Awards & Honours

Serena Villata, pioneering inventor of AI capable of argument and debate

Date:
Changed on 26/11/2021
The winner of the 2021 Inria Académie des sciences - Young Researcher Award, Serena Villata builds AI (artificial intelligence) tools capable of analysing the logical structure of any type of text, from a political speech to a Twitter message. This opens up promising perspectives in the fight against online harassment and fake news.
Portrait de Serena Villata
© Inria / Artiste Designer : Aurélie Bordenave

AI, a weapon in the fight against fake news

What if computers could help us to think, and not just process information? What if AI could identify arguments in a text, classify them and assess their reliability? What if it could enable us to prevent cyber-bullying or fake news? These are the questions that underpin the work of Serena Villata, a researcher at the Inria Sophia Antipolis -  Méditerranée Centre and a specialist in language processing and automatic argument generation, who was recently awarded the the Inria – Académie des sciences – Young Researcher Award.

She began tackling the question in 2007, after one of her thesis co-directors suggested she work on “computation argumentation models”. In other words, the way in which AI extracts and represents the logical structure of a text. “These models were very simple at the time”, says Serena Villata, “with arguments “for” or “against” a topic or decision.” 

Theoretical advances in natural language processing

The researcher developed two aspects of the theory. Firstly, the addition of a “support relation” to integrate arguments which support another. For example, photographs of the Earth from space support the argument that the Earth is round and not flat. The second aspect involves taking account of the reliability of sources that support arguments, which gain or lose credibility depending on whether or not the person cited is an expert in the subject.

“I managed to integrate these theoretical advances into algorithms”, she explains, “and then transpose them to the automatic analysis of “natural language”. In other words, text written by people.” The field was still in its early stages at the time, and the term “argument mining” had not yet been coined.

With the help of linguists, Serena Villata developed the very first algorithms for the automatic extraction of arguments. These were classified as premises or conclusions, and the links connecting them were defined as relationships of attack ("for" or "against") or support. Like magic, the logical structure of the text was revealed, without any human intervention.

Verbatim

I managed to integrate these theoretical advances into algorithms and then transpose them to the automatic analysis of “natural language”. In other words, text written by people.

Auteur

Serena Villata

Poste

Inria Académie des sciences - Young Researcher Award 2021

The field was still in its early stages at the time, and the term “argument mining” had not yet been coined.

With the help of linguists, Serena Villata developed the very first algorithms for the automatic extraction of arguments. These were classified as premises or conclusions, and the links connecting them were defined as relationships of attack ("for" or "against") or support. Like magic, the logical structure of the text was revealed, without any human intervention.

Watch the video (in French)

American election debates under close scrutiny

This provided a powerful analysis grid capable of multi-tasking which, after a learning phase, was applied to political speeches, medical articles and online discussions via Twitter or Facebook. It generated a great deal of interest. Serena Villata was soon invited to speak at international conferences and became involved in numerous French and international research projects.

In particular, her work formed the basis of Disputool, an argument mining project on candidates’ debates in American presidential elections from 1960 to 2016. Another initiative is the decision support tool Acta (Argumentative Clinical Trial Analysis), which automatically analyses reports on clinical drug trials and treatment. This eliminates the need to read the full report in order to draw a conclusion, as the report’s conclusion and its supporting arguments are presented in an argumentation graph which users can refer to.

Tools to identify victims of cyberbullying

However, it is in the field of social networks and the fight against bullying that Serena Villata’s work has had the greatest impact. By combining the analysis of links between users with the analysis of messages, she designed artificial intelligence tools capable of automatically identifying hateful messages on social networks. These tools identify internet users who are victims of misogyny, homophobia, racism or any other form of online hate. A similar approach can be used to detect emitters of fake news.

“I’m continuing to enhance these tools in collaboration with Italian and French schools, who use them to improve the management of students’ discussions on their intranet”, the researcher says. “I also use them in a project for the prevention of online hate, led in the framework of OTESIA*, which includes awareness actions for students.”    

 Serena Villata has also been a member of the Comité national pilote d’éthique du numérique (National Steering Committee for Digital Ethics) since December 2019. “This is fascinating work: the ethical issues around AI go far beyond the field of computer science. It is an opportunity to reflect on these issues with sociologists, philosophers, legal experts and elected representatives.”  

 

*Observatory for the technological, economic and societal impacts of artificial intelligence

Three highlights in her career

  1. Early works on the theory of formal argumentation, in 2007. “This was a fundamental topic, but it could lead to multiple applications in real life;  most discussions between humans are based on arguments. But at the time these applications seemed futuristic and highly ambitious...” 
  2. Beginnings in argumentation and natural language, in 2012 “I went from computational theory to practice and everyday life, the automatic extraction of arguments from messages on a forum, a political debate or a legal textn etc. We were starting from scratch and I was one of the pioneers, it was the right time to get into this.” 
  3. Projects on societal topics, as of 2018. “My research was starting to shift towards the prevention of online hate, propaganda, and disinformation. It's important for me to feel that my work has a real use in society.” 

 

What is the greatest challenge in terms of automatic argument generation?

“The biggest challenge in my field right now is to design a chatbot capable of speaking in a natural way with users such as schoolchildren, to help them develop their critical thinking. It could explain a topic to them, illustrate it with examples, put forward counter-arguments to the fake news it generates and describe the devastating effects of the spread of this fake news, etc.

This is a highly-ambitious project; the chatbot would be able to automatically process any theme, without preparation, by adapting to the knowledge of the user. What’s more, we have to prevent such a tool from being misappropriated for disinformation campaigns. It must be multi-disciplinary in design and bring together computer scientists, legal experts, sociologists, etc.”

Serena Villata in six key dates

Serena Villata

Short biography

2007  –  Began her thesis, under the co-tutorship of Turin and Luxembourg universities

2010  –  PhD in computer science (artificial intelligence)

2011  –  Joined the INRIA Sophia Antipolis -  Méditerranée Centre as a post-doctorate

 

<> Click on the arrows to scroll through its course.

Serena Villata

Express biography

2015  –  Class 1 research fellow at the CNRS

2018  –  Invited to present her results at one of the world’s most important conferences on AI, IJCAI (International Joint Conferences on Artificial Intelligence).

2021  –  Inria Académie des sciences Young Researcher Award.

 

Design: © Inria / Artiste Designer: Aurélie Bordenave

Inria 2021 Awards

Discover the awarded Inria researchers

Find out more