Tense Tagging for Verbs in Cross-Lingual Context: A Case Study

R. Dale et al. (Eds.): IJCNLP 2005, LNAI 3651, pp. 885 – 895, 2005. © Springer-Verlag Berlin Heidelberg 2005 Tense Tagging for Verbs in Cross-Lingual Context: A Case Study Yang Ye1 and Zhu Zhang2 1 Department of Linguistics, University of Michigan, USA 2 School of Information and Department of Electrical Engineering and Computer Science, University of Michigan, USA yye@umich.edu, zhuzhang@umich.edu Abstract. The current work applies Conditional Random Fields to the problem of temporal reference mapping from Chinese text to English text. The learning algorithm utilizes a moderate number of linguistic features that are easy and in- expensive to obtain. We train a tense classifier upon a small amount of manu- ally labeled data. The evaluation results are promising according to standard measures as well as in comparison with a pilot tense annotation experiment in- volving human judges. Our study exhibits potential value for full-scale machine translation systems and other natural language processing tasks in a cross- lingual scenario. 1 Introduction Temporal resolution is a crucial dimension in natural language processing. The fact that tense does not necessarily exist as a grammatical category in many languages poses a challenge on cross-lingual applications, e.g. machine translation. The fact that English tenses and Chinese aspect markers align at word level on one hand and sub-word level on the other hand poses a challenge for temporal reference distinction translation in a statistical machine translation (MT) system. A word-based alignment algorithm will not be able to capture the temporal reference distinction when mapping between Chinese and English. Being able to successfully map the temporal reference distinction in Chi- nese text through disparate features onto the most appropriate tenses for the parallel English text is an important criterion for good translation quality. Languages have vari- ous levels of time reference distinction representation: some have finer grained tenses than others, as typological studies have shown. When facing the unbalanced levels of temporal reference distinction between a pair of languages, we have to optimize the mapping between the two temporal systems through intelligent learning. Most machine translation systems do not have a separate temporal reference resolution module, but if we can integrate a special module into them, the temporal reference resolution of the system could be corrected accordingly and yield a better translation. Other than machine translation, in cross-lingual question answering (CLQA) with English as the target lan- guage, the ability to successfully formulate queries and maintain the temporal reference information in the original questions is desirable. 886 Y. Ye and Z. Zhang 2 Related Work 2.1 Temporal Reference Modeling in Cross-Lingual Scenario The nature of being past, present or future is highly relative and hence the information contained in tenses is often referred to as temporal reference distinction. While there is a large body of research on temporal reference in formal semantics and logic as well as in other disciplines of Linguistics, works in cross-lingual temporal reference mapping remain inadequate. Campbell et. al. [1] proposed a language-neutral framework for representing se- mantic tense. This framework is called …

Content truncated to 500 words