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

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

Voice and perfective allomorphy: an in-place hypothesis, not a proof against movement (T93a)

Working note. Kept as written (URL, data and status labels unchanged). Filed under The word as a spell-out domain; the consolidated argument is the essay word-edge. Rule IDs: rules index.

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). Light d, used below, is withdrawn (Grant, 2026-11-14); light n is an OPEN structural question. The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. Position-class template statements below describe the order of prefixes; they are not the derivation (§3). Derivations below in which the verb or subject does not raise are labelled contrast; the working default raises V→v→T and the subject to Spec,TP. The “roles not words” theory, where it appears below, is dropped. Status on the spine: see unified structure, §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. Local allomorphy is data, not proof that the stem or T remained stationary. The “without Lowering” argument assumes a particular locality and vocabulary-insertion theory. Our primary V→v→T comparator and PF/template rivals must predict the same attested allomorphs before either wins. The M/S-to-T and classifier-to-v identifications, and the full DM rule derivations, remain HYPOTHESIS / OPEN.

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.

Hale 2001 (n. 7) worried that one fact might force his "no movement" Navajo verb to use a Lowering rule. The fact is that the perfective Mode/Subject prefix has one set of forms next to the zero and ł classifiers and another next to d and l. This page checks that worry against three DM locality sources and against Young & Morgan's paradigm tables. The result: Mode/Subject and Voice are a head and the head of its complement, and nothing is ever spelled between them. Under cyclic, inside-out insertion the right form can be chosen where Mode/Subject is merged. Lowering would relate the same two heads and adds nothing. One condition stays OPEN: whether contextual allomorphy needs a single complex head.

1. The question

T93a (from subject-d-site): can the perfective Mode/Subject (M/S) allomorphy that depends on Voice (Hale 2001, n. 7) be chosen in situ under bottom-up insertion, or does it need Lowering? Hale's spine has no head movement (SD-1). If the choice needs M/S and Voice in one complex head, M/S has to Lower onto Voice before Vocabulary Insertion (VI), and Hale says his "nonmovement" claim would then be "considerably weakened".

2. Sources opened this run

  • SOURCED Hale, Ken. 2001. Navajo verb stem position and the bipartite structure of the Navajo conjunct sector. Linguistic Inquiry 32: 678–693. Footnote 7, p. 692 (re-read from the text layer used on subject-d-site).
  • SOURCED Embick, David & Rolf Noyer. 2001. Movement operations after syntax. Linguistic Inquiry 32: 555–595. Lowering (6), pp. 561–562; Late Linearization Hypothesis (8); Figure 1, p. 566; Local Dislocation Hypothesis (16), p. 566; Bulgarian definite article allomorphy, p. 573, and n. 22.
  • SOURCED Bobaljik, Jonathan David. 2000. The ins and outs of contextual allomorphy. University of Maryland Working Papers in Linguistics 10: 35–71. Assumptions (3) (separation, cyclicity, rewriting) and the conclusion on inward vs outward sensitivity.
  • SOURCED Merchant, Jason. 2015. How much context is enough? Two cases of span-conditioned stem allomorphy. Linguistic Inquiry 46: 273–303. Node Adjacency Hypothesis, p. 273; his report of Bobaljik 2012 and Embick 2010/2012, pp. 275–276 and n. 4; spans (28)–(29), p. 288; p. 289; n. 18, p. 287; Span Adjacency Hypothesis (38), p. 294.
  • SOURCED Young, Robert W. & William Morgan Sr. 1987. The Navajo Language: A Grammar and Colloquial Dictionary, revised edition. Verb entries yishą́ (p. 779) and yishdlą́ (p. 781), and the prefix paradigm tables. All read through the CU Boulder Y&M interface (verbs.colorado.edu/~mahu0110/ym/, show_paradigm.php), not from print.

Not opened: Embick 2010, Bobaljik 2012, Rice 2000, Svenonius 2012. Where they matter, this page cites them only as Merchant or Hale report them.

