Disambiguating “DE” for Chinese-English Machine Translation

Proceedings of the Fourth Workshop on Statistical Machine Translation , pages 215–223, Athens, Greece, 30 March – 31 March 2009. c⃝2009 Association for Computational Linguistics Disambiguating “DE” for Chinese-English Machine Translation Pi-Chuan Chang, Dan Jurafsky, and Christopher D. Manning Computer Science Department, Stanford University Stanford, CA 94305 pichuan,jurafsky,manning@stanford.edu Abstract Linking constructions involving � (DE) are ubiq- uitous in Chinese, and can be translated into En- glish in many different ways. This is a major source of machine translation error, even when syntax- sensitive translation models are used. This paper explores how getting more information about the syntactic, semantic, and discourse context of uses of � (DE) can facilitate producing an appropriate English translation strategy. We describe a finer- grained classification of � (DE) constructions in Chinese NPs, construct a corpus of annotated ex- amples, and then train a log-linear classifier, which contains linguistically inspired features. We use the DE classifier to preprocess MT data by explicitly labeling � (DE) constructions, as well as reorder- ing phrases, and show that our approach provides significant BLEU point gains on MT02 (+1.24), MT03 (+0.88) and MT05 (+1.49) on a phrased- based system. The improvement persists when a hi- erarchical reordering model is applied. 1 Introduction Machine translation (MT) from Chinese to En- glish has been a difficult problem: structural dif- ferences between Chinese and English, such as the different orderings of head nouns and rela- tive clauses, cause BLEU scores to be consis- tently lower than for other difficult language pairs like Arabic-English. Many of these structural differences are related to the ubiquitous Chinese �(DE) construction, used for a wide range of noun modification constructions (both single word and clausal) and other uses. Part of the solution to dealing with these ordering issues is hierarchi- cal decoding, such as the Hiero system (Chiang, 2005), a method motivated by �(DE) examples like the one in Figure 1. In this case, the transla- tion goal is to rotate the noun head and the preced- ing relative clause around �(DE), so that we can translate to “[one of few countries] � [have diplo- matic relations with North Korea]”. Hiero can learn this kind of lexicalized synchronous gram- mar rule. But use of hierarchical decoders has not solved the DE construction translation problem. We ana- lyzed the errors of three state-of-the-art systems (the 3 DARPA GALE phase 2 teams’ systems), and even though all three use some kind of hier- archical system, we found many remaining errors related to reordering. One is shown here: �� �� ���� � �� local a bad reputation DE middle school Reference: ‘a local middle school with a bad reputation’ Team 1: ‘a bad reputation of the local secondary school’ Team 2: ‘the local a bad reputation secondary school’ Team 3: ‘a local stigma secondary schools’ None of the teams reordered “bad reputation” and “middle school” around the �. We argue that this is because it is not sufficient to have a for- malism which supports phrasal reordering, but it is also …

Content truncated to 500 words