Merge and Agree inside the Navajo verb word: which word-internal facts are Merge sites, which are Agree, what is a phase, and is anything [+MFS]-like? (T40)
Verbs on this page → dictionary: yishxash ‘bite’ · dabiiłtsą́ ‘see’ · yiyííbį́į́ʼ ‘pick (round objects, e.g. berries)’ · yiyííłtaʼ ‘count (plural objects)’ · yíghááh ‘join; with -tah ‘among’: go among’
Restated on the unified structure (2026-11-15). Prefix order is restated as Grant’s hypothesis that headedness is fixed locally by Merge site, so the sublexical domain is linearized opposite to the phrasal domain and the prefixes are successive word-internal Merge and Agree sites (GRANT’S HYPOTHESIS, tested on the sourced template in §3.2; fit/misfit OPEN). A possible sublexical V-to-T is a separate, untested GRANT’S HYPOTHESIS (§3.3). The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Position-class template statements below describe the order of prefixes; they are not the derivation (§3). 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, §3. 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 1. MA-, EF-, TP- and related mid-level rules import hypotheses about phasehood, probes and feature inheritance; none follows from VPISH plus theta roles. “No movement needed” in the old comparison is not evidence against Grant’s primary V→v→T derivation. Existing step lists are retained as historical hypotheses; target creation, locality and LF must be re-derived on a left-specifier vP/TP skeleton in pass 2.
2026-11-13 audit pass 7. An earlier dated note on this page treats V→v→T and subject raising as Grant’s conjecture or a comparator to test against PF/template rivals. Since 2026-11-13 they are the working default in tensed clauses: VPISH frames where the arguments start so that syntax and semantics line up; it is not a word-order account. PF/template accounts are kept only as labelled contrast, and the open question is what else has to move. The earlier note is kept as written. See the 2026-11-13 note and the pass 7 report.
Grant's note “Typology of Merge/Agree Phenomena” sets three levels apart: primitives (Merge, Agree, phases, locality), mid-level configurations (Rudin's [±MFS], Multiple Agree, feature inheritance, Edge Features on particular heads, phasehood choices, wh-clustering) and surface constructions. His working notes “Merge Sites and Agree Sites” (August 2026) propose that in a template language like Navajo “ordered morphological slots function as successive Merge sites and Agree sites inside the complex word” (§2, §7). This page takes that idea and sorts the word-internal findings already on this site by mid-level setting. It keeps the two operations apart throughout. Merge builds structure: an element is placed at a position, either fresh (External Merge) or moved (Internal Merge). Agree values features: a probe on a head finds a goal and copies its value, and nothing moves. Written 2026-10-19, 1:52pm PT.
Update 2026-10-19, 8:46pm (T40b). The originals behind this page's mid-level terms have now been opened: Rudin 1988, Hiraiwa 2001 and Chomsky 2008 in full, and Grewendorf 2001 as an abstract only. The results are on the borrowed terms checked against their originals. In short: the picture here (Agree inside, Merge at edges, [−MFS]-like) survives, with five corrections. (1) The leftover deictic (MA-2b) is OPEN between Agree and Merge, because Chomsky's inheritance allows both readings. (2) The ‘start to’ placeholder matches Chomsky's expletive EM (2008 p. 157) directly. (3) The object prefix is not a Hiraiwa Multiple Agree candidate, and SH-6 is favoured. (4) The PP being a phase is our extension, and an inactive-goal alternative is added. (5) Rudin's decisive test is the clitic position, and the disjunct order is not her kind of order. Rules MA-0′ to MA-6′ and derivations MA1′ to MA10′ are on that page. The rules and derivations below are kept as written at 1:52pm.
Update 2026-10-20, 4:46am (T41e). After Chomsky 2000/2001 (the older definitions behind “edge feature”), the ‘start to’ placeholder (MA-2c) is still a Merge site, but its head is PINCH, low in the word, not the top. Both texts allow expletive insertion only at a non-phase head. The bumped object (MA-2a) is Move to the same position. Grant's “whether a head has an EPP-feature” is confirmed as the 2000/2001 notion.
Update 2026-10-20, 12:46pm (T41f). The n. 12 datum cited in §6 (“an enclitic can split a preverb off”) is a split between two words: Hale's ka ‘sick’ stands as its own word (where the yes/no enclitic -ísh can go). So it no longer “points to divisibility” of the disjunct zone, and §6's lean is dropped (OPEN, no lean). The clitic test of T40a cannot be run with -ísh if the enclitic never enters the verb word (EN-1′, a prediction). T40a is closed into T43a/T43b.
Ground rules (Grant's). Every Navajo form below is already on this site, copied from a named source with its example number. The page links to where it was read. No new Navajo form is introduced and none is segmented beyond what its source prints. Each source and each earlier page of this site may be partly right. Only contradicting data count against an analysis. The mid-level terms ([±MFS], Multiple Agree, feature inheritance, Edge Feature, phase) are general-syntax notions. The originals (Rudin 1988, Hiraiwa 2001, Chomsky 2008, Grewendorf 2001) were not opened for this page. (They were opened at 8:46pm: see the definitions side by side.) The definitions used here are the ones in Grant's note, and no Navajo source opened on this site uses any of these five terms. Every mapping of a Navajo fact onto one of them is therefore ours and is at most HYPOTHESIS. Grant's first notes contained no astrophysics analogies (he sent them separately later; see §11). The note named for this task (“Merge_and_Agree_Sites_Exploratory_Notes”, text and PDF, 6 pp.) was searched in full, and so were the two raw chat exports in the same folder. None mentions astrophysics.
