The spine: how every Navajo clause is built (Merge, theta roles, V→v→T, and what else moves)
Verbs on this page → dictionary: yiztał ‘kick’ · nishhozh ‘tickle’ · naa-ghá ‘walk around (one)’ · yi-sh-cha ‘cry’
2026-11-14 overhaul (approved by Grant for the live site, 2026-11-14). This page is rebuilt from Grant’s premises of 2026-11-13. VPISH lines up syntax and semantics: the arguments start inside the verb’s projections, where they get their roles. It is not an account of word order. In ordinary tensed clauses the verb moves V→v→T and the subject moves from Spec,vP to Spec,TP (on to Spec,CP or C when needed); that is near-universal and is assumed. So the question for every Navajo pattern is what else moves, and why. The older Navajo literature supplies the data; its analyses appear only as labelled contrast (§S6). The earlier text of this page, with all audit notes, is kept unchanged under History. No Navajo form, gloss, judgment or citation changed.
This page sets out the one clause structure, the spine, that every derivation on the site is built on. It says what is merged where, where each theta role is assigned, which movements are assumed in every tensed clause, how the verb word is pronounced, how logical forms are written, and what counts as a test. The findings of the first-principles reviews, restated on this spine, are on Results on the spine; every phenomenon placed on one structure for the clause, the noun phrase and the verb word is on One structure for Navajo (2026-11-15).
On this page: The spine in brief · S1 Premises · S2 Step by step · S3 Logical forms · S4 What else moves · S5 What counts as a test · S6 Older analyses as contrast · S7 Results · History
The spine in brief
A Navajo sentence is built from the bottom up. The verb is merged first with the participant it directly affects, which gets its role there (Theme, Patient). A light verb v is merged on top, and the participant that acts or experiences is merged in its specifier, Spec,vP, and gets its role there (Agent, Experiencer). This is VPISH: every argument starts inside the verb’s own projections, so the meaning can read each role off the place where the argument was merged. Then tense, T, is merged. In a tensed clause the verb moves up through v into T, and the subject moves up from Spec,vP into Spec,TP. If the sentence is negated, a negation head is merged above TP; if it is a question, a C head with Q is merged above that. The verb word is the V+v+T complex, pronounced where it ends up. Up to here nothing is special to Navajo. What is special to Navajo is what else has to move to give its orders: the verb last, the patient before the agent in the inverse, a phrase before doo that is still negated. Each of those extra steps is a hypothesis that has to be argued from sourced examples.
[CP [C′ [NegP [Neg′ [TP DP_subj [T′ [vP ⟨DP_subj⟩ [v′ [VP [V′ DP_obj ⟨V⟩]] ⟨v⟩]] V+v+T]] Neg⁰]] C⁰]] Specifiers on the left. NegP and CP only when the clause is negated or a question. Angle brackets = silent lower copies. Drawn here with T⁰, Neg⁰ and C⁰ final (HS-1); see S4.
S1. Premises
Only these are assumed. Everything else on the site, including every Navajo-specific rule, is a hypothesis to be tested.
- Theta roles (assumed). Each argument gets exactly one theta role, from the head it is merged with, and each role goes to exactly one argument.
- VPISH (assumed). The internal arguments are merged inside VP and the external argument in Spec,vP, so every argument starts within the scope of its verb and gets its role there. This lines up syntax and semantics. It is not a word-order claim, and VPISH is never a node label.
- The working default in tensed clauses (assumed; Grant, 2026-11-13). V⁰ moves to v⁰ and on to T⁰ (head movement, no head skipped), and the subject moves from Spec,vP to Spec,TP. When a question, focus or topic needs it, a phrase moves on to Spec,CP or a head to C. An example that seems to resist this is flagged OPEN, not smoothed over.
- Syntax matches logic (assumed). Every derivation ends in an explicit logical form, and an operator’s scope is what it c-commands where it is interpreted. A reading that needs different scope needs a different structure, or an explicitly stated extra step (reconstruction, covert movement).
- Specifiers on the left, real X-bar labels (CP, NegP, TP, vP, VP, DP, PP). Head direction is not assumed (S4).
- Two negations. Sentential negation (doo … da) and constituent negation (hanii) are kept apart unless data show they are one thing.
HYPOTHESIS Everything beyond S1 is labelled where it is used: the direction of each head; the inverse Move; NegP’s internal structure; Q placement; covert scope; the internal order of the verb word; any Navajo-particular rule. Light d is no longer among them: Grant withdrew it on 2026-11-14 (below). Telicity is OPEN unless a minimal pair decides it.
