An Algorithm for Learning Phonological Classes from Distributional Similarity

منو

" An Algorithm for Learning Phonological Classes from Distributional Similarity "
Mayer, Connor Hayes, Bruce

Document Type	:	Latin Dissertation
Language of Document	:	English
Record Number	:	904552
Doc. No	:	TL5jp6q2xn
Main Entry	:	Mayer, Connor
Title & Author	:	An Algorithm for Learning Phonological Classes from Distributional Similarity\ Mayer, ConnorHayes, Bruce
College	:	UCLA
Date	:	2018
student score	:	2018
Abstract	:	An outstanding question in phonology is to what degree the learner uses distributional information rather than substantive properties of speech sounds when learning phonological structure. This paper presents an algorithm that learns phonological classes from only distributional information: the contexts in which sounds occur. The input is a segmental corpus, and the output is a set of phonological classes. The algorithm is first tested on an artificial language with both overlapping and nested classes reflected in the distribution. It retrieves the expected classes, and performs well as distributional noise is added. It is then tested on four natural languages. It distinguishes between consonants and vowels in all cases, and finds more detailed, language-specific structure. These results improve on past approaches, and are encouraging given the paucity of the input. Further refined models may provide additional insight into which phonological classes are apparent in the distributions of sounds in natural languages.
Added Entry	:	Hayes, Bruce
Added Entry	:	UCLA