Identification of assertions in the medical texts: application to the relation {patient, medical problem} Identification des assertions dans les textes médicaux : application à la relation {patient, problème médical}

Scientific and technical texts are full of assertions (i.e., hypothesis, condition, possibility) that shade the discourse and give more or less certainty to information. We focus on the assertions that influence the relation {patient, medical problem} in patients’ records. We aim at automatically identifying certainty and the degrees of certainty in medical texts, as well as their polarity (positive/negative). We propose to exploit morphological, contextual and structural markers for identification and disambiguation of assertions. Our system has been evaluated within the I2B2 international challenge. The evaluation shows that serveral categories of assertion are correctly identified while others remain difficult to detect. We explore also the contribution of each marker and analyse the evolution of the results, as well as performances in the detection of different categories of assertions.