{"id":"d7cca566-4c8a-415d-9508-55f4b7c85790","arxiv_id":"2504.12020","paper_version":2,"verdict":"CONDITIONAL","confidence":"HIGH","novelty_score":6.0,"correctness_risk":"medium","formal_verification":"none","parameter_count":6,"one_line_summary":"MixSignGraph, with local, temporal, and hierarchical graph modules plus text-driven CTC pretraining, achieves state-of-the-art results on multiple sign language recognition and translation benchmarks.","lead":"This paper introduces MixSignGraph, a graph-based neural network for sign language recognition and translation that uses three graph modules to capture sign-related features across regions, frames, and feature scales. It also proposes a text-driven CTC pretraining method that improves gloss-free sign language translation by generating pseudo gloss labels from text.","discovery_kind":"extension","skeptic_critique":null,"referee_report":null,"author_rebuttal":null,"desk_editor":null,"rs_alignment":null,"lean_confirmation":null,"pith_extraction":null,"created_at":"2026-08-16T12:39:42.647070+00:00","model_set":{"reader":"deepseek-v4-flash"},"falsifier":null,"supporting_citations":[],"review_version":1}