ResearchBenchmarksAction Coreference Tracking on Tangram 5 utterance length Scone-derived (test)Action Coreference Tracking on Tangram 5 utterance length Scone-derived (test)Follow57.92AccuracyGraphTM31.171238.115645.0652.0044Jul 20, 2025Evaluation resultsMethodMethodLinksAccuracyGraphTMutterance length=5utterance length=52025.0757.92REINFORCEutterance length=5utterance length=52025.0737.3RANDOMERutterance length=5utterance length=52025.0737.1BS-MMLutterance length=5utterance length=52025.0732.2