Minimalism Agree-Conditioned Pronoun Spellout in Mam #
@cite{scott-2023} @cite{chomsky-2000} @cite{deal-2024} @cite{elkins-torrence-brown-2026} @cite{preminger-2014}
Connects the Agree operation (feature valuation) and probe restriction to the empirical distribution of overt vs. reduced pronouns in SJA Mam.
The Derivation #
In a Mam transitive clause with a 3SG agent and 3SG patient:
Voice probes for φ: Voice has [uPerson, uNumber]. It Agrees with the agent in Spec,VoiceP, valuing [Person:3, Number:sg]. Voice also assigns inherent ERG case to the agent. The valued φ-features spell out as Set A "t-" on Voice.
Infl probes for φ: Infl has a φ-probe with a disjunctive satisfaction condition [SAT: φ or Voice_TR]. In a transitive clause, the probe encounters transitive Voice and stops — no φ-features are copied to Infl. The Set B slot is filled by the Elsewhere form "tz'=" (default 2/3SG).
Object is case-licensed by Voice: The patient receives structural ACC from Voice (low-abs syntax; @cite{scott-2023}, §3.4). Infl's probe never reaches the object because it was satisfied by Voice_TR.
Pronoun realization follows: Agreed-with arguments (agent, S) undergo pronoun reduction — agreement morphology redundantly expresses their φ-features, triggering deletion of the pronominal base (@cite{scott-2023}, ch. 4). The patient, lacking φ-agreement, must be a full overt pronoun.
Two Paths to Set B "tz'=" #
A key insight of Scott's analysis is that the same Set B form "tz'=" arises via two distinct mechanisms:
- Intransitive: Infl's probe succeeds, copies S's φ-features (e.g., [3SG]) → no more specific Set B entry matches → Elsewhere "tz'=" is selected.
- Transitive: Infl's probe is blocked by Voice_TR → no φ-features on Infl → Elsewhere "tz'=" is selected.
Both yield "tz'=", but the intransitive case involves real agreement while the transitive case involves probe failure. This bridge demonstrates both paths.
These Vocabulary Insertion entries encode the Fragment's theory-neutral
marker tables as Minimalism feature bundles, enabling the Agree → Spellout
pipeline. The Fragment (Agreement.lean) stores the markers as simple
person × number → string tables; here they are parameterized by
FeatureBundle and Cat for use with spellout.
Set A (ERG) vocabulary entries: φ-features on Voice (.v) yield the morphological exponent (@cite{scott-2023}, Table 2.8).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Set B (ABS) vocabulary entries: φ-features on Infl (.T) yield the morphological exponent (@cite{scott-2023}, Table 3.5). The Elsewhere entry (no features, tz'=) surfaces when no more specific entry matches.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Which Minimalist head φ-Agrees with each argument position.
Equations
- Phenomena.Agreement.Studies.Scott2023.agreeProbe Fragments.Mayan.Mam.MamArgPosition.agent = some Minimalism.Cat.v
- Phenomena.Agreement.Studies.Scott2023.agreeProbe Fragments.Mayan.Mam.MamArgPosition.patient = none
- Phenomena.Agreement.Studies.Scott2023.agreeProbe Fragments.Mayan.Mam.MamArgPosition.intranS = some Minimalism.Cat.T
Instances For
Voice's probe features: [uPerson, uNumber].
Placeholder values (.third, .sg) are irrelevant — sameType matching
ensures any Person/Number goal is found regardless.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Infl's probe features: [uPerson, uNumber]. In intransitives, these are valued by S. In transitives, the probe is blocked by Voice_TR before reaching any DP.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A 3SG DP's features: [Person:3, Number:sg].
Equations
- One or more equations did not get rendered due to their size.
Instances For
Voice's [uPerson] is valued as [Person:3] from a 3SG agent.
After person agreement, Voice's [uNumber] is valued as [Number:sg]. This is the second step of φ-Agree: person first, then number.
Full φ-valuation of Voice by a 3SG agent: both person and number valued.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The two-step Agree pipeline produces a fully valued bundle.
Set A spellout: Voice's valued [Person:3, Number:sg] yields "t-" (A2/3SG).
Set A spellout for 1SG: Voice with [Person:1, Number:sg] yields A1SG marker.
Intransitive path: Infl Agrees with a 3SG intransitive S, copies [Person:3, Number:sg]. No Set B entry is specified for these features (1SG=chin, 1PL=qo, 2/3PL=chi — none match 3SG), so the Elsewhere entry is selected: "tz'=".
Transitive path: Infl's probe is blocked by Voice_TR → no φ-features are copied → the Infl node has an empty feature bundle. The Elsewhere entry matches (empty features are a subset of anything) and "tz'=" is selected.
This is the DEFAULT Set B — it appears in transitives regardless of the object's person/number features.
The two paths produce the same exponent — the surface form is identical even though the underlying mechanism differs (real agreement vs. probe failure).
Set B spellout for 1SG intransitive: Infl copies [Person:1, Number:sg] from S, yielding "chin" — NOT the Elsewhere form. This is real agreement, producing a distinct exponent.
In a transitive with a 1SG object, the default "tz'=" still appears — NOT "chin". This is because Infl's probe was blocked by Voice_TR, so the object's 1SG features are never copied to Infl.
Infl's probe has a disjunctive satisfaction condition [SAT: φ or Voice_TR]. In transitives, the probe encounters transitive Voice and stops before reaching any DP. This is modeled by the fact that the Infl node ends up with an empty feature bundle (no φ-features copied).
In intransitives, Voice is not transitive → the probe continues → finds S → copies φ.
