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

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

2026-11-11 audit pass 2. Apostrophe fixes only. Navajo glottal stops typed as ' or ’ (for example in prefix names like ʼa- and ʼeʼe-, or words like Awééʼ) now print as ʼ (U+02BC), and a build step no longer turns the closing quote of an English translation inside a Navajo row into a glottal stop (‘…the girls.’, not ‘…the girls.ʼ). No form, gloss or translation wording changed.

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

Verbs on this page → dictionary: jizloh ‘rope’

Restated on the unified structure (2026-11-15). Relative clauses are restated as a full CP, built like a main clause, under an overt D⁰ (=yę́ę, =ígíí) linked to one participant by Agree; nothing moves out (H27b, H28′, H31, H34). The head-direction mismatch argued below as rule sets A/B or a changeover is now stated as Grant’s Merge-site hypothesis and tested in §3.1–3.2. The “roles not words” theory, where it appears below, is dropped. Status on the spine: see unified structure, §2b and §4, relative rows. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.

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

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

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

1. Question

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

2. Sources

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

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

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

3. Rules first

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

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

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

4. The data

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

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

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

5. Answer to T71b

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

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

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

6. The head grid

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

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

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

7. What it does to /third-voice/

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

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

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

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

8. T71e: is the inverse agent internal?

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

9. Rulesets A and B (Focus 1)

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

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

10. Derivations

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

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

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

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

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

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

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

11. The role schema (Focus 2)

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

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

12. Fidelity

13. Open questions

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

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