A method, system, terminal and storage medium for generating synonymous sentences

Abstract

The invention provides a method, system, terminal and storage medium for generating synonymous sentences. The method includes: constructing a basic model for generating synonymous sentences; performing self-encoding on the sentences to be converted and sentences in all training corpus to obtain semantic vector and syntax vector; performing variational inference on the syntax vector to obtain the syntax vector after variational inference; decoupling the semantic vector and the syntax vector after variational inference; using the decoupled semantic vector and The grammatical vector after variational inference reconstructs the sentences in the training corpus and updates the training corpus; uses the updated training corpus to re-execute decoupling and optimize the vector decoupling; uses the semantic vector of the source text to splice the variation The inferred grammar vectors generate synonymous sentences. The invention decouples grammar and semantics, so that the grammar of the generated sentences is more controllable, and at the same time, the generated sentences have stronger semantic consistency.

Description

technical field [0001] The present invention relates to the technical field of text generation in natural language processing, in particular to a method, system, terminal and storage medium for generating synonymous sentences based on decoupling of semantics and grammar. Background technique [0002] Synonymous sentence generation is an important issue in the field of text generation, which refers to generating texts with the same or similar semantics for any given text. The importance of synonymous sentence generation is reflected not only in expressing the diversity of human language, but also in its role as a key component of other natural language processing tasks, such as question answering, machine translation, data augmentation and other tasks. Traditional models are usually rule-based, they replace a certain word or phrase in the input text, first find candidate synonyms or phrases from thesaurus dictionary, and then use synonyms or phrases to replace accordingly. A...

AI Technical Summary

Problems solved by technology

However, although the sequence-to-sequence learning framework based on the attention mechanism improves the effect of synonymous sentence generation compared with the traditional rule model, there are still two problems in this framework: (1) the model often pays too much attention to the original sentence and the generated sentence. The alignment of local information cannot guarantee that the semantics of the generated sentence are consistent with the original sentence; (2) The generation grammar of synonymous sentences is uncontrollable. When there are multiple synonymous sentences (the number is greater than 2), the generation process should be the original The semantic information of the target sentence is added to the grammatical information of the target sentence to generate the corresponding synonymous sentence

Images