Word Vector Generation for Multi-Corpus Translation Feature Retention
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing language processing techniques that use multiple corpuses to improve translation quality and response time often lose features of unselected corpuses or fail to reflect features of integrated corpuses in search results.
Innovation Solution
A generating device and method that combines word vectors from multiple corpuses using pre-generated corpus models, allowing features from each corpus to be retained and reflected in search results by generating a new vector based on the combined vectors.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Reliability
If multiple corpuses are integrated into a single corpus model, then translation quality improves, but features of individual corpuses are lost
Solution Approach 1:
The patent segments the corpus processing into multiple independent corpus models, each retaining features of individual corpuses. Instead of creating one integrated corpus model that loses individual features, the system maintains separate models (first corpus model, second corpus model, etc.) that can be independently processed and combined later through vector operations, thus preserving features while achieving integrated translation quality.
Solution Approach 2:
The patent implements a nested structure where multiple corpus models are embedded within a unified processing framework. Each corpus model contains features of its source corpus, and these nested models are combined through vector operations to produce translation results that benefit from all corpus features without losing individual characteristics.
2Productivity
If a single corpus model is used for processing, then processing speed improves, but translation accuracy decreases
Solution Approach 1:
The patent performs preliminary actions by pre-processing each corpus into separate corpus models with pre-computed word vectors before actual translation processing. This allows the system to quickly retrieve and combine pre-processed features during translation, maintaining high processing speed while utilizing multiple corpuses for improved accuracy.
Solution Approach 2:
The patent transitions from a single-dimension corpus model to a multi-dimensional vector space where each corpus contributes vectors along different dimensions. The vector combination process integrates features from multiple corpuses in this extended dimensional space, achieving both speed (through efficient vector operations) and accuracy (through comprehensive feature integration).
Data Source
AI summary
A non-transitory computer-readable recording medium stores therein a generating program that causes a computer to execute a process including: receiving a word; generating a first and a second vectors according to the received word by applying a first and a second conversion parameters each to the received word; and generating a new third vector according to the word based on the generated first and second vectors.