Title:   Treebanking and Parsing

Speaker: Dr. Keh-Jiann Chen
         Institute of Information Science, Academia Sinica
         115 Taipei, Taiwan, R.O.C.
         Phone: (886)-2-2788-3799 ext 1510
         e-mail: kchen@iis.sinica.edu.tw

Summary: In this tutorial, the issues regarding the Chinese language parsing
         and its relations to the grammar and threebanks will be addressed.
         In particular, focuses will be on the difficulties for processing
         Chinese language. Parsing natural language sentences makes use of
         many different knowledge sources, such as lexical, syntax, semantic,
         and common sense knowledge. Preparation of knowledge bank is a very
         difficult task, since there are vast amount of knowledge and they
         are not well organized. Corpus-based approach provided a way of
         automatically extract different knowledge. From part-of-speech tagged
         corpora to the syntactic structure annotated treebanks, each
         contribute more explicit linguistic knowledge at different level
         for better automation on knowledge extraction. Treebanks provide
         an easy way of extracting grammar rules and their occurrence
         probability. In addition, word-to-word relations are also precisely
         associated. Hence it raises the following important issues. 

         How were treebanks prepared? What are tag-sets suitable for Chinese
         treebanks? How many annotated tree structures are sufficient in a
         treebank for the purpose of grammar generation? What are tradeoffs
         between grammar coverage and ambiguities? We will try to answer the
         above questions in this tutorial.

         Regarding sentence parsing, some special characteristics of
         Chinese language cause difficulties for processing Chinese and make
         processing of Chinese different from other natural languages. For
         instances, there is no blanks to mark word boundaries and new words
         can be easily coined in Chinese text. No inflectional marking for
         grammatical functions and relatively free order of constituents
         cause the ambiguities and difficulties of the Chinese sentence
         parsing. We will try to provide the solutions to the above issues
         occurring at the stage of word segmentation, such as segmentation
         ambiguities, unknown word identification and the issues occurring
         at the stage of parsing, such as the problems of under coverage
         of grammars, structure ambiguities, and thematic role assignments.

Tutorial Outline:

         1. Introduction to Chinese treebanking and parsing
            a. Preparation of Chinese treebank
            b. Characteristics of Chinese language and their
               relations to Chinese language processing
         2. Uses of treebanks and grammar extraction
            a. Searching tools
            b. Grammar extraction from treebanks
            c. Grammar coverage and ambiguities
         3. Lexical analysis
            a. Word segmentation
            b. Unknown word identification
         4. Structure analysis and role assignment
            a. Knowledge for sentence parsing
            b. Resolution of structure ambiguities
            c. Thematic-role assignments

Bio:     His current research interests include Chinese language processing,
         lexical semantics, and corpus linguistics. He is developing research
         environments for Chinese natural language processing including
         lexical databases, tagged corpora, treebanks, and parsers.