Learning Both Variability and Exceptionality in Probabilistic OT Grammars
- Aleksei Nazarov (University of Huddersfield)
Abstract
The co-existence of variability and exceptionality in the same language, like in Modern Hebrew (Temkin-Martínez 2010), challenges OT-style learners. Probabilistic OT (e.g., Boersma 1998) captures variability, while exceptional words can be identified (e.g., Becker 2009) by inconsistency detection (Tesar 1995) in non-probabilistic OT; no previous proposal can do both. I propose a “soft inconsistency” criterion that identifies exceptional words in the probabilistic Expectation Driven Learning framework (Jarosz 2015), allowing learning of both variability and exceptionality. Tested on simplified Hebrew data, this model learns both the variable default pattern (>=95% accuracy) and the pattern of exceptions (>=95% overall accuracy on data).
Keywords: computational phonology, learning, exceptionality, variation, diacritics, grammar induction, constraint induction
How to Cite:
Nazarov, A., (2018) “Learning Both Variability and Exceptionality in Probabilistic OT Grammars”, Society for Computation in Linguistics 1(1), 221-222. doi: https://doi.org/10.7275/R5TH8JWG
Downloads:
Download PDF