S2. Building a clause, step by step
| Step | Operation | What it does | Status |
|---|---|---|---|
| 1 | Merge | Merge V⁰ with its internal argument(s) to form VP. V assigns their roles: Theme or Patient to its complement; a Goal or a postposition’s object where the verb selects one (in a ditransitive, which internal argument is higher is a hypothesis, H3-loc). | assumed (VPISH, theta) |
| 2 | Merge | Merge v⁰ with VP; merge the external argument in left Spec,vP. v assigns its role (Agent, Causer, Experiencer). In a passive this argument is silent and indefinite (H10b/H16a, a hypothesis). A causative adds a further v with its own causer (hypothesis). | assumed (VPISH, theta) |
| 3 | Merge | Merge T⁰ with vP. Which prefix, if any, spells out T is OPEN. | assumed |
| 4 | Head-move | V⁰→v⁰→T⁰, after both targets exist. Any head merged between v and T (for example the distributive head of H21a) is passed through, not skipped. | working default |
| 5 | Move | The subject moves from Spec,vP to left Spec,TP; the lower copy is silent. It keeps the role it got in Spec,vP; the Move assigns none. | working default |
| 6 | Merge | Sentential negation only: merge Neg⁰ above TP. On present evidence Neg⁰ is the final da, and doo is a lower marker adjoined inside its complement (H23). Constituent negation hanii is not merged here; it attaches to the phrase it negates (H20). | HYPOTHESIS |
| 7 | Merge | Questions and other clause types only: merge C⁰ above NegP or TP. Polar Q is merged at the clause edge (H8 Q-edge); a question word’s Q may be displaced to it (H12). | HYPOTHESIS |
| 8 | Move | Only when required: a phrase moves to Spec,CP (a fronted question word, H14; a host for =sh). Not obligatory in every question: question words can stay where they are merged. | working default when triggered; each trigger HYPOTHESIS |
| 9 | Spell-out | The verb word is the V+v+T complex, pronounced at T with its pronominal and other prefixes. Which exponent spells out which head, and why the prefixes come before the stem, is OPEN (see S4, last paragraph). Enclitics (-ísh, =sh) are written on a host at PF; the host is not their scope position. | OPEN |
| 10 | Interpret | LF: read each role at the argument’s first-merge position and each operator’s scope at its interpreted position. Generic negative yes/no question: ASK_POL(¬∃e[PRED(e) ∧ Agent(e,a) ∧ Theme(e,t)]). | assumed (S3) |
Analysis A: Working default with T⁰ final (HS-1): nothing else has to move for verb-final order Hypothesis
Analysis B: Working default with T⁰ initial (HS-2): the remnant VP must also move above T (HYPOTHESIS) Hypothesis

Bracket notation (copy)
A: [CP [C′ [NegP [Neg′ [TP [DP agent a] [T′ [AdvP lower negative marker] [T′ [vP ⟨DP⟩ [v′ [VP [V′ [DP theme t] [V⁰ predicate]]] v⁰]] T⁰]]] Neg⁰]] C⁰]] B: [CP [C′ [NegP [Neg′ [TP [DP agent a] [T′ VP [T′ T⁰ [vP ⟨DP⟩ [v′ v⁰ ⟨VP⟩]]]]] Neg⁰]] C⁰]]
A: [CP [C′ [NegP [Neg′ [TP [DP agent a] [T′ [AdvP lower negative marker] [T′ [vP ⟨DP⟩ [v′ [VP [V′ [DP theme t] [V⁰ predicate]]] v⁰]] T⁰]]] Neg⁰]] C⁰]] B: [CP [C′ [NegP [Neg′ [TP [DP agent a] [T′ VP [T′ T⁰ [vP ⟨DP⟩ [v′ v⁰ ⟨VP⟩]]]]] Neg⁰]] C⁰]]
Structure as a list
Analysis A
- CP
- C′
- NegP
- Neg′
- TP
- DP (id S-high; landing site of Move from subject): agent a
- T′
- AdvP (id negmark): lower negative marker
- T′
- vP
- DP (id subject; goal of Agree φ (optional); struck copy; moved to S-high): agent a
- v′
- VP (id lowVP)
- V′
- DP (id object): theme t
- V⁰ (head; id V; head-moves to v): predicate
- V′
- v⁰ (head; id v; head-moves to T)
- VP (id lowVP)
- T⁰ (head; id T; probe for Agree φ (optional) with subject; hosts the “verb word” (V + v + T))
- vP
- Neg⁰ (head; id Neg)
- TP
- Neg′
- C⁰ (head; id C; [Q, polar])
- NegP
- C′
Analysis B
- CP
- C′
- NegP
- Neg′
- TP
- DP (id S-high; landing site of Move from subject): agent a
- T′
- VP (id VP-high; landing site of Move from lowVP)
- T′
- T⁰ (head; id T; probe for Agree φ (optional) with subject; hosts the “verb word” (V + v + T))
- vP
- DP (id subject; goal of Agree φ (optional); struck copy; moved to S-high): agent a
- v′
- v⁰ (head; id v; head-moves to T)
- VP (id lowVP; struck copy; moved to VP-high)
- V′
- DP (id object): theme t
- V⁰ (head; id V; head-moves to v): predicate
- V′
- Neg⁰ (head; id Neg)
- TP
- Neg′
- C⁰ (head; id C; [Q, polar])
- NegP
- C′
S3. Logical forms
eranges over eventualities, states included (sis used for states). Event closure∃eapplies at vP.- Roles are written
Agent(e,x),Patient(e,y),Theme(e,y),Exp(e,x),Goal(e,z). Each argument bears one role; a result or path relation such asOUTcounts as a role when its argument is an individual. - Sentential negation contributes one
¬, from Neg⁰, scoping over the event:¬∃e[…]. A result with∃e[… ∧ ¬P(e)]is a different meaning and does not stand in for it. - Questions:
ASK_POL(p)asks whether p;ASK_WH(λx …)or?x.…asks for values of x. Q above Neg givesASK_POL(¬p). - Constituent negation: assertion
¬[focus = value]plus a presupposition for the rest of the clause. - Relatives and nominalized clauses: a linked enclitic gives
ιx …(the individual); an unlinked one names the fact (H34). - Scope = c-command at the interpreted position. Reconstruction (reading a moved phrase at a lower copy), covert movement (H22a) and choice functions (H22b) are extra steps and must be stated. Source translations are SOURCED; role labels and LFs are HYPOTHESIS.
S4. What else moves
With the working default in place, the Navajo-particular claims are all claims about additional movement, or its absence. The site’s current hypotheses, each argued on Results on the spine:
| Pattern | What else moves (HYPOTHESIS) | Main evidence | Status |
|---|---|---|---|
| Verb-final order | HS-1: nothing (T⁰ final). HS-2: the remnant VP, above T⁰ (T⁰ initial). | Headedness; clause spine | OPEN |
| Inverse (bi-) | The patient moves above the raised agent (H3), only if definite or presuppositional (H6′) | NX-029/030; NX-085 vs *NX-086 | HYPOTHESIS; trigger and landing OPEN |
| Fronted question word | Moves to Spec,CP when animacy keeps the roles apart (H14) | Willie 1991:206–207 | HYPOTHESIS |
| Displaced =sh | Q moves to the clause edge (H12-move), after the inverse Move | NX-085 vs *NX-086 | HYPOTHESIS; H12-pf OPEN |
| Phrase before doo, still negated | Moves above doo but stays under Neg⁰ da (H23) | NX-287, NX-288 | HYPOTHESIS |
| Missing argument with -í- | The argument moves into Neg⁰’s domain; written on the verb at PF (H18a) | NX-228 vs NX-229 | HYPOTHESIS |
| Wide-scope indefinite | Covert Move above the subject (H22a), or no Move and a choice function (H22b) | NX-267 | OPEN between the two |
| Relative clause | Nothing: D⁰ (=yę́ę, =ígíí) takes CP and is linked by Agree (H27b) | NX-314; NX-312/313 | HYPOTHESIS |
| Question with “you think / she said” | Nothing: two clauses side by side (H40a) | NX-378, NX-384, NX-385 | HYPOTHESIS; H40b not excluded |
| Single argument of ʼádin | Must reconstruct, or not raise | NX-292 | OPEN (resists the default) |
GRANT'S HYPOTHESIS The verb word (restated 2026-11-15 on Grant’s correction). Once V has raised through v to T, the verb word is one complex head. Grant’s hypothesis is that headedness is fixed locally by Merge site, and that Navajo linearizes the sublexical domain opposite to the phrasal one: the clause is head-final, the word’s prefix heads are initial. The order of heads inside the word is then predicted to mirror their order in the clause (a higher head further left), which is the order the Mirror Principle gives for prefixes; sourced orders that go the other way are misfits for it, not confirmations (headedness fact-check, row 14). Whether every sourced prefix order fits is OPEN: the conjunct core fits; aspect, object/subject, distributive and preverb orders do not yet (unified structure §3.2; Grant’s two-tier linking proposal in §3.3, and the earlier, untested sublexical V-to-T in §3.4). The older position-class template (S6) describes the order; it does not derive it.
