The inverse: when a lower argument outranks the agent
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.
Verbs on this page → dictionary: yiztał ‘kick’ · yiztsʼǫs ‘kiss’ · yishxash ‘bite’ · yiyiisxį́ ‘kill (one)’ · dabiiłtsą́ ‘see’ · yoolóós ‘lead (one animate)’ · yáłtiʼ ‘speak, talk (to)’
Restated on the unified structure (2026-11-15). The inverse is restated as one extra Move on the clause spine: the patient moves (via an outer Spec,vP) above the agent, which has already raised to Spec,TP by the working default (H3, H6′, H21b). The event LF is the same as the direct; the trigger and landing site are OPEN. Light d, used below, is withdrawn (Grant, 2026-11-14); light n is an OPEN structural question.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. 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, §4, inverse row. The data, judgments, translations and citations below are unchanged; the analysis below is kept as the record of what was argued then.
With two third persons, a Navajo transitive verb comes in two shapes: a yi- form (‘the horse kicked the man’) and a bi- form (‘the man got kicked by the horse’). Who did what to whom is the same in both. What changes is which participant is realized as the subject. This page states the choice as rules over proto-role entailments plus an animacy ranking, asks where in the derivation the choice is made, and derives one sourced pair step by step. This is open question T5. The other response to the same ranking clash — take the human off the scale and keep the direct form — is the fourth-person page (T8).
On this page: 1 Question · 2 Data · 3 Rules · 4 Which orders are impossible · 5 Where the choice is made · 6 Derivations I1–I5 · 7 Verdict · 8 Open
1. The question
GRANT'S CONJECTURE The project is argument ranking. The natural alignment is that the higher proto-agent argument is the subject. The edge case is when a lower argument outranks it. Then the passive-like expression (the Navajo bi- form) is the natural one, and some permutations are impossible. In Navajo the units being ranked and moved are theta roles, not lexical items. Dowty's proto-role entailments are fixed by the event. The inverse does not change them; it changes only which argument is realized where.
Open question T5, as posed on the open-questions page: does the inverse change the agent's position (Willie's "internal (Ergative) agent"), or only how the nominals are linked (Willie & Jelinek's 3ACC-3NOM gloss, which is the same in both voices)? The earlier pages state the yi-/bi- rule over the topic DP (schema §4, D4). This page reads Willie 1991 §3.3 in full and Sandoval & Jelinek 1989 against that rule.
2. Data SOURCED
Sources opened this run (2026-10-08). (a) Willie, MaryAnn 1991, Navajo Pronouns and Obviation, PhD dissertation, University of Arizona (UMI 9210298 scan). Page images of printed pp. 16, 27–28, 71–77, 86–88, 90–92, 94–95, 97 and 100 were read at 110 dpi. The Navajo lines of (37), (73)–(76) and (79)–(80) were re-read at 300 dpi. The typescript prints nasal hooks as cedilla-like marks; they are transcribed as hooks here. (b) Sandoval, Merton & Eloise Jelinek 1989, "The bi-construction and pronominal arguments in Apachean," in Rice & Cook (eds.), Athapaskan Linguistics, Mouton, pp. 335–377. This was read as page images (the scan has no text layer) from Carnie & Harley's 2012 compilation of Jelinek's papers. Printed pp. 335–361 and 368–371 were read at 90 dpi, and the Navajo examples on pp. 335–336 at 300 dpi. Most of their examples are Jicarilla Apache; only those they mark as Navajo are printed below. (c) Willie & Jelinek 2000, printed p. 268 at 300 dpi and p. 273 at 90 dpi (same compilation).
Glosses are copied exactly as printed. Willie glosses the verb as a whole (3.3.kicked, INV3.3.kicked). Sandoval & Jelinek separate only yi-/bi- (yi-kicked). No further segmentation is offered here.
2.1 The basic pair, and the edge case
| a. | łį́į́ʼ | dzaanééz | yiztał |
| horse | mule | 3.3.kicked | |
| ‘The horse kicked the mule.’ | |||
| b. | dzaanééz | łį́į́ʼ | biztał |
| mule | horse | INV3.3.kicked | |
| ‘The mule was kicked by the horse.’ | |||
Willie, pp. 71–72: "In (37a) the NP occurring immediately before the yi-form is a patient adjunct, and by default, the other NP is an adjunct to the agent argument. In (37b) the NP occurring nearest the bi-form is the agent adjunt [sic] of that sentence … Therefore, the immediately preceding NP, in both voice forms, is an adjunct to the "internal" argument, regardless of whether the theta role of that argument is agent or patient." Horse and mule are "of equal animacy", so here "the voice alternation is optional, that is, discourse determined" (p. 72).
