Middleton (2026) — Ordering of Impoverishment Rules in Taos and Basque #
@cite{middleton-2026} @cite{arregi-nevins-2012} @cite{halle-marantz-1993} @cite{harbour-2014} @cite{harbour-2016} @cite{kontak-kunkel-1987} @cite{watkins-1984} @cite{harbour-middleton-2026}
This file formalises the architectural argument of @cite{middleton-2026}. Working within Distributed Morphology (@cite{halle-marantz-1993}), @cite{arregi-nevins-2012} propose a strict modular postsyntax in which paradigmatic Impoverishment applies as a block before syntagmatic Impoverishment, and Metathesis follows all Impoverishment. Middleton shows from Taos verbal agreement that the second claim survives but the first does not (§§4.2.1–§4.2.5); the Basque half of the paper (§3.1) re-establishes the second claim — metathesis after impoverishment — using a different language and different rule shapes (whole-terminal deletion + adjacent-terminal swap).
Scope #
The full Taos paradigm involves dozens of Vocabulary Insertion rules and a large family of impoverishment / metathesis rules. We do not re-derive the entire paradigm. What lives here:
- The architectural pipelines (
runStrict,runInterleaved) live inTheories/Morphology/DM/PostsyntacticDerivation.lean. - This file gives one schematic pair of rules
paraAtomicRule/synMinimalRulethat exhibits the divergence predicted by the paper at a real-shaped Taos witness neighborhood. The conditioning features ([+author],[+atomic],[+minimal]) are drawn from the Harbour decomposition the paper uses, but the rules themselves are not literal transcriptions of paper rules — they are minimal witnesses to the para-vs-syn ordering interaction in @cite{middleton-2026} §4.2.1–§4.2.4. - The general claim — that
runStrictis strictly less expressive thanrunInterleavedwhenever a syntagmatic rule needs to feed a paradigmatic one — isrunStrict_forces_paraSyn_order/runInterleaved_admits_synParain the theory file; here we instantiate it on the witness. - §6 demonstrates how the postsyntactic output feeds Vocabulary Insertion (Subset Principle, @cite{halle-marantz-1993}), again schematically.
- §7 turns to the Basque half of the paper. Where §§1–5 work at the
focus level (rules modify a single bundle), Basque needs whole
terminal deletion (Participant Dissimilation, rule 16) and
adjacent terminal swap (Ergative Metathesis, rule 13). The
framework lift lives in
Theories/Morphology/DM/LinearPostsyntax.lean; the bundles live inFragments/Basque/Postsyntax.lean; the rules and the Ondarrus-endu-n(17a) divergence witness are stated here.
What is not modeled:
- The full Taos paradigm and the literal rule statements of @cite{middleton-2026} (rule numbers and conditioning environments vary across the four §4.2 cases).
- Harbour's Reciprocal Containment constraints on feature bundles.
- Real Taos VI competition — only enough VIs to demonstrate the pipeline.
A paradigmatic rule: deletes [+atomic] whenever the focus
contains both [+author] and [+minimal]. The condition refers
only to the focus, so the rule is paradigmatic by construction
(paradigmatic_isParadigmatic).
This is a minimal stand-in for the paradigmatic rules involved in @cite{middleton-2026} §4.2.1–§4.2.4 — it is not a transcription of any specific paper rule.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A syntagmatic rule: deletes [+minimal] when the focus
contains [+atomic] and there is at least one bundle of
object-context to the right (the schematic [O 3i] condition,
weakened to bare presence — sufficient for the bleeding/feeding
interaction the paper diagnoses). The dependence on rightCtx
is what makes the rule syntagmatic, and synMinimalRule_isSyntagmatic
proves it.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The two rules are genuinely in distinct phases: synMinimalRule
actually depends on its right-context (it is not paradigmatic).
Witness: two neighborhoods that share a focus but differ on
rightCtx.
Witness focus: a 1s-style bundle [+author, +atomic, +minimal]
(suppressing [+participant], which is irrelevant to either rule).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Witness neighborhood: the 1s-style focus, with a real 3s
bundle to the right standing in for the Taos object slot
that conditions synMinimalRule.
Equations
Instances For
Run para-then-syn (the order A&N's strict pipeline forces).
Equations
Instances For
Run syn-then-para (the order Middleton's interleaved pipeline can choose).
Equations
Instances For
Para-then-syn (= A&N): paraAtomicRule deletes [+atomic] first;
synMinimalRule then can't fire (no [+atomic] left in focus).
The [+minimal] survives.
Syn-then-para (= Middleton): synMinimalRule fires first (focus
has [+atomic], rightCtx non-empty), deleting [+minimal];
paraAtomicRule then can't fire (no [+minimal] left in focus).
The [+atomic] survives instead.
The two orders produce different feature bundles at this neighborhood.
The schematic A&N postsyntax that contains exactly paraAtomicRule
in the paradigmatic phase and synMinimalRule in the syntagmatic
phase, with no metathesis.