S5. What counts as a test
- A test is a sourced form with its judgment and reading, copied with its source and example number, set against a hypothesis that predicts that judgment and that reading. Minimal or near-minimal pairs count most.
- A hypothesis passes a test (“fits”, “consistent”), it is not “confirmed” by it. One starred example is weak evidence. Contradicting data count against a hypothesis; no older account is treated as a judge.
- Not evidence: a translation alone (an English passive does not show a Navajo passive), a mode or viewpoint label (not a telicity test), a silent head proposed to fit the order, an enclitic’s written host (not its scope position), or a grid cell with no example (not a judgment of ungrammaticality).
- Every review runs the question grid: polar or wh × affirmative or negative × active (direct), inverse and passive. Cells are filled only with sourced examples.
- No Navajo form or judgment is ever constructed. Where a test needs a sentence no source prints, it is listed as OPEN, for a speaker or another source.
S6. Older analyses, as labelled contrast
The Navajo literature was written before current syntax had these tools, so its analyses are not fully informed. Its data (forms, glosses, judgments, page and example numbers) are used exactly as printed. Its analyses appear on this site only as contrast, never as the structure a derivation is built on.
| Analysis | Where from | What we keep | On the spine instead |
|---|---|---|---|
| Position classes and the verb template (“Pos. IV”, “Pos. V”) | Young & Morgan 1987; Kari 1976 | The order of prefixes, as data | The order is something the spell-out of V+v+T must produce (S4, OPEN) |
| Qual > M/S > v > V, skeleton filling, no head movement | Hale 2001, 2003 | Examples, page numbers, segmentations as printed | V→v→T is assumed; “M/S” is not a node label |
| S-adjunction: nouns as adjuncts to S, “Maximal Verb Complex”, topic and focus adjuncts | Willie 1991; Willie & Jelinek 2000 | All examples, indices and judgments | Nouns are linked to participants (H4″); orders come from Move (H3, H14) |
| Pronominal Argument Hypothesis | Jelinek; Willie 1991 | Data on noun omission | Whether overt nouns are arguments or linked adjuncts is OPEN, not presupposed |
| NP-PRO constraint, parsing principles | Platero 1973, via Speas 1986 | The one-noun restrictions | H4″, a stated default with a theta override |
| Noun–clitic chains | Speas 1986 | (4.6)–(4.8) | Roles assigned at first Merge (S1) |
| A/B rule sets; the headedness changeover at the word; CYC | This site, 2026-10 (from Grant’s earlier proposal) | The sourced orders they were tested on | Head direction of T⁰, Neg⁰, C⁰ plus spell-out of the verb word (headedness) |
| ID-, RD-, EH-, FB-rules; “roles, not words” | This site, 2026-10 (thesis dropped 2026-11-02) | Data and rule IDs, as a record | Not used |
| Light d (d takes a DP, as v takes VP) | Grant’s proposal | A record of the proposal and its tests | Withdrawn by Grant, 2026-11-14. Never needed in reviews 1–13; the always-definite version is contradicted by NX-372 |
2026-11-14: light d withdrawn; light n an open question. Grant’s words (2026-11-14): light d “was just a proposal to account for facts … its nothing i’m deeply attached to … the existence of this unit was simply posited and i’m willing to unposit it.” Light d is therefore withdrawn, not a hypothesis still under test. In thirteen first-principles reviews it was never needed; the determiner-like elements that do work are all overt (results §13). The light-determiners essay and the dated reviews that test light d stay as a record.
OPEN Light n. Grant also noted that if DP and TP are built alike, there may be a light n in the noun phrase, parallel to light v in the clause. This is a structural question, separate from light d’s job of covering particular facts, and it is not a premise of the spine. No review so far has found data that bear on it (review 13 says so explicitly); a review raises it only when its data do.
S7. Results
The findings of first-principles reviews 1–13, restated on this spine with NX numbers and logical forms, are on Results on the spine: nouns and participants, the inverse, question words, sentential and constituent negation, questions with negation, quantifier scope, relative clauses, questions with “you think”, reported speech, =go clauses, passives, telicity, light d (withdrawn) and light n (OPEN), and the question grid. Open problems: what resists, and what is open.
History: the earlier versions of this page and the audit notes
Everything below is the page as it stood before the 2026-11-14 rebuild, kept unchanged as a record: the audit notes of passes 1–7 and the earlier sections 1–8. Where it differs from the spine above, the spine above supersedes it.
2026-11-11 audit pass 1. The earlier bracket tree mixed categories (vP/VoiceP, ApplP/PP), and light-d/Asp layers were built into the default. The core and both figures are redrawn with real X-bar categories, a left Spec,vP subject and explicit LF. GRANT’S CONJECTURE V→v→T and subject→Spec,TP are tested first; VPISH fixes only the base position, not either movement. Optional d/Appl/Asp and prefix realization stay OPEN. The old blanket objection to head movement depended on left adjunction and is corrected below.
2026-11-11 audit pass 3: syntax accuracy. All labels below are HYPOTHESIS unless marked. No Navajo form was added, changed or reconstructed in this pass.
