Machine Translation Accuracy via Bidirectional Feedback Loop
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine translation systems face challenges in achieving high accuracy, especially when translating sentences outside their predefined domains, and users lack easy methods to verify the meaning of translated sentences, particularly when translating into minor languages, leading to potential miscommunication.
Innovation Solution
A machine translation method that generates multiple forward-translated and backward-translated sentences, allowing users to select the most appropriate sentence, which is then output, enabling feedback for machine learning to improve translation accuracy without increasing computer resources or development costs.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple forward-translated and backward-translated sentences are generated to improve translation accuracy, then translation reliability is improved, but device complexity increases
Solution Approach 1:
The translation verification process is segmented into multiple independent translation steps (forward translation and backward translation), where each step generates separate translated sentences that can be independently evaluated. This segmentation allows the system to compare multiple translation paths without requiring a complete system redesign, thereby improving reliability while managing complexity.
Solution Approach 2:
The system implements feedback by generating backward-translated sentences from forward-translated sentences and comparing them with the original sentences. User selections of preferred backward-translated sentences provide feedback that is used to improve the translation model, creating a closed-loop system that continuously enhances translation accuracy.
2Reliability
If user feedback is collected through selection of backward-translated sentences, then translation accuracy is improved, but ease of operation deteriorates
Solution Approach 1:
The system performs automatic evaluation by comparing forward-translated and backward-translated sentences without requiring extensive user intervention. The machine automatically generates multiple translation variants, evaluates them through backward translation, and presents sorted results to the user, allowing the system to serve itself in the evaluation process while minimizing user burden.
Solution Approach 2:
The system collects minimal feedback from users (simple selection of preferred sentences) and uses this feedback automatically to improve translation accuracy through model retraining, transforming a potentially complex interaction into a simple feedback loop that benefits both accuracy and ease of use.
3Measurement precision
If multiple translation sentences are generated and evaluated, then measurement precision of translation quality is improved, but loss of time increases
Solution Approach 1:
The system performs preliminary action by pre-generating multiple forward-translated sentences and their corresponding backward-translated sentences before final evaluation. This preliminary generation allows the system to have multiple candidate translations ready for comparison, improving measurement precision without requiring time-consuming evaluation during the final selection phase.
Solution Approach 2:
The system skips detailed manual evaluation of each translation sentence by using automatic backward translation and comparison algorithms. Instead of time-consuming manual review, the system rapidly processes multiple translation candidates through automated evaluation and sorting, maintaining measurement precision while significantly reducing processing time.
Data Source
AI summary
A machine translation method is provided, in which a plurality of different forward-translated sentences is generated resulting from translation of a received translation-source sentence in a first language into a second language. A plurality of backward-translated sentences is generated resulting from backward translation of the plurality of different forward-translated sentences into the first language. As translation-source sentence, voice information or text information is received. The voice information is received via a microphone that receives a voice input performed by a user. The text information is received via a text input device that receives a text input performed by a user. An output form of the forward-translated sentence corresponding to the selected backward-translated sentence is changed in accordance with which of the voice information and the text information is received as the translation-source sentence.