Previous studies on the logical aspects of argumentation in natural conversations (see [dessalles 1985, 1990]) showed that people quite often introduce a new topic by uttering surprise. Consider for instance the following excerpt:
[ex_canteen]
context: A, B and C are choosing a table at their staff canteen. A is surprised because there are very few people, when it's normally quite difficult to find a table free. B gives an explanation: most of the students went to the so-called "forum" where companies present job opportunities.
A: What happens today? There 's nobody here. It's Wednesday...
B: Today is the forum day.
This kind of surprise, where one realizes that the situation does not match one's expectations and where reasons can be given to justify such expectations, appears as essentially logical. In this excerpt, A had reasons to think that students should have been more numerous: it was wednesday, and they had to attend classes. This expectation is in contradiction with the small number of students. A's surprise thus comes from an expectation.
It is possible to capture such expectations with a simple logical representation (such a logical representation can be made more precise at will, but it is not necessary here):
normal_workday ==> not students_are_absent
Some expectations are the result of logical reasonings, as was shown for instance by Inhelder & Piaget [1979] when they insisted on the difference between procedures and structures. In one experiment with children, they put pearls one after the other into two containers alternatively. The two containers were shown to contain initially a different number of pearls, and one of them was then hidden. Smaller children considered as possible that they contain the same number of pearls after some time, when older children considered such an event as impossible. One of them declared: "Oh yes: as soon as you know, you know for ever!" ("Ah oui: une fois qu'on sait, on sait pour toujours"). The authors concluded:
structures show themselves through inferences made by the child while procedures involve much more empiricism; and the structural nature of these inferences is best revealed by suppressions of contradictions and of incompatibilities.
So people have structures that allow them to draw inferences and thus to have expectations about the world (as was also recently emphasized by Ohlsson [1991]). For our purpose, we can consider these structure as logical models. Of course some expectations are not the result of logical inferences. They may follow for instance statistical measurements. But model-based surprise (caused by logical expectations) seems much more likely to be followed by an explanation, and thus we chose to focus on it.