Connectionist models of reading

2007 ◽  
pp. 234-250 ◽  
Author(s):  
Mark S. Seidenberg
Keyword(s):  
Computational Models ◽  
Distributed Processing ◽  
Causal Effects ◽  
Task Orientation ◽  
Connectionist Models ◽  
Grain Sizes ◽  
Skilled Reading ◽  
Simple Processing ◽  
Reading Impairments ◽  
Models Of Reading

Connectionist computational models have been extensively used in the study of reading: how children learn to read, skilled reading, and reading impairments (dyslexia). The models are computer programs that simulate detailed aspects of behaviour. This article provides an overview of connectionist models of reading, with an emphasis on the “triangle” framework. The term “connectionism” refers to a broad, varied set of ideas, loosely connected by an emphasis on the notion that complexity, at different grain sizes or scales ranging from neurons to overt behaviour, emerges from the aggregate behaviour of large networks of simple processing units. This article focuses on the parallel distributed processing variety developed by Rumelhart, McClelland, and Hinton (1986). First, it describes basic elements of connectionist models of reading: task orientation, distributed representations, learning, hidden units, and experience. The article then looks at how models are used to establish causal effects, along with quasiregularity and division of labor.

scholarly journals A review of computational models of word recognition and pronunciation

2020 ◽  
Author(s):  
M. Alex Kelly
Keyword(s):  
Computational Models ◽  
Distributed Processing ◽  
Model Development ◽  
Dual Processing ◽  
Activation Model ◽  
Development Models ◽  
Dual Route ◽  
Visual Attention Mechanism ◽  
Consistency Effects ◽  
Models Of Reading

How do we recognize words and assign a pronunciation? Computational models provide a formal description of the mechanisms and principles that guide the reading process. I review and evaluate the Interactive-Activation Model (IAM), Dual Route Cascaded (DRC) model, the Parallel Distributed Processing (PDP) model, and the Connectionist Dual Processing (CDP) model, as well as LEX, a variant of the MINERVA model of memory. I evaluate each model’s ability to account for consistency effects, serial effects, syllable effects, and phonological effects. Consistency effects pose a problem for the rule-based pronunciation of the DRC. Serial effects pose a problem for the purely parallel PDP models. Phonological effects pose a problem for all models save CDP. All models suffer from the distribution problem, weakening each model’s ability to learn spelling-to-sound relationships. LEX is the only model that handles polysyllabic words. As none of the models provide a complete answer to the question of ‘how do we read?’, ‘how do we pronounce?’, or ‘how do we recognize words?’, I outline a set of principles as guidelines for future model development. Models of reading should learn, include a visual attention mechanism, be sensitive to phonology, and account for meaning and spelling in addition to recognizing words and pronouncing them.

scholarly journals Neural Components of Reading Revealed by Distributed and Symbolic Computational Models

2020 ◽  
Vol 1 (4) ◽  
pp. 381-401
Author(s):  
Ryan Staples ◽  
William W. Graves
Keyword(s):  
Neural Activity ◽  
Neural Coding ◽  
Computational Models ◽  
Ann Model ◽  
Distributed Representations ◽  
Neural Data ◽  
Cognitive Components ◽  
Components Of Reading ◽  
Artificial Neural Network Ann ◽  
Models Of Reading

Determining how the cognitive components of reading—orthographic, phonological, and semantic representations—are instantiated in the brain has been a long-standing goal of psychology and human cognitive neuroscience. The two most prominent computational models of reading instantiate different cognitive processes, implying different neural processes. Artificial neural network (ANN) models of reading posit nonsymbolic, distributed representations. The dual-route cascaded (DRC) model instead suggests two routes of processing, one representing symbolic rules of spelling–to–sound correspondence, the other representing orthographic and phonological lexicons. These models are not adjudicated by behavioral data and have never before been directly compared in terms of neural plausibility. We used representational similarity analysis to compare the predictions of these models to neural data from participants reading aloud. Both the ANN and DRC model representations corresponded to neural activity. However, the ANN model representations correlated to more reading-relevant areas of cortex. When contributions from the DRC model were statistically controlled, partial correlations revealed that the ANN model accounted for significant variance in the neural data. The opposite analysis, examining the variance explained by the DRC model with contributions from the ANN model factored out, revealed no correspondence to neural activity. Our results suggest that ANNs trained using distributed representations provide a better correspondence between cognitive and neural coding. Additionally, this framework provides a principled approach for comparing computational models of cognitive function to gain insight into neural representations.

