CLIN 2005 Abstracts
  • The contours of a semantic annotation scheme for Dutch
    Ineke Schuurman (Centre for Computational Linguistics, KU Leuven)
    Paola Monachesi (Uil-OTS, Utrecht University)
    Within the STEVIN-project "Dutch Language Corpus Initiative" (D-Coi) we are developing a protocol for semantic annotation.

    For the moment, we are concentrating on two types of annotation, and their interaction: 1) semantic role assignment and 2) temporal and spatial semantics. The reason for this choice lies in the fact that semantic role assignment (i.e. the semantic relationships identified between items in the text such as the agents or patients of particular actions), is one of the most attested and feasible types of semantic annotation within corpora. On the other hand, temporal and spatial annotation was chosen because there is a clear need for such a layer of annotation in applications like information retrieval or question answering. Although the two types of semantic annotation interact, their annotation is kept separately for ease of maintenance.

    As far as semantic roles are concerned, we aim to reconcile the rather pragmatic PropBank approach to role assignment which is essentially corpus based and syntax driven with the more semantics driven FrameNet approach which is based on a network of relations between frames.

    There are not that many corpora around that are annotated for temporal and/or spatial semantics, TimeBank being one of the few examples of such a corpus. The situation with respect to spatial semantics is far less clear. Little has been done as far as corpus annotation is concerned. In our protocol we are taking parts of several schemes that are around and combining them to fit our needs.