Worst errors fixed.
- English particle-verb trees, panels A and C (phrasal verbs): the object sits in Spec,PP, not in the complement of throw, yet the LF gave it two theta roles (Theme of throw and the OUT role). It now has one (OUT); being thrown follows from Cause(e,s).
- Navajo prefix/phrasal and changeover trees, panels A and B (headedness, phrasal verbs): the directional AdvP is a VP adjunct but its LF OUT(e,t) gave the object a second role besides Theme. The LF is now OUT(e), an event modifier.
- Five working-note pages stated the A/B headedness changeover as a fact (“the word is the changeover point”, “the changeover … is visible in step 6”). Each sentence now says it is a hypothesis, not established.
- Three rule/derivation blocks written with Hale 2003’s M/S head (subject-head SH-7, pl-move-reason, deictic-order) were not marked as older notation; they are now labeled as a historical record, not our trees.
- The home page still framed the dropped “roles, not words” thesis as the flagship question, and its “Newest” item was stale. Both are fixed.
Counts by finding type. Theta-role errors in tree LFs: 6 panels in 3 trees. Changeover stated as fact: 7 sentences on 5 pages. Other overclaims (“confirmed”): 2. Dropped-thesis framing: 6 spots on 2 pages. Unmarked older notation: 3 blocks. “(VPISH)” step annotations: 3. Light-d pages given an explicit HYPOTHESIS note for dP: 10. Verdict-word notes on historical thesis pages: 3. Status badges brought into the label set: 59 on 20 pages. Local box file paths removed from public text: 105 on 37 pages. Undefined rule IDs: 0 (see below).
Checked and found sound. All 11 tree files: real categories only, binary branching, specifiers on the left, each head projecting its own category, subject first merged in Spec,vP, every panel with an LF. Every Move and head movement now passes a c-command test, every head movement passes the Head Movement Constraint (a skip must be flagged), and LF scope matches c-command for Neg over the event and for Q over Neg. In the questions trees, doo…da stays one clause-wrapping negation, consistent with review 4 (NX-203–217). V-to-T stays neither proven nor refuted; the changeover stays not established; light d is a hypothesis; telicity is OPEN.
Correction to the pass 2 report. Pass 2 listed about 65 “undefined” rule IDs (for example PL-1–4, OBJ-1–4, REFL-1–3). That was a false positive: those strings are Leipzig glosses (PL-1-mother, REFL-1.SBJ, INDF.OBJ-4.SBJ) or labels defined inline in a derivation. A stricter check finds 653 rule IDs referenced and 0 undefined.
New automatic checks. site/check_tree_categories.py now also enforces banned labels (VPISH, S, W, M/S, SC, Prt…), c-command of every movement, the HMC, LF scope order for Neg and Q, and one theta role per argument with every role filled; site/test_tree_categories.py has 9 new tests. Badges outside the label set now fail the build. Full report: site/audits/2026-11-11-pass3.md.
Deferred to pass 4. (1) 66 interlinear rows on 28 pages whose segmented form and gloss have different numbers of hyphens; a spot check shows they copy the source’s own finer gloss, so each needs its source page reopened before any change (list in the audit file). (2) Free translations are not yet machine-checked against glosses. (3) The theta check reads named roles only (Agent, Theme, Exp, Goal…); result and path relations such as OUT are checked by hand. (4) 450 older formal units in working notes remain OPEN until re-derived one by one. (5) Verdict words in running text on the historical thesis pages are explained by a note, not rewritten. (6) Forms damaged in the PDF (¬, ²) stay flagged and are not reconstructed.
2026-11-11 audit pass 4: syntax accuracy (gloss alignment, translations, theta, tree data)
Authorized by Grant (“continue to revise and check for syntax accuracy”). Same rules as passes 1–3: no Navajo form is invented or reconstructed; every change below either restores what a source prints or corrects our own gloss. Full list: site/audits/2026-11-11-pass4.md in the repository.
What was checked
- Form/gloss hyphen alignment. Re-extracted every interlinear row: 64 rows on 28 pages had a different number of hyphens or clitic marks in the form and the gloss. Each was reopened in its source. 53 rows (22 distinct cells) match their source exactly — Willie 1991, Hale & Platero 1996, Willie & Jelinek 2000, and the Navajo Language Academy notes sometimes gloss more finely (3.3.neck:sever) or more coarsely (Kinłání=jįʼ “to”) than the hyphens. These are kept as printed and marked with a † and a SOURCED note under the table. 4 rows (Hale 1997, y-/b-ní-Ø-Ø-¬-ʼ²í) segment a damaged print we cannot check: marked † OPEN, nothing reconstructed. 6 essay rows were a builder bug (below). 1 row was our own gloss missing hyphens (below). A new checker,
check_igt_alignment.py, now fails the build check on any unmarked mismatch: 0 remain. - Free translations against glosses (person/number, tense/aspect, negation scope, direct/inverse, passive): 625 rows screened, 9 flagged, all 9 confirmed against the source as correct (idioms such as doo ʼashtʼéʼégóó “extremely”, chʼééh “in vain”, Witherspoon’s “let” readings, the postposition-based alternation in possessor raising per Willie p.196). No translation changed.
- Theta checker extended to relation predicates (OUT, IN, UP, DOWN, AWAY, OFF, ON, INTO, ONTO, AT, TO, FROM, BECOME, CAUSE, RESULT). A relation whose argument is an individual, not an event or state, now counts as a role. Regression test: the pass-3 “throw the rat out” LF (Theme(e,r) plus OUT(s,r)) is caught. A panel can declare a deliberate composite role as a hypothesis (“θ-composite: r”), which Ramchand & Svenonius’s complex-predicate panel B now does. 0 errors on the current 11 trees.
- Trees against their cited NX entries. All 6 trees that cite NX entries (9 entries) were checked: every Navajo leaf is a printed token of the cited form, and subject/object roles in each tree and LF agree with the entry’s gloss and translation (e.g. NX-153 1-2-know = you know me; NX-086 inverse biztał = agent is the wh-word). 0 mismatches. A new automatic check makes any future non-cited leaf fail.