Bringing Computational Models of Word Naming Down to the Item Level

1997 ◽  
Vol 8 (6) ◽  
pp. 411-416 ◽  
Author(s):  
Daniel H. Spieler ◽  
David A. Balota
Keyword(s):  
Word Recognition ◽  
Computational Models ◽  
Predictive Power ◽  
Neighborhood Density ◽  
Word Naming ◽  
Connectionist Models ◽  
Item Level ◽  
The Individual ◽  
Level Performance ◽  
Naming Latencies

Early noncomputational models of word recognition have typically attempted to account for effects of categorical factors such as word frequency (high vs low) and spelling-to-sound regularity (regular vs irregular) More recent computational models that adhere to general connectionist principles hold the promise of being sensitive to underlying item differences that are only approximated by these categorical factors In contrast to earlier models, these connectionist models provide predictions of performance for individual items In the present study, we used the item-level estimates from two connectionist models (Plaut, McClelland, Seidenberg, & Patterson, 1996, Seidenberg & McClelland, 1989) to predict naming latencies on the individual items on which the models were trained The results indicate that the models capture, at best, slightly more variance than simple log frequency and substantially less than the combined predictive power of log frequency, neighborhood density, and orthographic length. The discussion focuses on the importance of examining the item-level performance of word-naming models and possible approaches that may improve the models' sensitivity to such item differences

The Structure-Oriented Approach in Ethology: Network Models and Sex-Ratio Adjustments in Parasitic Wasps

Behaviour ◽  
1990 ◽  
Vol 114 (1-4) ◽  
pp. 148-160 ◽  
Author(s):  
Felix Putters ◽  
Marijke Vonk
Keyword(s):  
Sex Ratio ◽  
Distributed Processing ◽  
Network Models ◽  
Animal Behaviour ◽  
Neural Nets ◽  
Parasitic Wasps ◽  
Connectionist Models ◽  
Parallel Distributed Processing ◽  
Oriented Approach

AbstractIt is argued that connectionist models (neural nets, parallel distributed processing systems) have a great potential for changing our perspectives on animal behaviour. The approach is more structure-oriented than the conventional explanation by analogy. The method is used to analyze the principles of organization underlying oviposition decisions of parasitic wasps.

scholarly journals Folk psychology, eliminativism, and the present state of connectionism

2021 ◽  
Vol 64 (1) ◽  
pp. 173-196
Author(s):  
Vanja Subotic
Keyword(s):  
Cognitive Science ◽  
Cognitive Processing ◽  
Folk Psychology ◽  
Short Term Memory ◽  
Distributed Processing ◽  
Cognitive Architecture ◽  
Propositional Attitudes ◽  
Connectionist Models ◽  
Short Term ◽  
Long Short Term Memory

Three decades ago, William Ramsey, Steven Stich & Joseph Garon put forward an argument in favor of the following conditional: if connectionist models that implement parallelly distributed processing represent faithfully human cognitive processing, eliminativism about propositional attitudes is true. The corollary of their argument (if it proves to be sound) is that there is no place for folk psychology in contemporary cognitive science. This understanding of connectionism as a hypothesis about cognitive architecture compatible with eliminativism is also endorsed by Paul Churchland, a radical opponent of folk psychology and a prominent supporter of eliminative materialism. I aim to examine whether current connectionist models based on long-short term memory (LSTM) neural networks can back up these arguments in favor of eliminativism. Nonetheless, I will rather put my faith in the eliminativism of the limited domain. This position amount to the following claim: even though that connectionist cognitive science has no need whatsoever for folk psychology qua theory, this does not entail illegitimacy of folk psychology per se in other scientific domains, most notably in humanities, but only if one sees folk psychology as mere heuristics.

Letter and grapheme perception in English and Dutch

2009 ◽  
Vol 12 (1) ◽  
pp. 116-139 ◽  
Author(s):  
Patrick Bolger ◽  
Susanne R. Borgwaldt ◽  
Emőke Jakab
Keyword(s):  
Time Course ◽  
Reading Acquisition ◽  
Strong Version ◽  
Letter Detection ◽  
Data Basis ◽  
Skilled Reading ◽  
Syllabic Structure ◽  
Models Of Reading

In the study of reading, there is a debate about whether letters or graphemes are the primary units of perception. A promising data basis for empirically contributing to this debate can be gained from measuring the perception of single vowel letters compared to vowel digraphs. We used letter detection with masked pseudoword primes on pseudoword targets among skilled native readers in order to test for the existence and time course of vowel digraph effects during reading in deep (English) and shallow (Dutch) orthographies. Selecting these two languages, which are similar in terms of syllabic structure, allowed us to use exactly the same pseudoword stimuli. Results indicate that whereas the Dutch readers show letter effects at short prime durations and digraph effects at longer prime durations, the English readers show only letter effects. These findings are inconsistent with a strong version of the claim that graphemes are perceptual in nature, but consistent with models of reading acquisition and skilled reading that predict that, although letter effects always precede grapheme effects, grapheme activation proceeds faster in relatively shallow orthographies than in relatively deep ones.

Sign in / Sign up

Export Citation Format

Share Document