Managing Linguistic and Terminological Variation in a Medical Dialogue System

Managing Linguistic and Terminological Variation in a Medical Dialogue System Leonardo Campillos-Llanos Dhouha Bouamor Pierre Zweigenbaum Sophie Rosset LIMSI, CNRS, Universit´e Paris-Saclay, 91405 Orsay, France firstname.lastname@limsi.fr Abstract We introduce a dialogue task between a virtual patient and a doctor where the dialogue system, playing the patient part in a simulated consultation, must reconcile a specialized level, to understand what the doctor says, and a lay level, to output realistic patient-language utterances. This increases the challenges in the analysis and generation phases of the dialogue. This paper proposes methods to manage linguistic and terminological variation in that situation and illustrates how they help produce realistic dialogues. Our system makes use of lexical resources for processing synonyms, inflectional and derivational variants, or pronoun/verb agreement. Specialized knowledge is used for processing medical roots and affixes, ontological relations and concept mapping, and for generating lay variants of terms according to the patient’s non-expert discourse. We report the results of a evaluation of the non-contextual analysis module—which supports the Spoken Language Understanding step—after 11 users interacted with the system. The annotation of domain entities obtained 91.8% of Precision, 82.5% of Recall, 86.9% of F-measure, 19.0% of Slot Error Rate, and 32.9% of Sentence Error Rate. Keywords: medical terminology, natural language understanding, virtual patient consultation 1. Introduction and Related Work Terminology management is a core component in med- ical informatics applications. While this need has long been identified for health professionals (Cimino, 1998), the needs of patients and lay people have only been addressed recently (McCray et al., 2000; Zeng-Treitler et al., 2007). Virtual patients (VP) are interactive systems and require managing terms—e.g. by formalizing ontological con- cepts (Nirenburg et al., 2008)—and a Natural Language Understanding (NLU) module. The NLU component may rely on text meaning representations for resolving para- phrases (Nirenburg et al., 2009) or a corpus of questions and answers curated by an expert (Kenny et al., 2008). We are developing a conversational agent to be used in a simulated consultation with a VP, where the system aims at training medical doctors (Campillos-Llanos et al., 2015). Users (medical students or doctors) interact with the VP to collect information that allows them to provide a correct diagnosis. Medical trainers define each e-learning case be- forehand by entering the VP profile data in a clinical record (e.g. symptoms or medical history). Managing linguistic and terminological variation is crucial to match a user’s question to a term in the clinical record and to select suitable terms for answer generation. This pa- per gives an overview of the difficulties (Section 2.) and strategies applied in both analysis (Section 3.) and genera- tion (Section 4.). We also report the results from an eval- uation (Section 5.) and conclusions (Section 6.). Although the system currently only supports French, the challenges found might be raised regardless of users’ language. 2. General Overview Each turn in a dialogue system includes steps of analysis, dialogue management, and generation. Additionally, a vir- tual patient must have a model of its (health) state, which is …

Content truncated to 500 words