| (1) | łį́į́ʼ | dzaaneez | yiztał | ‘The horse kicked the mule.’ |
| horse | mule | yi-kicked | ||
| (2) | łį́į́ʼ | dzaaneez | biztał | ‘The horse got kicked by the mule.’ |
| horse | mule | bi-kicked | ||
| (5) ?? | łį́į́ʼ | hastiin | yiztał | ?? ‘The horse kicked the man.’ |
| horse | man | yi-kicked | ||
| (6) | hastiin | łį́į́ʼ | biztał | ‘The man got kicked by the horse.’ |
| man | horse | bi-kicked | ||
| (7) | łį́į́ʼ | shiztał | ‘The horse kicked me.’ | |
| horse | me-kicked |
On (5), p. 336: it "is culturally "odd" and not employed, since it carries the implication that the horse has more volition and intelligence than the man. If the horse is understood as a supernatural being, with greater than human powers, then the sentence is acceptable; this fact demonstrates that the sentence is not ungrammatical, but pragmatically unacceptable." On (7), p. 337: "first and second person do not participate in the animacy hierarchy." Spelling: Sandoval & Jelinek print dzaaneez with no accents, and Willie prints dzaanééz. Both are recorded as printed. The two sources order the words of the bi- sentence differently: S&J (2) has horse–mule, Willie (37b) has mule–horse. Both follow the same rule, because in each one the first noun is the one kicked.
2.2 With one nominal or none
| (38a) | dzaanééz | yiztał | ‘It kicked the mule.’ |
| mule | 3.3.kicked | ||
| (38b) | łį́į́ʼ | biztał | ‘It was kicked by the horse.’ |
| horse | INV3.3.kicked | ||
| (39a) | yiztał | ‘It kicked it.’ | |
| 3.3.kicked | |||
| (39b) | biztał | ‘It was kicked by it.’ | |
| INV3.3.kicked | |||
p. 73: "When only one adjunct is present, it must be coindexed with the internal argument, and there is no adjunct to the external argument." And: "When no NPs are adjoined, the verb-sentences in the Direct and Inverse forms still differ in meaning." Sandoval & Jelinek print the same bare pair, (9a) yiztał yi-kicked ‘He kicked him.’ and (9b) biztał bi-kicked ‘He got kicked by him.’ (p. 338).
2.3 The ranking, and the stars it produces with no nominals at all
Willie 1991:86, (70): "If the agent is higher on the animacy scale than the patient, then the Direct voice is used: if the patient is higher on the scale, then the Inverse must be used. And if they are equal, the choice rests on discourse factors." Same page: "even when the sentence does not contain any nominals, the sentence must be interpreted with reference to the animacy scale."
| (72a) yishxash 3.3.bit | her label | (72b) bishxash INV3.3.bit | her label | |
|---|---|---|---|---|
| 1) | he bit him. | equal ranked | he was bitten by him. | equal ranked |
| 2) | he bit it. | pat. lower ranked | he was bitten by it. | patient higher |
| 3) | it bit it. | lower equals | it was bitten by it. | lower equals |
| 4) | *it bit him. | agent lower | *it was bitten by him. | higher agent |
pp. 87–88: "what is ruled out (72a.4) is a lower-ranked agent in the yi- verb, and what is ruled out in (72b.4) is the higher agent in the bi- verb … Each paradigm permits what the other excludes." Also p. 88: "on the preferred interpretation of these sentences, for both the Direct and Inverse voices, the internal argument is taken to be a lower or indefinite being."
2.4 Speaker variation, and fronting under the inverse
| (73a) | ʼiiʼni | łį́į́ʼ | yiyiisxį́ | ‘The lightning killed the horse.’ |
| lightning | horse | 3.3.killed sg. | ||
| (73b) | łį́į́ʼ | ʼiiʼniʼ | biisxį́ | ‘The horse was killed by lightning.’ |
| (74a) | iiʼniʼ | łį́į́ʼ | biisxį́ | ‘The lightning killed the horse.’ |
| (74b) | lííʼ | iiʼniʼ | biisxí | ‘The lightning killed the horse.’ |
| (74c) | *iiʼniʼ | łííʼ łaʼ | biisxį́ | |
| (75a) | łííʼ | iiʼniʼ | yiyiisxį́ | ‘The horse killed the lightning.’ |
| (75b) | iiʼniʼ | łííʼ | yiyiisxį́ | ‘The lightning killed the horse.’ |
| (76a) | *iiʼniʼish | łííʼ | biisxį́ | |
| (76b) | łííʼ-ísh | iiʼniʼ | biisxį́ | ‘Did the lightning kill the horse?’ |
Only (73a) has a gloss line in the source. Recorded as printed, not repaired: ‘lightning’ appears as ʼiiʼni (73a), ʼiiʼniʼ (73b) and iiʼniʼ (74)–(76). ‘Horse’ appears as łį́į́ʼ, łííʼ (no hooks) and, in (74b), lííʼ with a plain l. The verb in (74b) is biisxí with no hook. The translation printed under (76b) is the only one given for (76).
Willie's comments. On (73a), p. 90: "Actually I find (73a) objectionable, as an animacy violation, and (73b) more preferable." On (74), p. 90: "you can change the order of the nounphrases in the bi- form but not with yi- form -- and not change the meaning", and (74c) is out because "the external argument in a bi- construction must be definite". On (75), p. 91: "when the Inverse is used, the speaker presupposes that the agent is lower on the animacy scale … Therefore, it is possible to put the highly salient nominal 'ii'ni' in sentence initial position, by focus movement. Whereas, when the speaker hears a Direct construction, he does not presuppose a lower ranked agent; and Ex. (75a), though extremely odd pragmatically, can have only the interpretation shown." On (76), p. 92: "When there is focus movement, as in (76a), this clitic does not attach to the moved element." On variation, p. 16: Young (1977; 1991) "found only about half of the speakers obeying the animacy hierarchy" in an informal survey (secondhand through Willie; Young not opened).