Notation (the site's own). Σ = role-schema step; A = word level (cycle D1 = QualP + VP-internal PPs = Willie's MVS, MVS-D1); B = clause level (D2). Spine (Hale 2003): Qual > M/S > v > V, with lexical V, P, N head-final and the functional heads head-initial (p. 32). Slots ρ = ⟨E, pos, φ, (δ)⟩ (SH-0). CYC = C1 non-heads before heads, C2 non-heads higher-first, C3 heads higher-first (spell-out rules §3). VPISH: the subject slot is first-merged in Spec,v. In Hale 2003 “There is no head-movement in the traditional sense” (p. 30) and arguments are “licensed in their thematic positions” (pp. 29–30).
Bottom line, in plain words
HYPOTHESIS Most of what happens inside the Navajo verb word is Agree, not movement. Three facts are the clearest. The ordinary subject prefix is pronounced on the Mode/Subject head. The subject's slot stays where it was built, and that head copies its person and number (Hale 2003 n. 11's portmanteau condition, read under VPISH: subject on the head). The postposition's own pronoun, including the inverse bi- in W&J (50b) bitah, is valued inside the postpositional phrase. The choice between yi-, bi- and zero is made inside the word (D1), before any noun arrives. None of these needs an element to move inside the word.
HYPOTHESIS The word-internal Merge sites are few, and they are all at the left edge of a domain. There are three candidates. (1) An object that is bumped to the far left when the object position is already taken (Hale, Munro & Platero's “bumping” into “disjunct prefix Position I”, n. 2). (2) The deictic part (fourth person ji-, indefinite ’-, areal ho-) of a subject's bundle, which attaches to the highest head of the word. That is Merge by adjunction, not a specifier landing. (3) The role-less placeholder ’ that ‘start to’ adds when the verb supplies nothing for its object requirement. That last one behaves like an expletive filling an Edge Feature or EPP requirement. Grant's “Merge sites inside the word” therefore survives, but in a narrow form: it holds for edges, and the template's inner slots are mostly Agree sites.
HYPOTHESIS Feature inheritance is a good model for the split subject. If the qualifier (the top of the word, standing where C stands in Chomsky's model as Grant's note reports it) hands person and number down to Mode/Subject (where T stands, Hale 2003 p. 10: M/S = Infl/T), then two things follow. Ordinary person and number are pronounced on Mode/Subject. The deictic feature stays on top, on the qualifier. That is exactly what Hale found twice (the deictic is “a prefix to E”, 1997; it “prefixes to the Qualifier element”, 2001). It derives the same strings as the site's rule SH-4 and adds nothing, so it is a restatement, not new evidence.
HYPOTHESIS Where Navajo looks like [−MFS], not [+MFS]. Rudin's [+MFS] means a position that is already filled can host more. The two cases in the sources where a word-internal position is already filled both show the opposite. Under ‘start to’ the object position is full, so the real object is pushed out to the far left. When a thematic prefix (so-) already holds the far left, the bumped element does not join it. It stays and fuses (HMP n. 4). Neither position takes two occupants. The disjunct prefixes do co-occur, but in a fixed order by height, not in Rudin's attraction order, and Hale reports that an enclitic can split a preverb off (2003 n. 12). So disjunct “clustering” fails at least one of Rudin's diagnostics as far as it can be checked. Whether the others hold is OPEN.
HYPOTHESIS Phases inside the word. Two domains behave like phases. The word itself (D1) is spelled out once and then counts as a single block (the site's spell-out-twice rule, CYC). The postpositional phrase is opaque to Mode/Subject's probe: a participant inside a PP never becomes the subject (IN-3, supported by every ‘start to’ form). One older analysis lets a theme escape a PP through its edge (Hale 1997: it raises from Spec,PP to Spec,MP). That is phase-edge escape in modern terms, though Hale does not call it that.
OPEN Multiple Agree has candidates, but none is decided. The third-person object prefix depends on the subject (zero under a fourth-person subject in jishxash, yi- under a noun-linked subject in ʼashkii yishxash). The voice prefix compares two arguments' rank. Each could be one probe agreeing with two goals at once, or ordinary agreement plus a context-sensitive pronunciation rule. The strings are the same. No opened source has a form with one prefix that spells both the subject's and the object's person, which is the clean test.
GRANT'S CONJECTURE “Navajo has intra-lexical Merge sites, and possibly Agree too” is Grant's. The split proposed here (Agree for the inner slots, Merge for the edges) is our refinement, graded HYPOTHESIS. Nothing in the sources contradicts it, but no source states it.
