What kind of problems do protein interactions raise for anaphora resolution? - A preliminary analysis

In this preliminary study, we analyzed the kind of anaphoric expressions that occur in expressions describing protein interactions found in biological text. We also studied the impact of anaphora resolution on protein interaction extraction, when an off-the-shelf anaphoric resolver (i.e., not one specially developed for this domain) is used, and looking at full texts as well as abstracts. Our results suggest that about 5% of the descriptions of protein-protein interactions contain anaphoric expressions when full texts are considered. These anaphoric expressions are primarily pronouns, even though most anaphoric expressions are full NPs. The use of our anaphoric resolver gives a small improvement over our baseline system.