Automatic Semantic Role Annotation for Spanish

This paper describes and evaluates the automatic annotation of clause-level complements with semantic roles in a Spanish Web corpus, using a ruleand dependency-based approach. In all, 52 different role tags, like agent (§AG), experiencer (§EXP), location (§LOC) etc. are distinguished. The annotator uses a role grammar of 568 hand-written Constraint Grammar rules that take as input the syntactic analysis of the HISPAL parser. A rough evaluation of 5000 running words was performed, where the role annotation achieved an F1 of 81,6% on raw text and 90,0% on syntactically revised input. A Spanish Internet corpus of 11.2 million words has been compiled and automatically annotated with our semantic role grammar, allowing us to provide some linguistic and statistical interpretations about the relationship between semantic roles on the one hand and syntactic functions, part of speech and semantic prototypes on the other.