On this page: 1 Sources · 2 The mid-level terms, kept apart · 3 Classification table · 4 Formal statements · 5 Derivations MA1–MA9 · 6 Is anything [+MFS]-like? · 7 Phases inside the word · 8 What each option gets right and misses · 9 Grant's notes: what is kept, what is changed · 10 Tasks T40a–j · 11 Questions from the astrophysics notes
1. Sources
| Source | Access this run | Used for |
|---|---|---|
| Grant, “Typology of Merge/Agree Phenomena: A Takeaway Note on Mid-Level Configurations of Core Syntactic Operations” (PDF, 2 pp.; attachment of 2026-10-19) | Opened (full text) | The three levels (§4); the definitions of [±MFS], Multiple Agree, feature inheritance, EPP / Edge Feature, phasehood and wh-clustering (§§2–3). Its source list (Rudin 1988; Grewendorf 2001; Bošković 1997–2002; Hiraiwa 2001; Chomsky 2008; Krapova & Cinque; Billings & Rudin) was not opened |
| Grant's Grok notes, “Merge Sites and Agree Sites: Exploratory Notes Toward a Separate Research Project” (August 2026 working draft, 6 pp.) | Opened (text and PDF; same 3,268 words) | Core claims (§2), Navajo “word-level specialization” (§3.3), “Agree sites” (§4), the template as Merge/Agree sites (§7), open questions (§10). No astrophysics analogies are present in this file or in the two raw chat exports beside it |
| Grant, “Handoff Document: Hierarchical Sites, Residual Memory, and Quark Logic” (October 2026; attachment of 2026-10-19, 2:32pm) | Opened (full text) | Questions only (§11). Its astrophysical figures are not cited; §8 (quark logic) and the kanji remark tabled |
| This site's pages (Navajo forms and rules, each read earlier on page images or text, with sources listed there) | Re-read | Subject on the head (Hale 1997, 2001, 2003; Willie 1991; Kibrik 2005); ‘Start to’: HMP ms. and the ‘start to’ test; inverse as Agree/EPP; clause spine; postposition units; A/B as spell-out; feature bundles |
| Not opened | Rudin 1988; Hiraiwa 2001; Chomsky 2008; Grewendorf 2001; Bošković; Rice 2000 (known here only secondhand); Young & Morgan | Not used as evidence |
2. The mid-level terms, kept apart
Definitions as in Grant's note (§§2–3). The right-hand column says what would count as the configuration inside a Navajo word. That column is ours.
| Setting | Operation | Grant's note | What would show it inside the word (ours) |
|---|---|---|---|
| [±MFS] (Rudin) | Merge | “in [+MFS] languages an already-moved wh … can serve as a further landing site, so later whs Merge to it and form a cluster. In [–MFS] languages no such intermediate host is available” | A filled template position that takes a second occupant (or a cluster that moves and splits as a unit) |
| Edge Feature / EPP on a head | Merge (triggered) | “Whether a given head (T, v, C…) has an Edge Feature triggering Spec Merge” | A position that must be filled, by a placeholder if nothing else is available |
| Multiple Agree (Hiraiwa) | Agree | “Whether one Probe can Agree with multiple Goals in a single step” | One exponent valued by two slots |
| Feature inheritance (Chomsky 2008) | Agree (which head probes) | “Whether φ-features on C are inherited by T”; “Controls which head is the actual Agree site and what moves where” | One bundle's features pronounced on two heads, higher and lower |
| Phasehood | Locality for both | “Which heads count as phases (barriers to Agree and movement)” | A domain whose contents a higher probe cannot see, or that is spelled out as one block |
| Wh-clustering (Grewendorf) | Merge (feature-based [+MFS]) | “Whether a moved wh can itself host further whs” | As [+MFS] |
3. Classification table: each word-internal finding by mid-level setting
| # | Finding (already on the site) | Operation | Mid-level setting | Grade | Derivation |
|---|---|---|---|---|---|
| W1 | The prefix template is ordered non-heads first, then heads, each higher-first (CYC); this rests on Hale's head-initial functional spine (p. 32) | neither: linearization of a Merge-built structure | Not a mid-level setting. It is how a phase is pronounced (spell-out), as Grant's notes themselves say (§3.1: “linearization is local”) | HYPOTHESIS (CYC) | all |
| W2 | Ordinary subject φ is pronounced on M/S (Hale 2003 n. 11; SH-2 PORT) | Agree | Probe on M/S (the Agree site), goal ρA in Spec,v; no movement (VPISH) | SOURCED condition; Agree reading HYPOTHESIS | MA1 |
| W3 | Deictic subjects (4th ji-, indefinite ’-, areal ho-) are pronounced left of the qualifier (Hale 2001 (22a–c); Kibrik (8a)) | Agree for φ, then Merge (head-adjunction) of the leftover δ | Feature-inheritance model: Qual hands φ to M/S and keeps δ | HYPOTHESIS | MA2 |
| W4 | The third-person object exponent depends on the subject (Willie (23b) jishxash vs p. 129 ʼashkii yishxash; SH-6) | Agree, possibly with two goals | Multiple Agree candidate, or contextual allomorphy | OPEN | MA3 |
| W5 | The inverse on a postposition: W&J (50b) bitah yíghááh | Agree inside PP (IA-3); Internal Merge only under IA-1* | PP as an Agree domain; under IA-1*, escape needs a PP edge | HYPOTHESIS (IA-3 fits directly) | MA4 |
| W6 | ‘Start to’ bumps the object to Position I (HMP n. 2; (10a) shináʼniidlóóh; IN-5′) | Merge at the left edge | A filled position does not host: [−MFS]-like. Also, the PP is opaque to PORT: phase-like | SOURCED pattern; settings HYPOTHESIS | MA5 |
| W7 | A thematic Position I prefix blocks the bump (HMP n. 4 sodiʼniiszin) | blocked Merge; fusion | [−MFS]-like at Position I | SOURCED pattern; setting HYPOTHESIS | MA6 |
| W8 | ‘Start to’ needs an overt object; a verb with nobody low gets a role-less ’ (HMP (7b), (16a); IN-8, IN-9) | Merge of a placeholder | Edge Feature or EPP on INCH, satisfied by an expletive-like element | SOURCED requirement; EPP reading HYPOTHESIS | MA7 |
| W9 | da- counts the agent although it precedes bi- (Willie (64a) dabiiłtsą́) | Agree | Option e4 = da- shares the number M/S took (feature sharing); e2 = da- probes for a role | OPEN (trilemma on K5) | MA8 |
| W10 | A one-place verb whose participant starts low (areal (22c) hodildóóh; SH-8) | Internal Merge (U-1) or Agree with the closest goal (U-2) | The one place where Merge and Agree compete inside the word with the same string | OPEN (U-2 favoured on economy) | MA9 |
| W11 | The word is spelled out once and then ordered as one block (CYC, D1/D2; Hale n. 12: where the word stops is unsolved) | locality | Phase = D1 | HYPOTHESIS | §7 |
| W12 | The verb's arguments are feature bundles (FB-1 to FB-5) | (the goals) | Agree needs goals with features. Bundles are exactly that, with or without a noun | GRANT'S CONJECTURE formalized as HYPOTHESIS | §9 |
4. Formal statements (ours; all HYPOTHESIS)
Update 8:46pm: revised as MA-0′ to MA-6′ after reading the originals. MA-0's test would class subject agreement as Merge. MA-3 is not Chomsky's own version. MA-6's option is not Hiraiwa's Multiple Agree. MA-5 is our extension. See the revised rules and what changed.
