A discourse-annotated corpus of conjoined VPs

Abstract

English grammars indicate a variety of relations holding between conjoined VPs. VPs conjoined by and evince such senses as Result, Temporal Sequence and Concession. Although all these senses are ones associated with discourse relations, conjoined VPs have not been fully included in discourse annotation. Because of the value of discourse-annotated corpora for developing approaches to automated sense recognition, we have added their annotation to the Penn Discourse TreeBank. This paper describes how tokens were identified; how the process of span and sense annotation was modified and extended in order to keep the annotation of intra-sentential multi-clausal structures consistent with the rest of the corpus; and what the resulting corpus looks like, in terms of token frequency and common sense patterns.