Citation Context

...sitions (embedding them). In turn, embedding and atomic propositions act as arguments for discourse relations1 . Our main 1 Discourse relations, also referred to as coherence or rhetorical relations (=-=Mann and Thompson, 1988-=-), are not relevant to the shared task and, thus, we will not discuss them further in motivation in casting the problem of discourse interpretation in this structural manner is two-fold: a) to explore...

Citation Context

...cks. Our results demonstrate that such a general approach is viable and pinpoint some of its shortcomings. 1 Introduction In the past two years, largely due to the availability of GENIA event corpus (=-=Kim et al., 2008-=-) and the resulting shared task competition (BioNLP’09 Shared Task on Event Extraction (Kim et al., 2009)), event extraction in biological domain has been attracting greater attention. One of the crit...

Citation Context

...tries from the embedding dictionary ing heuristics involving the event types associated with the trigger2 . Some polarity values were assigned manually. Some strength values were based on prior work (=-=Kilicoglu and Bergler, 2008-=-), others were manually assigned. As Table (2) shows, in some cases, the semantic type (e.g., DEMONSTRATIVE, CAUSAL) is simply a mapping to the embedding categorization. In other cases, such as weakly...

Citation Context

...of the document. To obtain syntactic dependency relations, we segment documents into sentences, parse them using the re-ranking parser of Charniak and Johnson (2005) adapted to the biomedical domain (=-=McClosky and Charniak, 2008-=-) and extract syntactic 2 For example, if the most likely event type associated with the trigger is Negative regulation, its polarity is considered negative. 3 Note, however, that not all embedding pr...

Citation Context

...logical object or adjunct). Constraints on and exclusions from a rule can be defined, as shown in Table (3). We currently use about 80 such rules, mostly adapted from our previous shared task system (=-=Kilicoglu and Bergler, 2009-=-). After all the descendants of a non-leaf node are recursively processed for arguments, a semantic proposition can be composed. We define a semantic proposition as consisting of a trigger, a collecti...

Citation Context

...propositions, providing a discourse/pragmatic function. It is characterized by permeation of a limited set of discourse relations to the clausal level, often signalled lexically by “discourse verbs” (=-=Danlos, 2006-=-) (e.g., cause, mediate, lead, correlate), their nominal forms or other abstract nouns, such as role. We categorize the RELATIONAL class into CAUSAL, TEMPORAL, CORRELATIVE, COMPARATIVE, and SALIENCY t...