Automatic annotation of speculation in biomedical texts: new perspectives and large-scale evaluation

One emergent field in text mining tools applied to biological texts is the automatic detection of speculative sentences. In this paper, we test on a large scale BioExcom, a rule-based system which annotates and categorizes automatically speculative sentences (“prior” and “new”). This work enables us to highlight a more restrictive way to consider speculations, viewed as a source of knowledge, and to discuss the criteria used to determine if a sentence is speculative or not. By doing so, we demonstrate the efficiency of BioExcom to extract these types of speculations and we argue the importance of this tool for biologists, who are also interested in finding hypotheses.