The patent badge is an abbreviated version of the USPTO patent document. The patent badge does contain a link to the full patent document.
The patent badge is an abbreviated version of the USPTO patent document. The patent badge covers the following: Patent number, Date patent was issued, Date patent was filed, Title of the patent, Applicant, Inventor, Assignee, Attorney firm, Primary examiner, Assistant examiner, CPCs, and Abstract. The patent badge does contain a link to the full patent document (in Adobe Acrobat format, aka pdf). To download or print any patent click here.
Patent No.:
Date of Patent:
Jun. 16, 2026
Filed:
Oct. 31, 2024
Oracle International Corporation, Redwood Shores, CA (US);
Trung-Dung Hoang, Lausanne, CH;
Giulia Carocari, Zurich, CH;
Moein Owhadi Kareshk, Burnaby, CA;
Hesam Fathi Moghadam, Sunnyvale, CA (US);
Rhicheek Patra, Zurich, CH;
Sungpack Hong, Palo Alto, CA (US);
Hassan Chafi, Zurich, CH;
Oracle International Corporation, Redwood Shores, CA (US);
Abstract
Here is multitask finetuning of natural language (NL) interaction (NLI) to increase semantic accuracy of database statement generation. A computer associates a first natural language request with a correct database statement for a database schema. The correct database statement contains multiple distinct or repeated identifiers. A large language model (LLM) predicts, from the first natural language request and a strict subset of the database schema, multiple predicted identifiers. Finetuning the LLM entails neural backpropagation, into the LLM, of a loss that is based on a comparison of: a) identifiers in the correct database statement to b) the predicted identifiers. After finetuning, the LLM inferentially generates an inferred database statement from a second natural language request, and this statement is accurate even if a new (i.e. previously unseen) database schema is involved.