2.5 Goals and benefactives can invert; the theme cannot
| (79) | ʼashkii | ʼatʼééd | yichʼįʼ | yáłtiʼ | |
| boy | girl | 3.to | 3.speaking | ||
| ‘The boy is speaking to the girl.’ | |||||
| (80) | ʼashkii | ʼatʼééd | bichʼįʼ | yáłtiʼ | |
| boy | girl | 3.to | 3.speaking | ||
| ‘The boy is being spoken to by the girl.’ | |||||
| (85) | ʼashkii | ʼatʼééd | łį́į́ʼ | yá | yoolóós |
| boy | girl | horse | 3.for | 3.3.lead [sing. +animate] | |
| ‘The boy is leading the horse for the girl.’ | |||||
| (86) | ʼashkii | ʼatʼééd | łį́į́ʼ | bá | yoolóós |
| boy | girl | horse | 3.for | 3.3.lead [sing. +animate] | |
| ‘The girl is leading the horse for the boy.’ | |||||
The bi-forms bichʼįʼ (80) and bá (86) are glossed "3.to" and "3.for", with no INV, as printed. Willie, p. 94: "with an Inverse construction, the agent is always an internal argument; and the external argument is always an argument that has an "internal" theta role in the corresponding Direct construction." p. 95: "Note that the theme argument does not take part in the inversion process." p. 97: the Inverse "always makes the argument with an agent theta role the internal argument." Sandoval & Jelinek's Jicarilla rule (55b), p. 359, says the same for three-argument verbs: "Reverse coindexing of (1) and (2) [agent, goal]; coindexing in (3) [theme] remains unchanged."
2.6 Definiteness, case and topic
- SOURCED Willie 1991:76–77: (45) ʼashkii ʼatʼééd biztsʼǫs ‘The boy was kissed by the girl.’ vs (47) * ʼashkii léiʼ ʼatʼééd biztsʼǫs [A boy was kissed by the girl.]. Her conclusion: "with the Inverse, the external argument (the patient, with Absolutive case) must be definite." In the direct voice, either argument may be indefinite ((42)–(44), p. 76).
- SOURCED Case. On p. 72: "Case is assigned to the pronouns before they are incorporated into the verb. In the Inverse, where the agent is the "internal" argument, i.e., not the subject, I assume that the agent pronoun has Ergative case, and the patient pronoun has Accusative case (Marantz 1984)." Discrepancy in the source: pp. 77 and 100 instead give the inverse patient Absolutive case. Both are recorded; neither is chosen here.
- SOURCED Willie 1991:100 (summary): "The Inverse makes the agent an internal argument, with Ergative case, and places an argument with an "internal" theta role in the external argument position, with Absolutive case. With simple transitives, the Inverse subject is the patient … The subject of an Inverse construction is definite, and becomes the discourse topic." Sandoval & Jelinek 1989:339 (c): "the yi-/bi- alternation involves a change in the grammatical relations of the pronominal prefixes, such that a pronominal argument with the θ-role of Patient or Goal has the grammatical relation of SUBJECT in the bi- construction"; (d): it is "a transitive, inverse construction, and therefore not a passive".
- SOURCED Clause positions. Willie & Jelinek 2000:268: "The position immediately preceding the verb is the focus position … By default, an "outer" nominal must be coindexed with the pronoun that is not coindexed with the inner nominal; it is a "topic-adjunct"". Their (46) is ʼashkii łį́į́ʼ yiztał, boy horse 3ACC-3NOM-kicked (Direct), TOP-ADJ FOCUS Focus-Topic V, ‘The boy, he kicked [the horse]F.’ On p. 273: "The bi-pronoun marks a topic/focus inversion that is incompatible with focus movement of the Wh-word."
- SOURCED The prefixes. Willie 1991:27: "The subject is the innermost (rightmost) argument, closest to the verb stem, while the object is to the left of (external to) the subject." p. 28: "The yi- object pronoun appears only in Direct constructions, while the bi- pronoun appears only in Inverse constructions … The Inverse bi- is clearly cognate with the third person bi- pronouns that appear as possessors and postpositional objects".
3. The rules
Notation as on the schema page. Σ = {ρA, ρX}. ρA is the slot with the most Proto-Agent entailments (C5). ρX is the competing slot: a patient, or the object of a "grammatical" postposition (goal, benefactive). r(ρ) is the rank of the slot's referent on the animacy scale (Creamer's eight groups; Willie's "animacy scale"). EXT(Σ) is the slot realized as the external argument, i.e. the subject.
NAT Natural alignment. EXT = ρ_A. The slot with the most Proto-Agent entailments is the subject
(Dowty's ASP, 1991:576 (31)), spelled out yi-.
EDGE If r(ρ_X) > r(ρ_A), then EXT = ρ_X. ρ_A becomes internal, and the verb is spelled out bi-.
