Neural Machine Translation Correction Using Similar Source Sentences
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing neural machine translation (NMT) technologies suffer from incorrect translations due to neglecting the information of semantically similar sentences, leading to increased translation costs and inefficiencies.
Innovation Solution
A translation method that utilizes semantically similar source sentences to correct translations by determining target words based on probabilities of translation in both the original and similar sentences, without requiring a new encoder, thereby improving translation accuracy and efficiency.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Measurement precision
If neural machine translation is used for translation, then translation quality is improved, but incorrect translations still occur due to neglecting semantically similar sentences
Solution Approach 1:
The patent retrieves semantically similar source sentences and their correct translations beforehand, then uses this pre-acquired information to correct potential translation errors. By performing the retrieval and preparation of correction data in advance, the system can efficiently identify and correct incorrect translations without requiring complex real-time analysis.
Solution Approach 2:
The patent implements a feedback mechanism where the translation system uses semantically similar sentences and their known correct translations to provide corrective feedback. The system compares the initial translation with translations of similar sentences and uses this feedback to identify and correct errors, thereby improving translation reliability while maintaining quality.
2Measurement precision
If additional encoders are introduced to encode similar translated sentences, then translation accuracy is improved, but device complexity and translation costs increase
Solution Approach 1:
The patent makes the existing encoder multi-functional by using it to encode both the source sentence and the semantically similar source sentences. Instead of introducing a separate encoder for similar sentences, the same encoder handles both tasks, thereby improving translation accuracy through additional contextual information while avoiding increased device complexity.
Solution Approach 2:
The patent uses the existing encoder structure and applies it to similar sentences rather than creating new encoding components. By copying the encoding approach to handle similar sentences with the same encoder, the system achieves improved accuracy without the overhead of additional encoder complexity.
3Measurement precision
If semantically similar sentences are retrieved and used for correction, then translation accuracy is improved, but translation processing time increases
Solution Approach 1:
The patent retrieves only a limited number of semantically similar sentences (e.g., top-k similar sentences) rather than all possible similar sentences. This partial action approach provides sufficient correction information to improve accuracy while limiting the processing time and computational resources required, thus balancing accuracy improvement with time efficiency.
Data Source
AI summary
A translation method, a translation apparatus, a translation device, and a storage medium are provided. The method includes: firstly, determining the second source sentence semantically similar to the first source sentence, then determining the target source word used in both of the first source sentence and the second source sentence, and if translated words for the target source word in the first source sentence and the second source sentence are different, determining the target translated word for the target source word according to the probability of the target source word being translated into the first translated word or the second translated word. Thus, the translation method uses not only the information of a second translated sentence for the second source sentence but also the information of the second source sentence. The translation of the first source sentence is corrected according to the information of similar words.