Speculation and Negation detection in French biomedical corpora

Proceedings of Recent Advances in Natural Language Processing, pages 223–232, Varna, Bulgaria, Sep 2–4, 2019. https://doi.org/10.26615/978-954-452-056-4_026 223 SPECULATION and NEGATION DETECTION in FRENCH BIOMEDICAL CORPORA Cl´ement Dalloux and Vincent Claveau Univ Rennes, Inria, CNRS, IRISA, F-35000 Rennes, France name.surname@irisa.fr Natalia Grabar UMR 8163 STL CNRS, Universit´e de Lille, France natalia.grabar@univ-lille.fr Abstract In this work, we propose to address the detection of negation and speculation, and of their scope, in French biomedical doc- uments. It has been indeed observed that they play an important role and pro- vide crucial clues for other NLP applica- tions. Our methods are based on CRFs and BiLSTM. We reach up to 97.21 % and 91.30 % F-measure for the detection of negation and speculation cues, respec- tively, using CRFs. For the computing of scope, we reach up to 90.81 % and 86.73 % F-measure on negation and spec- ulation, respectively, using BiLSTM-CRF fed with word embeddings. 1 INTRODUCTION The detection of speculation and negation in texts has become one of the unavoidable pre-requisites in many information extraction tasks. Both are common in language and provide information on factuality and polarity of facts, which is partic- ularly important for the biomedical field (Elkin et al., 2005b; Denny and Peterson, 2007; Gindl et al., 2008; Chapman et al., 2001). Indeed, nega- tion and speculation provide there crucial infor- mation for detecting patient’s present, speculated or absent pathologies and co-morbidities, detect- ing whether a particular medication has been, may have been, or has not been prescribed or taken, defining the certainty of diagnosis, etc. In order to efficiently identify speculation and negation in- stances, it is first necessary to identify their cues, i.e., words (or morphological units) that express speculation and negation, and then their scopes, i.e., tokens within the sentence which are affected by the negation or speculation. In this paper, we present two French datasets annotated with nega- tion and speculation cues and their scope. We also propose machine learning and deep learning based systems to tackle the automatic detection of cues and scopes that achieve high performance. 2 EXPRESSION of NEGATION and SPECULATION In French, the expression of negation and specula- tion have some specifics that are described below. 2.1 NEGATION The negation cues may consist of one word or pre- fix, or of multiple words. Moreover, negation can be expressed via a large panel of cues which can be morphological (an, in, im, ir, dis), lexical (ab- sence de(absence of), `a l’exception de(excepting)), and grammatical (non, ne...pas, ni...ni). In the follow- ing examples, we present sentences with instances of negation (the cues are underlined). 1. En alternative des traitements locaux (chirurgie, radioth´erapie, radiofr´equence, cryoablation) peuvent ˆetre indiqu´es mais ils ne sont pas toujours faisables. (Alternatively, local treatments (surgery, radiotherapy, radiofre- quency, cryoablation) may be indicated but are not always feasible.) 2. Il n’existe toujours pas aujourd’hui de con- sensus quant `a une d´efinition pr´ecise de ce ph´enom`ene h´et´erog`ene ou des modalit´es de sa prise en charge. (There is still no consensus today on a …

Content truncated to 500 words