Ph.D Thesis

Ph.D StudentNelken Ran
SubjectQuestions, Time and Natural Language Interfaces to Temporal
DepartmentDepartment of Computer Science
Supervisor PROFESSOR EMERITUS Nissim Francez


Recent years have shown a surge in interest in temporal database systems, which allow users to store time-dependent information. Unfortunately, non-expert users are expected to have extreme difficulties in using temporal query languages developed for temporal databases, due to the added complexity introduced by handling temporality. To counter this problem, we suggest the design of a natural language interface to temporal databases, based on translating natural language temporal questions into SQL /Temporal, a recent temporal database query language. The interface is based on a two-stage translation process. In the first stage of the translation, questions are translated into a two-sorted first-order logic over temporal intervals. This translation is based on a combined syntactic and semantic grammatical analysis of a large class of questions. The syntactic analysis is done in the Type-Logical Grammar framework, highlighting its utility not only as a theoretical framework but also as a practical tool. The semantic analysis focuses on the role of temporal preposition phrases rather than the more traditional  focus on tense and aspect in previous approaches. In the second stage of the translation, we translate logical formulae into SQL/Temporal by extending standard methods for translating first order logic to standard SQL. Our translation method and reasoning about its correctness and logical properties are therefore considerably simpler than previous attempts in this direction. We have implemented a prototype natural language interface based on this translation method, and linked it with a prototype temporal database implementation.

Considering the semantics of natural language questions has lead us to develop an independently motivated novel theory of this topic. Previous theories have assumed that the meaning of a question must somehow encode the meaning of its answers. Consequently, these theories have suggested that questions must be interpreted as complex intensional objects. In particular, they predict an a-priori surprising asymmetry between indicative sentences, which can be interpreted extensionally and  questions which must be interpreted intensionally.  Our theory re-examines this assumption by suggesting an extensional theory based on a re-interpretation of the domain of propositions as a bilattice rather than the usual Boolean interpretation. We show that this theory not only satisfies a set of adequacy criteria imposed on the semantics of questions, but is also able to account in a straightforward manner for some complex issues in the semantics of questions including coordinated questions, combined indicative and interrogative sentences, questions with quantifiers, questions and negation, and the generalized quantifier interpretation of interrogative noun phrases. In addition, we develop a sound and complete logical calculus of questions.