Learning from Chinese-English Parallel Data for Chinese Tense Prediction

Proceedings of the 5th International Joint Conference on Natural Language Processing, pages 1116–1124, Chiang Mai, Thailand, November 8 – 13, 2011. c⃝2011 AFNLP Learning from Chinese-English Parallel Data for Chinese Tense Prediction Feifan Liu The University of Wisconsin-Milwaukee liuf@uwm.edu Fei Liu and Yang Liu The University of Texas at Dallas {feiliu,yangl}@hlt.utdallas.edu Abstract Tense prediction can be useful for many language processing tasks, such as tem- poral inference and machine translation. In this paper, we investigate using diverse contextual features for Chinese tense pre- diction under a statistical learning frame- work. Because of lack of annotated train- ing data, we propose to leverage Chinese- English parallel corpora to automatically generate reference tense for model train- ing. We also propose to use an itera- tive learning framework to deal with the noisy reference data to improve learning. Evaluation is performed using both auto- matically generated reference data and a manually annotated set with verb tense. Our results demonstrate the effectiveness of our proposed learning framework that maps annotation from one language to an- other using parallel data. Furthermore, we show better performance using our proposed iterative bootstrapping learning method compared to using the original au- tomatically created training data. 1 Introduction Tense is used in languages to indicate the time at which an action or event described by the sen- tence takes place. Lacking correct tense infor- mation can cause confusion and misunderstanding in communication. Predicting tense information is very useful for different natural language pro- cessing tasks: both monolingual applications such as speech recognition (Karlgren, 1996) and mul- tilingual applications such as machine translation (MT) (Buschbeck et al., 1991). In inflectional languages like English, tense is often expressed by verb inflections, which can be easily recognized. For example, “I worked till 5pm yesterday” uses the past-tense verb “worked” to describe an event in the past. However, in lan- guages such as Chinese, no verb inflections ex- ist to indicate any tense information (Xiao and McEnery, 2002). The morphology of a verb it- self never changes when used to express different tenses. The following examples (1a) and (1b) indi- cate a past-tense sentence and a future-tense sen- tence respectively; however, the form of the Chi- nese verb “�(arrive)” is the same in the two cases. 几天前 我 到 了(n/a)上海 今天晚上 点 我 将 到 上海 This lack of inflections in Chinese verbs im- poses some challenges in many applications. For instance, in a Chinese-English machine transla- tion (MT) system, it is almost impossible to deter- mine the tense for the corresponding English verb if solely based on the Chinese verb morphology. However, tense information does exist in Chinese language, and it is expressed lexically instead of morphologically. There are useful contextual cues that can help determine the tense for the whole Chinese sentence or individual verbs, such as tem- poral adverbs or phrases, aspect auxiliary words and prepositions. For instance, in example (1a), the aspect particle “�(a particle word in Chinese, there is no literal translation to English)” and tem- poral phrase …

Content truncated to 500 words