Paper

Stochastic harmonic grammars do not peak on the mappings singled out by categorical harmonic grammars

---

Abstract

A candidate surface phonological realization is called a peak of a probabilistic constraint-based phonological grammar provided it achieves the largest probability mass over its candidate set. Obviously, the set of peaks of a maximum entropy grammar is the categorical harmonic grammar corresponding to the same weights. This paper shows that the set of peaks of a stochastic harmonic grammar instead can be different from the categorical harmonic grammar corresponding to any weights. Thus in particular, maximum entropy and stochastic harmonic grammars can peak on different candidates.

Keywords: computational phonology, probabilistic phonology, maximum entropy

How to Cite:

Magri, G., (2023) “Stochastic harmonic grammars do not peak on the mappings singled out by categorical harmonic grammars”, Society for Computation in Linguistics 6(1), 269-277. doi: https://doi.org/10.7275/e6ev-jn97