MA-0 SEPARATE Merge(α, β) builds structure. Agree(P, G) values P's unvalued features from G.
A word-internal finding is classified as Merge only if an element is pronounced at
a position other than where its role puts it, or a position is filled by an element
with no role. Otherwise it is Agree plus CYC spell-out.
MA-1 AGREE-SITES The probes inside D1 are M/S (phi of the slot it locally c-commands and governs:
PORT, Hale 2003 n. 11), P (its own pronoun: AGR on P, Hale 1997), and the voice
position (the yi-/bi-/0 choice, SH-6). No probe in D1 triggers movement of a slot,
except under IA-1* (v's EPP in the inverse) and U-1.
MA-2 EDGE-SITES The Merge sites inside D1 are (a) the left edge of the V-level non-heads
(Position I: a bumped PP, IN-5'); (b) adjunction to the highest nuclear head (a
leftover delta, SH-4); (c) the object position of INCH, filled by a role-less rho_D
when no lower slot exists (IN-9). (a) and (c) are positions with a requirement.
(b) has no requirement; it is where leftover material goes.
MA-3 INHERIT (feature-inheritance restatement of SH-2 + SH-4.) The phase head of D1 (Qual)
bears the subject probe [uphi, udelta]. It passes [uphi] to its complement head
M/S and keeps [udelta]. M/S Agrees with rho_A (phi). Qual Agrees with rho_A's
delta if it has one. Exponents: phi in M/S; delta adjoined to Qual (or DETR if
present). Same strings as SH-2 + SH-4.
MA-4 NO-MFS A template position holds one exponent. A second candidate either goes to a
higher edge (IN-5') or fuses (' + ' -> ', HMP n. 4). No position is a host for a
cluster. (HYPOTHESIS; two positions tested.)
MA-5 PHASE-PP A PP inside D1 is a phase for M/S's probe (IN-3). Its edge (Spec,PP) is
accessible (Hale 1997's theme raising from Spec,PP), its complement is not.
MA-6 MA-CAND (OPEN.) Either the voice exponent of rho_X is valued by one probe that Agrees
with rho_A (delta, phi) and rho_X (phi) in one step (Multiple Agree), or rho_X is
valued alone and spelled by a context-sensitive rule that reads rho_A (SH-6).
5. Derivations MA1–MA9
MA1. Hale 2001 (24a) shidoołhosh ‘He/She will tickle me’: Agree only (as SD1)
- Σ Σ(√TICKLE) = {ρA, ρX}; φ(ρA) = 3, no δ; φ(ρX) = 1sg.
- A External Merge: V ghosh + ρX (Comp,V); v ł; ρA in Spec,v (VPISH); M/S; Qual d. Every slot is merged where its role puts it.
- A Agree (MA-1): M/S probes. ρA is the specifier of its complement and the closest goal, so M/S gets φ = 3 → w (“3rd:Perf”). No Internal Merge: ρA stays in Spec,v.
- A Spell-out D1 (CYC): sh ⌢ d ⌢ w ⌢ ł ⌢ ghosh = Hale's segment line ✓.
- Classification: one Agree site (M/S), no word-internal Merge site. The template order here is spell-out, not movement.
MA2. Hale 2001 (22b) shiʼdooʼniid ‘I was told (lit. someone told me)’: feature inheritance as a model of the split subject (as SD4)
- Σ ρA = ⟨φ = 3 (zero), δ = INDF⟩; ρX = 1sg.
- A External Merge: V niid + ρX; v d (Hale n. 6); ρA in Spec,v; M/S (w); Qual d with [uφ, uδ].
- A Inheritance (MA-3): Qual passes [uφ] to M/S. Agree M/S ⟶ ρA: φ = 3 (zero). Agree Qual ⟶ ρA: δ = INDF.
- A Merge by adjunction: the δ exponent ’ left-adjoins to Qual: [’ + d]. This is a word-internal Merge site (MA-2b), and the only one in this form.
- A Spell-out D1: sh ⌢ ’ ⌢ d ⌢ w ⌢ d ⌢ niid ✓ (Hale's segment line).
- Comparison. SH-2 + SH-4 give the same string with no inheritance step. MA-3 adds a reason why the leftover goes to the top head: that head held the probe first. It makes one further prediction: if a word has DETR above Qual (SD8, Kibrik (8a)), the phase head must be DETR, not Qual. Whether that is acceptable is T40c.
MA3. Willie (23b) jishxash ‘That person bit it’ vs p. 129 ʼashkii yishxash ‘The boy bit it’: a Multiple Agree candidate (as SD6)
- Σ Σ(√BITE) = {ρA, ρX}; φ(ρX) = 3. jishxash: ρA = ⟨3 (zero), δ = 4⟩. ʼashkii yishxash: ρA = ⟨3, no δ⟩.
