Post-Editing Model Segment Identification for Translation Accuracy
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current post-editing methods in machine translation have low accuracy and poor performance due to the inability to distinguish between source and target language segments, leading to inaccurate translation processing results.
Innovation Solution
A translation processing method that explicitly distinguishes source and target language segments using segment identifiers, performing embedding, vector fusion, encoding, and decoding to improve the accuracy of post-editing by considering cross-language information.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If current post-editing methods are used without segment distinction, then the processing is simpler, but the accuracy and performance of translation processing results deteriorate
Solution Approach 1:
The patent applies segmentation by dividing the translation processing into distinct segments: source language segments and target language segments. Segment identifiers are introduced to explicitly mark and differentiate these segments within the input sequence, allowing the model to process each segment type with appropriate attention mechanisms, thereby improving accuracy without overwhelming complexity
Solution Approach 2:
Segment identifiers serve as intermediaries between the raw translation text and the processing model. These identifiers act as mediators that carry structural information about segment boundaries and types, enabling the model to understand the hierarchical organization of translation components without requiring complex manual annotation or preprocessing
2Reliability
If source and target language segments are not distinguished, then the processing model is simpler, but the suggestion performance and post-editing accuracy deteriorate
Solution Approach 1:
The patent implements local quality by applying different processing characteristics to different segment types. Source language segments and target language segments receive differentiated attention and embedding treatments within the model, allowing each segment type to be processed with quality optimized for its specific linguistic properties and translation requirements
Solution Approach 2:
The patent adds a dimensional aspect to the translation processing by introducing segment identifiers as an additional layer of information. This dimensional enhancement transforms the flat sequence of tokens into a structured sequence with explicit segment boundaries and type information, enabling the model to operate in an enriched feature space that captures both linguistic and structural properties
Data Source
AI summary
Translation processing and training a post-editing model are performed. An input sequence including a plurality of segments and segment identifiers is obtained, where the plurality of segments includes a source language segment and a target language segment with a mask label, and the segment identifiers are configured to segment the source language segment and the target language segment. The mask label is located at a to-be-suggested position of the target language segment. An input vector of the input sequence is obtained by using a post-editing model based on a word vector, a position vector, and a segment vector corresponding to the input sequence. Encoding is performed by using the post-editing model based on the input vector to output an encoding result, and decoding is performed by using the post-editing model based on the encoding result to output a translation processing result corresponding to the to-be-suggested position.