Worst five
- Essays dropped the first Navajo word of every quoted example. The essay builder removed a cell that had already been removed, so e.g. Witherspoon’s Hastiin łį́į́ʼ yiztał showed as “łį́į́ʼ yiztał” with the glosses shifted one column. Fixed on all 11 essays: events, obviation and others.
- IMP for IMPF. Hale & Platero 1996 gloss the mode of naa-ghá as IMPF (imperfective); we printed IMP, which reads as imperative. Fixed on derivations and walk (sg).
- Our own prefix gloss did not line up with the prefixes in the nishhozh table on Navajo verbs. Hyphens added to the gloss.
- Hale 1997 segmentation is ours, not Hale’s, and one sign is damaged in the print. Now marked OPEN on inverse agree and the inverse/fourth-person topic.
- The pass-3 theta bug class is now caught automatically (object getting a role from both the verb and OUT): note on the prefix-order phrasal-verb section.
Still OPEN
- Hale 1997 y-/b-ní-Ø-Ø-¬-ʼ²í: segmentation and gloss need the original print.
- Gessner 2003 (cited for the walk-sg rows via H&P) is not available to us; the IMPF fix follows H&P 1996 directly.
Deferred to pass 5
- Translation screen used heuristics (person, negation, passive, inverse words); a full manual read of all 625 rows against sources is still due.
- Trees without NX citations use placeholders only; when real data is added they should cite NX entries so the new leaf check applies.
- Legacy pages’ prose glosses outside tables are not covered by the alignment checker.
2026-11-11 audit pass 5: translations read by hand, prose glosses, our own derivations, schematic trees
Authorized by Grant (“yes please keep correcting”). Same rules as passes 1–4: no Navajo form is invented, reconstructed or changed. Every change below either restores what a source prints or corrects our own wording. Full list: site/audits/2026-11-11-pass5.md in the repository.
What was checked
- Translations read by hand. All 164 interlinear rows on the syntax and derivation pages (derivations, syntax, clause spine, inverse as Agree, A/B spell-out, headedness, interrupted verb complex, =sh vs -ísh, and the where-nouns-sit and prefix-order topics) were read one by one for who acts on whom, person and number, negation, tense/aspect and passive. Every row used in a tree or verb-movement test was also checked against its source: NX-085 and NX-086 (Willie 1991 p.211, page image), NX-094 (Hale 2001 p.678), NX-153 and NX-159 (Perkins & Fernald ch.8 pp.194–195), NX-205 and NX-213 (Navajo Questions database entries 804 and 2199). Doubtful rows were reopened in the source: Willie & Jelinek 2000 (49), (50), (57)–(62) on page images; Willie 1991 (1), (18)–(21), (54)–(57), (67)–(69); Hale 2001 (15) on the page image. No translation was wrong. Two cases that look odd are printed that way in the source and are kept as printed: (59) and (60b) are the same sentence with different readings in W&J p.272–273, and Willie’s (1) glosses yiicha as “2d.crying” under ‘We … are crying’ (already marked [sic?]).
- Glosses outside tables and abbreviations. A count of IMP, IMPF, IPFV, PERF, PFV, OPT, PROG, FUT, ITER, 3i, 3o and 4 across the site found one real clash: our own “Leipzig” lines for Willie (24) and (26) used IMP (Leipzig imperative) for her impersonal. They now use IMPERS, the label the same page already uses for (94). Willie’s own category name IMP in prose, and Oxford’s Algonquian INV.IMP, are their authors’ labels and are kept. PERF and PFV both occur, each as its source prints it. 77 inline prose glosses of the shape form (gloss) were checked for alignment; none is misaligned once citation labels are excluded. One gloss cell used a quotation mark for the glottal stop (ʼnii); fixed. A new checker,
check_gloss_abbrev.py, now fails on IMP in a Navajo gloss row, a U+2019 glottal in an interlinear cell, or a misaligned inline gloss: 0 remain. - Our own derivations and LFs re-read (23 tree panels plus the formal blocks on derivation, clause spine and headedness). Scope matches c-command in every LF. In all 4 negation panels, NegP contains TP, the LF has one ¬ (doo…da is one clause-wrapping negation), and the question operator is above it. No page claims the headedness changeover is established. Fixed: 11 panels that move V to T now each say it is GRANT’S CONJECTURE, untested, neither proven nor refuted; the changeover figure now words the head-direction split as a hypothesis; the clause-spine formal statement says the same. The tree checker now enforces all of this (5 new tests).
- Placeholder-only trees. 4 trees draw only placeholders (the two derivation skeletons, the light-determiner comparison, the assembly rivals). Their pages do not name a single example to draw from, so instead of attaching words they are now titled “Schematic”. The assembly-rival title says that NX-094 motivates the test but its words are not drawn.
Worst five
- Tree panels moved V to T without saying it is untested. 11 panels now state that V→v→T is GRANT’S CONJECTURE, neither proven nor refuted: headedness, verb as derivation, the where-nouns-sit and prefix-order topics, and the light-determiners essay.
- The changeover figure stated the head-direction split flatly. It now says this is a hypothesis and not established: headedness.
- IMP used for “impersonal” in our Leipzig lines, where IMP means imperative: absolutive indefinite, indefinite person, indefinites topic.
- Placeholder trees looked like analyses of real sentences. Now titled “Schematic”: verb as derivation, headedness, light determiners.
- A glottal stop typed as a quotation mark in a gloss line: how high is da-?.
Still OPEN
- Hale 1997 y-/b-ní-Ø-Ø-¬-ʼ²í (from pass 4): needs the original print.
- Gessner 2003 is still not available to us; the D8 causative rows rest on the page-image check recorded earlier and on Hale & Platero 1996.
Deferred to pass 6
- Reopen the source for the remaining hand-read rows that were checked only for internal consistency this pass (NLA polar-question rows, Hale 2003 rows on A/B spell-out, Willie §3.2 postposition rows).
- The other ~460 translation rows outside the syntax/derivation pages.
- Prose glosses in other shapes (“glossed …”, quoted glosses) are not yet covered by the checker.
Tree panels that move V to T now each say that V→v→T is GRANT’S CONJECTURE: untested, neither proven nor refuted. (Superseded by audit pass 7, 2026-11-13: V→v→T is now the working default.)
Placeholder-only trees are now titled “Schematic” so they are not read as analyses of a particular Navajo example.
