Using Natural Language Descriptions to Improve the Usability of Databases

USING NATURAL LANGUAGE DESCRIPTIONS TO IMPROVE THE USABILITY OF DATABASES Carole D. Hafner John D. Joyce Computer Science Department General Motors Research Laboratories Warren, MI ~8090 ABSTRACT This paper describes the REGIS extended co~nd language, a relational data fan&c/age that allows users to name and describe database objects using natural language phrases. REGIS accepts multlple-word phrases as the names of tables and columns (unlike most systems, which restrict these names to a few characters). An extended command parser uses a network- structured dictionary to recognize multl-word names, even if some of the words are missing or out of order, and to prompt the user if an ambiguous name is entered. REGIS also provides facilities for attaching descriptive text to database objects, which can be displayed on- line or included in printed reports. Initial data from a few databases indicate that users choose to take advantage of the naturalness of multl-word descriptions when this option is available. I INTRODUCTION The REGIS extended command language is a relational data language that allows users to name and describe database objects using na- tural language phrases. REGIS [4] is an inter- active data management system that has been in use at General Motors since 1975. The system is designed to be easy for non-progrs~nmers to understand, and it has given many people their first hands-on experience with computers. A REGIS database consists of a hierarchical structure of named objects: one or more files, each containing zero or more tables, each com- posed of zero or more columns of data. REGIS users can create, query, or modify database objects interactively, using simple keyword- based relational conLmands, such as the following: BLUES = SUBSET TABLE1 WHERE COLOR = BLUE (creates a new table from selected rows of an existing one) RESULTS = PROJECTION BLUES TYPE ITEM COST (creates a new table from specified columns of an existing one) LIST RESULTS (lists a table at the terminal) Future research directions for REGIS are aimed at developing natural language capabilities that will make computer databases easier to understand and access. The usability of database query languages has been recognized as an important problem (Codd [i], Greenblatt and Wax~nan [2], Welty and Stemple [5]); however, a closely related issue that has not been addressed is the usability of the data itself. In order to interact with a database effectively, users must be able to understand and refer to the objects in the database. Current database systems restrict the names of database objects to a few char- acters, which can lead to cryptic abbreviations that are difficult to understand and remember. Documentation facilities (if they exist at all) are not designed to be accessed interactively. The need to refer to external sources for des- criptive information, and the need to remember cryptic abbreviations, are obstacles to usabi- lity that are especially disruptive to the new or occasional user of a database. To provide a more supportive environment for data management, a new commana interface has been added to REGIS, which …

Content truncated to 500 words