3. Structure assumed

[QualP Qual [M/SP M/S [vP ( dP_subj ) [v' Voice [VP dP_obj V ]]]]]      (as on /subject-d-site/#structure)
 nuclear:    Qual  M/S  Voice  V          (spelled into the stem skeleton, SD-1)
 nonnuclear: dP_obj, adjuncts, dP_subj    (proclitic, SD-2; deictic subject d to the top, SD-5)
 M/S selects VoiceP (here written vP), so Voice is the head of M/S's complement.
status: hypothesis (Hale's tree plus VPISH, unchanged from subject-d-site)

4. Rules first

VA-1  TWO PERFECTIVE SETS. In the perfective, the M/S paradigms "display two       SOURCED (Hale 2001 n. 7,
      sets of allomorphs depending upon the identity of the Voice element":         p. 692)
      one with ł and zero, one with d and l. Choosing in base position needs
      "Mode/Subj properly situated relative to Voice". Otherwise "it might be
      necessary to apply a rule of Lowering, prior to Vocabulary Insertion".
VA-2  LOWERING. A head lowers to the head of its complement, before VI.             SOURCED (Embick & Noyer 2001
      It is stated over hierarchy, not over the string, and it skips              (6), pp. 561–562)
      specifiers and adjuncts (English T to v).
VA-3  INSIDE-OUT VI. Insertion "proceeds from the inside-out" (A1). "Contextual    SOURCED (Merchant 2015 p. 276,
      allomorphy requires concatenation (linear adjacency)" (A2).                   quoting Embick 2012; n. 4 Embick
                                                                                   2010 pp. 15, 42)
VA-4  INWARD SENSITIVITY IS TO FORM. Because insertion is cyclic and              SOURCED (Bobaljik 2000, (3)
      morphosyntactic features are used up when an inner node is spelled, an      and conclusion)
      outer morpheme can look inward only at the inner morpheme's
      morphophonological or diacritic properties, not at its syntactic features.
VA-5  SPAN ADJACENCY. "Allomorphy is conditioned only by an adjacent span."         SOURCED (Merchant 2015 (38),
      A span is a sequence of heads in complement relation in one extended         p. 294; (28), p. 288; p. 289;
      projection. Spans "do not require head movement": "head movement may feed    n. 18, p. 287)
      Vocabulary Insertion, but it need not". "Contiguity is a relation that is
      defined over strings of elements, not just inside complex heads."
VA-6  M/S VOCABULARY (schematic).                                                  HYPOTHESIS (ours, written over
        M/S[PFV, 1SG] ↔ ...sh (si-type: ...sís) / __ Voice{d, l}                   Y&M's two columns, §5)
        M/S[PFV, 1SG] ↔ ...(no sh)  elsewhere
      Parallel pairs for the 3rd persons and 2DU. No D-L entry is needed for 2SG
      (yi-/ni- types) or 1DU (F2).
      The context names Voice's exponent class {d, l}, not a feature [passive].
VA-7  NO INTERVENER. Nothing is ever spelled between M/S and Voice. Nonnuclear     HYPOTHESIS (theorem from SD-1,
      material is procliticized to the left of the whole conjunct sector (SD-2),  SD-2, SD-5 and step 5e of
      and a deictic subject d is pronounced at the top, before Qual (SD-5). So     /subject-d-site/)
      in the nuclear string M/S and Voice are always concatenated.
VA-8  IN-SITU CHOICE (theorem). From VA-3, VA-4, VA-6, VA-7: when M/S is reached,  HYPOTHESIS
      Voice is already spelled, adjacent, and visible by form, so VA-6 applies
      with M/S in its base position. Lowering (VA-2) would relate the same pair
      of heads and adds nothing the choice needs.

5. Data

Forms and translations are Young & Morgan's, copied from the CU Boulder interface. The Leipzig segmentations are ours (HYPOTHESIS). We do not mark a 3rd-person object prefix because these forms have none on the surface.

