Building Concept Knowledge via Bidirectional Translation
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing lexical knowledge bases, such as WordNet, provide insufficient concept knowledge for specific application domains, failing to meet detailed requirements in practical applications, particularly in information retrieval and natural language processing, due to data sparseness and limited context coverage.
Innovation Solution
A computer system utilizing a machine-readable dictionary with bilingual or multilingual capabilities, enabling forward and backward translation to build concept knowledge by inputting a seed word and iteratively translating and merging related words across languages to expand the concept knowledge set, allowing users to select and delete irrelevant terms for improved specificity.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Adaptability or versatility
If lexical knowledge bases such as WordNet are used to extract concept knowledge, then the system can provide general-purpose concept knowledge, but the concept knowledge is insufficient in details for particular application domains
Solution Approach 1:
The patent applies local quality by transitioning from general-purpose concept knowledge to domain-specific concept knowledge. The system extracts concept knowledge tailored to particular application domains (e.g., medical, legal, technical fields) from domain-specific corpora, ensuring that the concept knowledge has the necessary detail and relevance for each specific domain while maintaining the ability to handle multiple domains through selective corpus extraction.
2Productivity
If statistical natural language processing method is used to determine word probabilities, then the system can process language data, but data sparseness occurs when word combinations appear with low frequencies or do not appear in the training corpus
Solution Approach 1:
The patent uses concept knowledge as an intermediary to bridge the gap between statistical NLP and domain-specific requirements. Instead of relying solely on raw frequency counts from training corpora, the system uses concept knowledge to represent and relate words and phrases, enabling probability estimation for low-frequency or absent word combinations by leveraging semantic relationships and contextual information from the concept knowledge structure.
3Productivity
If keyword-based search systems are used, then the system can retrieve information efficiently, but concept knowledge helps users construct and modify search queries which adds complexity to the search process
Solution Approach 1:
The patent implements self-service by enabling the search system to automatically utilize concept knowledge for query expansion and modification without requiring manual user intervention. The system autonomously extracts relevant concept knowledge, generates expanded search queries, and performs query modification based on the input query and domain context, thereby maintaining high retrieval efficiency while reducing the manual complexity burden on users.
Data Source
AI summary
This invention relates to a computer system and methods, apparatus and systems for building concept knowledge from a machine-readable dictionary. The machine-readable dictionary includes a plurality of words in a first language and a plurality of corresponding translated words in a second language, and a plurality of words in the second language and a plurality of corresponding translated words in the first language. The method comprises steps of providing a seed word in the first language; forward-translating said seed words to obtain a plurality of translated words corresponding to said seed word by looking up said machine-readable dictionary; and backward-translating said translated words to obtain a plurality of translated words in the first language corresponding to each of said plurality of translated words obtained by said step of forward-translating respectively, as words of the concept knowledge, by looking up said machine-readable dictionary.