Equiprobable mappings in weighted constraint grammars

Arto Anttila; Giorgio Magri; Scott Borgeson

arxiv: 1907.05839 · v1 · pith:CKUA2FZLnew · submitted 2019-07-12 · 💻 cs.CL · cs.LG

Equiprobable mappings in weighted constraint grammars

Arto Anttila , Scott Borgeson , Giorgio Magri This is my paper

Pith reviewed 2026-05-24 22:20 UTC · model grok-4.3

classification 💻 cs.CL cs.LG

keywords MaxEntStochastic HGequiprobable mappingsweighted constraint grammarsFinnish stressphonological variation

0 comments

The pith

MaxEnt assigns different probabilities to any two distinct mappings using some nonnegative weights, while Stochastic HG permits equiprobable mappings that can be completely characterized.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper shows that maximum entropy grammars can always separate any two different mappings by finding weights that make their probabilities unequal. Stochastic harmonic grammar instead allows cases where two mappings receive exactly the same probability, and the authors supply a full formal description of when those cases occur. The difference between the two frameworks is then examined on data from Finnish stress patterns. A reader would care because these models are used to capture variation and learning in phonology, so the presence or absence of forced equiprobability changes what each framework can represent.

Core claim

MaxEnt is rich enough that there always exists a nonnegative weight vector assigning different probabilities to any two different mappings. Stochastic HG admits equiprobable mappings and these are given a complete formal characterization.

What carries the argument

Weighted constraint systems with nonnegative weights, where each mapping is identified by its constraint violation profile; the existence result for MaxEnt and the characterization theorem for Stochastic HG.

If this is right

MaxEnt can model any pattern of unequal probabilities across mappings.
Stochastic HG restricts the set of possible probability ties in a fully describable way.
The two frameworks make different predictions about which mappings can share probability in variation data.
The Finnish stress test case illustrates concrete differences in the probabilities each framework can produce.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

If observed data never show equiprobable mappings, that would favor MaxEnt over Stochastic HG in practice.
The characterization could be used to test Stochastic HG against other languages that exhibit tied probabilities.
The richness result for MaxEnt raises questions about whether additional restrictions on weights are needed to match linguistic data.

Load-bearing premise

The two frameworks are correctly captured by weighted constraint systems with nonnegative weights and mappings are defined solely by their violation profiles.

What would settle it

A pair of distinct mappings for which no nonnegative weight vector produces unequal MaxEnt probabilities, or a pair that falls outside the stated characterization of equiprobable mappings in Stochastic HG.

read the original abstract

We show that MaxEnt is so rich that it can distinguish between any two different mappings: there always exists a nonnegative weight vector which assigns them different MaxEnt probabilities. Stochastic HG instead does admit equiprobable mappings and we give a complete formal characterization of them. We compare these different predictions of the two frameworks on a test case of Finnish stress.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Desk Editor's Note private letter to a colleague

The paper cleanly shows MaxEnt separates any two mappings via nonnegative weights while Stochastic HG has a nonempty class of equiprobable ones with an explicit characterization.

read the letter

The core result is that MaxEnt always finds weights making two distinct mappings have different probabilities, while Stochastic HG does not, and the paper gives the full characterization of the equiprobable pairs in the latter. This follows directly once mappings are identified with violation vectors and weights stay nonnegative, as the stress-test note confirms with the linear separation argument for MaxEnt and the explicit description for Stochastic HG. The Finnish stress example then shows the frameworks make different predictions on real data. The formal part is new relative to the cited literature and stands on its own without circularity or hidden parameters. The empirical test is secondary but useful for illustration. Minor soft spot is that the strength of the Finnish contrast depends on the specific candidate sets and constraint definitions chosen, though that does not affect the main theorems. This work is for phonologists and computational linguists already using weighted constraint models. It is technically grounded enough to merit peer review even if the empirical section needs tightening.

Referee Report

0 major / 2 minor

Summary. The paper claims that MaxEnt grammars with nonnegative weights can distinguish any two distinct mappings (identified with their violation profiles) by assigning them different probabilities, since for any v1 ≠ v2 there exists w ≥ 0 with w · v1 ≠ w · v2. Stochastic HG, by contrast, admits equiprobable mappings and the paper supplies a complete formal characterization of exactly those pairs. The differing predictions are compared on a Finnish stress test case.

Significance. If the formal results hold, the work isolates a sharp, model-internal difference between MaxEnt and Stochastic HG regarding the possibility of equiprobable mappings. The explicit characterization for Stochastic HG and the direct construction for MaxEnt are strengths; the Finnish stress comparison supplies a concrete empirical illustration.

minor comments (2)

§3: the definition of 'mapping' via violation vectors is clear, but a brief reminder of how candidate sets are generated for the Finnish example would aid readers unfamiliar with the data set.
The Finnish stress section would benefit from an explicit statement of the constraint set and the observed frequency counts used for the comparison.

Simulated Author's Rebuttal

0 responses · 0 unresolved

We thank the referee for their positive evaluation of the manuscript and for recommending acceptance. The summary provided in the report accurately captures the central claims regarding the distinguishability of mappings in MaxEnt versus the characterization of equiprobable mappings in Stochastic HG, along with the Finnish stress comparison.

Circularity Check

0 steps flagged

No significant circularity identified

full rationale

The central results—that MaxEnt separates any distinct violation vectors via some nonnegative w, while Stochastic HG admits a nonempty class of equiprobable pairs—are direct consequences of the model definitions (nonnegative weights, probability as normalized exp(−w·v), and mappings identified with violation profiles). No step reduces a prediction to a fitted parameter, renames a known pattern, or relies on a load-bearing self-citation whose content is itself unverified. The Finnish stress test case is an application, not a definitional input. The derivation is therefore self-contained against the stated assumptions.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Abstract-only review reveals no free parameters, invented entities, or non-standard axioms; results rest on standard definitions of MaxEnt and Stochastic HG.

axioms (1)

standard math Standard definitions of maximum entropy and harmonic grammar with nonnegative weights generate well-defined probability distributions over mappings.
The claims rely on the established mathematical properties of these two frameworks.

pith-pipeline@v0.9.0 · 5570 in / 1141 out tokens · 20222 ms · 2026-05-24T22:20:59.495891+00:00 · methodology

Equiprobable mappings in weighted constraint grammars

Core claim

What carries the argument

If this is right

Where Pith is reading between the lines

Load-bearing premise

What would settle it

discussion (0)