INV-0 Domain. VOICE is defined iff every slot in the competition is 3rd person, and Σ has ρ_A and some ρ_X.
INV-1 Fixity. E_ρ is the same under yi- and bi- for every ρ ∈ Σ. [= C2]
INV-2 Exponence. yi- iff EXT = ρ_A ; bi- iff EXT = ρ_X (and then pos ρ_A = internal).
INV-3 Ranking. EXT = the slot whose referent ranks higher:
r(ρ_A) > r(ρ_X) → EXT = ρ_A r(ρ_X) > r(ρ_A) → EXT = ρ_X
r(ρ_A) = r(ρ_X) → free (discourse decides)
INV-4 Definiteness. If EXT = ρ_X, the referent of ρ_X is definite. [= C7]
INV-5 Candidates. Only ρ_A and a slot with an "internal" role (patient, goal, benefactive) compete;
a theme in a three-slot frame never does.
INV-6 Linking (rule set B). The nominal immediately before the verb word links to the internal slot.
A lone nominal links to the internal slot. An outer nominal links to EXT (topic adjunct).
INV-7 Fronting (rule set B, optional). Under bi-, the internal-slot nominal may be focus-fronted over
the EXT nominal, with no change of roles. The fronted nominal cannot host -ísh.
| Rule | Status | Basis |
|---|---|---|
| NAT | HYPOTHESIS (the ASP is SOURCED; that it is the Navajo default is ours) | Dowty 1991:576. The direct voice is the one used when the agent is higher (Willie (70)). |
| EDGE | GRANT'S CONJECTURE, matched by SOURCED (70) | Grant's "a lower argument outranks it". Willie: "if the patient is higher on the scale, then the Inverse must be used" (p. 86). |
| INV-0 | SOURCED | Willie p. 86 ("only … if all arguments are third person"); S&J (7), p. 337; W&J 2000:264 (C4). |
| INV-1 | SOURCED in effect | Willie p. 71: the inverse "is a transitive sentence that has both agent and patient arguments"; the agent label survives the inversion (pp. 94, 97). S&J (d). The translations of (5)/(6) and (37a)/(37b) describe the same event. |
| INV-2 | SOURCED (restated) | Willie pp. 72, 94, 97, 100; S&J p. 339 (c). The formal statement is ours. |
| INV-3 | SOURCED as a generalization; its grammatical status is OPEN | Willie (70) "must". S&J on (5): "not ungrammatical, but pragmatically unacceptable". Willie (73): speakers vary. Young's survey: about half obey it (via Willie p. 16). Qualified 2026-10-14: the “equal → free” clause fails for two inanimates. Rock and tree (both Creamer group 7) allow no bi- in either order (Witherspoon 1977:66, (11)–(14)). Restated as INV-3′ with INV-CAP on the who acts on whom page. Restored 12:46pm: INV-CAP is falsified (Creamer 1974:37), and Creamer's movement rule makes the rock and tree unequal (1974:36), so INV-3 stands as stated here (Eight ranks §6.2). |
| INV-4 | SOURCED | Willie pp. 77, 88; (47), (74c). |
| INV-5 | SOURCED | Willie pp. 94–95, (79)–(80), (85)–(86); S&J (55b) for Jicarilla. OPEN Why the theme cannot compete (T5d). |
| INV-6 | SOURCED | Willie pp. 72–73, (37)–(38); W&J 2000:268, (46). |
| INV-7 | SOURCED (Willie); the link to W&J is HYPOTHESIS | Willie pp. 90–92, (74), (76). See §5 and T5e for how this squares with W&J's "incompatible with focus movement of the Wh-word". |
The cleanest restatement HYPOTHESIS. INV-3 is a rule about positions, not about roles: the external argument is the higher-ranked participant, and the internal argument is the lower one. That is Willie's p. 88 remark ("for both the Direct and Inverse voices, the internal argument is taken to be a lower or indefinite being"). The roles enter only through INV-2. bi- marks that the higher-ranked, external participant is not the one with the most Proto-Agent entailments. So the voice prefix reports a comparison between two orders over the same two slots. One is the entailment order (fixed by the event). The other is the rank order (fixed by who the participants are). This is the corollary on schema §4, bi- ⟺ ρrank ≠ ρA, now restated over EXT instead of over the topic DP.
4. Which orders are impossible
For a two-slot verb with two third persons there are three rank relations and two voices. SOURCED cells give the source's judgment. The schematic cells are labelled schematic, because no source prints that Navajo string.
