CUR - The Council on Undergraduate Research

Towards a Natural Language Database Query Processing System Using Montague Semantics

Aaron Cogbill and Dr. Suk-Chung Yoon, Computer Science, Widener University, 1 University Pl, Chester, PA 19013

Use of natural language interfaces to retrieve information stored in relational databases is a desirable alternative to complicated query languages such as SQL (Structured Query Language). In this project, we develop a system to transform natural language queries into SQL using Montague semantics. Montague semantics refers to a systematic method for transforming syntax trees of English sentences into formulas of quantified logic. Relational algebra is the formal language describing the operation of relational database systems. We outline a homomorphism (in fact, half of an isomorphism) between the formulas of quantified logic and relational algebra expressions which allows us to interpret English sentences as database queries, and we evaluate the soundness of the concept with a practical implementation. While Montague semantics faithfully captures the syntactic relationships of an English sentence, however, it does not tell us how to interpret the words themselves. Our implementation thus also explores techniques for inferring the meanings of individual words as well as their surrounding phrases in the context of a suitable database schema. In particular, we conceptualize the interpretation of nouns and verbs as a categorization problem, while prepositions' meanings can be automatically inferred from the relation schemas of the database itself. Since modern relational databases do not accept input written as relational algebra, we write the final queries as SQL expressions. Given a modestly-sized static configuration file and a database connection, our system can generate logical, relational algebra, and SQL expressions representing the meaning of a natural language query for an arbitrary database.

Additional Abstract Information