MUMBAI, India, Aug. 31 -- Intellectual Property India has published a patent application (202641100514 A) filed by Ms. Ilakkiya. B; and Ms. Hemanandhini. I. G on August 20, 2026, for Explainable Hallucination Detection With Dual-Stream Verification For Large Language Models.
Inventors include Ms. Ilakkiya. B; and Ms. Hemanandhini. I. G.
The application for the patent was published on August 28, 2026, under issue no. 35/2026.
Abstract: Explainable hallucination detection with dual-stream verification for large language models Abstract: The present invention addresses the explainable hallucination detection of Large Language Models (LLMs) for detecting and authenticating of dubious claims produced by LLMs. This invention utilizes a two-stream verification scheme: (i) semantic-contextual verification stream, and (ii) factual-evidence verification stream. The former verifies the consistency of the input prompt, output response and the contextual coherence of the output response while latter verifies the given statement by contrasting with one or more reliable knowledgebases. The latter output together with the former are synthesized to produce an individual claim hallucination probability and confidence score. The explainability module points out possible hallucinated claims and produces human interpretable explanations about why some claims were considered hallucinated: context inconsistency, lack of supportive evidence, or evidence contradiction. Finally, this invention produces a structural verification result with a reliability status for each claim analyzed.
Disclaimer: Curated by HT Syndication.
