Grant's “proto-word” a precise sense. The verb's argument is a φ-level bundle, smaller than a noun phrase (DP/NP), which the verb word can take in fully built. The noun is a full phrase that has to be linked to the bundle from outside.

2. Feature bundles in the literature

SourceWhat it saysAccessUse for FB
Jelinek 1984 (PAH)Pronominal clitics/affixes are the arguments; nominals are adjuncts (Warlpiri tree (28), 1984 pre-VPISH notation: CPP sisters of Verb–AUX under S)Opened 2026-10-15 (8:46pm) for attachment (page); earlier via Hale, Speas 1990:166FB-1 core; no verdict on A3 (notation cannot be translated directly; T26g)
Hale (handout; 2003 ms.)Person-number morphology is the full argument, “not ‘agreement’”. Category of true arguments: “D, an X-zero and X-max type” (handout (16)). An independent pronoun is a “contrastive adjunct” (ms.:5–6)OpenedFB-1 sourced. Note Hale says D, where FB-5 says φ (§3)
Déchaine & Wiltschko 2002, “Decomposing pronouns,” LI 33:409–442Three proform types. Pro-DP: argument only, definite, Condition C (R-expression). Pro-φP: argument or predicate, a variable (Condition B), “simply spell[s] out φ-features”. Pro-NP: predicate, constant (pp. 409–411, table (2)). The D/φ/N split “cuts across” phrasal pronouns, clitics and agreement (n. 1; §6). Obviation and different-subject marking are D-agreement, which obeys Condition C (§5)Opened (Georgetown copy)Gives FB-5 a typology and a binding prediction (§6, D6)
Baker 1996Morphological Visibility Condition: every argument must be visible on the head by agreement or incorporationSecondhandFB-1 is a version of it. But Navajo lacks verbal noun incorporation (Hale ms.:13)
Preminger (e.g. 2014, Agreement and Its Failures)φ-agreement as a probe that can fail without crashing; tests for distinguishing clitics from agreementOPEN (cited, not opened)His clitic-vs-agreement tests could tell whether Navajo prefixes are bundles in argument position (clitics) or agreement (Speas). Not yet applied
Classificatory stemsStem choice encodes the Theme's shape, rigidity, animacy and numberGrant's Notes §3; Willie's glosses (22), (24)The patient bundle's sortal features are built into the stem: FB-3's sortal match

3. Each account: what it gets right, what it misses

AccountGets right (FB keeps)Misses (on data, or not addressed)
Jelinek 1984 (PAH)Prefixes are the arguments. Nouns are optional and added for discourse (D1–D2 on the thesis page). 8:46pm: Warlpiri attachment is S-level (T25c page)Predicts free order of adjuncts. Navajo order is fixed by default (Willie (20), *(33c)). Rescued only by “discourse configurational” (W&J 2000). Paper is Warlpiri-centred; no Navajo Condition C
Willie 1991Nouns are licensed by a pronoun and linked by φ-match and position (pp. 51–70). The inverse is chosen inside the word with no nouns present (pp. 86–88). da- follows the agent (pp. 82–84)Some links rest on world knowledge (“horses typically are led”, p. 66) or on nothing visible ((25), p. 62). The inverse agent is stated as ergative, where W&J 2000 keep a constant NOM gloss (T7)
Hale (1973 via Willie; 2003)The PA Theory statement. Bare nominal arguments of postpositions are in base position (ms.:51). The Condition C problem (Perkins) as a testCalls the true arguments category D. That would make them R-expressions in D&W's sense, which sits badly with variable-like uses (fourth person, wh-binding: Willie p. 201). Leaves Condition C open himself
Speas 1986/1990Navajo has ordinary hierarchical structure and subject/object asymmetries. “There is no one parameter responsible for nonconfigurationality” (1990:195). The prefixes must be infixes, accessible to syntax (1990: Ch. 4 §4.5.2, p. 247; 1986: Ch. 3, pp. 200–202, opened). Her own Warlpiri analysis says “Verbal grid positions can only be discharged by pronominal clitics” and nominals (KPs) are modifiers linked to a grid position by theta-identification (pp. 167–170, (73))Checked 2026-10-14 (8:46pm) against the 1986 Navajo chapters (page): she declines the option that the prefixes are arguments that move into the verb at PF (1986:201) on two theory-internal grounds plus two Ch. 2 tests that do not bear on FB. Her (4.15d) Condition C case is confounded by rank. The abstract says overt nominals “behave like arguments” by standard tests (1986). How that squares with Willie's “a noun by itself cannot be an argument” (p. 52) is the open question (§5)
Baker 1996Arguments must be visible on the head. The CED and quantifier predictions make the adjunct analysis testableBuilt on Mohawk. Navajo lacks verbal noun incorporation, so Baker does not classify it as polysynthetic
Creamer 1974The ranking is a noun's “capacity to act upon” others (p. 30), and “the nature of the verb plays a secondary role” (p. 29)Verb-sensitive exceptions (lightning with ‘strike’, p. 33; the group-7 movement rule, p. 36) show that the capacity is read in the event, not just off the noun

