Machine Translation Model Using GAN and Dictionary Data
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Conventional machine-translation models using deep learning face difficulties in learning correspondence between significantly different languages, leading to poor translation quality and prolonged training times due to large differences in vector spaces.
Innovation Solution
A computing device and method that incorporates a generative adversarial network (GAN) with dictionary data to project source and target languages, utilizing a generator and discriminator network to reduce the distance between languages, enabling faster and more accurate translation by inputting source sentences and dictionary data to generate target sentences and calculating errors for network training.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If conventional deep learning projects source and target languages to vector space, then translation can be performed, but the large distance between significantly different languages makes it difficult to learn correspondence efficiently
Solution Approach 1:
The patent introduces an intermediary mapping layer that projects both source and target languages into a shared intermediate vector space before final translation. This intermediary space acts as a bridge that reduces the distance between significantly different languages, enabling more efficient learning of language correspondence while maintaining translation quality.
Solution Approach 2:
The translation model is segmented into multiple components: source language embedding, intermediate projection layer, target language projection, and output layer. This segmentation allows the intermediate layer to specifically handle the language distance problem while other layers focus on their specialized functions, improving overall training efficiency.
2Productivity
If dictionary data is incorporated into training, then convergence speed improves, but system complexity increases
Solution Approach 1:
The patent performs preliminary actions by pre-processing dictionary data and creating initial word embeddings before main training begins. This preliminary preparation organizes the vocabulary and semantic relationships in advance, allowing the training process to converge faster without requiring complex real-time processing during training.
Solution Approach 2:
The system uses dictionary data to create self-supervised learning signals that enable the model to learn from its own predictions and corrections. The dictionary provides inherent structure that the model can exploit without external intervention, improving convergence while avoiding additional system complexity.
Data Source
AI summary
A device and a method for generating a machine translation model and a machine translation device are disclosed. The device inputs a source training sentence of a source language and a dictionary data to a generator network so that the generator network outputs a target training sentence of a target language according to the source training sentence and the dictionary data. Then, the device inputs the target training sentence and a correct translation of the source training sentence to a discriminator network so as to calculate an error between the target training sentence and the correct translation according to the output of the discriminator network, and trains the generator network and the discriminator network respectively. The trained generator network is the machine translation model.