Equations
- Middleton2026.arregiNevinsPostsyntax = { paradigmatic := [Middleton2026.paraAtomicRule], syntagmatic := [Middleton2026.synMinimalRule], metathesis := [] }
Instances For
The schematic Middleton interleaved postsyntax, with the syntagmatic rule scheduled first — the order required by the §4.2.1–§4.2.4 Taos cases.
Equations
- Middleton2026.middletonPostsyntax = { impoverishment := [Middleton2026.synMinimalRule, Middleton2026.paraAtomicRule], metathesis := [] }
Instances For
A&N's pipeline computes the para-first answer at the witness.
Middleton's pipeline computes the (different) syn-first answer.
Architectural inadequacy of runStrict for Taos. At the
witness neighborhood, the strict A&N pipeline and Middleton's
interleaved one return different feature bundles. Hence no
ModularPostsyntax built from paraAtomicRule (paradigmatic) and
synMinimalRule (syntagmatic) — and no extension that adds
further rules to the same phases — can yield the syn-first output
that Taos requires in @cite{middleton-2026} §4.2.1–§4.2.4.
A metathesis rule that swaps [+author] with [+atomic] when the
focus contains all three of [+author], [+atomic], [+minimal].
Schematic of @cite{middleton-2026}'s metathesis rules: a metathesis
triggered in the presence of a particular number feature. The
dependence on [+minimal] is what couples this rule to
synMinimalRule (which deletes [+minimal]), so that the IM/MI
orders diverge — the empirically motivated witness of "metathesis
after impoverishment, not before."
Equations
- One or more equations did not get rendered due to their size.
Impoverishment-then-metathesis (Middleton's and A&N's shared order).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Metathesis-then-impoverishment (the order both authors reject).
Equations
- One or more equations did not get rendered due to their size.
Instances For
The two orders of impoverishment vs. metathesis genuinely diverge at the witness, witnessing the architectural fact that metathesis must follow impoverishment.
The post-postsyntactic focus bundle from A&N's strict pipeline at the witness — extracted as a top-level def so it is the input to Vocabulary Insertion below.
Equations
Instances For
The post-postsyntactic focus bundle from Middleton's interleaved pipeline at the witness.
Equations
Instances For
A small schematic Vocabulary Item set keyed on FeatureVal. The
Subset Principle (@cite{halle-marantz-1993}) selects the longest
matching entry. We use Morpheme.surface for the exponents to
keep the connection to the Taos morpheme inventory in the
Fragment.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A&N's output feeds VI as [+author, +minimal]; the Subset
Principle selects the [+author, +minimal] entry over the
elsewhere ô. Surface form: n.
Middleton's output feeds VI as [+author, +atomic]; the Subset
Principle selects the [+author, +atomic] entry. Surface form:
o. The two architectures predict different surface forms at
the same input — the empirical bite of the architectural
divergence.
The architectural divergence shows up at the level of surface exponents, not just feature bundles: the same input neighborhood yields different morphemes under the two pipelines.
Participant Dissimilation (@cite{middleton-2026} (16),
@cite{arregi-nevins-2012} §4.6). Delete a 1p absolutive clitic
([CL +participant +author]) when there is a participant
ergative clitic somewhere to the right in the same auxiliary.
The rule operates at the terminal level — it deletes a whole
bundle, not a feature within one.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Ergative Metathesis (@cite{middleton-2026} (13),
@cite{arregi-nevins-2012} §3.2). Swap T with an immediately
following ergative clitic when T is leftmost in the auxiliary.
The leftmost requirement (left.isEmpty) is what lets
Participant Dissimilation feed Ergative Metathesis: only
after PD deletes the absolutive clitic does T become leftmost.
Equations
- One or more equations did not get rendered due to their size.
Instances For
The Ondarru witness phrase from @cite{middleton-2026} (17a):
s-endu-n [1pABS, T:past, 2sERG]. The complementizer is
suppressed — it does not participate in either rule.
Equations
Instances For
PD-then-Meta surface form (the grammatical s-endu-n order).
PD deletes the absolutive 1p, leaving [T, ERG]; with T now
leftmost, Ergative Metathesis swaps to [ERG, T] — the order
that surfaces as s-endu-n.
*Meta-then-PD surface form (the rejected 17b order).
Ergative Metathesis cannot fire at the input — T is not
leftmost (the absolutive clitic precedes it). PD then deletes
the absolutive clitic, but it is too late to feed metathesis;
the result is the T-leftmost order [T, ERG], the form
@cite{middleton-2026} marks ungrammatical (would require
L-Support repair *d-endu-s-n).
The two phrase-level pipelines diverge on the Ondarru
witness. This is the Basque counterpart to
arregiNevins_neq_middleton_at_witness (Taos §4); together they
are the two empirical legs of @cite{middleton-2026}'s claim that
metathesis must follow impoverishment.