HYPOTHESIS What FB keeps from each. From Jelinek, Willie and Hale: the argument is the prefix, and the noun links to it. From Speas: the bundle sits in a real argument position in a hierarchical structure (Spec,v / Comp,V, as in this site's derivations), and noun linking is theta-identification with a grid position, her Warlpiri mechanism. From D&W: the bundle is φ-level. From Creamer and Dowty: κ is a capacity feature. From Willie: linking by φ, sortal class and position.

4. The capacity feature κ (from Creamer 1974)

SOURCED Creamer: higher status comes from “capacity for having intent or purpose; … intelligence; … strength, vigor, aggressiveness, or special potency; … usefulness to man, or relatedness to man; and … animation, or capacity for movement”, “summed up in an inherent ‘capacity to act upon’ in relation to other beings” (1974:30). Equal rank means “either noun may be chosen … The subject chosen by the speaker is the object of major interest or emphasis” (p. 32). A newborn ranks below other humans “only as long as the infant retains the helplessness characteristic of the newborn” (p. 33). Personified nouns are treated as equal in status to whatever they interact with (p. 37). Full data on eight ranks.

HYPOTHESIS Why this fits a bundle. κ is a general capacity of the participant, not a per-event entailment count. That is shown by old age, which causes the death and still ranks below the tree ((c28), p. 37, and eight-ranks §6.3). A general capacity is exactly the kind of inherent feature a bundle can carry before any noun is present, and it is assigned by construal. The same word ‘baby’ gets a lower κ when it means a helpless newborn (p. 33), and a personified tree gets a raised κ (p. 37). So κ belongs to the bundle's referent as construed, not to the word. That is what Grant's “not quite a word” predicts: the verb ranks what fills its slots, and a noun's lexical identity does not fix the rank. OPEN κ is verb-sensitive in two places: lightning with ‘strike’ (p. 33), and the mover in group 7 (p. 36). FB-4 has to let Σ(√) adjust κ (movement entailed by the verb raises it). That is a constrained interaction, but it needs to be stated (FB-4′, §7).

5. The configurational objection, steelmanned

SOURCED (book opened; Navajo chapter not available in the copy we could read) Speas 1990 argues that “there are no configurationality parameters” (p. 2). Every language projects arguments by the same lexicon-to-syntax mapping, and Navajo's surface freedom comes from independent, lexically stated properties (pp. 195–198). Ch. 4 (pp. 202–281) covers subject/object asymmetries (§4.2.3, p. 209), null objects and null topics (§4.3.3, p. 217), relative clauses and the Parallel Function Constraint (§4.3.4, pp. 221–237), “On Navajo Nominals as Adjuncts” (§4.4, pp. 237–240), pronominal clitics as infixes (§4.5.2, p. 247) and bi- as an incorporated pronoun (§4.5.3, pp. 260–275). Through Hale & Platero’s negative-polarity ms. (secondhand; earlier labelled Hale & Perkins), her Parallel Function Constraint (p. 232) explains Perkins's Condition C facts without adjunction. Her 1986 abstract says overt nominals “behave like arguments” by standard tests.

