Shieber (1985) @cite{shieber-1985} #
Evidence against the Context-Freeness of Natural Language. Linguistics and Philosophy, 8(3), 333–343.
Core Argument #
@cite{shieber-1985} proves that Swiss German is not weakly context-free, using a purely string-based argument that makes no assumptions about constituent structure or semantics. The proof rests on four empirical claims about Swiss German subordinate clauses, plus the closure of context-free languages under homomorphism and intersection with regular languages.
The Four Claims #
- Swiss German subordinate clauses have structures where all Vs follow all NPs.
- Among such sentences, those with all DAT-NPs before all ACC-NPs, and all DAT-Vs before all ACC-Vs, are grammatical.
- The number of DAT-Vs must equal the number of DAT-NPs (and similarly for ACC).
- An arbitrary number of Vs can occur (subject to performance).
The Proof #
Define a homomorphism f mapping Swiss German words to an abstract alphabet:
- DAT-NPs (e.g., em Hans) →
a - ACC-NPs (e.g., d'chind, de Hans) →
b - DAT-Vs (e.g., hälfe) →
c - ACC-Vs (e.g., lönd, aastriiche) →
d - Framing material →
w,x,y
Intersect f(L) with the regular language r = w a* b* x c* d* y.
By Claims 1–4, f(L) ∩ r = {w aᵐ bⁿ x cᵐ dⁿ y}.
A further homomorphism removing w, x, y yields {aᵐ bⁿ cᵐ dⁿ},
which contains {aⁿ bⁿ cⁿ dⁿ} (setting m = n). Since {aⁿ bⁿ cⁿ dⁿ}
is not context-free (anbncndn_not_pumpable), and CFLs are closed under
homomorphism and intersection with regular languages, Swiss German is not
context-free.
Contrast with @cite{bresnan-etal-1982} #
@cite{bresnan-etal-1982}'s earlier argument for Dutch non-context-freeness relied on linguistic assumptions about constituent structure, which @cite{gazdar-pullum-1982} contested. @cite{shieber-1985}'s argument is purely formal — it rests entirely on the string set of Swiss German and the case-marking facts, making no claims about phrase structure.
A Swiss German subordinate clause token, abstracting over specific lexical items to their role in the cross-serial construction.
@cite{shieber-1985}'s proof only needs to distinguish NPs and Vs by case.
- datNP : Token
Dative NP (e.g., em Hans)
- accNP : Token
Accusative NP (e.g., d'chind, de Hans)
- datV : Token
Dative-subcategorizing verb (e.g., hälfe "help")
- accV : Token
Accusative-subcategorizing verb (e.g., lönd "let", aastriiche "paint")
Instances For
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
The case that a token bears or requires.
Equations
- Phenomena.WordOrder.Studies.Shieber1985.Token.datNP.caseValue = Core.Case.dat
- Phenomena.WordOrder.Studies.Shieber1985.Token.accNP.caseValue = Core.Case.acc
- Phenomena.WordOrder.Studies.Shieber1985.Token.datV.caseValue = Core.Case.dat
- Phenomena.WordOrder.Studies.Shieber1985.Token.accV.caseValue = Core.Case.acc
Instances For
Whether a token is an NP (vs a verb).
Equations
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Equations
- One or more equations did not get rendered due to their size.
- Phenomena.WordOrder.Studies.Shieber1985.instBEqCrossSerialClause.beq x✝¹ x✝ = false
Instances For
Equations
- One or more equations did not get rendered due to their size.
Instances For
Claim 3: case matching — the number of dative verbs equals the number of dative NPs, and similarly for accusative.
Instances For
A grammatical cross-serial clause satisfies case matching.
- matching : self.caseMatches
Instances For
Claim 4: any combination of dative and accusative verb counts can occur
(we can produce a GrammaticalClause for any m, n).
Equations
- Phenomena.WordOrder.Studies.Shieber1985.arbitraryDepth m n = { datNPs := m, accNPs := n, datVs := m, accVs := n, matching := ⋯ }
Instances For
Shieber's homomorphism f: maps Swiss German cross-serial clause tokens to the abstract alphabet {a, b, c, d}.
