Machine Translation Using Probabilistic Synchronous Dependency Insertion Grammars

Proceedings of the 43rd Annual Meeting of the ACL, pages 541–548, Ann Arbor, June 2005. c⃝2005 Association for Computational Linguistics Machine Translation Using Probabilistic Synchronous Dependency Insertion Grammars Yuan Ding Martha Palmer Department of Computer and Information Science University of Pennsylvania Philadelphia, PA 19104, USA {yding, mpalmer}@linc.cis.upenn.edu Abstract Syntax-based statistical machine transla- tion (MT) aims at applying statistical models to structured data. In this paper, we present a syntax-based statistical ma- chine translation system based on a prob- abilistic synchronous dependency insertion grammar. Synchronous depend- ency insertion grammars are a version of synchronous grammars defined on de- pendency trees. We first introduce our approach to inducing such a grammar from parallel corpora. Second, we de- scribe the graphical model for the ma- chine translation task, which can also be viewed as a stochastic tree-to-tree trans- ducer. We introduce a polynomial time decoding algorithm for the model. We evaluate the outputs of our MT system us- ing the NIST and Bleu automatic MT evaluation software. The result shows that our system outperforms the baseline sys- tem based on the IBM models in both translation speed and quality. 1 Introduction Statistical approaches to machine translation, pio- neered by (Brown et al., 1993), achieved impres- sive performance by leveraging large amounts of parallel corpora. Such approaches, which are es- sentially stochastic string-to-string transducers, do not explicitly model natural language syntax or semantics. In reality, pure statistical systems some- times suffer from ungrammatical outputs, which are understandable at the phrasal level but some- times hard to comprehend as a coherent sentence. In recent years, syntax-based statistical machine translation, which aims at applying statistical mod- els to structural data, has begun to emerge. With the research advances in natural language parsing, especially the broad-coverage parsers trained from treebanks, for example (Collins, 1999), the utiliza- tion of structural analysis of different languages has been made possible. Ideally, by combining the natural language syntax and machine learning methods, a broad-coverage and linguistically well- motivated statistical MT system can be constructed. However, structural divergences between lan- guages (Dorr, 1994)，which are due to either sys- tematic differences between languages or loose translations in real corpora，pose a major chal- lenge to syntax-based statistical MT. As a result, the syntax based MT systems have to transduce between non-isomorphic tree structures. (Wu, 1997) introduced a polynomial-time solu- tion for the alignment problem based on synchro- nous binary trees. (Alshawi et al., 2000) represents each production in parallel dependency trees as a finite-state transducer. Both approaches learn the tree representations directly from parallel sen- tences, and do not make allowances for non- isomorphic structures. (Yamada and Knight, 2001, 2002) modeled translation as a sequence of tree operations transforming a syntactic tree into a string of the target language. When researchers try to use syntax trees in both languages, the problem of non-isomorphism must be addressed. In theory, stochastic tree transducers and some versions of synchronous grammars pro- vide solutions for the non-isomorphic tree based transduction problem and hence possible solutions for MT. Synchronous …

Content truncated to 500 words