Querying the French yellow pages: Natural language access to the directory

摘要

ERLI was asked by the French TELECOM to develop a specific system to query the professional headings of the French Yellow Pages directory. Approximately 4 million end users now have access (via their “Minitel” terminals) to some 6 million professionals registered under 2500 different headings. (A second application has also been developed using a similar system: the Minitel Applications Directory, which gives information on all the available applications in the Minitel network.)Although the retrieval of a heading is a necessary step in accessing data, it is of no real interest to the user, who wishes only to retrieve the phone number of a given professional or tradesperson.The general aims of the Natural Language System (NLS) are to facilitate access to headings by intelligent query processing (or even to bypass completely the necessity of choosing between headings).This is done through: • The association of a specific knowledge base to the list of headings, • The construction of a “grammar” ensuring a consistent interpretation of the queries.ERLI's system is as an alternative to the existing one, which is based on a key-word indexing technique. The weaknesses and insufficiencies of such a technique are well known, especially in this context, where queries are expressed by unqualified users, who are unfamiliar with the data (i.e., the headings of the directory).Finally, it is important to note that the NLS was developed with regard to industrial considerations (in particular, the minimizing of the average processing time per query). The system is not a prototype. Extensive on-side testing is scheduled to begin in July 1988 and a complete installation will be carried out at the end of the year.