Steelman. If nouns were adjuncts outside the clause, Navajo should look like Mohawk: free order, no subject/object asymmetries, no binding into or out of nominals. Instead it shows a fixed default order (Willie (20)), asymmetries (only the external slot raises; inverse definiteness, pp. 76–77), and the relative-clause facts. So the simplest grammar puts the nouns in argument positions, with the prefixes as clitics (or, for bi-, incorporated pronouns) linked to them.

HYPOTHESIS Where FB stands. Most of the steelman is compatible with FB. FB has hierarchical slots, so asymmetries and fixed default linking are expected (the bundles are ordered in Σ, and nouns link by position default). Speas's Warlpiri mechanism, nominals as modifiers theta-identified with a grid position that only a pronominal discharges (pp. 167–170), is FB-3 in her own vocabulary. She applies it to Warlpiri and, on the abstract's evidence, rejects it for Navajo. If the Navajo chapter shows nouns behaving as arguments on a test where Warlpiri KPs behave as modifiers, that would be contradicting data. Until it has been read, FB stands.

HYPOTHESIS Update 2026-10-14 (8:46pm). The 1986 Navajo chapters contain no such test (datum table). Her final structure (4.124, p. 324) assigns the θ-role to a CHAIN of noun plus clitic. FB assigns it to the bundle and links the noun. No datum in the dissertation tells those two apart.

6. The argument diagnostics under FB

DiagnosticConfigurational predictionFB predictionData so farVerdict
D6 Condition CNouns in A-positions should show Condition C. Perkins's relative clause is explained by the Parallel Function Constraint (Speas p. 232)The bundle is φ-level (a variable, D&W). It cannot c-command a noun as a binder because the noun is linked, not merged, in the argument position. So no Condition C between a prefix and a noun, as in Perkins's case (Hale handout (17))Perkins's sentence: coreference allowed (Hale handout (17))Compatible with FB; also compatible with Speas + PFC. 8:46pm: Speas 1986 (4.15a), (4.119)–(4.121) show the same exemption, and her one Condition C exclusion is confounded by rank: leans SUPPORTS FB (Speas 1986). 2026-10-15 (12:46pm, restated 1:07pm): Hale 1999 (10), (11a,b) give an equal-rank case with i ≠ j, which is in tension with FB-5's exemption if assumptions A1–A10 hold; Speas's (4.119) has the opposite judgment. COMPATIBLE, UNDECIDED. LINK-IN, FB-5′, FB-6′ are conditional additions (page, audit)
Obviation-type disjointness—D&W: disjoint-reference morphemes are D-level. Prediction: the fourth person / bi- marking “someone different” should show Condition C-like disjointnessThe fourth person means “someone different” in all four constructions studied (relatives, clause chains)SUPPORTS (a D/φ split inside the bundle system)
D7 ExtractionExtraction from argument nouns possible (subject to islands)Wh-words are adjunct-like and link to a bundle (Willie p. 201). No extraction out of a noun. No long-distance whNo long-distance questions; no extraction from wh + noun (Willie Ch. 8)SUPPORTS FB. Speas 1986 (2.96c) multiple questions: COMPATIBLE (page)
D8 QuantifiersTrue NP quantifiers with scope from their positionScope is computed on bundles. A quantifier on a noun takes scope via its bundle, and da- (on the agent bundle) shows thisda- follows ρA in both voices (Willie 82–84). The scope of (58)–(59) is unstatedPartly supports; OPEN (T5c)
Possessor yi- bindingLexical DPs in A-positionsBinding of possessor yi- by the bundle, not the nounUnopened thesis argues A-position (abstract)OPEN. Only its data would count

7. Derivations

F1. (8) shí bi-chidí yílchxǫʼ ‘I (am the one who) ruined his car’ (Willie 1991:56)

Σ     Σ(√RUIN) = {β1[ρA], β2[ρTh]}                              HYPOTHESIS
Merge β2 Comp,V ; β1 Spec,v                                     VPISH base positions
Value β1 = [π1, #sg, κ person, ρA] ; β2 = [π3, κ ?, ρTh]          SOURCED (gloss 3s.1s)
Move  V-P: β1 → Spec,M/S (portmanteau with mode, Hale ms.:10)   HYPOTHESIS
A     spell-out yílchxǫ'  (complete clause)                     SOURCED (pp. 51–52)
B     shí [π1] matches only β1 ; bi-chidí [π3] matches only β2   SOURCED (ADJ-PHI)
Result: linking by feature match; no position needed.

