Extended Abstract

Investigating the Consequences of Iterated Learning in Phonological Typology

---

Abstract

This work builds on previous investigations of the effects of learning biases on gradient typological predictions in phonology. Our previous work (e.g. Hughto and Pater 2017) used an interactive, agent-based learning model and found robust biases against cumulativity effects in weighted-constraint grammars, and towards more deterministic grammars, where one output accumulates majority probability. This work compares the results of using an iterated learning model, in which “parent” agents teach “child” agents in a generational chain, and finds that these biases are present, but less robust across parameter settings. The deterministic bias was only present with longer learning times; the anti-cumulativity bias was more robust, but only emerged with shorter learning times if child agents\' initial weights were set to zero (rather than randomly sampled).

Keywords: phonology, typology, learning, computational phonology

How to Cite:

Hughto, C., (2018) “Investigating the Consequences of Iterated Learning in Phonological Typology”, Society for Computation in Linguistics 1(1), 182-185. doi: https://doi.org/10.7275/R5WH2N63