Contributo in atti di convegno, 2011, ENG

T2HSOM: Understanding the Lexicon by Simulating Memory Processes for Serial Order

Ferro, Marcello; Marzi, Claudia; Pirrelli, Vito

Institute for Computational Linguistics - National Research Council (CNR-ILC, Pisa)

Over the last several years, both theoretical and empirical approaches to lexical knowledge and encoding have prompted a radical reappraisal of the traditional dichotomy between lexicon and grammar. The lexicon is not simply a large waste basket of exceptions and sub-regularities, but a dynamic, possibly redundant repository of linguistic knowledge whose principles of relational organization are the driving force of productive generalizations. In this paper, we overview a few models of dynamic lexical organization based on neural network architectures that are purported to meet this challenging view. In particular, we illustrate a novel family of Kohonen self-organizing maps (T2HSOMs) that have the potential of simulating competitive storage of symbolic time series while exhibiting interesting properties of morphological organization and generalization. The model, tested on training samples of as morphologically diverse languages as Italian, German and Arabic, shows sensitivity to manifold types of morphological structure and can be used to bootstrap morphological knowledge in an unsupervised way.

First International Workshop on Lexical Resources, pp. 32–41, Ljubljana Slovenia, 1-5 Agosto 2011

Keywords

Mental Lexicon, Self-organizing Maps, Morphology

CNR authors

Pirrelli Vito, Marzi Claudia, Ferro Marcello

CNR institutes

ILC – Istituto di linguistica computazionale "Antonio Zampolli"

ID: 205490

Year: 2011

Type: Contributo in atti di convegno

Creation: 2013-02-25 16:15:59.000

Last update: 2014-11-24 15:13:30.000

External IDs

CNR OAI-PMH: oai:it.cnr:prodotti:205490

PUMA: cnr.ilc/2011-A2-005