A neural network-based accurate translation method for legal terms

By constructing a logical dependency chain graph of legal texts and introducing a logical constraint prediction head into the translation model, the problem of the lack of explicit modeling of logical relationships between clauses in existing technologies is solved, thereby improving the logical coherence and consistency of legal text translation.

CN122287658APending Publication Date: 2026-06-26LIAONING UNIVERSITY
View PDF 0 Cites 0 Cited by

Patent Information

Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
LIAONING UNIVERSITY
Filing Date
2026-03-30
Publication Date
2026-06-26

AI Technical Summary

Technical Problem

Existing neural network-based legal text translation methods lack the ability to explicitly model the logical references and dependencies between clauses, resulting in defects in the logical coherence and system consistency of the translation results.

Method used

By identifying legal terminology units and citation markers, a logical dependency chain graph is constructed to form an extended context sequence. A logical constraint prediction head is introduced into the neural network translation model to dynamically inject the constraints of the logical dependency chain to ensure the accurate transmission of logical semantics.

Benefits of technology

It significantly improves the logical coherence and systematic consistency of legal text translation, ensuring the usability and reliability of the translation results in a rigorous legal context.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN122287658A_ABST
    Figure CN122287658A_ABST
Patent Text Reader

Abstract

This invention discloses a method for accurate translation of legal terminology based on neural networks, belonging to the field of natural language processing technology. Specifically, it includes: first, identifying legal terminology units and article citation markers in the source language legal text, and constructing logical dependency chains by parsing the citations; then, fusing each terminology unit with its local context and the remote article content associated with the dependency chain to form an extended context sequence; this sequence is input into a neural network translation model integrating a logical constraint prediction head; during decoding and generation, the prediction head determines whether word generation should be subject to logical constraints, and if constrained, extracts target language constraints from the dependency chain, injecting them into the probability distribution of the decoder in the form of a dynamic bias vector, thereby guiding the generation; finally, it outputs the target language translation and the association annotations between the translated words and the source language logical elements; this invention achieves the preservation and accurate transmission of the cross-article logical structure of legal text.
Need to check novelty before this filing date? Find Prior Art