Explorations in Synthetic Pragmatics
(with Christian Balkenius)
We explore a number of pragmatic principles of communication in a series of computer simulations. These principles characterize both the environment and the behavior of the interacting agents. We investigate how a common language can emerge, and when it will be useful to communicate rather than to try the task without communication. When we include the cost of communicating, it becomes favorable to communicate only when expectations are not met.
This paper is published. Please quote as follows:
Short note on the paper:
The last paper in the thesis deals with the formal dynamics of a process that can be called lexicalization. It was written together with Christian Balkenius for a conference "New Trends in Cognitive Science - Does Representation Need Reality?" in Vienna, May 1997, where it was presented. It was published in the Proceedings.
We have made a computer simulation of a simple language game, which is informally illustrated as follows:
A person comes to a place where he knows that a reward is placed behind one of two doors. He looks behind one of the doors, and let us say he finds the reward. To be nice to his friend, he writes a note saying "A" (or "B") and puts it between the doors.
His friend comes along, finds the note, and wonders what "A" might mean. He follows any weak preferences he may have, chooses a door, opens it, and if he finds what he wants, strengthens the preference that "A" indicates that door and not the other. To help the first person, he returns the service of putting a note between the doors. The first person comes by again, and so they continue. When the game is repeated, the matching of doors and labels will (hopefully) stabilize, and we get a common language with a small lexicon of two words and two meanings.
The simulations illustrate the stabilization of a lexicon, and we also show how both the agents contribute to the process. We show results from the simple game described above, and also the more complicated cases where the reward is moved with a certain probability. If the reward stays behind one of the doors, communication is of limited use. Also, if the place changes very often, it is no use to report the change, because the probability that the location changes again is too high.
Finally, we show that when a cost is added to the communication, it is preferable only to communicate when expectations are not met, concerning where the reward is located.
In section 1 in the Introduction, I presented language as a gradual conventionalization from pragmatics to semantics to syntax, where each level has a certain autonomy. It is this autonomy in the process of lexicalization that we intend to model in the simulations.