Discourse understanding and response planning in a chinese language front-end for databases
Martha W. Evens, Da-Jinn Wang · 1999
In this research project, our goal was to build a domain-independent Chinese language interface FCIDB (Friendly Chinese Interface for DataBase management system). By interacting with FCIDB, users can communicate with a DBMS in the Chinese language. FCIDB is a natural language interface system. It performs lexical analysis, syntactic analysis, and semantic analysis to understand the Chinese queries and translate them to SQL commands. Then, FCIDB communicates with the DBMS to retrieve the required information. Finally, FCIDB decides whether to respond to the user with a table or a paragraph. My research focus is on the lexicon building, word segmentation, and response planning. The lexicon in FCIDB is essential to word segmentation and query translation. It helps FCIDB perform lexical analysis. As a first step, we constructed two different databases, a school database and a warehouse database, and we designed questionnaires to collect queries. The results helped us to identify which words we needed in the domain-independent and domain-specific lexicons. To make FCIDB transportable, I also designed a mapping dictionary for the FCIdB system. The mapping dictionary is used to help users provide the specification of a new database. The initial interaction between a user and FCIDB is designed to inform the system about the new domain. The information includes column names, table names, their corresponding Chinese words, and relationships between tables. The goal of FCIDB is to communicate with users in a way that seems natural to humans. Sometimes it is better to put the response in a table, sometimes in a paragraph. Which format is the best format to display the information that the user requires? There is no absolute answer for this question. The dialogue component in the FCIDB system will place the response in a table or respond with a paragraph. The response format depends on the information that the user requires. In any case, the goal of the dialogue component is to make the response in a format that users can understand easily. FCIDB can be used with any DBMS that uses SQL. We have tested it with ACCESS, Visual dBASE, and FoxPro.