SMTPOST: Using statistical machine translation approach in Filipino part-of-speech tagging

College

College of Computer Studies

Department/Unit

Software Technology

Document Type

Conference Proceeding

Source Title

Proceedings of the 30th Pacific Asia Conference on Language, Information and Computation, PACLIC 2016

First Page

391

Last Page

396

Publication Date

1-1-2016

Abstract

The field of Natural Language Processing (NLP) in the country has been continually developing. However, the transition between Tagalog to the progressing Filipino language left tools and resources behind. This paper introduces a Statistical Machine Translation Part-of-Speech (POS) Tagger for Filipino (SMTPOST), with the purpose of reviving, updating and widening the scope of technologies in the POS tagging domain, catering to the changes made by the Filipino language. Resources built are comprised mainly of a tagset (218 tags), parallel corpus (2,668 sentences), affix rules (59 rules) and word-tag dictionary (309 entries). SMTPOST was tested to different tagsets and domains, producing 84.75% as its highest accuracy score, at least 3.75% increase from the available Tagalog POS taggers. Despite SMTPOST's utilization of Filipino resources and good performance, there are room for improvements and opportunities. Recommendations include a better feature extractor (preferably a morphological analyzer), an increase in scope for all of the resources, implementation of pre- And/or postprocessing, and the utilization of SMTPOST research to other NLP applications.

html

Disciplines

Computer Sciences

Keywords

Computational linguistics; Filipino language—Machine translating; Tagalog language—Machine translating

Upload File

wf_no

This document is currently not available here.

Share

COinS