Phrase rule extracting method based on combination

Abstract

The invention relates to a phrase rule extracting method based on combination. The phrase rule extracting method comprises the following steps of: configuring a 'minimum phrase rule' in a bilingual corpus; configuring a combined phrase rule set through the combination; generating a minimum phrase rule set in a given bilingual parallel corpus and storing into a hash data structure; configuring a combined phrase rule and judging that the combined phrase rule is formed by several minimum phrase rules through the minimum phrase rule set; if the phrase rule in the combination is formed by less than or equal to n minimum phrase rules in the minimum phrase rule set, putting the phrase rule into a new hash data structure; and outputting a new minimum phrase rule set and phrase rules of the combined phrase rule set, and finishing one time of a phrase rule extracting process based on the combination. According to the phrase rule extracting method based on the combination disclosed by the invention, a high-quality phrase rule set containing abundant contextual information is effectively generated; and under the condition that a translation performance is not reduced, the phrase rule set extracted by the method is reduced by 56.5% when being compared with the phrase rule set extracted by a standard method.

Description

technical field [0001] The invention relates to a phrase processing technology in a phrase-based statistical machine translation system, in particular to a combination-based phrase rule extraction method. Background technique [0002] Phrase-based statistical machine translation systems have demonstrated very competitive performance in the field of machine translation. A large part of the reason why the phrase-based approach works is that it relies on a high-quality phrase rule set. In a phrase rule set, each source language phrase is mapped to one or more different target language phrases. In the phrase system, a phrase is composed of a series of consecutive words, and the phrase has no linguistic meaning. At present, some researchers in the field of machine translation have proposed some effective phrase rule extraction methods. In these phrase rule extraction methods, heuristic methods have been widely used. This extraction method extracts all the phrase rules that ar...

AI Technical Summary

Problems solved by technology

[0006] However, the benchmark phrase rule extraction method has an inevitable problem, that is, during the rule extraction process, the phrase length needs to be mechanically adjusted to obtain the optimal phrase rule set

The extracted phrase rule table is very large, takes up a lot of hard disk space, and contains a lot of noise data

Images