Luc Steels


Most day-to-day language use is based on routinized patterns which are gradually acquired from early childhood. But every speaker and hearer of a language must occasionally be creative. A new concept may need to be expressed for which there is no precise word and hence speakers need to be able to exapt an existing word for a new purpose (as with "twitter") or to invent a brand new word (like "skype").
An existing word may not fit with a chosen grammatical construction and the speaker may then have to coerce the word so that it can be used in a novel syntactic context (as in "Just google me", where the noun "google" became coerced to be a transitive verb).
New sounds may also enter the inventory of a language, for example, a vowel may become nasalized (as in French "non") due to the influence of the "n" on the preceeding vowel. Many more examples can be given.
Some of them lead to deeper systemic innovations, for example when articles entered in a language (which happened in the transition from Latin to French, Italian, Catalan and other romance languages) or when the copula "is" started to function as an auxiliary (as in "he is mailing me"). Other changes are motivated by the tendency of language users to optimize some aspect of the language. For example, speakers try to simplify the pronunciation of a word, or leave out a complex morpheme, or no longer pronounce a function word. The historical evolution of human languages shows the gradual accumulation of all these inventions and further insights about the lexicalisation and grammaticalisation processes underlying them can be gleaned from recent, new emergent sign languages or from creole language development.

This talk addresses the question how we can construct agent-based models of the creative processes that drive this type of language change and language innovation. I will introduce the concept of a LANGUAGE STRATEGY which is a series of steps that speakers and hearers use to expand, adopt or align some part of their language system. I will show several examples of language strategies and experiments with robotic agents demonstrating that these strategies lead to shared language systems. From these examples emerges a theory of cultural language evolution based on transposing the idea of evolution by selection to the domain of cultural systems.
Speaker and hearers create variation in the language but only those words and constructions that lead to more reliable communicative success and decreased cognitive effort proliferate.



+ Steels, L. (2011) Modeling the cultural evolution of language. Physics of Life Reviews. 8(4) 330-356.
+ Steels, L. (2012) Experiments in cultural language evolution. John Benjamins, Amsterdam.
+ Beuls, K., & Steels, L. (2013). Agent-Based Models of Strategies for the Emergence and Evolution of Grammatical Agreement.
PLOS ONE, 8(3), e58960.