2026-11-11 audit pass 6 report
Scope. Same rules as passes 1–5: forms are never invented, reconstructed or changed; only the labels SOURCED, HYPOTHESIS, SPECULATIVE, OPEN, NONE and GRANT’S CONJECTURE are used.
1. Sources reopened for rows pass 5 checked only for consistency. NLA yes/no grammar (version 3.2022) exx. (1)–(4), (3a–c), (9), (10a, b), (11a, b), (15) and the NLA Postposition Lexicon sentence Q2; Hale 2003 spell-out rows N2–N5; Willie 1991 §3.2 postposition rows (26)–(36). All match their sources (forms, glosses, translations, brackets). 0 corrections.
2. Translation rows outside the syntax and derivation pages. 1,117 distinct rows with a translation were read: 234 first shown on the dictionary, 165 on topic pages, and 718 on the other notes and essays. About 45 doubtful-looking rows were checked against the sources (Willie 1991, Saxon 2003 via the note, the NLA lexicons, Smith, Perkins & Fernald 2007): who acts on whom, person and number, negation, tense/aspect and the inverse/passive. No translation differed from its source; every odd-looking one (e.g. Willie (55) ‘John was kissed by two girls’, (74b) lííʼ, Speas’s ‘bit’ under biztał) is printed that way and already flagged. One correction: instrument-voice and the postpositions essay said the instrument bee occurs “in both 3/3 sentences”, but NLA 92 is 3/3 only by its translation (its gloss line marks a first-person subject). Both now say so; the point is OPEN (T51a).
3. Checker. check_gloss_abbrev.py rule 4 applies the alignment test to the other prose-gloss shapes after a segmented Navajo form (quoted gloss, “glossed …”, “is glossed …”, “= …”). 38 such glosses on the built site; 0 mismatches. 5 new tests (13 in all).
4. Labels on our own claims, non-syntax pages. A paragraph screen for V-to-T, the changeover, light d and telicity found 377 unlabelled mentions; most are source reports, headings or text under a labelled section. Fixed: three visible tree-panel titles that move V to T now say GRANT’S CONJECTURE (the changeover and prefix/phrasal figures); §8 of absolutive-indef and indef-person and §12 of fourth-person-uses (statements of the A→B changeover) are now labelled HYPOTHESIS; the inchoative derivations that called ‘get cold’ telic now say this is our reading of the English gloss, with telicity OPEN. Light d and the changeover stay labelled as Grant’s proposals (GRANT’S CONJECTURE) or hypotheses.
Plateau. Pass 6 found no translation or form errors against the sources and only labelling slips. Corrections have plateaued for the translation layer. OPEN carried forward: NLA 92 subject (T51a); NLA A1 Jáan Mary bił baghan was not found again in the live lexicon this pass (unchanged, not re-verified).
2026-11-13: V-to-T and subject raising are the working default
What changed, in plain words. Until today the site treated two movements as Grant’s conjecture, to be tested against an analysis in which nothing moves: the verb moving up to T (V→v→T), and the subject moving from where it starts (Spec,vP) up to Spec,TP. Grant corrected this on 2026-11-13. VPISH, the VP-Internal Subject Hypothesis, is not a way to account for word order. It frames the verb’s arguments so that syntax and semantics line up: the arguments start inside the verb’s phrase, where they get their roles. In ordinary sentences with tense, in most languages, the verb’s head moves to T and the subject moves from Spec,vP to Spec,TP (and on to Spec,CP or C when a question or focus needs it). That is close enough to universal that it should be assumed.
What the site now does.
- In every tensed clause, V→v→T and subject Spec,vP→Spec,TP are the working default. They are drawn in the trees and used in the derivations, not labelled GRANT’S CONJECTURE, “untested” or “neither proven nor refuted”.
- The useful question is no longer whether the verb moves but what else has to move to give the Navajo order, where the verb comes last. If T⁰ is final, the raised verb is already last and nothing else has to move. If T⁰ is initial, the material left in the verb phrase (object, postpositional phrases, adverbs) has to move above T, for example as a remnant VP. Each such extra movement is labelled HYPOTHESIS; which head direction is right is OPEN, and the headedness changeover is still not established.
- The analyses in which nothing moves (PF lowering, template filling, “no-movement rival”) are kept only as labelled contrast, mainly for how prefixes end up inside the verb word. Tree panels that only repeated a clause with no movement were removed or redrawn; they gave the same string and the same meaning, so no data were lost. On Grant’s instruction (2026-11-13) the last three panels that drew the verb word with no V raising were removed too, so no tree now draws a tensed clause without V in T.
- Sourced examples that seemed to bear on verb movement are kept, now as tests of the default. *NX-246 (hanii after verb + object) is what the default predicts, conditional on H20; its two follow-up tests are OPEN. NX-292 (ʼádin “not exist”) resists the default unless its one argument is read in its lower position or does not raise; that is OPEN.
- Unchanged: sentential negation (doo … da) is kept apart from constituent negation (hanii); telicity is OPEN; light d is a tested hypothesis; LF scope must match c-command; specifiers are on the left and VPISH is never a node label. No Navajo form, gloss, judgment or source citation changed.
The tree checker now enforces this: every T⁰ in a tensed-clause tree must be the target of V→v→T and the subject must be raised to Spec,TP, unless the panel carries an explicit “V not in T … OPEN” note explaining why. Full list of changes: the pass 7 report.
2026-11-13 audit pass 7 report: V→v→T as the working default
Authorized by Grant (2026-11-13: “This isn’t my syntax if you don’t do that for real”). Same rules as passes 1–6: no Navajo form, gloss, judgment or citation is changed. Full list: site/audits/2026-11-13-pass7.md in the repository. Plain-words summary: the 2026-11-13 note.
- Trees (14 of 15 tree files; 28 panels). 8 panels that left V in place now draw V→v→T with the subject in Spec,TP (the second panels of the negative-polar NPI and passive-glossed wh figures; both panels, both clauses, of the long-distance wh figure; both panels of the inverse Q-host and negative-question figures). 3 panels that only repeated a clause with no verb or subject movement were removed (negative disjunct, the two relative-clause figures); each gave the same string and LF. 14 panels that already raised V are relabelled “working default” and say what else has to move. 3 panels that drew the verb word with no V raising (prefix/phrasal B; assembly B, PF lowering, and C, template filling) were first kept as labelled contrast with a “V not in T … OPEN” note, then removed on Grant’s instruction, 2026-11-13. They carried no Navajo form, gloss or judgment not shown in the remaining panels (prefix A; assembly A and D, which keep their letters so older dated notes still match). The trees now have 22 panels in these 14 files. The PF/template accounts remain only as table rows and prose contrast on the headedness page.
