Bio-inspired Grammatical Inference - Foundations on Natural and Artificial Computation - page 319

Our second approach is based on semantic corrections . Corrections have a
semantic component that has not been taken into account in previous works.
Hence, we have proposed a new computational model of language learning that
takes into account semantics. This model is bio-inspired by studies on children's
language acquisition, and more concretely, by the results obtained by Chouinard
and Clark in [12]. Our final goal has been to find a formal model that gives an
account of meaning preserving corrections , and in which we can address the
following questions: What are the effects of corrections on learning syntax? Can
corrections facilitate the language learning process? Can semantic information
simplify the problem of learning formal languages? It is worth noting that this
has been the first attempt to incorporate a robust notion of semantics in the
field of GI. Such a model has allowed us to investigate aspects of the roles of
semantics and corrections in the process of learning to understand and speak a
natural language. Our main results can be found in [4,3,2].
Taking into account that GI studies have been focused on learning regular
and context-free languages, but, as linguistic studies suggest, these classes have
limited expressive power to describe natural language syntax, we propose that
GI studies focus on other classes such as Mildly context-sensitive .Moreover,
we also support the idea that natural languages occupy an orthogonal position
in the Chomsky hierarchy. Therefore, we propose to study formalisms that are
able to generate MCS languages (i.e., they generate multiple agreement, crossed
agreement and duplication structures, and they are computational feasible), and
that occupy an orthogonal position in the Chomsky hierarchy (i.e., they contain
some regular, some context-free, but they are included in context-sensitive). We
have also done some works in this direction. We studied a mechanism that have
such interesting properties, called Simple External Contextual . Our main results
can be found in [9,25,6].
5 Conclusions
The field of GI provides a good theoretical framework for investigating the pro-
cess of natural language acquisition. However, studies in this field have been
focused on the mathematical aspects of the formal models proposed, without
exploiting their linguistic relevance. Therefore, the linguistic motivation that
originated GI studies has been left aside.
In this paper, we have discussed some linguistic studies on children's language
acquisition and we have proposed to use them in order to improve models and
techniques used in GI. Concretely, we have proposed that GI studies take into
account corrections and semantics during the learning process, and they focus
on classes of languages that are MCS and occupy an orthogonal position in the
Chomsky hierarchy.
We have also present some works in this line. These works show that new
challenging results in the field of GI can be obtained. Moreover, models in GI
can be improved by using these linguistics ideas.
It is worth noting that a formal model bio-inspired by all these linguistic
ideas could also help us to better understand natural language acquisition. By