This mechanism replaces the older "closest-goal intervention" account: it is NOT that the agent intervenes between Infl and the object, but that the probe is STOPPED by the VoiceP phase boundary.
In an intransitive clause, Infl probes for φ and Agrees with S. This is REAL agreement — the probe copies S's φ-features. The resulting valued features spell out as Set B.
The complete prediction for a 3SG transitive clause:
- Voice Agrees with agent → [Person:3, Number:sg] on Voice
- [Person:3, Number:sg] on Voice spells out as Set A "t-"
- Infl's probe is blocked by Voice_TR → empty bundle on Infl
- Empty Infl spells out as Elsewhere Set B "tz'="
- Patient is not φ-Agreed-with → overt pronoun required
The pipeline generalizes: for every argument position, the predicted pronoun reduction matches the observed pattern.
Set A and Set B have different contexts: Set A on Voice (.v), Set B on Infl (.T). The same valued features produce different exponents depending on which head hosts them.
Set A vocabulary does NOT match Infl context.
Set B vocabulary does NOT match Voice context (for specified entries). Only the Elsewhere entry could match (it has no features), but it requires Infl context.
The three argument positions each have distinct agreement marking patterns, yielding morphological tripartite alignment (Scott p. 113):
- Agent (ERG): Set A agreement from Voice
- Intransitive S (ABS): Set B agreement from Infl
- Patient (ACC): default Set B (Infl probe blocked)
These three cases each have distinct underlying syntactic case values, assigned by different heads (Voice for ERG/ACC, Infl for ABS).
Agreement probes are on different heads: Voice for Set A, Infl for Set B. The patient's lack of agreement is NOT because both heads target the agent — it's because Infl's probe is blocked by VoiceP.
The transitive Set B default is an instance of Preminger's probe failure:
Infl's probe searches an empty domain (blocked by Voice_TR) and finds no
DP with matching φ-features. attemptAgree maps the none result from
applyAgree to ProbeOutcome.unvalued.
The intransitive case is real agreement: Infl's probe finds S and copies
its φ-features. attemptAgree maps the some _ result to .valued.
Under Preminger's obligatory-no-crash model, probe failure converges
and produces Elsewhere morphology — exactly the Set B "tz'=" we observe
in Mam transitives. This connects the abstract failure model to the
concrete spellout: ProbeOutcome.unvalued → PFRealization.elsewhere
→ the Elsewhere Vocabulary entry → "tz'=".
The Fragment file (Agreement.lean) stipulates isPhiAgreed := false for
patients. Here we DERIVE that result from the SatisfactionCond machinery
in Agree.lean: Infl's disjunctive probe [SAT: φ or Voice_TR] encounters
transitive Voice and stops without copying features.
This closes the gap between stipulation and derivation: the patient's
lack of φ-agreement is not an axiom but a consequence of probe
satisfaction theory.
In a transitive clause, mamInflSatisfaction is satisfied by Voice_TR
(head encounter .v) and copies no features — matching the Fragment's
patient.isPhiAgreed = false.
In an intransitive clause, mamInflSatisfaction is satisfied by
φ-features and DOES copy them — matching intranS.isPhiAgreed = true.
The satisfaction condition's copiedFeatures result aligns exactly with
the Fragment's isPhiAgreed for both Infl-probed positions:
- patient (transitive): copiedFeatures = false = isPhiAgreed
- intranS (intransitive): copiedFeatures = true = isPhiAgreed
Voice⁰ in Mam carries two independent probes:
- φ-probe [uPerson, uNumber]: Agrees with agent in Spec,VoiceP, yielding Set A morphology (e.g., "t-" for 3SG).
- Oblique probe [uOblique]: Agrees with a passing Ā-moved oblique, yielding =(y)a' on Voice⁰.
Both are instances of the same abstract Agree operation: probe searches c-command domain, finds closest matching goal, copies features, and the valued features are spelled out by Vocabulary Insertion. They differ only in which features they probe for and what vocabulary entries match.
This section makes the unity explicit by running both pipelines in parallel and showing they produce different exponents from the same mechanism.
See also: Phenomena.FillerGap.Studies.ElkinsTorrenceBrown2026 for the
full =(y)a' analysis.
Both probes on Voice are unvalued features — both act as probes in the sense of Agree.
φ-Agree and oblique-Agree are parallel instances of the same operation: probe, value, spellout. They differ only in which features are probed and which vocabulary entries match.
- φ-Agree: Voice + 3SG agent → [Person:3, Number:sg] → Set A "t-"
- Oblique-Agree: Voice + oblique → [+oblique] → =(y)a'
Both pipelines use applyAgree for valuation and spellout for
morphological realization.
Scott's impoverishment rule (ex. 84/94):
`[+/−singular] → ∅ / [+author]^F`
Deletes [±singular] from first person pronouns that have been
agreed with (marked by the F diacritic). This bleeds insertion of
the pronominal base morphemes *qin* ([+author,+singular]) and *qo*
([+author,−singular]), yielding reduced pronouns.
We model this using `Morphology.DM.Impoverishment.ImpoverishmentRule`.
The condition checks for [+author] (= first person in our feature
system), and the target is [±singular] (= number).
The Mam first-person impoverishment rule: delete [±singular] (number) when the bundle contains [+author] (first person) features that have been agreed with.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The impoverishment rule fires for 1st person bundles.
The impoverishment rule does NOT fire for 3rd person bundles.
After impoverishment, the number feature is deleted from 1st person bundles, bleeding insertion of the base morpheme qin.
Without impoverishment (3rd person), the number feature survives.