- Dated review sections (9): first-principles reviews 4–11 and the phrasal-verbs update. Each has a pass-7 note; “GRANT’S CONJECTURE … tested against a no-movement rival” is now the working default, verdicts say what else has to move, and the no-movement analyses are labelled contrast.
- Method and formal statements: this page’s core, step table and surface-order section; the clause-spine formal statement; the headedness movement section (§9); background and home-page summaries. Pages whose dated pass-1 notes called V→v→T a comparator to test get a pass-7 note; the old notes are kept as written.
- Checker: the rule “V-to-T panels must be labelled GRANT’S CONJECTURE” is replaced by “tensed-clause trees show V in T and the subject in Spec,TP, or carry an explicit OPEN note”; panels that raise V must call it the working default, and calling V→v→T a conjecture, untested, or neither proven nor refuted now fails. 5 new tests replace 1.
- Also fixed: the MIT manuscript hale014 on negative polarity is by Hale & Platero, not “Hale & Perkins”, on the queue and sources pages.
Left OPEN under the default
- Head direction of T⁰: final T⁰ gives verb-final order with nothing else moving; initial T⁰ needs remnant-VP movement above T (HYPOTHESIS). The headedness changeover is not established.
- NX-292 (ʼádin): its lone argument must reconstruct or not raise.
- *NX-246: the default’s two follow-up predictions (whole-clause focus; hanii after the object) need speakers or a source.
- How a low directional exponent (chʼí) and other prefixes reach the verb word once V has raised: the PF step, if any, is OPEN.
- What triggers the second subject move above doo (NX-223/229, NX-288).
This page describes the Navajo verb word as a minimalist derivation. The subject DP starts in left Spec,vP. The core builds VP, vP and TP. In a tensed clause, V→v→T and subject raising from Spec,vP to Spec,TP are the working default (see the 2026-11-13 note); VPISH only fixes where the subject starts and gets its role. Additional aspect, applicative and nominal structures are hypotheses, not part of this core. Position classes such as Young & Morgan's "Pos. IV" and "Pos. V" are treated as data that the derivation has to account for, not as the description itself. The earlier skeleton came from a 2026-11-04 from-memory sketch without opened sources; the current generic core replaces it. HYPOTHESIS throughout, unless a claim is marked otherwise.
On this page: 1 Base structure · 2 Tree · 3 Which step is what · 4 The surface-order problem · 5 Word and clause · 6 Position labels as data · 7 References · 8 Open questions
1. Base structure, bottom to top
- VP. The root V merges with its internal argument. The core uses an object DP. An extra light determiner selecting a DP/NP, as in Light determiners (analysis D), remains an optional analysis rather than a required layer. GRANT'S CONJECTURE / HYPOTHESIS
- P/Appl (optional). When the verb has a postpositional object, a low head sits between V and v and takes its own dP from the same d series. Whether this head is an applicative or an incorporated postposition is OPEN (T92b).
- v/Voice. Introduces the external argument. The template's "classifier" (its voice and valence element) is the best candidate for the exponent of v/Voice, and it is listed with the root. HYPOTHESIS
- Asp (optional hypothesis). The aspect/mode head above vP. The stem-set alternation (the stem changes shape with aspect and mode) is a dependency between Asp and the stem that is listed with each stem. HYPOTHESIS
- T, then independently motivated C. HYPOTHESIS
Where the d heads sit. In the older light-d proposal, object d heads the nominal selected by V. The two subject-d proposals were: (a) It is the same kind of head as the object d, and its dP sits in Spec,Voice. (b) It is a light head in the verbal spine, like Voice: it introduces the subject and is listed with the stem, as light verbs are listed with their nominals. The older draft chose a split: object d taking DP and subject d as a spine head. That choice is not established. The corrected core does not identify d with either nominal D or verbal v. HYPOTHESIS, OPEN (T92a). The stem-listing of indefinite ʼa- (listed-d, Young & Morgan 1987 pp. 362, 364) fits option (b) for subjects SOURCED (listing) / HYPOTHESIS (fit).
2. Tree (schematic, no Navajo forms)
WORKING DEFAULT — generic transitive core, not a source tree [TP [DP a_i] [T′ T⁰ [vP [DP a_i (lower copy)] [v′ v⁰ [VP [V′ [DP t] V⁰]]]]]] Specifiers are left. T⁰ and v⁰ are initial; lexical V⁰ is final in this hypothesis. V⁰ → v⁰ → T⁰; subject DP: Spec,vP → Spec,TP, lower copy unpronounced (working default in tensed clauses). With T⁰ initial, the raised verb precedes the object; a verb-final clause then needs the remnant VP to move above T (HYPOTHESIS). With T⁰ final, nothing else moves. Which is right is OPEN. LF: ∃e[PRED(e) ∧ Agent(e,a) ∧ Theme(e,t)] No telicity, light d, Asp, applicative, C position or prefix exponent order is inferred.