- A External Merge, Agree M/S ⟶ ρA, as MA1. For jishxash, MA2's steps 3–4 (δ = 4 on Qual).
- A Option MA (Multiple Agree): one probe at the voice position Agrees in a single step with ρX (φ = 3) and with ρA (δ present or absent). Its exponent is y- if ρA is φ = 3 with no δ, and Ø if δ = 4.
- A Option CA (contextual allomorphy, SH-6): the probe Agrees with ρX only. At spell-out the exponent rule reads whether ρX is c-commanded in D1 by a δ-less third person.
- A Spell-out D1, either option: j ⌢ … ⌢ shxash; yi ⌢ … ⌢ shxash ✓ (not segmented further).
- B ʼashkii links to ρA in D2. It cannot change the D1 choice (“B is inert”, WF-5), so the deictic δ must already be in D1.
- What would separate MA from CA: a case where ρX is not third person but its exponent still varies with ρA's δ. (MA predicts the probe can still read ρA. CA ties the rule to the third-person exponent only.) No such form is in an opened source (T40d).
MA4. W&J (50b) dibé łį́į́ʼ bitah yíghááh: Agree inside the PP; Merge only under IA-1* (as L5, PU4)
- Σ ρA (the one who goes), ρX = P's object; both π = 3. Inverse: ρX [top], ρA [foc].
- A External Merge: P -tah + ρX (head-final PP). Agree inside PP: P's own agreement (Hale 1997: AGR on P) with ρX. Under IA-3 ρX already carries [top], so the exponent is b- → bitah.
- A External Merge: PP inside VP; v; ρA in Spec,v; M/S; Agree M/S ⟶ ρA (MA-5: ρX is inside the PP, not a goal); Qual.
- A IA-1* branch: v's EPP feature in the inverse should attract the topic slot (Internal Merge). ρX is in the PP's complement, which MA-5 makes inaccessible. So either (i) ρX first moves to Spec,PP, the PP edge (Hale 1997's position for a theme in a PP), or (ii) IA-1* fails for postpositional objects. Under (i), bi- is pronounced on P only if the lower copy is the one pronounced (T31e).
- A Spell-out D1: bitah ⌢ yíghááh.
- B łį́į́ʼ → [foc] = ρA; dibé → [top] = ρX. D2: dibé ⌢ łį́į́ʼ ⌢ [bitah yíghááh] ✓.
- Classification: Agree site inside the PP. The PP behaves as an Agree domain closed to M/S. A word-internal Merge step is needed only by IA-1*, and then it must go through the PP edge. That is one more reason the site's in-situ option IA-3 is cheaper here.
MA5. HMP (10a) shináʼniidlóóh ‘I start to get cold (about dawn)’: a left-edge Merge site, a filled position that does not host, and a PP opaque to Agree (as HD2)
- Σ Σ(√GET.COLD) = {ρ1}, first-merged in Spec,V; φ = 1sg.
- A External Merge: V dlóóh, ρ1; REP ná; INCH puts the thematic ’ (ρI) in the object position (HMP: “the Position IV indefinite object prefix … functioning here as a thematic element”).
- A MA-4 NO-MFS: the object position is full. ρ1 cannot join it. IN-5′: [PP ρ1 PINCH] is merged as the highest V-level non-head (MA-2a), above ná. In HMP's words the object is “bumped” “into the disjunct prefix Position I” (n. 2).
- A Agree M/S: Spec,v is empty. ρ1 is inside a PP (MA-5, IN-3), and ρI is thematic, so there is no goal. M/S takes the default (IN-7 (D); HMP: the subject position is “entirely empty … as it is regularly in persons other than first or second”).
- A Spell-out D1: shi ⌢ ná ⌢ ʼnii ⌢ (M Ø) ⌢ d ⌢ dlóóh ✓ = shináʼniidlóóh. HMP do not segment (10a); the alignment is the site's.
- Classification: one Merge site (left edge), a [−MFS]-like refusal (object position), a phase-like PP (opaque to M/S), and Agree failing to find a goal. It is the richest single form for this page.
MA6. HMP n. 4 sodiʼniiszin ‘I started to pray’: the left edge is full too (as HD6)
- Σ ‘pray’ = theme so=d-l-zin; ρA = 1sg in Spec,v; nothing below v, so INCH's requirement is met by ρD (IN-9 (c)).
- A MA-4: the thematic Position I prefix so holds the left edge. The candidate does not join it (no cluster). It stays in the object position and fuses with ρI: ’ + ’ → ’ (HMP: “hypothetically at least”).
- A Agree M/S ⟶ ρA (1sg). Spell-out: so ⌢ di ⌢ ’ ⌢ nii ⌢ … ⌢ zin ✓ (HMP give no further segmentation).
- Classification: the escape site is itself one-occupant. That is a second [−MFS]-like position. Fusion is not hosting: one exponent results.
MA7. HMP (16a) ʼiʼniichííl ‘it started to snow’ and (7b) ʼawééʼ ʼináʼniichah ‘the baby starts to cry’: an EPP-like requirement met by a placeholder (as HD3, HD7)
- Σ ‘snow’: no participant. ‘cry’: ρA in Spec,v only; nothing below v.
- A IN-8: INCH “must always be preceded by overt object agreement morphology” (HMP). Read as an Edge Feature or EPP on INCH: a position that must be filled.
- A IN-9: no slot below v, so a fresh role-less ρD = ⟨E = none, φ = 3, δ = INDF⟩ is Merged, spelled ’. It is expletive-like: it fills a position and has no role (HMP: “use of indefinite agreement is a reasonable alternative”).
