For a variety of discourse level analyses and tasks performed on scientific literature, it is necessary to identify which (if any) cited paper the discourse entities in focus are attributable to. In this paper we introduce a scientific attribution task that aims to associate a range of linguistic expressions such as definite descriptions, pronouns and “work ” nouns with specific cited papers. We report human agreement of Krippendorff’s Alpha greater than 0.8 on our scientific attribution task, based on written guidelines with ten rules for common systematic problem cases. The high alpha suggests that our task is well defined and fairly intuitive to annotators. Our machine learning approach achieves Krippendorff’s Alpha of 0.67 and percentage agreement of 85 % with a manually constructed gold standard, suggesting that the task is simpler than traditional anaphora resolution tasks.
|Title of host publication||Proceedings of the 6th Discourse Anaphora and Anaphor Resolution Colloquium (DAARC'07)|
|Publication status||Published - 2007|
|Event||6th Discourse Anaphora and Anaphor Resolution Colloquium (DAARC'07) - Logos, Portugal|
Duration: 29 Mar 2007 → 30 Mar 2007
|Conference||6th Discourse Anaphora and Anaphor Resolution Colloquium (DAARC'07)|
|Period||29/03/07 → 30/03/07|