Hello, I’m
Phuong Trung Le.

We study the generation of patent claim sets for a target language, jurisdiction, and examination stage. Using 884,179 patent families linked through Patent Cooperation Treaty applications, we build a Japanese–English dataset connecting Japanese and US patent records. The benchmark evaluates machine translation, prompted language models, and supervised adaptation for drafting claims that reflect the target patent office and examination stage.

This work uses neural machine translation to construct SentiWordNets: lexical resources that associate word senses with sentiment scores. It proposes ways to select translated synset members and assign their sense order, helping build sentiment resources for languages supported by machine translation.