Natural language inference method and device fused with semantic analysis

Abstract

The invention provides a natural language inference method and device fused with semantic analysis. The method comprises the following steps: acquiring a premise T and a corresponding hypothesis Q; respectively carrying out semantic analysis on the premise T and the hypothesis Q to obtain a plurality of features; carrying out unification and feature fusion on the plurality of features to obtain a fusion feature input; and inputting the fusion feature input into a JudgeNLIModel model, and outputting an answer corresponding to the hypothesis. According to the method, natural language understanding is combined, and specific features in the premise and the hypothesis are extracted, so that the relation between the premise and the hypothesis is deduced, the accuracy of natural language deduction is improved, and an improved reference direction is provided for a subsequent natural language deduction method.

Description

technical field [0001] The present invention relates to the technical field of language recognition, in particular to a natural language inference method and device integrated with semantic analysis. Background technique [0002] Natural Language Inference (NLI), also known as textual entailment recognition, is one of the most important problems in natural language processing, which requires inferring the logical relationship between two given sentences. The task is given two sentences, called premises and assumptions, and the goal is to determine whether the logical relationship between them is necessary, neutral, or contradictory. [0003] Traditional natural language inference methods include text similarity, text alignment, logical calculation, and text conversion. Text similarity is to calculate the similarity between the premise and the hypothesis to determine whether they constitute an implication relationship; text alignment is to align the similar parts of the prem...

AI Technical Summary

Problems solved by technology

[0004] However, the above methods rely heavily on transformation rules, some of which come from the knowledge base and some from the corpus

In addition, traditional natural language inference methods, such as text similarity, cannot find the difference between the numerical comparisons of two texts when given a premise and associated assumptions

For example: the premise is "If you are taller than 1.3 meters, you need to buy a full-price ticket", the question is "Don't buy a full-price ticket if your height is less than 1.2 meters", there is no obvious difference in the text, only a difference in the number, in many cases the model will misjudge that the two are the same

Images