V1. Y&M 1987 p. 779, entry yishą́ 'to eat it' (classifier Ø).
Shimá sání bighandi naayízí tʼóó báhádzigo łikango yíyą́ą́ʼ ʼadą́ą́dą́ą́ʼ
'I ate a really delicious squash yesterday at my grandmother's.'
yíyą́ą́ʼ: yí-Ø-yą́ą́ʼ PFV.1SG.SBJ-VOICE(Ø)-eat.PFV

V2. Y&M 1987 p. 781, entry yishdlą́ 'to drink it, to quaff it, to eat it (liquid, as soup)' (classifier d; Y&M: "*d-dlą́ = dlą́").
ʼAhbínídą́ą́ʼ shimá sání bighandi tóshchiin yishdlą́ą́ʼ
'I ate (drank) some gruel yesterday at my grandmother's.'
yishdlą́ą́ʼ: yi-sh-(d)-dlą́ą́ʼ PFV-1SG.SBJ-VOICE(d)-drink.PFV

V3. Y&M 1987 p. 779, the same entry's perfective row: "(P) yíyą́ą́ʼ, yíníyą́ą́ʼ, yiyíiyą́ą́ʼ, jííyą́ą́ʼ, yiidą́ą́ʼ, wooyą́ą́ʼ (yidą́ą́ʼ/biʼdoodą́ą́ʼ)", with the note "Stem-initial y- = d- in the presence of D-Classifier, and in the 1st person dpl." The forms in parentheses are Y&M's passive forms. No sentences are given for them.

SOURCED The pair. V1 and V2 are the same frame from the same pages: transitive, yi-perfective, 1st singular subject, object noun outside the verb. The perfective prefix with the Ø classifier is yí- with no sh. With the d classifier it is yish-. V2's d is the verb's own listed classifier. 'Drink it' is active and transitive, not passive.

The paradigm tables

Y&M's prefix paradigm tables have two PERFECTIVE columns, headed "Ø - Ł" and "D - L". This run read every table the interface serves (table ids 1–410; the last non-empty one is 385). 140 distinct tables (by page and title) have both perfective columns and parsed cleanly. The others have no D - L column, or a header our script could not align. Counts below are over cells where both perfective columns are non-blank. The third column compares D - L with the IMPERFECTIVE cell, because Rice says (as Hale reports her, 2000: 169, 178) that M/S next to d/l is realized "as if imperfective".

Cell (Y&M row)Tables with both cellsD - L ≠ Ø - ŁD - L = IMPERFECTIVE
1sg13212760
2sg13218 (24 counting accent and apostrophe differences)0
311678 (87)46
3a134105 (106)46
3i2917 (18)9
1du13316 (17)85
2du13312324
3, distributive-plural block11087 (90)19

1sg endings: the D - L form ends in sh in 101 of 132 tables. Most of the rest are si-perfective tables where Ø - Ł -sé stands against D - L -sís (e.g. p. 154 béésé / béésís; p. 688 si-: se / sis). The Ø - Ł 1sg ends in sh in 1 of 132 (p. 332 di- ((2,5)): deesh / díí, perhaps with the columns swapped). 3rd person: Ø - Ł -z against D - L -s in 33 of 116.

Example, Y&M p. 50, ʼahééʼa- (cells copied as shown; IMPERFECTIVE / PERFECTIVE Ø - Ł / PERFECTIVE D - L): 1sg ʼahééʼnísh- / ʼahééʼní- / ʼahééʼnísh-; 2sg ʼahééʼí- / ʼahééʼííní- / ʼahééʼííní-; 3 ʼahééʼí- / ʼahééʼní- / ʼahééʼí-; 2du ʼahééʼnóh- / ʼahééʼnoo- / ʼahééʼnooh-. Y&M p. 63, ʼahi- ((1)): 1sg ʼahish- / ʼahíí- / ʼaheesh- (D - L differs from the imperfective). Y&M p. 48, ʼahééhi-: 3 ʼahééheez- (Ø - Ł) vs ʼahééhees- (D - L).

