Semantic Annotation of CESS-ECE: Lexical Semantic Structures

The correspondence between syntactic functions and thematic roles is carried out following the predicate analysis presented in Levin & Rappaport-Hovav [9] and Rappaport-Hovav & Levin [12]. We consider that their proposal is appropriated for our work for a number of reasons. First, because in this model converge lexical semantic, event and argument information and diathesis alternations. And second, because similar works in corpus and computational linguistics have been carried out following this approach, such as the construction of VerbNet (Kipper et al. [7]), a lexicon with lexical semantic, argument and diathesis information for English predicates. VerbNet follows Levin’s semantic classification (Levin [8]) and adopts PropBank semantic annotation (Palmer et al. [11]). We characterize predicates by means of a limited number of LSS and Event Structure Patterns, according to the four basic event classes: states, activities, accomplishments, and achievements (Vendler [17], Dowty [4]). These general classes can be split in subclasses, as we will see in section 3. Thematic roles are determined by the event class that the predicate belongs to and the type of diathesis alternation that the predicate presents. Thus, not only thematic roles are assigned, but also predicates are characterized both from the aspectual and argument perspective. In fact, the semantic classes determine the mapping between syntactic functions and semantic roles. This information is currently being stored in the lexicon CESS-LEX for both languages.