The Processing & Recognition of Symbol Sequences
Year: 2001
Type: article
Abstract: It is proposed that learning a language (or more generally, a sequence of symbols) is formally equivalent to reconstructing the state-space of a non-linear dynamical system.Given this, a set of results from the study of nonlinear dynamical systems may be especially relevant for an understanding of the mechanisms underlying language processing.These results demonstrate that a dynamical system can be reconstructed on the basis of the data that it emits.They imply that with minimal assumptions the structure of an arbitrary language can be inferred entirely from a corpus of data.State-Space reconstruction can implemented in a straightforward manner in a model neural system.Simulations of a recurrent neural network, trained on a large corpus of natural language, are described.Results imply that the network sucessfully recognizes temporal patterns in this corpus.
Author: Mark Andrews
Institution: Cornell University
Language: English
Subfield: Artificial Intelligence
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
OpenAlex ID: W2624935667
