Method and device for processing feature templates for syntactic analysis

Abstract

The invention provides a method and device for processing feature templates used for syntactic analysis. The method includes: obtaining a set of first feature templates, wherein the set of first feature templates includes basic templates and combined templates, and each of the combined templates includes at least two basic templates; from the first feature templates Extract at least one basic template from the set; respectively use the at least one basic template to decode the training corpus, and count the number of first feature values ​​extracted from the training corpus for each of the basic templates; A set of feature templates deletes only the basic templates that extract a single first feature value and the basic templates included in the combined templates. By deleting the feature templates that do not affect the test accuracy from the set of feature templates, the feature templates are automatically screened to obtain a set of feature templates with good versatility.

Description

technical field [0001] The present invention relates to the technical field of natural language processing, in particular to a method and device for processing feature templates for syntactic analysis. Background technique [0002] Dependency analysis is currently the most widely used method of syntactic analysis. The syntactic relationship between two words is represented by a dependency relationship, such as figure 1 As shown, it is an example diagram of the sentence dependency relationship in the syntactic analysis of a sentence. For "male protagonist" and "who", "Dragon and Babu" is dependent on "de", and at the same time, the clauses formed by these two words are dependent on "male protagonist" in the relationship of modifier (ATT), and "yes" is The core word (HED) of this sentence, "ROOT" is a virtual node, representing the root of the syntax tree, representing the root of the syntax tree, each sentence will add this node, the specific dependency analysis method can r...

AI Technical Summary

Problems solved by technology

Generally, the feature screening of dependency analysis refers to the screening of feature templates. Existing feature screening methods use iterative increase to improve the accuracy of feature template methods, which have the following disadvantages: 1) The speed is slow, and it takes a long time to judge whether a feature template is retained. At least one experiment, and the number of feature templates in the general dependency analysis system is relatively large, so a large number of experiments are required to screen; 2) The versatility is poor, and the feature templates screened on one corpus cannot be directly transplanted to the new corpus

Images