ResearchBenchmarksAction Coreference Tracking on Tangram 3 utterance length Scone-derived (test)Follow68.5AccuracyREINFORCE62.36463.95765.5567.143Jul 20, 2025Evaluation ResultsMethodMethodLinksAccuracyREINFORCEutterance length=3utterance length=32025.0768.5RANDOMERutterance length=3utterance length=32025.0765.8GraphTMutterance length=3utterance length=32025.0764.14BS-MMLutterance length=3utterance length=32025.0762.6