Text inference method based on limited semantic dependency analysis

Abstract

The invention discloses a text inference method based on limited semantic dependency analysis. The method comprises the following steps of: 1, according to a semantic dependency representation mechanism, establishing a Chinese text inference marking resource, wherein a text inference marking instance comprises a text T, a hypothesis H, a semantic dependency graph of the hypothesis and an inference type; 2, on the basis of the Chinese text inference marking resource, performing semantic dependency analysis on the hypothesis H which is newly input under the limitation of the text T which is newly input so as to judge whether the text T can infer the hypothesis H; and 3, evaluating, analyzing and summarizing an analysis process and a judgment result of the step 2, and improving the performance of the analysis process with feedback. By adoption of the method, the judgment of text inference is formalized into the problem of the limited semantic dependency analysis, and the text inference extends from entailment to preset and implication in type. The method contributes to processing of the inference of discourse and dialogue corpora which are relatively complicated.

Description

technical field [0001] The invention belongs to the field of natural language processing, in particular to a text reasoning method based on limited semantic dependency analysis. Background technique [0002] In recent years, the research on textual reasoning has received extensive attention in the field of natural language processing, and textual reasoning has transformed from the earliest purely theoretical discussion to a large-scale theoretical research group and empirical platform construction. From 2005 to 2007, the EC research platform PASCAL organized three text reasoning competitions RTE (Recognizing Textual Entailment), which is now organized by NIST (National Institute of Standards and Technology, the United States National Institute of Standards and Technology). So far, the text reasoning competition RTE series evaluation competition has been held for 6 sessions. The Association for Computational Linguistics (ACL) has also organized several seminars to discuss th...

AI Technical Summary

Problems solved by technology

[0019]First, as far as the classification strategy is concerned, the two categories of entailment and non-implication of text reasoning are relatively complex, and the similarity between instances is difficult to guarantee. Their distinction The characteristics are not easy to determine, so the performance of the classifier established based on it is not ideal; in terms of deep semantic analysis and reasoning strategies, the acquisition of reasoning knowledge is the main bottleneck. Without sufficient reasoning knowledge support, the performance of the reasoning process is difficult. ensure

[0020]Second, the transformation strategy based on implication rules is the main strategy of text reasoning at present, and the core of deep semantic analysis and reasoning strategies and performance-driven strategies is also the automatic discovery of implication rules , but the overall performance of automatic discovery of entailment rules needs to be improved

[0021]Third, global factors, such as resolution of anaphora, are generally used as a prerequisite for judging implication relations, and their errors may spread in subsequent operations[8- 10]

According to the previous analysis report of RTE (Recognizing Textual Entailment), RTE1 has 17 submission systems with an accuracy rate between 50% and 60%; RTE2 has 23 submission systems with an accuracy rate between 49% and 80%. between 45% and 74% of the 26 submitted systems in RTE3; most of the submitted systems in RTE5 and RTE6 are below 75% accurate

Images