The goal of our research is to develop a theory for the origins and evolution of language. Such a theory necessarily involves three aspects: social, cultural and biological. The social aspect should give us answers to the question “Why did humans start to talk?. The cultural aspect looks to explain how new language forms arise in language and keep on changing over time. The biological aspect addresses how the biological foundations for language may have arisen. We focus mostly on the cultural aspect, developing and testing agent-based models to explain how features of language, such as agreement systems, arise and culturally evolve.
Origins and evolution of grammatical structures
Although there is a lot of data about the historical change in language, there is virtually no theory of the fundamental processes underlying this kind of evolution. We try to understand the cognitive mechanisms, interaction patterns, and collective dynamics that could explain how grammatical structures arise in human language by building agent-based models and using the hypothesis that self-organization and (linguistic) selection are the primary driving forces. We analyze the behavior of our models using the tools of complex systems science, and compare the results with phenomena observed in human languages. At this point we focus in particular on the origins of agreement systems and grammatical patterns (such as noun phrases).
Fluid Construction Grammar (FCG)
In order to conduct agent-based experiments in language evolution it is necessary to have a computational formalism that is capable of handling variation, flexibility and change. We are therefore working in collaboration with other research centres on the development of such formalism. The formalism takes a construction grammar viewpoint, which is more appropriate for modelling language evolution. It consists of data structures for representing linguistic knowledge and mechanisms for parsing, production and language learning. FCG has been released as open source and has a growing community of users (http://www.fcg-net.org/).
Neural implementations of Fluid Construction Grammar
To bridge the gap between computational models and neurobiology, we are investigating how a replicator dynamics model of the brain could potentially be used to implement the highly complex operations that Fluid Construction Grammar demands.
Lab website: Language Evolution Lab
Global experts agree on declaration for proper use and development of artificial intelligence in Europe
International experts debate the most recent breakthroughs in Artificial Intelligence at the CosmoCaixa
BEE DEVICES and other curious observations
Steels L.; Szathmáry E. 2018. The evolutionary dynamics of language. BioSystems. 164:128-137 doi: 10.1016/j.biosystems.2017.11.003
Steels L., Szathmáry E. 2018. The evolutionary dynamics of language. BioSystems. 164( ): 128-137 doi: 10.1016/j.biosystems.2017.11.003
Steels L. 2017. Requirements for computational construction grammars.. AAAI Spring Symposium - Technical Report. SS-17-01:251-257
On the "Artificial Intelligence: Dreams, Risks, and Reality" B-Debate, co-organized by Luc Steels:
- 12th March, 2017. El Punt Avui. "Científics alerten del mal ús de la intel·ligència artificial"
- 9th March, 2017. La Vanguardia. "Barcelona define las bases de la inteligencia artificial en Europa"
29th December, 2015. Playground. "La IA no es una amenaza, jamás podrá sustituir a la inteligencia humana". Interview with Luc Steels.
November, 2015. Jot Down. "Luc Steels: “Science is a cultural activity, it’s like arts: it gives us insights into the world". Interview with Luc Steels.