Word sense disambiguation in Thai using decision list collocation

Abstract

To develop a prototype of word sense disambiguation program in Thai by using the decision list collocation algorithm. หัว /hua4/ is chosen as a representative of nouns and เก็บ /kep1/ as a representative of verbs. We analyzed the senses of หัว /hua4/ and เก็บ /kep1/ for preparing manually sense-tagged corpus based on the Thai dictionary of "The Royal Institute" and 1,800 samples of หัว /hua4/ and 1800 samples of เก็บ /kep1/ collected from a corpus of Bangkok Business newspaper. Twenty senses of หัว /hua4/ and nine senses of เก็บ /kep1/ were found. To test for the optimal span and the location of sense indicators, we trained the algorithm at twenty spans of collocation and obtained twenty decision lists and tested the algorithm with these decision lists. The algorithm made the decision by choosing the sense with the highest collocational weight. The result suggests that the span +-2 is sufficient for the disambiguation of both words and the sense indicators of both words are mostly on the right side. The optimal span for disambiguating หัว /hua4/ is one-word-to-the-right-and-left with the precision rate of 87%, while เก็บ /kep1/ is two-words-to-the-right, with the precision rate of 80.25%. The high precision rate suggests that the decision list algorithm used in this study is applicable to the task