N-gram Based WSD for Improving Accuracy of Machine Translation using TM

Author Name(s): Sunita Rawat, Manoj Chandak, Tabassum Khan
Author Email: ssunitarawatt@gmail.com

Abstract

Word Sense Disambiguation (WSD) concerns with selecting an accurate sense of a term automatically in the given situation. It is very significant and challenging problem in several applications of natural language processing. We have used a probabilistic model in our system. Word Sense Disambiguation is done based on n-grams (bigrams and trigrams). The objective is to evaluate the performance of the system with various kinds of an input string which have an ambiguous word. For resolving the probability of the sense of an ambiguous word in the given sentence we use a Naïve Bayes classifier. We have use translation memory (TM) to speed up the translation process. A classical translation memory chooses contender translations into a target language (Hindi) by getting similar translations to an entered text from available pairs of earlier translated segments.

Introduction

In natural language processing Word Sense Disambiguation is a topic that has been learned from so many years. The objective of word sense disambiguation is to choose the accurate sense of an ambiguous word in a specified context. In reality automatic word sense disambiguation persists to be an open problem has raised an immense attention in the community of computational linguistics, hence, various techniques has been commenced in the previous decades [1]. Rivalries like Senseval and lately SemEval1 have moreover encouraged the invention of innovative methods for word sense disambiguation, offering a motivating atmosphere for checking those methods. In spite of the word sense disambiguation task has been learned for a decade of time, the probable sentiment is that word sense disambiguation has to be included into actual applications for example information retrieval, lexicography, machine translation systems, knowledge mining, automatic answer machines,
semantic interpretation, etc. [1]. So many studies on this matter have confirmed that those applications have advantage from word sense disambiguation [2, 3]. To decide the probability of a sense of an ambiguous word, the statistical dictionary is given as input to Naïve Bayes classifier. We make a classification model which depends on the probability of finding correct sense of each ambiguous word as well as the words which enclose it. We know that some other classification models are also present for instance, support vector machine [4] and conditional random field [5]. On the other hand, while we have preferred a probabilistic model founded on independent description so as to determine the accurate target sense, we consider that the Naïve Bayes classifier entirely suits with this type of technique.

Conclusion

WSD is the task of deciding accurate meaning of the word (among available all meanings) in the specified context. In this paper, we presented a word sense disambiguation technique which uses n-grams for input sentences containing ambiguous word. Specifically, we used a Naive Bayes classifier for finding the probability of the accurate sense of an ambiguous word in the source language. Naïve Bayes is the simplest classification algorithm. It is easy to implement and requires only small amount of training data. It exhibits high accuracy and speed when applied to large databases. We have also used the concept of translation memory to speed up the translation process. Advantage of using TM is reduced translation cost as well as time.

236 total views, 1 views today

Download PDF File

About the author: admin