| Rank relation | yi- (EXT = agent) | bi- (EXT = non-agent) |
|---|---|---|
| agent > patient (e.g. horse kicks rock) | ✓ (72a.2); W&J (61a) | ✗ (72b.4) "*it was bitten by him"; W&J (61b) ?* |
| agent = patient (horse, mule) | ✓ (37a), (72a.1) | ✓ (37b), (72b.1); the choice "rests on discourse factors" |
| patient > agent (horse kicks man) | ✗ (72a.4) "*it bit him"; S&J (5) ??; Willie (20b) *; W&J (62a) ?* | ✓ (6), (72b.2), (20c), (62b) |
HYPOTHESIS Two of the six cells are excluded, and they are the two where EXT is the lower-ranked participant. Add the order of the two nominals and there are 3 × 2 × 2 = 12 strings. INV-6 assigns each string exactly one reading: the nominal next to the verb is the internal participant. So the order of the nominals adds no new impossible cells. It only decides which noun is which slot. The one extra freedom is INV-7: under bi- a focus-fronted internal nominal can come first, as in (74a), and the reading is unchanged. Under yi- the reverse order is read literally, as in (75a) ‘The horse killed the lightning’. Schematically:
[NP_high] [NP_low] yi-V NP_low = internal = patient → high one is agent ✓ [NP_low] [NP_high] yi-V NP_high = internal = patient → low one is agent ✗ (72a.4) / ?? (5) [NP_high] [NP_low] bi-V NP_low = internal = agent → high one is patient ✓ (6) [NP_low] [NP_high] bi-V NP_high = internal = agent → low one is patient ✗ (72b.4) [NP_low]F [NP_high] bi-V (fronted from [NP_high] [NP_low]F bi-V) ✓ (74a); only under bi- (schematic; the sourced instances are cited in the right column)
Grant's edge, stated exactly. HYPOTHESIS The natural alignment (NAT) is overridden in one configuration only: patient > agent in rank. In that configuration the bi- form is not just allowed but the expected one (Willie's "must"; S&J's "the bi-construction is employed"), and the NAT form is excluded or marked. What is never possible is a lower-ranked participant as the external argument. The entailments are the same in every cell (INV-1). Only EXT moves.
5. Where in the derivation the choice is made
5.1 Inside the word, not in the clause HYPOTHESIS (from SOURCED facts)
- The choice is made with no nominals at all. (39a) yiztał ‘It kicked it.’ and (39b) biztał ‘It was kicked by it.’ "still differ in meaning". (72a.4) and (72b.4) are starred with no noun in the sentence. So the voice cannot be decided by rule set B, because B (topic/focus positions, linking of nominals) has nothing to work on here. It is fixed at Σ, by the slots and their referents, and spelled out inside the word by rule set A, as yi- or bi- to the left of the subject marker (Willie p. 27).
- Willie's own model puts it there. "Case is assigned to the pronouns before they are incorporated into the verb" (p. 72). The inverse "has scope over the entire MVS" (p. 100), her Maximal Verb Sentence.
- Rule set B only reads the result. INV-6 links nominals to slots once EXT is known. INV-7 (focus fronting) is possible only because bi- has already fixed which slot is the agent (Willie p. 91). B can be blocked by the voice, as in (76a) -ísh and W&J's wh-facts, but it never chooses the voice.
- Correction to the earlier pages. Schema §4 and D3/D4 define VOICE by the slot of the clause-initial topic DP (a rule-set-B notion). Because of (39) and (72), that definition is too late in the derivation. The rule should be stated over EXT (INV-2, INV-3), and the topic facts should follow from INV-6. In sentences that have two nominals the two statements give the same results, so no earlier derivation changes its output.
5.2 What VPISH predicts
Update 2026-10-16 (8:46pm): restated as probe/Agree + EPP, with Hale 1997's own two-step EPP analysis and Bruening 2005's Voice-[EPP] mechanism, on the inverse as Agree and EPP (T27k). Four options are tested there (IA-0 linking, IA-1* raising in the inverse only, IA-2 Hale 1997, IA-3 in situ). da- plus the word's spell-out order (CYC) constrains raising, and the theme exclusion (INV-5) follows from closeness under raising if the PP is above the theme.
- HYPOTHESIS Under V-P (VPISH raising), EXT is simply the slot that raises to Spec,M/SP. In the direct voice ρA raises from Spec,v and nothing crosses. In the inverse, ρX raises from Comp,V (or from inside a grammatical PP) past ρA, which stays in Spec,v. That is Willie's "internal (Ergative)" agent. So the inverse is the one construction on this site where V-P gives the first Spec-to-Spec movement to a non-agent. The "second type" of movement that Grant allows for arguments beyond the first becomes, under the inverse, the movement to subject.
- HYPOTHESIS Under V-H (Hale 2003, licensing in situ), nothing moves. EXT is a case designation on the pronouns (Ergative agent, Absolutive or Accusative patient; Willie pp. 72, 77). The two variants give the same string, as in D1–D5. They differ in structure: V-P makes ρX c-command ρA in the inverse; V-H does not.
- OPEN So the inverse is where VPISH makes its sharpest testable prediction: scope or binding of the patient over the agent under bi- only. Willie's quantifier pairs (58)–(60), pp. 79–80, are the place to start. Update 2026-10-09: they are now page-checked and printed on the number & scope page. They do not settle T5c, because Willie does not say which scope readings (58) allows (T10a). That page also finds that the distributive da- counts the agent in both voices. This constrains V-P: Pl must Agree with ρA before the patient raises, or else Pl must be role-indexed (§5.1 there). It also bears on step 6 of I1 below.
- GRANT'S CONJECTURE + HYPOTHESIS If V-P is right, what moves in the inverse is a role-slot (ρX), not a word: the nominals are adjuncts and stay where rule set B puts them (INV-6). The S&J horse case points the same way. The same word łį́į́ʼ in (5) gives ?? for an ordinary horse and ✓ for "a supernatural being, with greater than human powers". So the rank belongs to the participant as construed, not to the lexical item.
