Neural Network Machine Translation Using Supplement Sentence Retrieval
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Current machine translation technologies face challenges in achieving high accuracy, particularly when training data is insufficient, leading to errors in translating words or sentences not learned during the training process, and require substantial costs and resources to acquire large parallel corpora for effective translation.
Innovation Solution
A neural network-based machine translation method that extracts information associated with keywords from source sentences, searches for supplement sentences from databases or the internet, and combines vector values from source and supplement sentences to enhance translation accuracy, using techniques like TF-IDF and gated sum methods to determine when to incorporate additional context for improved translation.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If large parallel corpora are acquired to improve translation accuracy, then translation accuracy is improved, but costs and resources increase substantially
Solution Approach 1:
The patent introduces an intermediary mechanism (attention module and supplement sentence retrieval system) that bridges the gap between limited training data and accurate translation. The attention module acts as a mediator to selectively focus on relevant source sentence components, while the supplement sentence retrieval system retrieves additional contextual information from external sources, reducing dependence on large parallel corpora while maintaining translation accuracy.
Solution Approach 2:
The patent performs preliminary actions by retrieving supplement sentences and extracting keywords before the actual translation process. The system pre-processes the source sentence by identifying key components and retrieving relevant supplement sentences from external sources, preparing enhanced contextual information that improves translation accuracy without requiring extensive parallel corpus training.
2Reliability
If neural network model processes both source sentence and supplement sentence, then translation accuracy is improved, but computational complexity increases
Solution Approach 1:
The patent segments the translation process into distinct functional modules: an encoder for processing the source sentence, an attention module for selecting relevant components, a supplement sentence retrieval system, and a decoder for generating the translation. This segmentation allows each module to perform its specific function efficiently, managing computational complexity while improving translation accuracy through specialized processing.
Solution Approach 2:
The patent implements dynamic processing through the attention module, which dynamically adjusts its focus based on the input source sentence. The attention weights are computed dynamically to emphasize relevant source sentence components and de-emphasize irrelevant ones, allowing the system to adapt computational resources to the specific requirements of each translation task rather than processing all information uniformly.
3Reliability
If supplement sentence is always incorporated, then translation accuracy is improved, but processing time increases
Solution Approach 1:
The patent applies partial action by selectively incorporating supplement sentences only when necessary. The attention module evaluates the source sentence and determines the degree to which supplement sentences are needed, retrieving and processing only the relevant portion of supplement information rather than always incorporating complete supplement sentences, thus balancing accuracy improvement with processing time constraints.
Data Source
AI summary
A machine translation method and a machine translation apparatus using a neural network model are provided. The machine translation apparatus extracts information associated with a keyword from a source sentence, obtains a supplement sentence associated with the source sentence based on the extracted information associated with the keyword, acquires a first vector value from the source sentence and a second vector value from the supplement sentence using neural network model-based encoders, and outputs a target sentence corresponding to a translation of the source sentence based on any one or any combination of the first vector value and the second vector value using a neural network model-based decoder.