Tag
Presents a systematic methodology for converting Hindi WordNet into 1.25 million instruction-response pairs to fine-tune a 12B-parameter language model using LoRA, demonstrating improved pedagogical effectiveness for specialized conversational systems in low-resource languages.
This paper proposes a resource-light algorithm to automatically assign part-of-speech tags to senses in the Al-Mawrid Arabic-English bilingual dictionary by transferring tags from English WordNet after disambiguation, achieving high accuracy with minimal cost.
This paper introduces a semantic-timescale analysis pipeline to study how generic vs. specific content is distributed over time in human and AI-generated speech, revealing that autocorrelation-window measures capture temporal organization of semantics beyond static lexical distributions.