5.3 How INV-7 fits W&J's wh-facts HYPOTHESIS
W&J 2000:273 say the bi-form is "incompatible with focus movement of the Wh-word": in (60b) háí-sh Jáan biztsʼǫs, reading 2) *‘John was kissed by whom/which one?’ is out (D7). Willie (74a) allows focus fronting of the agent nominal under bi-. The two are consistent if what blocks (60b-2) is the question clitic: a focus-moved element cannot host -ísh (76a), and the wh-word in (60b) carries -sh. Then the rule is "under bi-, a fronted internal nominal cannot carry the question clitic". It is not "under bi-, nothing fronts". This also refines prediction (b) on the open-questions page (multiple-wh). OPEN (T5e).
Update 2026-10-20, 8:46pm (T43f). Willie p. 92 and pp. 210–212 and W&J pp. 272–273 were re-read on page images (the two question endings compared). One statement (QC-5, HYPOTHESIS) now covers *(76a), *(60b) reading 2 and Willie *(17) Díí=sh háí biztał: under bi- a question ending may sit only on a host in its unmoved position. *(17) involves no fronting, so the rule above (“a fronted internal nominal cannot carry the question clitic”) is too narrow. (76b) łííʼ-ísh and W&J (59) háí-sh Jáan biztsʼǫs show that the inverse subject itself can carry either ending, so the ban is not on the landing site. T5e is partly answered. Whether a fronted wh-agent with no -sh is possible is still OPEN.
6. Derivations I1–I5, step by step
Conventions as on the derivations page. Tags: Σ schema step, A word-internal linearization, B clause level. V-H = in-situ licensing (Hale 2003:29). V-P = VPISH raising. Verb words are not segmented beyond the sources' glosses. Entailment assignments are ours, from the English translations. They match D3: for ‘kick’, EρA ⊇ {vol, caus, mov} and EρP ⊇ {aff} (HYPOTHESIS).
I1. S&J (6) hastiin łį́į́ʼ biztał ‘The man got kicked by the horse.’ (edge case, inverse)
- Σ Σ(√KICK) = {ρA, ρP}. posρA = Spec,v; posρP = Comp,V (Hale 2003 tree (8)). C5: ρA has three P-A entailments and ρP has none, so ρA is the agent slot. These entailments are fixed here and are never touched again (INV-1).
- Merge V (the ‘kick’ stem; not segmented) with ρP. A lexical V is complement-final: ρP⌢V.
- Merge v (Voice/classifier; its value is not given by the sources). A v⌢VP. Merge ρA in Spec,v. φ unvalued.
- Merge M/S. A M/S⌢vP.
- Σ π-valuation: π(ρA) = π(ρP) = 3, so INV-0 holds and VOICE is defined. Rank: the referent of ρP is a man and the referent of ρA is an ordinary horse, so r(ρP) > r(ρA). INV-3 gives EXT = ρP, and INV-2 gives VOICE = bi-. No nominal has been used yet.
- Agree M/S ⟶ EXT. (Update 2026-10-16, 8:46pm: superseded under the raising version IA-1* on the inverse as Agree and EPP. There the patient slot moves first and M/S then agrees with ρA left below it (Bruening 2005 n. 22), so M/S agrees with ρA in both voices, as W&J's 3NOM gloss has it.) (Caveat 2026-10-09: if da- number agreement rides on M/S, M/S must agree with ρA, not EXT, in the inverse; see number & scope §5.1, T10b.) In Willie's terms, the external argument gets Absolutive case (p. 100; Accusative on p. 72) and the internal agent gets Ergative case.
- Move. V-P HYPOTHESIS: ρP goes from Comp,V to Spec,M/SP, crossing ⟨ρA⟩, which stays in Spec,v as the "internal (Ergative)" agent. V-H: no movement.
- Spell-out A. The position-IV exponent is bi-, to the left of the (null) 3rd-person subject marker and the stem. The word is biztał, glossed "bi-kicked" by S&J. Which slot bi- itself indexes, the external patient or the internal agent, is OPEN (T5a).
- B Linking (INV-6). łį́į́ʼ, immediately before the verb, links to the internal slot, which under bi- is ρA. hastiin, the outer nominal, links to EXT = ρP as the topic adjunct (W&J 2000:268). INV-4: hastiin is unmarked, so its preferred reading is definite (Willie p. 76) ✓. Order: DPEXT ⌢ DPint ⌢ V.
I2. S&J (5) ?? łį́į́ʼ hastiin yiztał ?? ‘The horse kicked the man.’ (the natural alignment, marked)
- Σ Same Σ, same entailments, same referents as I1 (INV-1). ρA = the horse, ρP = the man.
- Merge as in I1, steps 2–4.
- Σ NAT: EXT = ρA, so VOICE = yi-. INV-3: r(ρP) > r(ρA) requires EXT = ρP. Conflict: NAT and INV-3 pick different slots, and INV-3 is violated.
