Semantic parsing method and semantic parser

Abstract

The invention discloses a semantic parsing method and a semantic parser. The method comprises: receiving a natural sentence to be parsed; determining a canonical sentence corresponding to the natural sentence to be parsed; inputting the canonical sentence into a pre-trained naive semantic parser to obtain a logical expression corresponding to the canonical statement. In the embodiment of the present invention, the natural sentences are converted into canonical sentences with the same meaning before parsing, and the naive semantic parser (trained from the canonical sentence-logical form sample set in advance) is used for parsing, so that the The parsing of natural sentences is transformed into the parsing of canonical sentences with the same meaning, which solves the problem of poor generalization performance that directly uses naive semantic parser to parse natural sentences.

Description

technical field [0001] The invention relates to the technical field of artificial intelligence, in particular to a semantic parsing method and a semantic parser. Background technique [0002] Semantic parsing refers to the task of converting natural language questions into logical forms. Logical form is a structured semantic expression, usually an executable statement, such as Lambda expression and SQL query language, which can be directly executed by the program, retrieved from the database and return the answer. Due to the tight coupling with knowledge bases, semantic parsing is often used in the field of automatic question answering based on knowledge graphs or databases. [0003] To build a semantic parser in an entirely new domain, researchers need to first acquire a large amount of training data, usually starting from writing template rules for tuples (canonical questions, logical forms). [0004] Template rules: Grammar rules manually written by researchers that map...

AI Technical Summary

Problems solved by technology

[0006] However, because only the corpus generated by template rules is used, and due to the obvious difference in data distribution between canonical sentences and natural sentences, the trained naive semantic parser is less effective on real questions (natural language questions), and the generalization performance not good

Images