Analysing
the ontogeny of children language
Besides the
analysis of large, well-structured corpuses and their evolution through
historical times, an additional pattern of change is obtained by
exploring the architecture of language learning in children. The
resulting network of word-word co-occurrences reveals a diferent,
fascinating aspect of language organization: how it develops and how
grammar, lexicon and topology relate to each other as the child grows.
FIGURE: left, children reading.
Through
language acquisition, both lexical and syntactical complexity increase.
Right: language network at some intermediate step during the learning
process in a child.
The process
of lexical growth is associated to a constant input of words, but the
process of linking words together involves a deeper level of
complexity. During language acquisition, a number of physiological
contraints are present at early stages, but the emergence of true
syntactic relations is the main player, Single word utterances dominate
at the beginning, but at some point two-word combinations start to
increase and word sucessions appear as grammar
develops.
In terms of
network complexity, the increase in the number of components (the
lexicon) goes in parallel with an increase in network complexity that
can be properly measured. Even at early stages, patterns of word-word
interactions can be seen, whose complexity and heterogeneity increase
over time.
We are exploring this problem at several levels, from co-occurrence graphs
(see an example below) to syntactic ones.
Possible extensions of this work might
involve
the study of the ontogeny of robot languages. Under different
conditions of learning and individual complexity, we can compare the
growth of both lexicon and grammar in artificial-based collectives of
robotic agents. Such a scenario might help understanding the possible
presence of universal patterns of language evolution.