- Agree, Move (V-P: ρA to Spec,M/SP, nothing crosses), and Spell-out A yiztał all go through. No step of the word fails.
- B INV-6: hastiin, next to the verb, links to the internal slot = ρP; łį́į́ʼ links to EXT = ρA. The reading is coherent (the man is kicked), so linking does not fail either.
- Result. The only violation is INV-3, at the Σ step, and S&J mark it ??, not *. If the horse "is understood as a supernatural being", then r(ρA) > r(ρP), INV-3 is satisfied, and the same string is acceptable (S&J p. 336). HYPOTHESIS INV-3 reads the rank of the participant as construed in context. It does not read a fixed property of the word łį́į́ʼ.
I3. Willie (37a)/(37b): equal rank, both voices
- Σ Σ(√KICK) as in I1. ρA = the horse, ρP = the mule. r(ρA) = r(ρP): "of equal animacy" (Willie p. 72).
- Σ INV-3 leaves EXT free, and discourse chooses. (37a): EXT = ρA → yi-. (37b): EXT = ρP → bi-.
- Move (V-P). In (37a) ρA raises. In (37b) ρP raises past ⟨ρA⟩. V-H: none.
- Spell-out A yiztał / biztał.
- B INV-6. (37a) dzaanééz (next to V) → internal = ρP; łį́į́ʼ → EXT = ρA. (37b) łį́į́ʼ (next to V) → internal = ρA; dzaanééz → EXT = ρP. In both, the horse is the kicker.
- One-nominal variants (38): the lone nominal links to the internal slot, so (38a) dzaanééz yiztał ‘It kicked the mule.’ and (38b) łį́į́ʼ biztał ‘It was kicked by the horse.’ ✓ INV-6.
I4. Willie (72a.4) *yishxash ‘it bit him’ and (72b.4) *bishxash ‘it was bitten by him’: stars with no nominals
- Σ Σ(√BITE) = {ρA, ρP}. (72a.4): the referent of ρA is an "it" and the referent of ρP is a "him", so r(ρP) > r(ρA).
- Σ VOICE = yi- means EXT = ρA, the lower-ranked one. INV-3 is violated: *.
- (72b.4): the referent of ρA is "him" and the referent of ρP is "it", so r(ρA) > r(ρP). VOICE = bi- means EXT = ρP, the lower-ranked one. INV-3 is violated: *.
- B No B step exists: there are no nominals to link and nothing to front. HYPOTHESIS Both stars are therefore made by Σ and rule set A alone (§5.1).
I5. Willie (73b) → (74a), and (76a): fronting under the inverse (rule set B)
- Σ Σ(√KILL) = {ρA, ρP}. EρA ⊇ {caus} (no vol or sent: lightning). EρP ⊇ {chg, aff} (ours). C5: ρA = the lightning, because it has the only P-A entailment.
- Σ The horse outranks the lightning (Willie calls (73a) "an animacy violation"). INV-3 gives EXT = ρP, so VOICE = bi-. A biisxį́.
- B Base linking (INV-6), (73b) łį́į́ʼ ʼiiʼniʼ biisxį́: ʼiiʼniʼ, next to V, is the internal slot = ρA; łį́į́ʼ is EXT = ρP.
- Move (A′, optional; INV-7) B. iiʼniʼ focus-fronts to the left edge, leaving a copy: (74a) iiʼniʼ łį́į́ʼ ⟨iiʼniʼ⟩ biisxį́ ‘The lightning killed the horse.’ The roles do not change, because bi- already fixed them in step 2. With yi- no such parse is used. (75a) łííʼ iiʼniʼ yiyiisxį́ can only mean ‘The horse killed the lightning.’ (p. 91).
- B (76a): adding -ísh to the fronted element gives *iiʼniʼish łííʼ biisxį́. The clitic "does not attach to the moved element", and the question is formed instead on the unmoved order (76b) łííʼ-ísh iiʼniʼ biisxį́.
What I1–I5 show HYPOTHESIS
- The entailments never change. I1–I3 share Σ step 1. Only EXT differs, which is exactly Grant's "the inverse changes only which argument is realized where".
- The voice is decided at Σ and spelled out by rule set A. I4 has stars with no rule-set-B input at all. Rule set B links nominals afterwards (INV-6) and may front one (INV-7), which only bi- licenses.
- The impossible cells are the ones where the external argument is the lower-ranked participant. Whether that is a grammatical * (Willie) or a pragmatic ?? (S&J) is not settled (T5b).
- Under VPISH raising, the inverse is the movement of a patient slot over the agent slot into subject position. No lexical item moves. That is the testable part (T5c).
7. Verdict on T5
- Position, not only linking. SOURCED Willie 1991's answer is explicit. The inverse "makes the agent an internal argument … and places an argument with an "internal" theta role in the external argument position" (p. 100). Sandoval & Jelinek 1989:339 agree that the patient or goal becomes SUBJECT. HYPOTHESIS W&J's constant "3ACC-3NOM" gloss labels the prefix string. It is not a claim that positions stay constant. Which prefix indexes which slot is still OPEN (T5a), and Willie's own case labels disagree (Accusative on p. 72, Absolutive on pp. 77 and 100).
