Bio-inspired Grammatical Inference - Foundations on Natural and Artificial Computation - page 312

Bio-inspired Grammatical Inference

Leonor Becerra-Bonache

Laboratoire Hubert Curien, UMR CNRS 5516
Universite de Saint-Etienne, Jean Monnet
Rue du Professeur Benoit Lauras, 42000 Saint-Etienne, France
[e-mail]

Abstract. The field of Grammatical Inference was originally motivated
by the problem of natural language acquisition. However, the formal
models proposed within this field have left aside this linguistic motiva-
tion. In this paper, we propose to improve models and techniques used
in Grammatical Inference by using ideas coming from linguistic studies.
In that way, we try to give a new bio-inspiration to this field.

1 Introduction

The problem of how children acquire their first language has attracted the at-
tention of researchers for many years. The desire to better understand natural
language acquisition has motivated research in formal models of language learn-
ing [27,26]. Such models are of great interest for several reasons. On one hand,
these models can help us to answer several key questions about natural language
learning. On the other hand, these formal models can provide an operational
framework for practical applications of language learning; for example, language
learning by machines.
Grammatical Inference (GI) is a subfield of Machine Learning that deals with
the learning of formal languages. The initial theoretical foundations of GI were
given by E.M. Gold [18], who tried to formalize the process of natural language
acquisition. After Gold's seminal work, research in this field has been specially
focused on obtaining formal results (e.g., formal descriptions of the languages
to be learned, formal proofs that a concrete algorithm can eciently learn ac-
cording to some concrete denitions, etc) [13]. Several formal models of language
learning have been proposed in this field [17], however, such models do not take
into account some important aspects of natural language acquisition, and as-
sume idealized conditions as compared to the conditions under which children
learn language (as we will see in Section 2). Therefore, although GI studies were
motivated by the problem of natural language acquisition, its mathematization
has left natural approaches aside.
Since the study of formal models of language learning is of great interest to
better understand natural language acquisition, it is important that such models
are inspired by studies of natural language acquisition. In that way, models can
be more realistic, and can better simulate the human processing and acquisition
of language.