Keenan & Comrie (1977) @cite{keenan-comrie-1977} #
Noun Phrase Accessibility and Universal Grammar. Linguistic Inquiry 8(1): 63–99.
Formalizes the three Hierarchy Constraints (HCs) and the derived Primary Relativization Constraint (PRC) from @cite{keenan-comrie-1977}, verified against a subset of the paper's Table 1 data (pp. 76-79).
Architecture #
This file derives K&C's typological theorems directly from
Fragments.{Lang}.relMarkers : List RelClauseMarker, the per-language
data layer encoding actual linguistic markers (particles, pronouns,
verbal suffixes). No intermediate KCProfile/StrategyEntry schema —
predicates and aggregations are stated over List RelClauseMarker
directly, projecting through RelClauseMarker.{positions, bearsCaseMarking, rcPosition} as needed.
The Fragment files cite @cite{keenan-comrie-1979} (the per-language exemplification appendix originally intended for publication with K&C 1977 — Language 55(2): 333–351) inline where its sentence-level examples back the descriptive marker data.
Hierarchy Constraints #
The paper proposes three constraints on how languages form relative clauses, building on the Accessibility Hierarchy (AH):
SU > DO > IO > OBL > GEN > OCOMP
- HC₁ (p. 67): A language must be able to relativize subjects.
- HC₂ (Continuity) (p. 67): Any RC-forming strategy must apply to a continuous segment of the AH.
- HC₃ (Cut-off) (p. 67): Strategies that apply at one point may cease at any lower point.
From HC₁ + HC₂, the Primary Relativization Constraint (p. 68) follows: if a language's primary strategy (one that covers subjects) can apply to a low position N, it can apply to all positions above N. Non-primary strategies need not satisfy this — they may cover a continuous segment that excludes subjects (e.g., the +case strategy covering IO–OCOMP but not SU–DO in Welsh and Arabic, p. 70 + Table 1 p. 76).
Multi-Strategy Profiles #
The paper's key empirical contribution is showing that languages typically have multiple relativization strategies, each covering a different contiguous segment of the AH. The ±case distinction (whether the relative element bears case marking) is the primary parameter distinguishing strategies.
Sample #
Nine languages cover the key patterns: gap-to-resumptive split (Welsh, Hebrew, Arabic, Toba Batak), multi-strategy with prenominal RCs (Korean, Finnish), single-strategy (Malagasy), and per-position strategy split with serial-verb-mediated obliques (Yoruba).
HC₁: a language can relativize subjects iff some marker covers SU.
Equations
- Phenomena.Relativization.Studies.KeenanComrie1977.SatisfiesHC1 markers = ∃ (m : Core.RelClauseMarker), m ∈ markers ∧ m.Covers Core.AHPosition.subject
Instances For
Equations
- Phenomena.Relativization.Studies.KeenanComrie1977.instDecidableSatisfiesHC1 markers = id inferInstance
HC₂: every marker covers a contiguous segment of the AH.
Equations
- Phenomena.Relativization.Studies.KeenanComrie1977.SatisfiesHC2 markers = ∀ (m : Core.RelClauseMarker), m ∈ markers → m.IsContinuous
Instances For
Equations
- Phenomena.Relativization.Studies.KeenanComrie1977.instDecidableSatisfiesHC2 markers = id inferInstance
PRC: every primary marker is upward-closed on the AH. If marker m
is primary and covers position pos, then m covers every position
above pos. This is the paper's Primary Relativization Constraint
(p. 68), which follows from HC₂ for primary strategies (see
prc_from_hc2 below for the general derivation).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- Phenomena.Relativization.Studies.KeenanComrie1977.instDecidableSatisfiesPRC markers = id inferInstance
The lowest AH position covered by any marker in the list (i.e., the
deepest the language can reach). Returns .subject if even SU is
uncovered (vacuously, since HC₁ would be violated).
Equations
- One or more equations did not get rendered due to their size.
Instances For
Per-language abbrevs over Fragment marker lists. The original 8-language sample from the paper plus Yoruba (added later via @cite{awobuluyi-1978} + @cite{keenan-comrie-1979}).
Instances For
Instances For
Equations
Instances For
Instances For
Equations
Instances For
Instances For
Instances For
Instances For
Instances For
The 8-language sub-sample from the original paper Table 1 (pp. 76-79).
Equations
- One or more equations did not get rendered due to their size.
Instances For
The original 8-language sample plus Yoruba (the only post-1977
addition; refutes one of the paper's implicit ±case generalizations
— see yoruba_refutes_minus_case_covers_subjects below).
Equations
Instances For
HC₁ holds: every language in the sample can relativize subjects.
HC₂ holds: every marker in every sampled language covers a contiguous AH segment.
PRC holds: every primary marker satisfies upward closure on the AH.
Restating HC₁ in terms of RelClauseMarker.IsPrimary: every language
has at least one primary marker.
