Browsing by Author "Santos, Rui Pedro Talhadas dos"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Automatic semantic role labeling for European PortuguesePublication . Santos, Rui Pedro Talhadas dos; Baptista, Jorge Manuel EvangelistaThis thesis addresses the task of Semantic Role Labeling (SRL) in European Portuguese. SRL can be used in a number of NLP application, namely Anaphora Resolution, Question Answering, Summarization, etc. A general-purpose, consensual set of 37 semantic roles was defined, based on a survey of the relevant related work, and using highly reproducible properties. A set of annotation guidelines was also built, in order to clarify how semantic roles should be assigned to verbal arguments in context. A SRL module was built and integrated in a fully-fledged Natural Language Processing (NLP) chain, named STRING, developed at INESC-ID Lisboa. For this module, the information from a lexicon-syntactic database, ViPEr, which contains the relevant linguistic information for more than 6,000 European Portuguese full (or lexical, or distributional) verbs, was used and the database manually enriched with the information pertaining to the semantic roles of all verbal arguments. The SRL module is composed of 183 pattern-matching rules for labeling of subject (N0), first (N1) and second (N2) essential complements of verbal constructions and also allows the attribution of SR to other syntactic slots in the case of time, locative, manner, instrumental, comitative and other complements (both essential and circumstantial). This module was tested in a small corpus that was specifically annotated for this purpose. After this manual annotation, the corpus containing 655 semantic roles was used as a golden standard for automatic comparison with the system’s output. Considering that the SRL module operates at the last stages of the processing chain, a relatively high precision was achieved (69.9% in a strict evaluation and 77.7%, when evaluation included partial matches), though the recall was low (17.9%), which calls for future improvements.