Applying Maximum Entropy to Robust Chinese Shallow Parsing

Applying Maximum Entropy to Robust Chinese Shallow Parsing Shih-Hung Wu*†, Cheng-Wei Shih†, Chia-Wei Wu†, Tzong-Han Tsai†, and Wen-Lian Hsu† †Institute of Information Science, Academia Sinica, Taiwan, R.O.C (shwu,dapi,cwwu,thtsai,hsu)@iis.sinica.edu.tw *Dep. Of CSIE, Chaoyang University of Technology, Taichung County, Taiwan, R.O.C shwu@cyut.edu.tw Abstract Recently, shallow parsing has been applied to various information processing systems, such as information retrieval, information extraction, question answering, and automatic document summarization. A shallow parser is suitable for online applications, because it is much more efficient and less demanding than a full parser. In this research, we formulate shallow parsing as a sequential tagging problem and use a supervised machine learning technique, Maximum Entropy (ME), to build a Chinese shallow parser. The major features of the ME-based shallow parser are POSs and the context words in a sentence. We adopt the shallow parsing results of Sinica Treebank as our standard, and select 30,000 and 10,000 sentences from Sinica Treebank as the training set and test set respectively. We then test the robustness of the shallow parser with noisy data. The experiment results show that the proposed shallow parser is quite robust for sentences with unknown proper nouns. 1. Introduction Parsing is a basic technique in natural language processing; however, a full parser is usually costly and slow. Recently, shallow parsing has been applied to various information processing systems [12]. Compared to the performance of full parsers, a shallow parser is much faster and the parsing result is more useful for various applications, such as information retrieval and extraction, question answering, and automatic document summarization. In this paper, we adopt a machine learning approach to the Chinese shallow parsing problem. Chinese full parsing is very challenging,[18, 22] because it is difficult to achieve high accuracy, and the performance is not suitable for online applications. Shallow parsing of Chinese, on the other hand, is promising and desirable in terms of efficiency. Researchers in Beijing, Harbin, Shenyang, and Hong Kong have also developed related techniques [10, 15, 16, 20, 21]. Most of these works use machine learning approaches, instead of the rule-based approach used in full parsing. Popular machine learning methods such as SVM, CRF, and ME, have been tested. The parsing speed of each approach is fast and the parsing accuracy is acceptable. Currently, there is no standard for Chinese shallow parsing. Li et. al. [9] developed a Chinese shallow parsed treebank to extract Chinese collocations automatically and built a large collocation bank. There are also some works on a standard for Chinese shallow parsing [9, 19, 20]. Nevertheless, the POS standard and vocabulary in each approach are different; thus, between simplified Chinese and traditional Chinese, we cannot adopt their standard for simplified Chinese to traditional Chinese. Instead, we use the first level of the parsing results of Sinica Treebank as our shallow parsing standard [4]. Originally, Sinica Treebank was designed to provide full parsing results, whereby sentences could be labeled with POS tags and the full parsing structure. There are 54,000 sentences in Sinica Treebank, from which we randomly …

Content truncated to 500 words