Learning to Map English User Queries to Graph-Based Query Languages via Semantic Parsing

open_access
Open Access
Author:
Gearhart, Noah
Area of Honors:
Computer Science
Degree:
Bachelor of Science
Document Type:
Thesis
Thesis Supervisors:
Rui Zhang, Thesis Supervisor
John Joseph Hannan, Thesis Honors Advisor
Keywords:
GraphQL
Natural Language Processing
Semantic Parsing
T5
Transformer Neural Networks
SQL
Abstract:
Semantic parsing is the process of mapping language input to some formal meaning. Due to its prevalence in industry and usefulness of modeling real-world data, most existing research addresses semantic parsing to structured query language (SQL). However, many other applications of semantic parsing exist to other areas outside of SQL. In this paper, we attempt to apply the techniques of semantic parsing to SQL to a modern graph-based language, GraphQL. GraphQL can model relational data, enabling models to train using existing relational datasets with simple modification. We transform the extensive Spider dataset and apply a basic semantic parsing model to determine that this technique is possible. This approach can be generalized to other query languages provided that a comparable dataset exists.