Bracket notation (copy)
[TP [DP agent a] [T′ T⁰ [vP ⟨DP⟩ [v′ v⁰ [VP [V′ [DP theme t] [V⁰ predicate]]]]]]]
[TP [DP agent a] [T′ T⁰ [vP ⟨DP⟩ [v′ v⁰ [VP [V′ [DP theme t] [V⁰ predicate]]]]]]]
Structure as a list
- TP (id TP)
- DP (id S-high; landing site of Move from subject): agent a
- T′ (id Tbar)
- T⁰ (head; id T; probe for Agree φ (optional) with subject)
- vP (id vP)
- DP (id subject; goal of Agree φ (optional); struck copy; moved to S-high): agent a
- v′ (id vbar)
- v⁰ (head; id v; head-moves to T)
- VP (id lowVP)
- V′ (id Vbar)
- DP (id object): theme t
- V⁰ (head; id V; head-moves to v): predicate
- V′ (id Vbar)
The same head chain with a word box

Bracket notation (copy)
[TP [DP agent a] [T′ T⁰ [vP ⟨DP⟩ [v′ v⁰ [VP [V′ [DP theme t] [V⁰ predicate]]]]]]]
[TP [DP agent a] [T′ T⁰ [vP ⟨DP⟩ [v′ v⁰ [VP [V′ [DP theme t] [V⁰ predicate]]]]]]]
Structure as a list
- TP (id TP)
- DP (id S-high; landing site of Move from subject): agent a
- T′ (id Tbar)
- T⁰ (head; id T; probe for Agree φ (optional) with subject; hosts the “V+v+T” (V + v + T))
- vP (id vP)
- DP (id subject; goal of Agree φ (optional); struck copy; moved to S-high): agent a
- v′ (id vbar)
- v⁰ (head; id v; head-moves to T)
- VP (id lowVP)
- V′ (id Vbar)
- DP (id object): theme t
- V⁰ (head; id V; head-moves to v): predicate
- V′ (id Vbar)
3. Which step is what
| Step | Operation | What it does | Status |
|---|---|---|---|
| 1 | Merge | Merge V⁰ with the theme DP, project V′ and VP. | HYPOTHESIS |
| 2 | Merge | Merge v⁰ with VP; merge the agent DP in left Spec,vP. | HYPOTHESIS |
| 3 | Merge | Merge T⁰ with vP, creating T′ and the left Spec,TP landing position. | HYPOTHESIS |
| 4 | Agree | Optional T-to-subject Agree copies already interpretable subject person/number; it creates no branch. The identification of T with a particular prefix is OPEN. | HYPOTHESIS |
| 5 | Head-move | Only after both targets exist: V⁰→v⁰→T⁰. This is the working default in tensed clauses (2026-11-13), not an axiom: an example that resists it is flagged OPEN. | HYPOTHESIS |
| 6 | Move | Subject DP raises from left Spec,vP to left Spec,TP; strike the lower copy. | HYPOTHESIS |
| 7 | Merge | If a construction independently requires C, merge it above TP first. Only then can a separately motivated V→C or DP→Spec,CP operation apply. | HYPOTHESIS |
| 8 | Spell-out | Spell out the V+v+T complex at T. If T⁰ is initial, a verb-final clause first needs the remnant VP above T (HYPOTHESIS; OPEN against a final T⁰). PF-lowering/template accounts are kept only as labelled contrast. No rule here derives all Navajo prefix positions. | HYPOTHESIS |
| 9 | Interpret | LF: ∃e[PRED(e) ∧ Agent(e,a) ∧ Theme(e,t)]. No aspectual endpoint is entailed. | HYPOTHESIS |
The fourth person ji- is proposed to be a d of the same series (Light determiners). Where it attaches in the tree is not decided here. OPEN
4. The surface-order problem
In the verb word, the stem comes last and the higher heads come first, with the classifier next to the stem. HYPOTHESIS (stated from memory of the template, not yet checked against Kari 1976). If the spine C T Asp v V were head-initial and nothing moved, this would already give prefixes followed by the stem.
Problem 1 (Mirror Principle). The object prefix is farther from the stem than the subject prefix (Pos. IV before Pos. V; biʼdi-, Young & Morgan 212/3, on pos-labels) SOURCED. But the object merges lower than the subject. A head-initial spine with no movement would put the subject prefix farther left, which is the opposite order. So the object-before-subject order cannot come straight from the order of Merge. HYPOTHESIS
Problem 2 (head movement). A V-first complex follows if each raising head is always left-adjoined and no later reordering applies. Those extra assumptions would give the wrong prefix order. This does not refute V→v→T as such: adjunction direction and PF realization must be specified independently. Head raising is the working default; post-syntactic merger is kept only as labelled contrast for word-internal exponent order, which stays OPEN. HYPOTHESIS
Minimal template statement (a description, not an explanation). The core builds V < v < T. Optional P/Appl, Asp and light-d layers in the older model have not been established. A rule that applies after syntax (a position-class template) orders the spelled-out forms of these heads and of the d heads, and puts the heads listed with the stem right next to it. This keeps the syntax minimalist but gives up the Mirror Principle for the prefix chain. It states the order. It does not derive it. HYPOTHESIS
Getting the order to fall out of the derivation instead would take extra machinery, such as moving d_obj above Voice and building the word by moving heads out of specifiers (Baker 1988's problem). That is unverified and heavy, and this page does not adopt it. OPEN
5. Word and clause: an observed asymmetry
Many declaratives have arguments before the verb word, and many exponents precede its stem. Surface order alone does not establish that those exponents are selecting heads, or that every clause-level head is final. This page records that mismatch as an observation still to be explained, not as a setting of a head-direction parameter. It is the same A (word) / B (clause) split as on driver-3i. HYPOTHESIS
6. Position labels as data
Young & Morgan's "Pos. IV, obj. pron." and "Pos. V, subj. pron." are labels on prefix entries, not on verbs (pos-labels) SOURCED. In this framework they are the facts the derivation has to produce: one surface slot (IV) for a d that merges low, and one (V) for a d that merges high. The labels are not the analysis.
7. References
- Opened earlier on this site: Young & Morgan 1987 (pp. 212, 362, 364, via listed-d and pos-labels). Aissen 1997 and Willie 1991, as cited in Light determiners.
- Suggested by the Syntax bot, not opened: Kari 1976; Rice 2000; Hale 1973; Witherspoon 1977; Baker 1988; Speas 1990.
8. Open questions
T92a–d are on Open questions: (a) is the subject d a spine head or a dP in a specifier? (b) Appl or incorporated P? (c) head movement is the working default for V→v→T (2026-11-13); does word-internal prefix order still need post-syntactic merger? (d) does Navajo need phrasal movement of its argument dPs?
Update (2026-11-04, afternoon PT): Where the subject d sits tests options (a) and (b) against Hale 2001's spell-out procedure. Option (b) as stated here (a Voice-like spine head) gives the wrong order for (22b) shiʼdooʼniid. So does a plain dP in a specifier. Historical result under that particular procedure: a dP in Spec,vP with d displacement, or a spine head above Mode/Subj. Neither is a current result under all rivals; the dated audit above supersedes treating this procedure as the standard. HYPOTHESIS
