arrow drop search cross
arrow drop search cross

Serena Villata, 3IA Chair: Artificial Argumentation for Humans


Serena Villata, 3IA Chair: Artificial Argumentation for Humans


Publication : 29/05/2020
Partager cet article :

Artificial Argumentation for Humans

Since the early years of the field, Artificial Intelligence (AI) has sought to understand the principles governing intelligent behavior and to encode such principles in so-called intelligent machines. Artificial argumentation plays an important role in AI research. Argumentation-enhanced intelligent machines require the use of argumentation technologies to support the interactive explanation of the outcome of the deliberation process (why the machine deliberated in a certain way) taking into account the user feedback through natural language argumentative explanations, and to mine, analyze, summarize, and generate natural language argumentation structures from different settings (e.g., clinical trials, political debates, legal cases).  Humans argue. Machines should be able to argue too if we aim to achieve mixed human-machine teams in a hybrid society.

Success story!

Two years ago, we started to work on mining argumentative structures to ease evidence-based analysis from clinicians on clinical trials, together with Elena Cabrio (UCA) and Tobias Mayer (UCA). This work lead to a machine learning architecture relying on deep bidirectional transformers to extract, classify and link argumentative components in clinical trials. The results we obtain with deep bidirectional transformers in combination with different neural architectures (i.e., LSTM, GRU and CRF) outperform current state-of-the-art end-to-end argument mining systems, and they were published at the 24th European Conference on Artificial Intelligence, Europe’s premier AI research venue. Furthermore, we also presented ACTA, an inline tool (http://ns.inria.fr/acta/) for argumentative analysis of clinical texts.

See above: a screenshot of the ACTA system for the argumentative analysis of clinical trials based on a set of articles retrieved on PubMed.

Collaborations with several research teams from both the Nice / Sophia Antipolis region and foreign universities

This project is pursued in collaboration with several research teams from both the Nice / Sophia Antipolis region and foreign universities. In particular, we already collaborate closely with the GREDEG research team of UCA on the argument-based analysis of legal documents, and the Institut National Supérieur du Professorat et de l'Education of the Académie de Nice on mining and classifying arguments in abusive text. Some of our international collaborations include: FBK Trento (Italy) on the analysis of emotions in arguments about the Brexit news and the classification of hate arguments in social networks, Imperial College London (UK) on deep learning for argument mining, the University of Luxembourg on mining and analyzing political debates, and the University of Cordoba (Argentina) on mining arguments from legal cases.

The region provides an invaluable environment to foster and structure cross-disciplinary exchanges

The advantages of developing this project in the Nice / Sophia Antipolis region are numerous. First of all, the region provides an invaluable environment to foster and structure cross-disciplinary exchanges in academic research and public-private research partnerships for innovation. Second, the newly created AI Institute, 3IA Cote d’Azur, integrates this environment making this place one of the most prestigious venues for AI in France.

Serena Villata, 3IA Chair, Axe 1 Core elements of AI

Current Position

  • She is a member of the SPARKS-WIMMICS Research Team.
  • Since October 2015, she is a researcher (CR1) at CNRS. She defended her HDR (habilitation) in July 2018.
  • She is affiliated with the I3S research centre in Sophia Antipolis (France).

Research Interests

  • Artificial Intelligence
  • Knowledge Representation and Reasoning
  • Argumentation Mining
  • Artificial Intelligence and Law
  • Computational Linguistics
  • Semantic Web