F2. (c28) Tʼiis są́ biisxį́ ‘The tree died of old age’ (Creamer 1974:37)

Σ     {β1[ρ causer], β2[ρ undergoer]}                            HYPOTHESIS
Value κ(β2: tree) = group 7 > κ(β1: old age) = group 8            SOURCED (ranks, pp. 36–37)
FB-4  higher κ = β2 → external; β2 ≠ ρA-like causer → bi-         SOURCED (bi- preferred)
B     t'iis links to external β2 (first noun = topic) ; są́ to β1   SOURCED (p. 32 topic rule)
Result: rank is a capacity of the bundle's referent, not an
event-entailment count (old age causes but ranks lower).

F3. FB-4′, the verb's adjustment to κ (group 7, Creamer p. 36)

FB-4′  If Σ(√) entails movement of one participant, raise that bundle's
       κ above a stationary co-argument of the same group.          HYPOTHESIS
       (Creamer: the moving noun "takes higher status than the more
        stationary object", p. 36 — SOURCED)

8. Verdict

8:46pm (T25c), revised 9:20pm: Jelinek 1984's Warlpiri tree (28) is pre-VPISH notation (flat S: V, AUX, CPP; no VP or Spec,vP), so it gives no verdict on A3; the 8:46pm wording “supports A3 for that source” is withdrawn. Baker's later n. 16 contemplates PP/vP adjunction for Navajo (against A3). Attachment height for Navajo stays open; restating the PAH in a VPISH/TP/CP structure is T26g — page.

No Navajo datum on the site contradicts FB. It inherits the confirmed parts of Jelinek, Willie and Hale: prefixes as arguments, nouns linked by φ and sortal match, rank computed inside the word. It adds a precise sense for “not quite a word” (a φ-level bundle, D&W). It gives a home to Creamer's capacity-based ranking (κ). It can be stated inside Speas's own configurational machinery, as nominal linking by theta-identification. It survives Speas as far as we can currently check: her general framework and her Warlpiri analysis are compatible with it. But her Navajo-specific arguments (pp. 209–281) and Barss, Hale, Perkins & Speas 1989 have not been read, so “survives Speas” is provisional. 8:46pm update: the 1986 Navajo chapters are now checked, and no datum in them contradicts FB (Speas 1986 on Navajo). FB adds FB-3a (LINK-NEAR) and FB-6 (PAR), restated from Platero and Speas. The 1990 revision and Barss et al. 1989 remain. The tests that could contradict FB are Condition C minimal pairs, extraction out of nominals, and the scope of quantified nouns under the inverse.

9. Next steps

#StepStatus
FB-aGet Speas 1990 Ch. 4 (pp. 202–281): Internet Archive lending copy phrasestructurei0000spea, or a library. Check §4.3.4 and §4.4 test by test against FBPARTLY DONE 2026-10-14 (8:46pm): the 1986 dissertation versions read (page); 1990 Ch. 4 OPEN (T22d)
FB-bBarss, Hale, Perkins & Speas 1989, “Aspects of Logical Form in Navajo” (Cook & Rice, pp. 317–334): quantifier and binding dataOPEN
FB-cApply Preminger's clitic-vs-agreement diagnostics to Navajo prefixes (do they double, and can they fail?)OPEN
FB-dTest FB-5: can a 3rd-person prefix be a bound variable under a quantified noun (D&W's φ diagnostic)? Needs dataOPEN (speaker judgments)
FB-eRead the possessor-yi- binding thesis for its dataOPEN