Bio-inspired Grammatical Inference - Foundations on Natural and Artificial Computation - page 316

acquire their native language. This proposal is based on the Chomsky's poverty
of stimulus argument: there are principles of the grammar that cannot be learnt
from only positive data, and since children do not receive negative data (i.e., evi-
dence about what is not grammatical), one can conclude that the innate linguistic
capacity is what provides the additional knowledge that is necessary for language
learning. Moreover, work presented by Brown and Hanlon in [11], has been used
as an argument to support the unavailability of negative data to children. Con-
cretely, they analyzed adult approval and disapproval of child utterances (for ex-
ample, adult's answers as “That's right”, “Correct”, “That's wrong”, “No”), and
they found no relation between this type of answers and the grammaticality of the
sentences produced by the children. However, this approach raises several ques-
tions: Should only explicit disapproval count as negative evidence? Could adults
correct children in a different way?
The second proposal is that children receive negative data in the form of dif-
ferent reply-types given in response to grammatical versus ungrammatical child
utterances. Hirsh-Pasek et al. [19] and Morgan and Travis [24] studied this type
of negative evidence and proposed that parents respond to ungrammatical child
utterances by using different types of answers from those they use when respond-
ing to grammatical utterances. Under this view, the reply-type would indicate to
the child whether an utterance was grammatically correct or not. The problem
of this second approach is that it does not take into account whether the adult's
replies contain corrective information [12]. Moreover, under this approach, chil-
dren would learn what utterances are correct only after complex statistical com-
parisons [23].
The third proposal is that children receive negative evidence in the form of re-
formulations , and they not only can detect them, but also they can make use of
that information. Reformulations are sentences adults use in checking up on what
their children intended to say (for example, a child says ”milk milk” and the father
answers ”you want milk”?). Chouinard and Clark [12] proposed this new view of
negative evidence. The main properties of this kind of corrections are the follow-
ing: i) Adult's correction preserves the same meaning of the child; ii) Adult uses
the correction to keep the conversation on track (adult reformulates the sentence
just to make sure that he has understood the child's intentions); iii) Child's utter-
ance and adult's correction have the same meaning, but different form.
It is worth noting that reformulations are often provided to children during
the early stages of children's language acquisition. Moreover, semantics seems
to play an important role in the first stages of children's language acquisition,
concretely in the stage known as the two-word stage, in which children go through
the production of one word to the combination of two elements [29,30].
3.2 Location of Natural Languages in the Chomsky Hierarchy
The question of where natural languages are located in the Chomsky hierarchy
has been a subject of debate for a long time. This question was posed by Chomsky
in the 50's. The debate was focused on whether natural language are context-free
or not.