Abstract: . . .  Computational Psycholinguistics - Winter 2006 27 Learning in a nutshell Patterns are vectors on [0,1] Input pattern is passed through a weight matrix Net values are summed and squashed to [0,1] Output pattern is compared to target pattern Error between output and target is propagated back through weight matrix Weights are changed to minimize error Page 10 10 © Mat hew W. Crocker Computational Psycholinguistics - Winter 2006 28 Summary Connectionism is inspired by information processing in the brain Models typical y contain several layers of processing units Units correspond to a neuron (or group of neurons) Units sum weighted inputs from previous layers, and compute activation Output activation is passed to units of the next layer An input stimulus . . .
. . .  Psycholinguistics - Winter 2006 27 Learning in a nutshell Patterns are vectors on [0,1] Input pattern is passed through a weight matrix Net values are summed and squashed to [0,1] Output pattern is compared to target pattern Error between output and target is propagated back through weight matrix Weights are changed to minimize error Page 10 10 © Mat hew W. Crocker Computational Psycholinguistics - Winter 2006 28 Summary Connectionism is inspired by information processing in the brain Models typical y contain several layers of processing units Units correspond to a neuron (or group of neurons) Units sum weighted inputs from previous layers, and compute activation Output activation is passed to units of the next layer An input stimulus causes a “pattern . . .
. . .  psycholinguistics Page 1 1 Computational Psycholinguistics Lecture 9: Introduction to Connectionist Models Marshall R. Mayberry Computerlinguistik Universität des Saarlandes © Mat hew W. Crocker Computational Psycholinguistics - Winter 2006 2 Connectionist language learning: contents Connectionist Information Processing Simple connectionist models and their properties: The perceptron Multi-layer perceptrons: feed-forward networks and . . .
. . .  representations The encoding problem: Localist and distributed representations Generalisation and association Connectionist Models of Language Model ing acquisition of the English Past-Tense and reading aloud Processing sequences: Simple recurrent networks Model ing acquisition of hierarchical syntactic knowledge Tutorials: tLearn neural network simulator (Win/Mac/Linux) Introduction and Learning with tLearn Autoassociation and cluster analysis The English past-tense SRNs © Mat hew W. Crocker Computational Psycholinguistics - Winter 2006 3 Theories of Language Theories of the human language faculty: Knowledge: what is the nature of our knowledge of language? Rules and Representations Symbolic versus Distributed Explicit versus Implicit Acquisition: . . .
--3000,4,375,3009,15315