A Sense-Based Translation Model for Statistical Machine Translation

Abstract

The sense in which a word is used determines the translation of the word. In this paper, we propose a sense-based translation model to integrate word senses into statistical machine translation. We build a broad-coverage sense tagger based on a nonparametric Bayesian topic model that automatically learns sense clusters for words in the source language. The proposed sense-based translation model enables the decoder to select appropriate translations for source words according to the inferred senses for these words using maximum entropy classifiers. Our method is significantly different from previous word sense disambiguation reformulated for machine translation in that the latter neglects word senses in nature. We test the effectiveness of the proposed sensebased translation model on a large-scale Chinese-to-English translation task. Results show that the proposed model substantially outperforms not only the baseline but also the previous reformulated word sense disambiguation.

Sense-Aware Statistical Machine Translation using Adaptive Context-Dependent Clustering

“…Xiong and Zhang (2014) attempted to answer this question by performing word sense induction for large scale data. In particular, they proposed a topic model that automatically learned sense clusters for words in the source language.…”

Word Sense Induction for Better Lexical Choice

“…Recent work in Machine Translation ( [7] and [8]) and Information Retrieval [9] indicates that induced senses can lead to substantial improvement in performance where methods based on a fixed sense inventory such as HowNet have previously failed ( [10] and [11]). Therefore, We adopt the similar approach of Xiong and Zhang [8] by resorting to Word Sense Induction (WSI) that is related to but different from WSD.…”

“…Although the redefined WSD has proved helpful for SMT, recently, Xiong and Zhang [8] re-investigated the question of whether pure senses are useful for SMT by using WSI. They proposed a sense-based translation model to integrate word senses into SMT which enables the decoder to select appropriate translations for the source words according to the inferred senses for these words using Maximum Entropy classifiers.…”

“…To obtain word senses for any source words, we build a sense tagger that relies on the nonparametric Bayesian model based word sense induction ( [17], [18]) similar to Xiong and Zhang [8]. We used HDP-based WSI 4 [19] to predict sense clusters and to annotate source words in our training/development/test sets with these sense clusters.…”

“…word senses that are obtained using the HDP-based WSI. We adopt the same approach of Xiong and Zhang [8] to build the sense-based Translation Model.…”