Human language is a mystery. In a society where information has value, what justifies our talking to others without compensation? Even more intriguing: what processes govern our communications, whether it is a deep debate or a spontaneous conversation with an acquaintance? These are the issues that drive the research of Jean-Louis Dessalles, a computer science researcher at Télécom ParisTech. His work has led him to reconsider the vision of information adopted by Claude Shannon, a pioneer in the discipline. The result is original theories and models of conversation that describe both futile discussions and the most argued debates.
Phy are we talking about? And what are we talking about? Nourished by the young researcher's optimism, Jean-Louis Dessalles thought he could answer these two questions within a few months of completing his thesis in 1993. Nearly 25 years have passed since then, and the subject of his research has not changed. From his office in Télécom ParisTech's Computer Science and Networks department, he continues to take an interest in language. His work breaks with the classical approach adopted by researchers in information and communication sciences. "Discipline is primarily concerned with how a message is sent, but not with what is sent or why a message is sent." he stresses, thus opposing the approach of the communication as described by Claude Shannon in 1948.
The value of communications, and the motivations underlying these exchanges of information, are, however, legitimate and complex issues. As the researcher explained in the film The great novel of Manreleased in 2014, communication is a contradictory act of certain behavioural theories. Game theory, for example, which is sometimes used in economics to describe and analyse behavioural mechanisms, struggles to justify the contribution of communication between humans. According to this theory, and by attaching a value to any information, the expected communication situations would consist in each actor giving a minimum of information, and trying to get a maximum of information from others. However, this is not the logic that drives humans in everyday discussions. "We must therefore plunge the communication game into a social game." deducted Jean-Louis Dessalles.
By dissecting the scientific corpus of communication situations (interviews, attitudes on online forums, discussions...), he then tried to find an explanation for the gift of useful information. The hypothesis that he puts forward today is compatible with all the communications he observed: for him, the gift of quality information is not motivated by economic gain - as game theory would suppose - but by a gain in social representation. "On online technical forums, for example, experts do not respond out of altruism or monetary interest. They compete to give the most complete answer in order to establish their status as experts. In this way, they earn a social existence. points out the researcher. Talking, demonstrating our ability to stay informed, means positioning ourselves in the hierarchy of a society.
When the unexpected loosens tongues...
Once the question of "why are we talking? "...has been clarified, the question of what we're talking about remains. Jean-Louis Dessalles was not interested here in the subjects of the discussions, but rather in the general mechanisms that dominate the action of communication. After a detailed analysis of tens of hours of recording, he came to the conclusion that a large part of the spontaneous exchanges are structured around the unexpected. Triggers for spontaneous conversations are indeed events that humans consider improbable, abnormal - in the sense that the normality of a situation is broken. A person measuring more than 2 m, a series of cars of the same colour parked next to each other, a lottery draw with all the numbers in sequence are all examples that can cause a person to be surprised and spontaneously engage in a conversation with someone.
To explain this commitment based on the unexpected, Jean-Louis Dessalles has developed a theory of simplicity. Thus, according to him, what is unexpected corresponds above all to something simple to describe. Simple, because it is always easy to describe a situation that is out of the ordinary, simply by comparing it with what is expected. Thus, describing a person measuring more than two metres is easy, since this criterion in itself is enough to establish a point of narration. On the other hand, describing an individual of normal height, with normal weight, common clothes and a face without any particular distinguishing feature requires a much more complex description to arrive at a definition.
If simplicity is therefore a driving force behind spontaneous conversations, there is also a second major category of discussion: argumentative conversations. Here, it's no longer a matter of the unexpected. This type of exchange follows a model defined by Jean-Louis Dessalles and called CAN, for "conflict, abduction, negation". "To start an argument, there must be a conflict, an opposition of points of view. Abduction is the next step, it consists in going back to the cause of the conflict in order to move it there and deploy its arguments. And finally, negation allows us to move on to the counterfactual in order to reflect on solutions to resolve the conflict. "Beyond the simple description, the CAN model could allow the development of artificial intelligence (see box).
When artificial intelligence takes an interest in language theories...
"The machine has to be able to have a sensible conversation in order to appear intelligent." assures Jean-Louis Dessalles. For the researcher, the test invented by Alan Turing, which consists in decreeing an intelligent machine if a human cannot discern it from another human by conversing with it, has all its legitimacy. His work therefore has its place in the development of artificial intelligences capable of passing this test. It is indeed essential to understand the mechanisms of human communications in order to transmit them to machines. A machine integrating the CAN model would thus be better able to debate with a human. In the case of a GPS, this would make it possible to define routes on parameters other than time or kilometres. To argue with a GPS about what we expect from a route - beautiful landscapes for example - in a logical way, would considerably extend the quality of the human-machine interface.
In the hours of conversations recorded by the researcher, the distribution between spontaneous discussions induced by the unexpected and argumentation is 25 % and 75 % respectively. But he notes that the boundary is not necessarily strict, as spontaneous narration can lead to deeper debate, which will shift to the CAN model. These results are thus in response to the question "what are we talking about? "and are the culmination of years of research. For Jean-Louis Dessalles, it is proof that "naivety pays". The carelessness of his beginnings finally led him to theorize over the course of his career about the models on which humans base their communications, and on which he will probably continue to rely for a long time to come.
Jean-Louis Dessalles, human language computer scientist Telecom ParisTech
The original of this article was published in IM'Tech.
A graduate of Polytechnique and Télécom ParisTech, Jean-Louis Dessalles became a de facto researcher in computer science after completing his thesis in 1993. It was difficult then to see the link with questions of human language and its origins, usually reserved for linguists or ethnologists. "I chose to take an interest in the subject with the weapons I had at my disposal, which are information sciences," he says.
The result is research that is a counterpoint to Claude Shannon's probabilistic approach, as he presented at the Institut Henri Poincaré symposium in October 2016 on the occasion of the centenary of the father of information theory.
His reflections on information are notably the subject of the book "The thread of life"published by Odile Jacob in 2016. He is also the author of several books on the question of the emergence of language.
