Word Sense Subjectivity for Cross-lingual Lexical Substitution

Human Language Technologies: The 2010 Annual Conference of the North American Chapter of the ACL, pages 357–360, Los Angeles, California, June 2010. c⃝2010 Association for Computational Linguistics Word Sense Subjectivity for Cross-lingual Lexical Substitution Fangzhong Su School of Computing University of Leeds, UK scsfs@leeds.ac.uk Katja Markert School of Computing University of Leeds, UK scskm@leeds.ac.uk Abstract We explore the relation between word sense subjectivity and cross-lingual lexical substitu- tion, following the intuition that good substi- tutions will transfer a word’s (contextual) sen- timent from the source language into the target language. Experiments on English-Chinese lexical substitution show that taking a word’s subjectivity into account can indeed improve performance. We also show that just using word sense subjectivity can perform as well as integrating fully-fledged fine-grained word sense disambiguation for words which have both subjective and objective senses. 1 Introduction Cross-lingual lexical substitution has been proposed as a Task at SemEval-2010.1 Given a target word and its context in a source language (like English), the goal is to provide correct translations for that word in a target language (like Chinese). The trans- lations must fit the given context. In this paper, we explore the relation between the sentiment of the used word in the source language and translation choice in the target language, focus- ing on English as the source and Chinese as the tar- get language. Our work is motivated by the intuition that most good word translations will be sentiment- invariant, i.e. if a source word is used in a subjec- tive (opinion-carrying) sense it will be often trans- lated with a subjective sense in the target language whereas if it used in an objective sense, it will be 1http://lit.csci.unt.edu/index.php/ Semeval_2010 translated with an objective sense. As an exam- ple, consider the two words positive and collaborate with example senses from WordNet 2.0 below. (1) positive—greater than zero; “positive numbers” (objective) (2) plus, positive—involving advantage or good; “a plus (or positive) factor” (subjective) (3) collaborate, join forces, cooperate—work together on a common enterprise of project; “We joined forces with another research group”(objective) (4) collaborate—cooperate as a traitor; (subjective) In most cases, if the word positive is used in the sense “greater than zero” (objective) in an English context, the corresponding Chinese trans- lation is “ ”; if “involving advantage or good”(subjective) is used, its Chinese translations are “È4, Д. Similarly, for the word collab- orate, the sense “work together on a common en- terprise of project” (objective) corresponds to “Ü Š, Š” in Chinese translation, and “cooperate as a traitor” (subjective) corresponds to “�(, H r”. Therefore, subjectivity information should be effective for improving lexical translation for what we previously (Su and Markert, 2008) termed subjectivity-ambiguous words, i.e. words with both subjective and objective senses such as positive and collaborate above. We therefore incorporate subjectivity word sense disambiguation (SWSD) as defined in Akkaya et al. (2009) into lexical substitution. SWSD is a binary classification task that decides in context whether a word occurs with one of its subjective or one …

Content truncated to 500 words