Learning to Map Context-Dependent Sentences to Executable Formal Queries
This addresses the challenge of understanding natural language in multi-turn dialogues for database query systems, representing an incremental improvement over existing methods.
The paper tackles the problem of mapping context-dependent utterances to executable formal queries in interactive systems, achieving improved performance on the ATIS flight planning dataset by incorporating interaction history and explicit reference modeling.
We propose a context-dependent model to map utterances within an interaction to executable formal queries. To incorporate interaction history, the model maintains an interaction-level encoder that updates after each turn, and can copy sub-sequences of previously predicted queries during generation. Our approach combines implicit and explicit modeling of references between utterances. We evaluate our model on the ATIS flight planning interactions, and demonstrate the benefits of modeling context and explicit references.