Bootstrapping named entity canonicalizers from English using alignment models

摘要

Methods, systems, and apparatus, including computer programs encoded on a computer storage medium, for training recognition canonical representations corresponding to named-entity phrases in a second natural language based on translating a set of allowable expressions with canonical representations from a first natural language, which may be generated by expanding a context-free grammar for the allowable expressions for the first natural language.

主权项

1. A computer-implemented method, the method comprising:
receiving a set of acceptable expressions, each acceptable expression being a string that identifies a value of a variable entity in a first natural language, each acceptable expression being associated with a canonical representation of the value identified by that expression; performing, a first machine translator that translates expressions from the first natural language to a second natural language, machine translation on each acceptable expression in the first natural language to obtain a translated expression of the acceptable expression in the second natural language; associating the canonical representation associated with each acceptable expression with the corresponding translated expression in the second natural language; providing a set of training data for training a second machine translator that translates expressions in the second natural language that each include a respective translated expression to expressions in the second natural language that each include a respective canonical representation, the set of training data comprising the translated expressions and the canonical representations that are associated with the translated expressions; and using the second machine translator to translate a particular expression that includes a particular translated expression into a particular translated expression that includes a particular canonical representation.