- A Spell-out: ‘cry’ ʼi ⌢ ná ⌢ ʼnii ⌢ … chah ✓; ‘snow’ ʼi ⌢ ʼnii ⌢ … chííl ✓ (not segmented further than HMP).
- B ʼawééʼ links to ρA, not to ρD (ρD has no role to match).
- Classification: the clearest word-internal EPP effect in the sources. The contrast with (10d) biʼniichxʼǫʼ ‘(this meat) has started to spoil’, where the participant itself fills the position, is the one English shows between there-insertion and raising. That parallel is ours.
MA8. Willie (64a) dabiiłtsą́ ‘They/he (was)were seen by them/*him’: da- as shared agreement (as K5)
- Σ ρA plural (W&J gloss PL-3ACC-3NOM-saw); ρX 3; inverse.
- A External Merge; Agree M/S ⟶ ρA (φ including plural number).
- A e4 as feature sharing: da- has an unvalued number feature and Agrees with M/S, not with a slot. It receives the agent's number second-hand, so its position above b- does not matter. This is a chain of two Agree steps, not Multiple Agree. e2: da- probes for ρA's role directly.
- A Spell-out D1: da ⌢ b ⌢ heads → dabiiłtsą́ ✓.
- Classification: Agree. Under e4 this is the inheritance pattern again (a value passed from head to head), applied to number. The site's own derivation does not decide between e2, e3 and e4 (T30 tasks).
MA9. Hale 2001 (22c) hodildóóh ‘It (e.g., cloud) starts to drift’: where Merge and Agree compete (as SD5)
- Σ ρ1 = ⟨φ = 3 (zero), δ = AREAL⟩. Whether ρ1 starts in Spec,v or Spec,V is unknown (SH-8).
- A U-1 (Internal Merge): if ρ1 starts in Spec,V, it raises to Spec,v inside D1, where M/S locally governs it, and then Agree.
- A U-2 (Agree only): M/S Agrees with the closest slot in its domain, wherever it is (IN-4 PORT-CLOSEST).
- A Either way, δ = AREAL is left over and adjoins to Qual (MA-2b): [hw + d]. Spell-out: hw ⌢ d ⌢ Ø ⌢ l ⌢ dóóh ✓.
- Classification: same string, different operation. The ‘start to’ facts favour U-2 on economy: there the low participant's subjecthood is bled with no movement. Only a movement diagnostic would decide (T37b, T40e).
6. Is anything [+MFS]-like?
Update 8:46pm: checked against Rudin 1988. [+MFS] needs adjunction to a filled SpecCP (pp. 480–481). Her decisive [−MFS] test is a clitic right after the first element (pp. 461–466), which Hale 2003 n. 12 reports for ka. Her fixed order is subject before non-subject (p. 475), so the disjunct order below counts for neither type. See Rudin's diagnostics applied.
| Candidate | What the sources show | Rudin's diagnostics (as in Grant's note) | Verdict |
|---|---|---|---|
| Object position (IV) already filled by INCH's ’ | The object does not join it. It is bumped to Position I (HMP n. 2; (2b) bi-ná-ʼnii-sh-…, (10a)) | No second occupant | HYPOTHESIS [−MFS]-like |
| Position I already filled by thematic so- | The bumped element stays and fuses (HMP n. 4; sodiʼniiszin) | No second occupant; fusion yields one exponent | HYPOTHESIS [−MFS]-like |
| Hale 2003 n. 13: an oblique takes “the highest prefixal position when not in competition with a preverb” | A preverb at the edge blocks the oblique's jump (cited on spell-out rules (f2)) | Competition, not clustering | HYPOTHESIS [−MFS]-like (third case, same pattern) |
| Several disjunct non-heads in one word (bumped PP + ná-; preverb + iterative + plural, Hale's trees (11), (31b)) | They co-occur at one level, ordered by height (C2), each adjoined to the projection of the highest nuclear head (Hale pp. 16–17) | Order: fixed by height and class, not by attraction order (no Superiority analog in any opened form). Indivisibility: Hale 2003 n. 12 reports that the preverb ka “can be separated from the rest of the verb by the polar interrogative enclitic” (forms printed in a broken font, not reproduced). Clitic placement after the whole group: unknown. Multiple extraction: not applicable | OPEN, leaning “not a cluster”. Multiple adjunction to one projection has the shape of multiple specifiers, but the one diagnostic that can be checked (n. 12) points to divisibility |
HYPOTHESIS Summary. Wherever the sources show a word-internal position that is already filled, the second candidate goes elsewhere or fuses. That is the [−MFS] behaviour Grant's note describes (“no such intermediate host is available”), transposed to template positions. The disjunct zone allows several elements, but as separate adjuncts in a fixed order, not as a cluster built on a first-moved element. The test that would show a genuine cluster: does the polar enclitic go after the first disjunct or after the whole disjunct group? It needs forms with two or more disjuncts and the enclitic (T40a).
7. Phases inside the word
| Domain | Behaves like a phase because | Against, or unknown | Grade |
|---|---|---|---|
| D1 = QualP + VP-internal PPs (the word / MVS) | Spelled out once, then ordered as one non-head in D2 (CYC); nouns link after it and cannot change its choices (WF-5; MA3 step 6); W&J: the MVS is “a noninterruptible constituent … the domain of the voice alternation” (p. 270) | The boundary is stipulated; Hale: “when to stop building onto the skeleton” is unsolved (n. 12); preverbs can be separate words | HYPOTHESIS |
| PP inside D1 | Opaque to M/S's probe (IN-3; every ‘start to’ form); has its own Agree (P's pronoun) | Whether lexical PPs are in D1 at all is T31f; the edge (Spec,PP) is accessible only in Hale 1997's analysis | HYPOTHESIS |
| vP | — | Nothing on the site needs vP to be a phase inside the word. M/S Agrees into vP (with Spec,v and, under U-2, below it) | OPEN: no evidence either way |
HYPOTHESIS Grant's notes ask (§10, Q4): “Are phases themselves the natural domains that delimit the possible heights of Merge and Agree sites?” On the Navajo word, a partial yes. Every Merge site found (MA-2a, b, c) is at the edge of D1's V-level material or on D1's highest head. Every Agree site (M/S, P, voice) is inside D1 and probes down. That is the distribution a phase-edge account predicts. It is a fit, not a test.
