Grounded Sentence Generation With Hallucination Filtering
Find Innovative SolutionsGenerate Solutions
Solution Overview
Problem
Existing sentence generation systems, such as generative AI models, often produce hallucinations or unreliable information that is not directly based on the input or related information, reducing the reliability of generated sentences.
Innovation Solution
A sentence generation apparatus and method that includes a search unit to find related information, a generation unit to create sentences, a determination unit to assess the relation between generated sentences and related information, and a processing unit to modify or delete sentences not based on the related information, ensuring the output is reliable.
Engineering Contradictions & Design Principles
Engineering Contradiction Analysis
1Extent of automation
If generative AI models are used to automatically generate sentences, then sentence generation capability is improved, but reliability of generated information deteriorates due to hallucinations
Solution Approach 1:
The patent introduces an intermediary verification mechanism between the generative AI model and the final output. A determination unit checks whether generated sentences are based on the input sentence and related information, acting as a mediator to filter out hallucinations while preserving the automated generation capability
Solution Approach 2:
The system implements feedback by using the input sentence and related information as reference standards to verify generated sentences. The determination unit provides feedback on whether each generated sentence is grounded in the provided information, enabling reliability control in automated sentence generation
2Measurement precision
If sentences are generated based on input character string and related information, then accuracy of information is improved, but complexity of processing increases
Solution Approach 1:
The patent segments the sentence generation process into distinct functional units: a search unit for retrieving related information, a generation unit for creating sentences, and a determination unit for verifying accuracy. This segmentation enables systematic accuracy control while managing processing complexity through modular design
Solution Approach 2:
The system performs preliminary action by searching for and retrieving related information before sentence generation. This pre-prepared information base enables the generation unit to create accurate sentences with proper grounding, improving information accuracy while organizing processing steps in advance
Data Source
AI summary
A sentence generation apparatus includes processing circuitry. The processing circuitry searches for related information on an input character string. The processing circuitry generates a set of sentences based on the input character string and the related information. The processing circuitry determines a relation between the set of sentences and the related information. The processing circuitry processes a set of output sentences including the set of sentences based on a result of the determination. The processing circuitry outputs the processed set of output sentences.