SOURCED The facts the derivation must produce (counts from the table above). (F1) With the D - L classifiers the perfective M/S has its own forms in the 1st singular (127/132), mostly ending in sh (101/132), and mostly also in the 3rd persons and the 2nd dual. (F2) Two cells mostly do not change: the 2nd singular (same in 114/132) and the 1st dual (same in 117/133). Most 2sg differences are in si-perfective tables (p. 155 bííní / bísíní), or in rows that look misaligned (p. 121 shows 2du-like ʼanoo / ʼanooh in the 2sg row). (F3) The D - L form equals the imperfective in only 60/132 1sg cells, 46/116 3rd cells and 0/132 2sg cells. So Rice's "as if imperfective" holds literally for under half. (F4) A verb with a lexical d classifier and active meaning (V2) takes the D - L form.

HYPOTHESIS Why the 1st dual doesn't change. Y&M p. 779 say the stem-initial consonant shows the d-classifier effect "in the presence of D-Classifier, and in the 1st person dpl" (V3: 1dpl yiidą́ą́ʼ with d although the verb is Ø-class). If the 1dpl already brings the d-effect on its own, a D - L Voice adds nothing, and the two columns coincide (F2). That would make the 1dpl d a second source of the same context (T94d).

6. Derivations

Same notation as subject-d-site: B = syntax/clause steps, A = word spell-out steps. Targets: V1 yíyą́ą́ʼ and V2 yishdlą́ą́ʼ.

DERIVATION I: IN SITU (no Lowering)
1  Merge   [VP DP_obj V]; V = √yą́ą́' 'eat' (V1) or √dlą́ą́' 'drink' (V2)          B
           DP_obj = naayízí 'squash' / tóshchiin 'gruel', a full DP (analysis D)
2  Merge   Voice: Ø (V1); d (V2). The stem is listed with its classifier          B
           (Y&M p. 781 "(*d-dlą́ = dlą́)"); this is selection, not a passive.
3  Merge   dP_subj (1SG pro) in Spec,vP (SD-6)                                    B
4  Merge   M/S [PFV(yi), uφ]                                                      B
5  Agree   M/S probes, finds dP_subj, gets φ = 1SG                                  B
6  Merge   Qual (none overt here); clause continues                               B
   ---- the word is handed to spell-out: B → A ----
7  VI      V first (A1): yą́ą́' / dlą́ą́'                                             A
8  VI      Voice: Ø / d  (d + dl → dl: Y&M's "*d-dlą́ = dlą́", not derived here)   A
9  VI      M/S[PFV, 1SG]. Voice is spelled (A1), concatenated with M/S (VA-7,     A
           A2), and visible by form (VA-4). VA-6 picks:
             V1: elsewhere → yí-      V2: / __ Voice{d} → yish-
10 Spell   nuclear string: yí–Ø–yą́ą́' ;  yi–sh–(d)–dlą́ą́'                           A
11 Spell   nonnuclear: no overt object prefix; DP_obj pronounced as a phrase      A/B
           before the word: naayízí ... yíyą́ą́' ;  tóshchiin yishdlą́ą́'   ✓ V1, V2

DERIVATION II: WITH LOWERING (Hale's fallback, VA-2)
1–6        as in I                                                               B
7' Lower   M/S → Voice (head of its complement), before VI: [Voice Voice M/S]     B→A
           (skips Spec,vP: VA-2)
8' VI      V, then Voice, then M/S inside the complex head; VA-6 picks as in 9    A
9' Linear  the complex must be linearized M/S before Voice (prefix), against      A
           the usual right-adjunction for lowering T to v (English -ed)
10'–11'    as in I                                                               A

HYPOTHESIS Comparison. I and II pick the same allomorph from the same two heads. Lowering goes to "the head of its complement" (VA-2), and that head is Voice. So Lowering can't reach anything that I can't already see. II costs two things. One is the Lowering step itself. The other is an extra ordering statement (9'), because Hale's spine puts M/S to the left of Voice, while English T lowers and is pronounced to the right. Step 9 of I needs no new order: SD-1 already gives it.