8. What each option gets right and misses
| Option | Gets right | Misses or costs |
|---|---|---|
| Grant's notes as stated (“ordered morphological slots function as successive Merge sites and Agree sites”, §7) | The template is the place where the decisions are made; the inverse “is fixed at a specific height inside the conjunct domain” (§4) fits IA-3 and SH-6 | “Successive” Merge sites slot by slot would mean each prefix is a landing site. The sources show most prefixes are pronounced where their slot was built (no head movement, Hale p. 30). The notes also say “the classificatory stem … carrying … Proto-Patient properties … must be valued” by the classifier and prefixes (§4). That is not in any opened Navajo source (T40h) |
| This page (Agree inside, Merge at the edges) | Separates the two operations; places every word-internal finding; keeps all strings | Adds no new prediction except MA-3's DETR case (T40c) and the enclitic test (T40a); the Multiple Agree question is left open |
| A movement-heavy alternative (every prefix raises to its template position, as antisymmetry requires: spell-out rules §5) | Gives every prefix a Merge site, as Grant's “successive Merge sites” suggests | The object must raise above the subject, which Hale considered and rejected (2003:11–12), and da- then counts the wrong slot (K5) |
9. Grant's notes: what is kept, what is changed
- GRANT'S CONJECTURE → HYPOTHESIS “Specialized Merge sites and Agree sites are local decisions fixed at particular heights or morphological domains” (§11). Kept: §3 places each word-internal finding at a height. Changed: inside the Navajo word the sites are mostly Agree. The Merge sites are edges.
- GRANT'S CONJECTURE “The yi-/bi- alternation is the morphological reflex of an Agree relation that reads the relative ranking of two arguments” (§4). Kept as a Multiple Agree candidate (MA3, W4). It is equally compatible with the in-situ topic analysis (IA-3), where one slot carries [top].
- GRANT'S CONJECTURE Feature bundles (FB-1 to FB-5). Agree needs goals with φ and δ features, present whether or not a noun is linked. Every derivation above values features on bundles in D1 before any noun arrives. So the bundle picture is what makes the Agree-heavy classification work.
- HYPOTHESIS Grant's notes distinguish the classificatory stem series from the position-IX classifier (§7). Nothing on this page depends on that distinction; it is recorded, not tested.
- Grant mixes Merge and Agree in places (e.g. the template slots as “Merge sites and Agree sites” without saying which is which). Rule MA-0 is the separation used here.
11. Questions from the astrophysics notes
What this section is. Grant later sent his astrophysics handoff, “Hierarchical Sites, Residual Memory, and Quark Logic” (2026-10-19 attachment, opened). It is used here only as a source of questions. Its astrophysical figures (for example the share of second-generation black holes and the age of a galactic merger) are not sourced by this site and are not cited as findings. Its quark-logic section (§8) and its kanji remark stay tabled. Three of its claims are turned into questions about the Navajo verb word: early/low sites “tend to be restrictive: they fix core properties” while late/high sites “tend to be selective: they organize or adjudicate already-fixed material without fully rewriting it” (§4); “combination events leave residual properties that later events can read” (§2); and “Phases function analogously to causal horizons”, since “The Phase Impenetrability Condition freezes the complement while leaving the edge accessible” (§7). The handout itself names Navajo: “classificatory stem lexicalizes Proto-Patient properties early; conjunct slots handle inverse, agreement, and aspect” (§3). That is Grant's claim, tested below. A chemistry counterpart of each question is on the Chemical·Theta Roles scout page.
11.1 Early-restrictive vs late-selective sites
| Layer (inside out) | What the sources show it doing | Restrict, select, or add? | Grade |
|---|---|---|---|
| Stem / theme (the classificatory stem; the role schema Σ) | Fixes properties of the patient slot that limit which noun can link: Willie's (22) yiyííbį́į́ʼ needs “round objects” and (24) yiyííłtaʼ “plural objects” (pp. 61–62; thesis D9). The roles in the action are fixed here. Under ‘start to’, “the role in the action is unchanged by INCH” (HD2) | Restricts | SOURCED for linking restrictions (Willie); “Proto-Patient properties lexicalized early” as a general claim is GRANT'S CONJECTURE, supported here only by these two glosses |
| Event layer (v, the ł-causer, the inceptive qualifier, INCH, DETR) | Adds participants or slots: the ł-causer adds a causer (causative split); ‘start to’ adds a thematic ’ and, when nothing low exists, a role-less placeholder (MA7); it also turns an unaccusative's sole participant into an object (HMP (11); MA5) | Adds and re-maps (not merely selects) | HYPOTHESIS; the facts are SOURCED (HMP; Hale & Platero 1996 via Gessner) |
| Agreement and voice (M/S, P, the yi-/bi-/Ø choice, da-) | Picks which already-built slot is pronounced on the head (PORT) and which is marked as the topic (IA-3); da- counts the agent in both voices (Willie pp. 82–85), so voice does not change roles | Selects | HYPOTHESIS (SOURCED facts) |
| Clause (D2: nouns, topic/focus) | Nouns link to existing slots by feature match and position; they cannot change the word's choices (WF-5; MA3 step 6) | Selects | HYPOTHESIS (SOURCED linking rules: Willie p. 53; H&P 1974 (28)) |
HYPOTHESIS Result. The handout's asymmetry holds at the two ends and fails in the middle. The innermost layer restricts (Willie's sortal glosses) and the outer layers select (voice, agreement, noun linking). But the event layer does not just select among fixed material: it adds a causer, adds a placeholder slot, and re-maps an unaccusative participant to object. So the Navajo word is “restrict, then add, then select”, not “restrict, then select”.
