Cyclic Agree @cite{bejar-rezac-2009} #
@cite{bejar-rezac-2009} derive person hierarchy (PH) effects from three independently motivated mechanisms:
- Articulated φ-features: person is decomposed into hierarchical segments — [π] ⊂ [participant] ⊂ [speaker]/[addressee] — each of which can independently enter Agree.
- Feature-relativized locality: a probe segment [uF] only sees goals bearing [F]; a goal lacking [F] is bypassed, leaving an active residue.
- Cyclic expansion: v's probe first Agrees with the IA (cycle I), then with the EA (cycle II). The IA is preferred because cyclicity puts it in the search space first.
The interaction of probe articulation and cyclic Agree derives:
- Agreement displacement: IA agreement bleeds EA agreement when the IA fully checks the probe. When it doesn't, residue reaches the EA.
- Direct/Inverse contexts: Inverse = core probe doesn't Agree with EA (IA fully checks, or EA can't match the residue). Direct = EA Agrees with residue segments.
- Crosslinguistic variation: languages differ in probe articulation — flat [u-3] (no PH sensitivity), partial [u-3-2] (SAP vs 3P), or full [u-3-2-1] (all persons distinguished).
Person Geometries #
Two attested geometries for the innermost feature:
- Standard (1>2>3): [speaker] distinguishes 1st from 2nd. Used by Basque, Georgian (@cite{bejar-rezac-2009} Table 2B).
- Addressee (2>1>3): [addressee] distinguishes 2nd from 1st. Used by Nishnaabemwin, Mohawk (@cite{bejar-rezac-2009} Table 2C).
Person Licensing and Repair #
The Person-Licensing Condition (PLC) requires every π-feature to be licensed by Agree. In inverse contexts, the EA's π-features are not licensed by the core probe, triggering repair strategies:
- Added probe: an extra probe is inserted on v_II in inverse contexts (Nishnaabemwin, Mohawk, Basque)
- R-Case: the IA receives a special oblique Case (Kashmiri)
Both strategies are derivationally bounded: they occur in all and only inverse contexts.
Integration #
Connects to PersonGeometry.lean via bridge theorems. The direct/inverse
classification predicts which languages show differential P indexing
(Basque, Georgian fragments).
A single segment in an articulated person feature structure.
Segments are privative features organized in a containment hierarchy. Every person value bears [π]; SAPs additionally bear [participant]; the innermost feature ([speaker] or [addressee]) distinguishes 1st from 2nd person.
Instances For
Equations
- Minimalism.CyclicAgree.instBEqSegment.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Equations
- Minimalism.CyclicAgree.instBEqGeometry.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
Equations
Equations
- One or more equations did not get rendered due to their size.
Instances For
The person specification for a given person value under a geometry.
Returns the list of segments a DP of this person bears.
Equations
- One or more equations did not get rendered due to their size.
- Minimalism.CyclicAgree.personSpec geom Core.Prominence.PersonLevel.third = [Minimalism.CyclicAgree.Segment.pi]
- Minimalism.CyclicAgree.personSpec Minimalism.CyclicAgree.Geometry.addressee Core.Prominence.PersonLevel.first = [Minimalism.CyclicAgree.Segment.pi, Minimalism.CyclicAgree.Segment.participant]
- Minimalism.CyclicAgree.personSpec Minimalism.CyclicAgree.Geometry.standard Core.Prominence.PersonLevel.second = [Minimalism.CyclicAgree.Segment.pi, Minimalism.CyclicAgree.Segment.participant]
Instances For
The most specified person under a given geometry.
Equations
Instances For
An articulated probe: a list of unvalued segments, ordered from outermost (most general) to innermost (most specific).
Languages vary parametrically in probe articulation:
- Flat [u-3]: no PH sensitivity (e.g., Swahili, Abkhaz)
- Partial [u-3-2]: distinguishes SAP from 3P (e.g., Basque, Georgian)
- Full [u-3-2-1]: distinguishes all persons (e.g., Nishnaabemwin, Mohawk)
Instances For
Flat probe: [uπ]. Any DP fully matches.
Instances For
Partial probe: [uπ, uParticipant]. Distinguishes SAP from 3P. Geometry-independent: [participant] is the same in both geometries.
Equations
Instances For
Full probe in standard geometry: [uπ, uParticipant, uSpeaker].
Equations
Instances For
Full probe in addressee geometry: [uπ, uParticipant, uAddressee].
Equations
Instances For
A language's agreement system: the geometry and characteristic probe.
@cite{bejar-rezac-2009} parameterize crosslinguistic variation by two choices: (1) which geometry organizes the innermost feature, and (2) how articulated the probe is. The full probe depends on the geometry (standard uses [speaker], addressee uses [addressee]).
- geometry : Geometry
- probe : ProbeArticulation
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Swahili/Abkhaz: flat probe, no PH sensitivity.
Equations
- Minimalism.CyclicAgree.swahili = { geometry := Minimalism.CyclicAgree.Geometry.standard, probe := Minimalism.CyclicAgree.flatProbe }
Instances For
Basque/Georgian: partial probe, standard geometry.
