作者: Jérémy Perret , Stergos Afantenos , Nicholas Asher , Mathieu Morey
DOI: 10.18653/V1/N16-1013
关键词:
摘要: In this paper we present the first, to best of our knowledge, discourse parser that is able predict non-tree DAG structures. We use Integer Linear Programming (ILP) encode both objective function and constraints as global decoding over local scores. Our underlying data come from multi-party chat dialogues, which require prediction DAGs. dependency parsing paradigm, has been done in past (Muller et al., 2012; Li 2014; Afantenos 2015), but formal framework SDRT exploit SDRT's notions left right distributive relations. achieve an F-measure 0.531 for fully labeled structures beats previous state art.