- DAT-NPs →
a - ACC-NPs →
b - DAT-Vs →
c - ACC-Vs →
d
The framing material (Jan säit das mer, es huus haend wele, etc.) maps to fixed delimiter symbols that are removed by regular intersection.
Equations
- Phenomena.WordOrder.Studies.Shieber1985.tokenToSymbol Phenomena.WordOrder.Studies.Shieber1985.Token.datNP = FourSymbol.a
- Phenomena.WordOrder.Studies.Shieber1985.tokenToSymbol Phenomena.WordOrder.Studies.Shieber1985.Token.accNP = FourSymbol.b
- Phenomena.WordOrder.Studies.Shieber1985.tokenToSymbol Phenomena.WordOrder.Studies.Shieber1985.Token.datV = FourSymbol.c
- Phenomena.WordOrder.Studies.Shieber1985.tokenToSymbol Phenomena.WordOrder.Studies.Shieber1985.Token.accV = FourSymbol.d
Instances For
The string image of a grammatical clause under the homomorphism.
Equations
- One or more equations did not get rendered due to their size.
Instances For
A grammatical clause with m DAT-pairs and n ACC-pairs maps to
aᵐ bⁿ cᵐ dⁿ.
Setting m = n in the clause image gives aⁿbⁿcⁿdⁿ.
The diagonal clause images are in {aⁿbⁿcⁿdⁿ}.
CFL closure (contrapositive). If a language L can be mapped by a homomorphism f and intersected with a regular language r to produce a non-context-free language, then L is not context-free.
This is the contrapositive of the standard closure theorem for CFLs (Hopcroft & Ullman 1979, pp. 130–135): CFLs are closed under homomorphism and under intersection with regular languages.
We state this as a proposition rather than proving it from first principles, since linglib does not formalize the full theory of CFLs. The pumping lemma proof of the specific non-CFL witness ({aⁿbⁿcⁿdⁿ}) IS fully verified.
Equations
- One or more equations did not get rendered due to their size.
Instances For
Main result. The image of Swiss German cross-serial clauses under @cite{shieber-1985}'s homomorphism contains {aⁿbⁿcⁿdⁿ}, which is not context-free. Combined with CFL closure properties, this proves Swiss German is not context-free.
The conjunction packages the two independently verified facts:
- The homomorphism maps Swiss German data to {aⁿbⁿcⁿdⁿ} (by construction)
- {aⁿbⁿcⁿdⁿ} violates the CFL pumping property (by proof)
Cross-serial dependencies with case-marking require at least mildly
context-sensitive power — the same classification used by
Phenomena.WordOrder.CrossSerial.
Corollary: Swiss German is not strongly context-free either.
@cite{shieber-1985} §3: "As a trivial corollary, Swiss German is not strongly context-free either, regardless of one's view as to the appropriate structures for the language." Since strong context-freeness implies weak context-freeness, weak non-context-freeness implies strong non-context-freeness.
hälfe is a dative verb — its case requirement matches the DAT token.
lönd is an accusative verb — its case requirement matches the ACC token.
aastriiche is an accusative verb.
The formal–processing dissociation: crossed dependencies are formally harder (not CF) but psycholinguistically easier.
@cite{shieber-1985} establishes the formal side; the processing side is
in Phenomena.WordOrder.Studies.BachBrownMarslenWilson1986.
Example (1): mer em Hans es huus hälfed aastriiche "we helped Hans paint the house"
em Hans (DAT) → hälfed (DAT-verb "helped") es huus (ACC) → aastriiche (ACC-verb "paint")
Equations
Instances For
Example (5): triply embedded cross-serial clause mer d'chind em Hans es huus lönd hälfe aastriiche "we let the children help Hans paint the house"
d'chind (ACC) → lönd (ACC-verb "let") em Hans (DAT) → hälfe (DAT-verb "help") es huus (ACC) → aastriiche (ACC-verb "paint")
With case sorting: 1 DAT-NP, 2 ACC-NPs, 1 DAT-V, 2 ACC-Vs
Equations
Instances For
The homomorphic image of example (5) is abcdd = a¹b²c¹d².