- Where. HYPOTHESIS The choice is made at Σ (which slot is external, given rank) and spelled out inside the word by rule set A. This follows from (39) and (72), where the forms differ or are starred with no nominals. Rule set B only links and fronts. The earlier pages' topic-DP definition of VOICE is restated over EXT (§5.1, item 4).
- Grant's ranking picture fits the data. HYPOTHESIS NAT (agent as subject) holds unless the non-agent outranks the agent. Then bi- is the expected form, and the two cells with a lower-ranked external argument are out ((72a.4), (72b.4)). Theme arguments never compete (INV-5).
- Not settled. OPEN Whether VPISH raising is real here. That needs scope or binding evidence that the inverse patient is structurally above the agent (T5c). Also open: whether the ranking is grammar or usage (T5b). (2026-10-23: Oxford 2023's tests are run on deep or shallow inverse?; deep-like where runnable, binding still open, T53a.)
8. Open questions this page adds
| # | Question | What would settle it |
|---|---|---|
| T5a | In biztał, does bi- index the external (patient) slot or the internal (agent) slot? W&J gloss the string 3ACC-3NOM in both voices. Willie calls the inverse patient Accusative on p. 72 and Absolutive on pp. 77 and 100. | Young & Morgan 1987 paradigms; Willie 2000 (IJAL 66) and Aissen 2000, neither opened. Also inverse forms with a dual or plural on just one argument (Willie (63)–(69), pp. 81–85: da- scope. Page-checked 2026-10-09 on the number & scope page. da- counts the agent in both voices. That does not settle T5a, but it ties it to T7.) OPEN |
| T5b | Is INV-3 a grammatical rule or a usage preference? Willie says "must" and stars (72a.4). S&J call (5) "pragmatically unacceptable" and acceptable for a supernatural horse. Speakers vary (Willie (73); Young's survey via Willie p. 16). HYPOTHESIS The supernatural-horse fact is Dowty's (ii)-type diagnostic (meaning-based, extendable in context) from the plain ł-causer page, so it favours a constraint that reads meaning over a fixed class of nouns. | Native judgments on (5)-type pairs in plain and personifying contexts, by speaker and age (with T6). OPEN |
| T5c | VPISH test: under bi-, does the external patient take scope over, or bind into, the internal agent, and not under yi-? V-P predicts yes; V-H predicts no difference. | Willie (58)–(60), pp. 79–80. Page-checked 2026-10-09 on the number & scope page. They are not decisive, and the question is sharpened there as T10a. New elicitation with reflexives or quantifiers is still needed. OPEN |
| T5d | Why can a goal or benefactive become external under bi- but a theme cannot, even an animate theme (‘horse’ in (85)–(86))? Is the candidate set fixed by an entailment of the slot, or by structure (object of a "grammatical" postposition)? If it is an entailment, this is a T7 case: a generalization over Eρ that pronouns alone do not state. | More three-participant verbs with animate themes; Jelinek & Willie 1996 ("Psych verbs in Navajo"), not opened. OPEN |
| T5e | Is the clitic restriction (76a) the whole reason why W&J (60b) reading 2) is out? If so, a fronted wh-agent under bi- without -sh, or with -sh placed elsewhere, should be possible. | Elicitation; W&J 2000 pp. 270–273 re-read together with Willie ch. 8. OPEN |
| T5f | The forms are spelled inconsistently: łį́į́ʼ / łííʼ / lííʼ and ʼiiʼni / ʼiiʼniʼ / iiʼniʼ within Willie §3.3.3, and dzaanééz (Willie) vs dzaaneez (S&J). Which spellings are typescript errors? | Young & Morgan 1987 dictionary entries; a Navajo speaker. OPEN |
| T5g | Hale 1973 ("A note on subject-object inversion in Navajo"), the original statement that the higher-ranked noun comes first, is known here only through Willie (p. 12) and S&J (p. 336). Update 2026-10-14: Witherspoon 1977:69–70, 72, 80–81 quotes Hale 1973:303–306 (the animacy table, the ranking hypothesis, “cannot be formulated in absolute terms”, “relative acceptability”); page-checked there, see who acts on whom §4. | A library copy of Kachru et al. (eds.) 1973, pp. 300–309. OPEN (the quotations are secondhand) |
Added 2026-10-14 (8:46pm): with one overt noun, Atʼééd yiiłtsą́ is only ‘Sh/he saw the girl’ and Atʼééd b-iiłtsą́ only ‘The girl saw him/her’ (Speas 1986:250, (4.8)). So a lone noun always links to the bundle that is internal after the voice choice (FB-3a, Speas 1986 on Navajo). In a bi- relative under a yi- verb, (4.54), the horse is external in both clauses although it is the patient of one and the agent of the other.
Compared 2026-10-23: the Algonquian inverse (T52). Algonquian has two inverses: an obligatory, agreement-only one (3→1/2) and an optional voice that raises the patient (3′→3). Theta roles are unchanged in both. yi-/bi- lines up with the second only.
Follow-up (2026-10-09): number, quantifiers and da- across the two voices, on the number & scope page (T5c, T10).
Next increment (done 2026-10-08 evening): the fourth person (T8) — escape hatch from the same ranking clash, not a second inverse.
