Target-Side Paraphrase Augmentation for Sign Language Translation with Large Language ModelsWorkshop paper

Pedro Dal Bianco, Jean Paul Nunes Reinhold, Oscar Stanchi, Facundo Quiroga, Franco Ronchetti, Ulisses Brisolara Corrêa

GenSign Workshop @ CVPR 2026

A large language model (GPT-4o) writes paraphrases of the target sentences to augment sign language translation data: pre-train on the augmented corpus, then fine-tune on the original references. Evaluated on PHOENIX14T, GSL and LSA-T, with lexical metrics and an LLM-as-a-Judge semantic evaluation.