When one probabilistic grammar is not restrictive enough: The curious case of an L2 learning simulation

Abstract

This article explores an intriguing case described by Zhou and Hamann (2026), who provided a Noisy Harmonic Grammar account for how perception and orthography together shape the L2 acquisition of the Spanish palatal approximant. Although the proposed grammar successfully generates the attested outputs in a categorical manner, an accompanying computer simulation showed a mismatch: a single grammar produced more variation than was observed empirically, while a two-grammar simulation exactly mirrored the data. Through a close analysis of the constraints and candidate evaluation, we lay bare the underlying contradictory requirements on the constraint weights, and demonstrate how such a contradiction can be identified and resolved by two separate grammars.