Equations
- Minimalism.CyclicAgree.basque = { geometry := Minimalism.CyclicAgree.Geometry.standard, probe := Minimalism.CyclicAgree.partialProbe }
Instances For
Nishnaabemwin/Mohawk: full probe, addressee geometry.
Equations
Instances For
Active residue: unmatched probe segments after Agree with a goal DP.
Each probe segment that has a corresponding segment in the goal's person specification is deactivated (matched). Segments without a match remain active and can participate in further Agree on the next cycle.
This is the core operation of @cite{bejar-rezac-2009}: partial matching of articulated probes drives agreement displacement.
Equations
- Minimalism.CyclicAgree.activeResidue probe goal = List.filter (fun (s : Minimalism.CyclicAgree.Segment) => !goal.contains s) probe
Instances For
Equations
- Minimalism.CyclicAgree.instBEqController.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Does the EA Agree with residue segments on cycle II?
Returns true iff the EA matches at least one segment that the
IA left unmatched.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Determine which argument controls the core agreement slot.
Cycle I: probe Agrees with IA. If residue remains, cycle II: probe Agrees with EA. EA controls iff it matches any residue segment; otherwise IA controls (either it fully checked the probe, or it left residue the EA couldn't match).
Equations
- One or more equations did not get rendered due to their size.
Instances For
The person value that the core agreement slot realizes.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Convenience: controller using an AgreementSystem.
Equations
- sys.controller ea ia = Minimalism.CyclicAgree.agreementController sys.geometry sys.probe ea ia
Instances For
Convenience: agreement value using an AgreementSystem.
Equations
- sys.value ea ia = Minimalism.CyclicAgree.agreementValue sys.geometry sys.probe ea ia
Instances For
Which cycle valued the probe's remaining segments.
@cite{bejar-rezac-2009} §3.2: when the probe is valued on two different cycles, the morphological realization can differ. Georgian uses m- (first cycle) vs v- (second cycle) for 1sg agreement; Nishnaabemwin uses -in (1P, cycle II) vs -igw (3P, cycle II) vs -aa (default/cycle I only).
Returns (cycleI, cycleII) — the segments deactivated on each cycle.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Was the probe valued on two distinct cycles?
True when both the IA (cycle I) and EA (cycle II) matched at least one segment. This is the configuration that creates second-cycle morphological effects in languages like Georgian and Nishnaabemwin.
Equations
- Minimalism.CyclicAgree.hasSecondCycleEffect geom probe ea ia = match Minimalism.CyclicAgree.cycleSegments geom probe ea ia with | (c1, c2) => !List.isEmpty c1 && !List.isEmpty c2
Instances For
Inverse context: the core π-probe on v does NOT Agree with the EA.
This occurs when either (a) the IA fully checks the probe, leaving no residue, or (b) residue exists but the EA cannot match any of it. Inverse contexts trigger PLC violations on the EA's π-features, requiring repair strategies (added probe or R-Case).
Equations
- Minimalism.CyclicAgree.isInverseContext geom probe ea ia = !Minimalism.CyclicAgree.eaAgrees geom probe ea ia
Instances For
Direct context: the EA Agrees with at least one residue segment.
Equations
- Minimalism.CyclicAgree.isDirectContext geom probe ea ia = Minimalism.CyclicAgree.eaAgrees geom probe ea ia
Instances For
Convenience: inverse using an AgreementSystem.
Equations
- sys.isInverse ea ia = Minimalism.CyclicAgree.isInverseContext sys.geometry sys.probe ea ia
Instances For
The Person-Licensing Condition (PLC).
@cite{bejar-rezac-2009} eq. (13): "A π-feature [F] must be licensed by Agree of some segment in a feature structure of which [F] is a subset."
In inverse contexts, the EA's π-features are not licensed by the core probe (because the probe either has no residue or residue the EA can't match). This drives repair strategies.
Returns true if the EA is person-licensed (its π-features were
checked by the core probe on cycle II).
Equations
- Minimalism.CyclicAgree.eaIsLicensed geom probe ea ia = Minimalism.CyclicAgree.eaAgrees geom probe ea ia
Instances For
PLC violation: EA is NOT person-licensed. Exactly characterizes inverse contexts — this is the paper's key insight connecting syntactic derivation to morphological repair.
Repair strategies for PLC violations in inverse contexts.
@cite{bejar-rezac-2009} §4 identifies two strategies:
addedProbe: an extra probe is inserted on v_II, creating an additional agreement slot for the EA (Nishnaabemwin, Mohawk, Basque)rCase: the IA receives a special oblique Case distinct from the regular v-assigned Case (Kashmiri)
- addedProbe : RepairStrategy
- rCase : RepairStrategy
Instances For
Equations
- Minimalism.CyclicAgree.instBEqRepairStrategy.beq x✝ y✝ = (x✝.ctorIdx == y✝.ctorIdx)
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Does a given EA→IA combination require repair under this system?