HYPOTHESIS Against Grant's STRATA conjecture (theta roles, then event roles, then lexical selection): the order matches the three outcomes exactly. The theta-role layer restricts. The event-role layer is the one that adds, which is what an event-structure layer is expected to do (it introduces causers and phases of the event). Lexical selection (nouns, D2) is the last and purely selective layer. On this reading STRATA is the better fit and the handout's two-way split is a simplification of it. Nothing opened contradicts either; the one thing that would is a later (D1-top or D2) operation that changes a role in the action. None is known on the site.
11.2 Residual memory read by later operations
| Residual | Where it is left | What later reads it | Grade |
|---|---|---|---|
| The deictic δ of a fourth-person / indefinite / areal subject | At the top edge of the word, adjoined to Qual or DETR (SH-4; MA2) | Inside D1, the object exponent reads it (Ø, not yi-, in jishxash: MA3). Across clauses, the fourth person is what introduces a new agent in a =go chain, and only from the first clause (Willie Ch. 5 §5.1, pp. 132–140, on 4th across clauses). So a later clause reads it | SOURCED facts; “left at the edge, read later” is HYPOTHESIS |
| Voice (yi- vs bi-) | In the word's exponents (the finished D1) | Noun linking at D2: “the NP immediately preceding” a yi- verb is its object and before a bi- verb its subject (Hale & Platero 1974 (28), p. 17); W&J: the verb complex “determines the interpretation of the adjoined nominals” (p. 270) | SOURCED |
| The inverse on a postposition (bitah) | On the PP, inside D1 | D2 links dibé to the bi- object ((50b); W&J: coindexed but “not a constituent of the PP”) | SOURCED (W&J (50b)) |
| Whether ρX moved inside D1 (IA-1* vs IA-3), or the low participant raised (U-1 vs U-2) | Interior of D1 | Nothing later reads it: both options give the same strings, and no D2 rule is sensitive to the difference | HYPOTHESIS |
HYPOTHESIS Fit with the PIC. What later operations read is always something pronounced on the word's exponents or at its edge (the deictic on the top head, the voice prefix, P's pronoun). What they never read is how the interior was derived. That is the PIC pattern (“freezes the complement while leaving the edge accessible”, in the handout's words) applied to D1. One caveat: the voice prefix is not at the top edge of the word; it is an object exponent inside it. So the generalization that fits is “later operations read D1's output, not D1's derivation”, which is weaker than an edge-only PIC.
11.3 Phases as horizons
HYPOTHESIS D1 acts as a horizon in the handout's sense for derivational history (11.2, last row), not for content: its exponents stay visible to D2. Hale's caution applies: where the word stops is unsolved (2003 n. 12), so the position of the horizon is stipulated. The PP inside D1 is a horizon for M/S's probe (IN-3) but not for noun linking at D2, which reaches P's object (50b). OPEN whether any Navajo operation is blocked by D1 in the way the PIC blocks Agree into a complement: T40g.
Changes made elsewhere
Links added from clause spine §11 and open questions (row T40a–j), and from the home page. Sources for this page: sources, added 2026-10-19 (1:52pm).
10. Tasks T40a–j
| Task | What it needs |
|---|---|
| T40a | Rudin's indivisibility / clitic diagnostic: forms with two or more disjunct prefixes and the polar interrogative enclitic (Hale 2003 n. 12, Platero p.c.; legible page or Young & Morgan) |
| T40b | Answered 2026-10-19, 8:46pm on the originals page. Open Rudin 1988, Hiraiwa 2001, Chomsky 2008 and Grewendorf 2001 and check the definitions used here against them |
| T40c | Conditionally answered 8:46pm (selection condition; → T41a). MA-3 with DETR: is the phase head DETR when di- is present (Kibrik (8a) ʼá-zh-dí-l-zhééh)? Does anything else treat DETR as the top of the word? |
| T40d | Narrowed 8:46pm: Hiraiwa's Multiple Agree is excluded (MA3′). Multiple Agree vs contextual allomorphy: a form where the object is not third person and its exponent still varies with the subject's δ; or a single prefix spelling both subject and object person |
| T40e | U-1 vs U-2 (= T37b): a movement diagnostic for a low participant inside the word |
| T40f | PP edge: is there any Navajo form where a postpositional object behaves as if at the PP edge (Hale 1997's Spec,PP theme) under the inverse? |
| T40g | Partly answered 8:46pm: in Chomsky 2008, v*P is a phase and unaccusative vP is not (MA9′). Is vP a phase inside the word? Needs a probe on M/S that is blocked by vP, or one that is not |
| T40h | Grant's notes §4: valuation of the classificatory stem's “Proto-Patient” features by prefixes. Is there a Navajo source for agreement between a classificatory stem and a prefix? |
| T40i | A third test of MA-4: any template position that holds two exponents at once (a counterexample would make that position [+MFS]-like) |
| T40j | Person Case Constraint, listed in Grant's note as a related family member: do Navajo combinations of a local-person object with a deictic subject show gaps? Not checked |
