The sensorimotor origins of linguistic categories: Experiments with
grounded neural network models

Angelo Cangelosi

Adaptive Behaviour and Cognition Research Group, University of
Plymouth

acangelosi@plymouth.ac.uk


The scientific investigation of the origins of language has
significantly benefited from the use of computer simulation approaches
(Cangelosi & Parisi, 2002; Kirby 2002).  Computer models based on
evolutionary neural networks and artificial life can be used to simulate
the emergence of grounded language in populations of sensorimotor
agents.  These models make it possible to investigate the fine
interactions, during evolution, between language and other cognitive,
behavioural and neural abilities.  In grounded evolutionary models, the
communicative behaviour of agents is directly grounded in their
cognitive and sensorimotor abilities.  The evolution of language depends
on the concurrent (or preceding) emergence of an ability to interact
with the environment and to build a categorical representation of it. 
All behaviours are controlled by the same neural network or by a set of
interconnected, modular networks.  The analyses of the agents' neural
networks allow us to highlight the neural mechanisms responsible for the
integration of communicative, cognitive and motor abilities and the
evolution of language. 

In this paper new data on the relationship between the evolution of
basic word categories and that of the sensorimotor abilities upon which
they are built will be presented.  The analyses of the neural networks
were performed through using the Synthetic Brain Imaging method (SBI:
Arbib et al.  2000; Cangelosi & Parisi, in press).  This technique
supports qualitative comparisons between empirical neuroimaging data on
the neural processing of language (Pulvermuller, 2003; Cappa & Perani,
2003) and the control of linguistic behaviour in the neural networks of
simulated agents.  In our computational model, we evolve populations of
agents that use two categories of communication signals: names of
objects and names of actions.  These respectively share some of the
properties of the categories of nouns and verbs, albeit in a very
simplified fashion.  These signals can actually be considered
proto-linguistic categories, in an evolutionary sense. 

SBI analyses show that the neural representations of behavioural
categories and that of word classes are sensitive to the level of
integration of linguistic information and sensorimotor knowledge.  The
neural networks show a functional modular organisation that closely
reflects that observed in humans through brain imaging studies (Cappa &
Perani, 2003).  The names of actions (nouns) are more active in the
parts of the network that process sensory and visual information only
(corresponding to the left dorsolateral areas in the human cortex). 
Words related to the names of actions produce more activity in the parts
of the network where sensorimotor information is integrated
(corresponding to the left prefrontal motor areas where verbs produce
more activity).  Overall, the model shows that the evolution of early
proto-linguistic categories is dependent on the sensorimotor
organisation of the agents' behavioural and cognitive abilities.  The
implications of this model and data for assessing the role of
sensorimotor knowledge in the evolution of language and syntax will be
discussed.  In particular, this model supports hypotheses on the
re-organisation of the brain for the origins of symbolic abilities
(Deacon, 1997) and the sharing of neural structures for linguistic and
motor tasks (Greenfield, 1991; Reilly, in press). 

References

Arbib M.A., Billard A., Iacoboni M.  & Oztop E.  (2000).  Synthetic
brain imaging: grasping, mirror neurons and imitation.  Neural Networks,
13, 975-997

Cangelosi A.  & Parisi D.  (Eds.) (2002).  Simulating the Evolution of
Language.  London: Springer-Verlag. 

Cangelosi A., Parisi D.  (in press).  The processing of verbs and nouns
in neural networks: Insights from Synthetic Brain Imaging.  Brain and
Language

Cappa, S.F., & Perani, D.  (2003).The neural correlates of noun and verb
processing.  Journal of Neurolinguistics, 16 (2-3), 183-189

Deacon T.W.  (1997).  The Symbolic Species: The coevolution of language
and human brain, London: Penguin. 

Greenfield P.  (1991).  Language, tool and brain: The ontogeny and
phylogeny of hierarchically organized sequential behavior.  Behavioral
and Brain Sciences, 14, 531-595. 

Kirby S.  (2002).  Natutal language and artificial life.  Artificial
Life, 8, 185-215. 

Pulvermuller F.  (2003) The neuroscience of language.  On brain circuits
of words and serial order, Cambridge University Press. 

Reilly R.G.  (in press).  The relationship between object manipulation
and language development in Broca's area: A connectionist simulation of
Greenfield's hypothesis.  Behavioral and Brain Sciences.