Effects of Empty Categories on Machine Translation

Proceedings of the 2010 Conference on Empirical Methods in Natural Language Processing, pages 636–645, MIT, Massachusetts, USA, 9-11 October 2010. c⃝2010 Association for Computational Linguistics Effects of Empty Categories on Machine Translation Tagyoung Chung and Daniel Gildea Department of Computer Science University of Rochester Rochester, NY 14627 Abstract We examine effects that empty categories have on machine translation. Empty categories are elements in parse trees that lack corresponding overt surface forms (words) such as dropped pronouns and markers for control construc- tions. We start by training machine trans- lation systems with manually inserted empty elements. We find that inclusion of some empty categories in training data improves the translation result. We expand the experiment by automatically inserting these elements into a larger data set using various methods and training on the modified corpus. We show that even when automatic prediction of null ele- ments is not highly accurate, it nevertheless improves the end translation result. 1 Introduction An empty category is an element in a parse tree that does not have a corresponding surface word. They include traces such as Wh-traces which indi- cate movement operations in interrogative sentences and dropped pronouns which indicate omission of pronouns in places where pronouns are normally expected. Many treebanks include empty nodes in parse trees to represent non-local dependencies or dropped elements. Examples of the former include traces such as relative clause markers in the Penn Treebank (Bies et al., 1995). An example of the lat- ter include dropped pronouns in the Korean Tree- bank (Han and Ryu, 2005) and the Chinese Tree- bank (Xue and Xia, 2000). In languages such as Chinese, Japanese, and Ko- rean, pronouns are frequently or regularly dropped when they are pragmatically inferable. These lan- guages are called pro-drop languages. Dropped pro- nouns are quite a common phenomenon in these lan- guages. In the Chinese Treebank, they occur once in every four sentences on average. In Korean the Treebank, they are even more frequent, occurring in almost every sentence on average. Translating these pro-drop languages into languages such as En- glish where pronouns are regularly retained could be problematic because English pronouns have to be generated from nothing. There are several different strategies to counter this problem. A special NULL word is typically used when learning word alignment (Brown et al., 1993). Words that have non-existent counterparts can be aligned to the NULL word. In phrase-based translation, the phrase learning system may be able to learn pronouns as a part of larger phrases. If the learned phrases include pronouns on the target side that are dropped from source side, the system may be able to insert pronouns even when they are miss- ing from the source language. This is an often ob- served phenomenon in phrase-based translation sys- tems. Explicit insertion of missing words can also be included in syntax-based translation models (Ya- mada and Knight, 2001). For the closely related problem of inserting grammatical function particles in English-to-Korean and English-to-Japanese ma- chine translation, Hong et al. (2009) …

Content truncated to 500 words