Associate 2007-08

Richard W Sproat



A key task of linguistics is to explain the evolution of a language’s morphology (its word formation including inflection, derivation, and compounding). During his Center appointment, Professor Sproat will use computational simulations to develop an innovative model of language change. He will simulate a simple society of artificial agents where “children” learn language from “parents.” As the children make mistakes in their learning, they will introduce errors that change the language over time.

Professor Sproat’s research group will initialize the simulation with a language, and parent agents will talk to child agents, thereby teaching them the language. Agents will be assigned a number of characteristics, including age (which will change over time), a location, a social status, and a probability of communicating with other agents as a function of physical distance and relative social status. Each agent will also have a learned grammar. The model will set an upper bound on the distance between any pair of speakers in the simulated society, with the members of the society converging to keep their differences within that bound. The language of some agents will be more highly valued than others, and other agents will tend to converge toward the language of those individuals. Groups of agents will be allowed to migrate, and in these cases Professor Sproat hopes to be able to simulate the evolution of different dialects.

A hypothesis for the project is that mistakes will result from two main sources: (a) incorrect generalization to combinations of forms that a child agent has never heard before and (b) over-generalization of natural phonetic changes introduced into the language. How such mistakes come about, are or are not perpetuated, and evolve over time will form the basis of Professor Sproat’s model of language change. He views the project as the first step in a long-term program that could position the University of Illinois as a leader in this approach to linguistics research.