HYPOTHESIS Why the subject does not get in the way. In the tree, Spec,vP lies between M/S and Voice (the left-headed spine: M/S [vP dP_subj [Voice ...]]). If VI looked at the whole linearized clause, the subject phrase would separate the two heads, and A2 would fail without Lowering. That is the strongest case for Hale's worry. It does not arise, because Hale's spell-out builds the nuclear string from the nuclear heads only (SD-1). Phrases are added afterwards (SD-2). That separation is needed anyway to put the object and subject DPs outside the word.

7. Answer to T93a (in part)

  1. HYPOTHESIS In situ is enough. M/S and Voice are a head and the head of its complement. They are concatenated in the nuclear string (VA-7). Voice is spelled first (A1). So the Voice-dependent perfective can be chosen with M/S in its base position (VA-8). Hale's n. 7 condition, "Mode/Subj properly situated relative to Voice", is met by the head–complement relation itself.
  2. SOURCED The DM sources agree that head movement is not required for this. Embick's A2, as Merchant quotes it, asks for linear adjacency. Merchant says spans "do not require head movement" and that contiguity is "not just inside complex heads" (VA-5). Merchant reports Bobaljik 2012: 13 as suggesting that the strict no-maximal-projection-boundary condition "holds only of root allomorphy ... not of affixal allomorphy", and M/S is affixal.
  3. HYPOTHESIS The conditioning is by form. Hale groups ł with Ø and l with d. Y&M's columns are named by the classifiers themselves, and V2's lexical d triggers the D - L form with an active meaning (F4). All three point to the context being Voice's exponent {d, l}. That is what Bobaljik 2000 allows for inward sensitivity (VA-4). A context "[passive]" would be the wrong kind of feature for an outer morpheme to see, and it would misclassify V2.
  4. OPEN What is not settled. Embick & Noyer 2001's own allomorphy cases (Bulgarian Def, p. 573; English T, n. 22) occur after Lowering or adjunction. If Embick 2010 (not opened) requires the two pieces to be in one complex head (M-word), Lowering returns. Merchant's report of Embick (A2 as "linear adjacency") does not require it. T94a.
  5. HYPOTHESIS Rice's "as if imperfective". F3 refines it. The D - L perfective is a third set, not the imperfective set, in a little over half the tables (F3). The robust facts are F1 and F2: the D - L set differs from the Ø - Ł set in the 1sg, the 3rd persons and the 2du, and mostly not in the 2sg and the 1du.

8. Word and clause; prefixes and nouns

9. Limits

10. Open questions

IDQuestionWhat would settle it
T94aDoes contextual allomorphy need the conditioning head in the same complex head (M-word) as the target? If so, M/S has to Lower to Voice after all.Embick 2010, chapters 2–3 (the concatenation definition), opened first-hand.
T94bWhy is the 2nd singular perfective the same with both classifier sets (F2), except in some si-perfective tables (p. 155 bííní / bísíní)? Is 2SG underspecified, so that VA-6 has no D - L entry for it, or does one 2SG exponent win everywhere?Y&M print tables. An analysis of the 2sg perfective -íní- in Hale or Rice.
T94cAre the 3rd-person z (Ø - Ł) vs s (D - L) pairs (e.g. p. 48) the same allomorphy as the 1sg sh, or phonology (z + d → s)?A sourced statement of the Navajo d-effect (Y&M grammar section; Kari 1976).
T94dY&M p. 779 says stem-initial y becomes d "in the presence of D-Classifier, and in the 1st person dpl". Is the 1st-dual-plural d the same Voice-like d? The 1du column identity (F2, 117/133) suggests that it is. If so, a subject φ value conditions the same d-effect, and how that d is placed between M/S and V bears on VA-7.Hale or Rice on the 1dpl d-effect; Y&M grammar text.
T94eDoes the perfective type (yi-, si-, w-) depend on the stem V, across Voice? If so, M/S is conditioned by the span ⟨V, Voice⟩ (VA-5), not by Voice alone.Y&M stem entries listing perfective type, checked against classifier.

Answered in part here: T93a (in situ is enough under A1/A2 and spans; the complex-head version is OPEN as T94a). T93b–d are untouched.