This paper describes three RoBERTa based systems. The first one recognizes adverse drug events (ADEs) in English tweets and links themwith MedDRA concepts. It scored F1-norm of 40 for the Task 1. The next one extracts pharmacovigilance related named entities inFrench and scored a F1 of 0.4132 for the Task 2a. The third system extracts pharmacovigilance related named entities and their relationsin Japanese. It obtained a F1 of 0.5827 for the Task 2a and 0.0301 for the Task 2b. The French and Japanese systems are the best performing system for the Task 2

Cet article a pour but de montrer la faisabilité d’un système de fouille de texte pour alimenter un moteur d’inférences capable de construire, à partir de prédicats extraits des articles scientifiques, un réseau de signalisation en biologie systémique. Cette fouille se réalise en deux étapes : la recherche de phrases d’intérêt dans un grand corpus scientifique, puis la construction automatique de prédicats. Ces deux étapes utilisent un système de cascades de transducteurs.

Unlike Entity Disambiguation in web search results, Opinion Disambiguation is a relatively unexplored topic. RevOpiD shared task at IJCNLP-2107 aimed to attract attention towards this research problem. In this paper, we summarize the first run of this task and introduce a new dataset that we have annotated for the purpose of evaluating Opinion Mining, Summarization and Disambiguation methods.