Q-TRANS: Query Translation Into English

Q-TRANS, which stands for Query-TRANSlation System, translates formal database queries into English to enhance the usability of both natural and formal language database access systems. Q-TRANS is designed for the database query language SQL, whose query expressions serve as an abstract representation from which an English paraphrase is generated. Q-TRANS is also intended to be part of the Transformational Question Answering System (TQA system), which provides a natural language interface for database query, analyzing and ultimately translating the English queries into SQL expressions. The concepts and methods used in Q-TRANS to arrive at a query translation are, however, independent from the TQA system except for compatibility of lexical and grammatical coverage of the paraphrar.es produced. The paraphrases generated are true translations of the SQL expressions which are the input to Q-TRANS and serve in a sense as deep structures that get mapped into English imperatives. The grammatical English structures Q-TRANS produces obey somewhat conflicting constraints in that they preserve as much of the SQL structure as necessary to reflect the internal logic to the user, and at the same time represent as natural English sentences as possible. I. PURPOSE A N D GOALS OF Q-TRANS Formal database query languages represent a welldefined, syntactically constrained, yet powerful means of accessing databases. Once users have mastered the syntax and lexicon of such languages, however, they still may not be successful in retrieving the desired data due to the amount of detail that needs to be specified, due to the requirement for exact matching of variables, or due to misconceptions about the contents and structure of the database. Should users be fortunate enough to have a natural language interface for database accessing, different problems arise, caused by the inherently vague and ambiguous nature of natural language. That is, the system's interpretation of English queries in terms of database queries can yield ambiguous or otherwise inappropriate results. In any case, whether a formal database language or English is the basis for communication with the database, users might benefit from an independent means of verifying whether the data retrieved is really what they wanted or asked for. Q-TRANS has been designed to help users resolve such uncertainties by providing an English paraphrase of the formal query, regardless of whether it was produced by the user directly or indirectly via a natural language front end system. Q-TRANS is currently being developed at IBM Research in conjunction with the Transformational Question Answering System (TQA) (Petrick 1981, Plath 1976). TQA provides a natural language interface to databases, allowing users to pose their queries in English rather than in some formal language. The English input is processed in the TQA system by computing its underlying structure using a lexicon, a transformational grammar, and a transformational parser. The underlying structure is translated into logical form, a relational calculus expression, which is then mapped into SQL. SQL is a relational database language used independently of TQA for accessing, maintaining and updating databases (Astrahan et al. 1976). The output of TQA, the natural language front end, is a formal SQL expression. The latter serves as input to the back end. System R. which executes the SQL query against the database. The SQL expression also constitutes the only direct link between TQA and Q-TRANS. Q-TRANS translates SQL expressions into English requests without utilizing the original English query or any intermediate analysis produced by TQA. The grammatical and lexical coverage of the paraphrase, however, is designed to be compatible with that of TQA. Given the Q-TRANS paraphrase, users now can check whether their original query was interpreted correctly by the system or phrased correctly by them, or in the rare event that TQA produces more than one formal query from the English input they can choose the appropriate one on the basis of the QTRANS paraphrases. The fact that the English paraphrase is directly generated from the formal database query distinguishes this work from most other research on database response generation, which tends to rely on the natural language front end and its analysis and semantic representation of the input query (McKeown 1979, Grishman 1979, Kaplan 1977) or on template matching (Codd 1978, Waltz 1978). Whether the INTELLECT system, which is commercially available, uses either of these methods for its paraphrase is not clear. The paraphrase appears to be produced by simply inserting English-like names for the column names of the database into a sentence skeleton. 11. Q-TRANS PROCESSING Three major steps are involved in producing the English paraphrase from the SQL expression. First, the incoming SQL expression is parsed by Q-TRANS, producing for the input string a hierarchical description which serves as the underlying structure in the translation process. This is accomplished by using a modified version of the SQL grammar which is provided to every user in the SQL manual. The original grammar is unambiguously context-free, and processing with it is extremely fast. The modifications that have been implemented consist of changes in category names and some regrouping of the original structures in the grammar so as to make them more compatible with the English structures to be generated. The end result after the modification is still unambiguously context-