@inproceedings{rajpoot-2026-tifin,
    title = "Tifin {I}ndia at {S}em{E}val-2026 Task 5: Semantic Bridge: Augmented Encoding for Word Sense Plausibility",
    author = "Rajpoot, Pawan",
    editor = "Kochmar, Ekaterina  and
      Ghosh, Debanjan  and
      North, Kai  and
      Komachi, Mamoru",
    booktitle = "Proceedings of the 20th {I}nternational {W}orkshop on {S}emantic {E}valuation (2026)",
    month = jul,
    year = "2026",
    address = "San Diego, California, USA",
    publisher = "Association for Computational Linguistics",
    url = "https://preview.aclanthology.org/ingest-acl-workshops/2026.semeval-1.388/",
    pages = "3095--3102",
    ISBN = "979-8-89176-414-9",
    abstract = "We present a hybrid system for SemEval 2026Task 5: Rating Plausibility of Word Senses inAmbiguous Stories. Our approach reframesLLMs as feature generators rather than directpredictors. We combine two subsystems: onethat appends LLM-generated hints to the in-put context and trains an encoder-based regres-sion model, and another that feeds structuredhints from multiple LLM configurations into alightweight regression ensemble. We generatemultilingual enrichments to probe LLMs forcomplementary signals and take advantage ofthe fact that translation into certain languagesimplicitly disambiguates word senses, makingthe encoder more robust. The 50/50 ensem-ble achieves 859/930 (92.37{\%}) accuracy withSpearman {\ensuremath{\rho}}= 0.8384 on the test set, exceed-ing the estimated human ceiling of 89.2{\%}. Thesame LLM enrichments, processed through fun-damentally different paradigms (tabular regres-sion vs. full-text encoding), produce comple-mentary errors that cancel under ensembling.Notably, simple 50/50 averaging captures thisgain without any learned combiner, suggest-ing that"
}