Documentation

Linglib.Core.Computability.Subregular.Function.WeaklyDeterministic

Weakly Deterministic (WD) Functions #

The Weakly Deterministic function class sits between Subsequential and the full Non-Deterministic regular relations in the function-level subregular hierarchy (Function/{ISL,OSL,Subsequential}.lean; @cite{aksenova-rawski-graf-heinz-2020} Fig. 1; @cite{meinhardt-mai-bakovic-mccollum-2024} Fig. 1):

Non-Deterministic (regular relations)
       ↓
Weakly Deterministic ← THIS FILE
       ↓
Subsequential (Left and Right)
       ↓
OSL (Left and Right)
       ↓
ISL
       ↓
Finite

WD captures bidirectional iterative phonological maps — a left-to-right pass plus a right-to-left pass — and was introduced by @cite{heinz-lai-2013} to delimit attested patterns (Maasai/Turkana ATR harmony) from unattested ones (sour-grapes-style spreading; @cite{wilson-2003}, @cite{wilson-2006}).

Substrate scope: structural decomposition only #

This file defines IsWeaklyDeterministic as the structural backbone shared by every formalisation of the class: a function decomposes as the composition of two subsequential functions reading from opposite directions, without alphabet enrichment (the intermediate alphabet equals the input/output alphabet).

The "no alphabet enrichment" condition (β = α in the type signature of the inner subsequential) is the @cite{heinz-lai-2013} no-markup framing made structural. Under this signature, no separate non- interaction predicate is needed for the basic class membership: any decomposition into two contradirectional subseq passes that share the input/output alphabet IS WD by the no-markup criterion.

Open conjecture: HL2013 vs MMBM2024 #

@cite{meinhardt-mai-bakovic-mccollum-2024} argue that the structural "no-markup" criterion above is too permissive: some patterns satisfy it but still require unbounded bilateral information at output time (paper §§4.3, 5). Their patched definition (Def. 6, p. 1219) strengthens the criterion via a bimachine output-function condition: the bimachine implementing f must have an output function ω such that for all q^L, x, y, q^R, ω(q^L, x, q^R) = y iff y is licensed by q^L alone OR by q^R alone (never simultaneously).

The open conjecture: under the strengthened MMBM2024 definition, some functions in IsWeaklyDeterministic (per this file's structural definition, equivalent to the HL2013 framing) would NOT count as WD — the MMBM2024 patch refuses them. The paper's Maasai analysis is WD-by-both-definitions; the unattested counterexamples that distinguish them are sour-grapes-style functions.

Formalising this distinction in Lean requires bimachine substrate (Schützenberger's sequential bimachines; see @cite{mohri-1997} §6 for the canonical reference), which is non-trivial new substrate. Land in a follow-up Function/Bimachine.lean once a Studies file needs the distinction; the current file's structural IsWeaklyDeterministic suffices for both:

Out of scope #

Weakly Deterministic function class. A function f : List α → List α is WD iff it decomposes as the composition of two subsequential functions reading from opposite directions, with no alphabet enrichment (intermediate alphabet equals input/output alphabet — the structural form of @cite{heinz-lai-2013}'s no-markup condition).

Per the file docstring, this is the structural backbone shared by both HL2013 and MMBM2024 formalisations. The MMBM2024 patch adds a bimachine output-function condition that further refines the class; that refinement is deferred until bimachine substrate exists.

Equations
  • One or more equations did not get rendered due to their size.
Instances For

    Every Left-Subsequential function is WD: take the inner function as itself and the outer function as the identity (Right-Subsequential trivially via the identity FST). The composition id ∘ f = f.

    Every Right-Subsequential function is WD: dual of the left case. Take the inner function as itself and the outer as the identity.