CLIN 2005 Abstracts
  • Automatic Markup and Editing of Anaphora Relations
    Christer Johansson (University of Bergen)
    Anders Nøklestad (University of Oslo)
    This article presents work on developing an online tool for quick automatic markup and editing of anaphora relations. The system accepts a text, analyses the text looking for anaphora relations, and shows the text with markables (e.g. noun phrases) as highlighted boxes and the automatically found relations as lines between boxes. Highlighting is dynamic, triggered by the user moving the mouse cursor over the relevant and available phrases. The relations the machine has found are easily edited. Selecting two markables with a link between them unlinks them, and any intermediate candidates between them keep their internal links. For example, A links to B links to C links to D. Selecting D and A dissolves the link between them, and results in B links to C and 'free' A and D. Linking two candidates, or two chains, is equally easy. The current rules do not allow chains to branch, which then implies unchaining intermediate candidates when needed. After the editor is satisfied with the result, the annotated text can be stored in a format that includes the efforts of the used resources (e.g. parts-of-speech tagging, and functional roles) as well as the annotation of co-reference. The final result is presented as an xml-tagged file. We hope that sharing the resources will lead to co-operation on creating larger databases of co-reference in multiple languages. We think it feasible to utilize examples from multiple languages, since most of the features we are currently using are based on matching the available features (e.g. gender, number, functional role) and not on storing the feature value itself.