Proto-phoneme reconstruction as naive Bayes inference
dc.contributor.advisor
Kirby, James
en
dc.contributor.author
Maftei, Dan
en
dc.date.accessioned
2014-03-26T15:09:39Z
dc.date.available
2014-03-26T15:09:39Z
dc.date.issued
2012-11-27
dc.description.abstract
The comparative method is the standard technique by which historical linguists reconstruct ancestral languages from their descendants. The method, however, has received little attention from the computational linguistics community. We present a principled method by which sound change plausibility can be encoded, and a probabilistic framework for learning about sound change and using this knowledge for reconstruction. Our techniques are entirely probabilistic, and leverage the wealth of data that is becoming available in a machine-readable format. We show that a Naive Bayes classifier, combined with phoneme-conditioned categorical distributions over phonemes, learned via Maximum a Posteriori with smoothing based on phonetic similarity, can be used to accurately reconstruct proto-words from their descendants. Our system out-performs previous approaches to language reconstruction.
en
dc.identifier.uri
http://hdl.handle.net/1842/8593
dc.language.iso
en
dc.publisher
The University of Edinburgh
en
dc.subject
linguistics
en
dc.subject
comparative method
en
dc.subject
computational linguistics
en
dc.subject
naive bayes
en
dc.title
Proto-phoneme reconstruction as naive Bayes inference
en
dc.type
Thesis or Dissertation
en
dc.type.qualificationlevel
Masters
en
dc.type.qualificationname
MSc Master of Science
en
dcterms.accessRights
RESTRICTED ACCESS
en
Files
This item appears in the following Collection(s)