In the original 8-language sub-sample, every -case marker covers subjects. The -case (gap/deletion) strategy is always primary when present in those languages.
REFUTED by Yoruba — see yoruba_refutes_minus_case_covers_subjects
below: Yoruba has gap markers for DO and OBL that do not cover SU
because subject relativization independently uses pronoun retention
(ó, per @cite{awobuluyi-1978} §6.19).
@cite{keenan-comrie-1979} effectively documents Yoruba as a refutation
of the gap-implies-subject correlation. Yoruba's IO/OBL relativization
is mediated by serial-verb DO recasting (K&C 1979 p. 349), producing
-case markers that do not cover SU. SU relativization independently
uses pronoun retention (ó, K&C 1979 p. 350 analyzes as verb
agreement; descriptive surface form per @cite{awobuluyi-1978} §6.19).
Most languages in the sample use more than one marker, with markers covering different segments.
+case markers that are non-primary (don't cover SU) never cover SU in our sample. This reflects the typological generalization that pronoun retention is used for lower, not higher, AH positions. Holds across all 9 languages including Yoruba.
Toba Batak has a genuine gap at DO: neither marker can relativize direct objects. This is consistent with the HCs because each individual marker is contiguous — the gap exists between markers, not within one. The paper notes this explicitly (p. 68-69: "direct objects cannot be relativized using this or any other strategy in Toba").
Despite the DO gap, Toba Batak satisfies HC₂: both individual markers are contiguous (SU alone; IO–GEN alone).
English (Table 1 p. 76): -case that/∅ covers SU/DO (2 positions);
+case who/whom covers IO/OBL/GEN/OCOMP (4 positions).
Welsh (Table 1 p. 76; paper §1.3.2 p. 70): markers split at DO/IO. -case (particle a) covers SU/DO; +case (particle y + resumptive) covers IO/OBL/GEN/OCOMP.
Arabic (MSA) (Table 1 p. 76): the relative pronoun alladhī/allatii used alone (-case strategy) covers SU only; alladhī/allatii with a resumptive pronoun (+case strategy) covers DO–OCOMP.
Malagasy (Table 1 p. 78; paper §1.3.1 p. 69-70): single marker, SU only.
Korean (Table 1 p. 78; paper §1.3.4 p. 74): -case adnominal verb suffix covers SU/DO/IO/OBL but not GEN; +case genitive marker covers GEN only.
Finnish (Table 1 p. 76; paper §1.3.2 p. 70-71): the +case marker joka is the broader/primary one (covers SU–GEN); the -case participial marker also covers SU but is narrower (SU/DO only).
Yoruba: 4 per-position markers. relTiSubject (-case, primary, only SU); relTiObject (-case, NOT primary, only DO); relTiOblique (-case, NOT primary, IO/OBL); relTiGenitive (+case, NOT primary, GEN only). All 4 individually contiguous on the AH, so HC₂ holds.
K&C 1977 Table 1's per-position coverage and RelativizationProfile's
WALS-derived lowestRelativizable encode complementary views of the same
data. Bridge theorems below verify agreement on the lowest position
covered, language by language. K&C's Table 1 is strictly more detailed
than WALS Ch 122/123 (which only ask about subjects and obliques), so
the K&C lowestCovered is at least as deep as the WALS
lowestRelativizable.
Systematic coverage agreement: K&C is at least as detailed as WALS for every sample language. The WALS profile never claims a language can relativize a position that K&C Table 1 doesn't cover.
HC₂ ("any RC-forming strategy must apply to a continuous segment of the
AH") is a paper-anchored claim. The contiguity machinery (contiguousOnAH,
AHPosition.rank) lives in Core/Relativization/Hierarchy.lean because it
mirrors Core/Case/Hierarchy.lean::validInventory and is genuinely
framework-agnostic. The specific contiguous-segment witnesses below
exemplify HC₂ on the AH and are part of @cite{keenan-comrie-1977}'s core
argumentation.
The full hierarchy [SU, DO, IO, OBL, GEN, OCOMP] is contiguous.
A single position is trivially contiguous.
[SU, DO] is contiguous.
[IO, OBL, GEN] is contiguous (a non-primary segment).
[SU, DO, OBL] is NOT contiguous (skips IO at rank 4).
The PRC is the paper's main derivation: it follows from HC₁ + HC₂ rather
than being an independent stipulation. The general proof lives here (paper
content), not in Core/Relativization/Hierarchy.lean (substrate).
Primary Relativization Constraint (general proof).
If a list of AH positions is contiguous (HC₂) and contains .subject
(i.e., the strategy is primary), then the list is upward-closed:
for any covered position p, all positions above p on the AH
are also covered.
This proves that the PRC is a logical consequence of HC₂ + being primary, not an independent constraint — the paper's core derivation (@cite{keenan-comrie-1977} p. 68: "PRC₂ follows directly from HC₂ and the definition of primary").
All 6 canonical primary strategy segments are upward-closed.
These are the only possible contiguous segments containing .subject.