Using Deep Learning to Explore Phonotactic Learning

What phonetic information (if any) is required and/or helpful for learning phonotactic generalizations? Does prior phonotactic knowledge (e.g., an already-acquired language) affect the helpfulness of certain phonetic information for learning phonotactic generalizations?


Project Status

This project is currently on hold, as I am primarily focusing on my thesis work. However, if you have any questions about the codebase, the findings, or anything else relating to this project, feel free to reach out!


Findings So Far

  • Pre-specificed phonological distinctive features are not required for phonotactic learning
  • For an LSTM, distinctive features may even be detrimental, constraining the solution space unhelpfully

Presented at

  1. MidPhon 23 (Mid-continental Phonetics and Phonology Workshop), October 2018: Computationally Assessing the Usefulness of Phonotactic Distinctive Features (oral presentation)
  2. Association for Computational Linguistics, July 2019: Using LSTMs to Assess the Obligatoriness of Phonological Distinctive Features for Phonotactic Learning (poster presentation)