Separating Syntax and Semantics for Semantic Parsing

Study of language disorders, theoretical linguistics, and neuroscience suggests language competence involves two interacting systems, typically dubbed syntax and semantics. However few state-of-the-art deep-learning approaches for natural language processing explicitly model two different systems of representation. While achieving impressive performance on linguistic tasks, they commonly fail to generalize systematically. For example, unlike humans, learning a new verb like jump in isolation is insufficient for models to combine it with known words (like jump twice or jump and run). We continue the line of investigation recently started by Russin et al, whereby deep learning models are encouraged to learn separate representations for syntactic and semantic aspects of the input. Our goal is to achieve better systematic generalization on the task of semantic parsing, which requires transforming natural language utterances into executable programs. Building better models of semantic parsing is of high practical importance as they enable many important applications, such as e.g. natural language interfaces to databases.

Faculty Supervisor:

Timothy J. O’Donnell;Siva Reddy

Student:

Emily Goodwin

Partner:

Element AI

Discipline:

Languages and linguistics

Sector:

Professional, scientific and technical services

University:

McGill University