Repair is needed iff the context is inverse.
Equations
- Minimalism.CyclicAgree.needsRepair sys ea ia = sys.isInverse ea ia
Instances For
In the standard geometry, a person value has [participant] as a segment
iff DecomposedPerson.hasParticipant is true.
In the standard geometry, 3rd person has exactly one segment ([π]).
Segment count reflects the person hierarchy: 1st (3 segments) > 2nd (2) > 3rd (1) in standard geometry.
Entailment: more specified persons' segments are supersets.
Flat probe: IA always fully checks the probe, regardless of person.
Flat probe: IA always controls — no PH effects.
Flat probe: all contexts are inverse (no EA agreement).
Partial probe: same behavior in both geometries, since [participant] is geometry-independent.
Basque: 3→1 = 1 (IA controls, inverse).
Basque: 2→1 = 1 (IA controls, inverse).
Basque: 1→2 = 2 (IA controls, inverse).
Basque: 3→2 = 2 (IA controls, inverse).
Basque: 1→3 = 1 (EA controls, direct).
Basque: 2→3 = 2 (EA controls, direct).
Basque: 3→3 = 3 (IA controls, inverse — residue [uParticipant] unmatched by either argument).
Basque: SAP IA → inverse context (agreement displacement to IA).
Basque: 3P IA + SAP EA → direct context (EA controls).
Basque: 3P IA + 3P EA → inverse (neither fully checks).
Nishnaabemwin: 3→2 = 2 (IA controls, inverse — IA fully checks).
Nishnaabemwin: 1→2 = 2 (IA controls, inverse — IA fully checks).
Nishnaabemwin: 3→1 = 1 (IA controls, inverse — EA can't match residue).
Nishnaabemwin: 2→1 = 2 (EA controls, direct — EA matches [uAddressee]).
Nishnaabemwin: 1→3 = 1 (EA controls, direct).
Nishnaabemwin: 2→3 = 2 (EA controls, direct).
Nishnaabemwin: 3→3 = 3 (IA controls, inverse — residue [uParticipant, uAddressee] unmatched by 3P EA).
Nishnaabemwin: 2P IA → always inverse (2nd is most specified).
Nishnaabemwin: 3P IA with SAP EA → direct.
Nishnaabemwin: 3→1 is inverse despite residue, because EA (3P) can't match [addressee].
Nishnaabemwin: 3→3 is inverse.
Georgian second-cycle effect: 1sg agreement is m- when valued on cycle I (IA=1, any EA), but v- when valued on cycle II (EA=1, IA=3).
Georgian: when IA is SAP, no second cycle (IA fully checks).
Nishnaabemwin: 2→1 has a second-cycle effect (IA checks [π,participant], EA checks [addressee] on cycle II).
Nishnaabemwin: the second cycle in 2→1 values exactly [addressee].
When EA and IA have the same person, IA always controls (the EA contributes nothing new — every segment it could match was already matched by the identical IA).
The proof proceeds by showing that activeResidue is idempotent
when applied with the same goal: if some segments survive matching
against personSpec(p), applying the same filter again removes nothing,
so eaAgrees returns false.
The most specified person always controls against 3P (standard).
The most specified person always controls against 3P (addressee).
The direct/inverse split exhaustively partitions the paradigm: every EA→IA combination is either direct or inverse, never both.
In a partial-probe language like Basque/Georgian, object (P) agreement is person-conditioned: the object controls agreement iff the context is direct (EA > IA on the person hierarchy). This predicts that SAP objects trigger agreement displacement to EA, while 3P objects don't.
Formally: for a fixed 3P EA (the "subject" in a canonical transitive), a SAP IA triggers direct context (EA controls → "no P indexing"), but this is backwards from the traditional description. Let's think about it from the P indexing perspective:
In Basque, the agreement slot tracks the controller. When IA is SAP, IA controls (inverse) — the agreement tracks IA, reflecting the object. When IA is 3P, EA controls (direct) — agreement tracks EA, not the object.
So pIsIndexed ↔ the IA controls (the agreement slot shows the
object's features) ↔ inverse context.
3P IA yields direct context when EA is SAP — the agreement slot tracks the EA, not the (3P) object.
Differential P indexing prediction: the SAP/3P split in object
agreement (Basque pIsIndexed, Georgian pIsIndexed) is exactly
the inverse/direct split of the partial probe.
An object (IA) is "indexed" when the core agreement slot tracks the IA's person features, which happens iff the context is inverse (IA controls). SAP IAs always produce inverse contexts; 3P IAs produce direct contexts (when EA is SAP).
This theorem proves the key direction: SAP IA → inverse (indexed).
Count inverse contexts in a 3×3 paradigm.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Swahili (flat): all 9 cells are inverse (no PH sensitivity).
Basque (partial): 7 inverse, 2 direct (only SAP EA + 3P IA).
Nishnaabemwin (full): 6 inverse, 3 direct.
More articulated probes yield more direct contexts.