Cognitive Neuropsychology Deep Dyslexia
Total Page:16
File Type:pdf, Size:1020Kb
This article was downloaded by: [University of Toronto] On: 16 February 2010 Access details: Access Details: [subscription number 911810122] Publisher Psychology Press Informa Ltd Registered in England and Wales Registered Number: 1072954 Registered office: Mortimer House, 37- 41 Mortimer Street, London W1T 3JH, UK Cognitive Neuropsychology Publication details, including instructions for authors and subscription information: http://www.informaworld.com/smpp/title~content=t713659042 Deep dyslexia: A case study of connectionist neuropsychology David C. Plaut ab; Tim Shallice c a Carnegie Mellon University, Pittsburgh, USA b Department of Psychology, Carnegie Mellon University, Pittsburgh, PA, USA c University College London, London, UK To cite this Article Plaut, David C. and Shallice, Tim(1993) 'Deep dyslexia: A case study of connectionist neuropsychology', Cognitive Neuropsychology, 10: 5, 377 — 500 To link to this Article: DOI: 10.1080/02643299308253469 URL: http://dx.doi.org/10.1080/02643299308253469 PLEASE SCROLL DOWN FOR ARTICLE Full terms and conditions of use: http://www.informaworld.com/terms-and-conditions-of-access.pdf This article may be used for research, teaching and private study purposes. Any substantial or systematic reproduction, re-distribution, re-selling, loan or sub-licensing, systematic supply or distribution in any form to anyone is expressly forbidden. The publisher does not give any warranty express or implied or make any representation that the contents will be complete or accurate or up to date. The accuracy of any instructions, formulae and drug doses should be independently verified with primary sources. The publisher shall not be liable for any loss, actions, claims, proceedings, demand or costs or damages whatsoever or howsoever caused arising directly or indirectly in connection with or arising out of the use of this material. COGNITIVE NEUROPSYCHOLOGY. 1993.20 (5) 377-500 Deep Dyslexia: A Case Study of Connectionist Neuropsychology David C. Plaut Carnegie Mellon University, Pitisburgh, USA Tim Shallice University CoIIege London, London, UK Deep dyslexia is an acquired reading disorder marked by the Occurrence of semantic errors (e.g. reading RIVER as “ocean”). In addition, patients exhibit a number of other symptoms, including visual and morphological effects in their errors, a part-of-speech effect, and an advantage for concrete over abstract words. Deep dyslexia poses a distinct challenge for cognitive neuropsychology because there is little understanding of why such a variety of symptoms should CO-OCCur in virtually all known patients. Hinton and Shallice (1991) replicated the co-occurrence of visual and semantic errors by lesioning a recurrent connectionist network trained to map from orthography to semantics. Although the success of their simulations is encouraging. there is little understanding of what underlying principles are responsible for them. In this paper we evaluate and, where possible, improve on the most important design decisions made by Hinton and Shallice, relating to the task, the net- work architecture, the training procedure, and the testing procedure. We identify four properties of networks that underly their ability to reproduce the deep dyslexic symptom-complex: distributed orthographic and semantic representations, gradient descent learning, attractors for word meanings, and greater richness of concrete vs. abstract semantics. The first three of these are general connectionist principles and the last is based on earlier theorising. Requests for reprints should be sent to Dr. David C. Plaut, Department of Psychology, Carnegie Mellon University, Pittsburgh, PA, 15213-3890, USA (email: [email protected]). Most of the work reported in this paper was carried out while the authors were visiting Downloaded By: [University of Toronto] At: 19:49 16 February 2010 scientists in the Departments of Psychology and Computer Science at the University of Toronto. The work could not have been carried out without the generous support and guidance of Geoff Hinton, whom we believe deserves to be a co-author of this paper but would not be persuaded to be included. We would like to thank Gus Craik for his help with many aspects of our visits to Toronto. We also wish to thank Marlene Behrmann, Max Coltheart, Argye Hillis, and two anonymous reviewers for their detailed suggestions on an earlier version of the paper. All of the simulations were run on a Silicon Graphics Iris-4DR4os using an extended version of the Xenon simulator developed by Tony Plate. This research was supported by grant 87-2-36from the Alfred Sloan Foundation. Much of the work was carried out while Tim Shallice was a member of the Medical Research Council’s Applied Psychology Unit. @ 1993 Lawrence Erlbaum Associates Limited 378 PIAUT & SHALLICE Taken together, the results demonstrate the usefulness of a connectionist approach to understanding deep dyslexia in particular, and the viability of connectionist neuropsychology in general. INTRODUCTION Despite its familiarity as a concept in cognitive neuropsychology, deep dyslexia remains controversial. It was first suggested as a symptom-complex by Marshall and Newcombe (1973), who described two patients, GR and KU. These patients both made semantic errors in attempting to read aloud and also made visual and derivational errors. Coltheart (1980a) was able to add another 15 cases. Kremin (1982) added another 8 and over 10 more are referred to by Coltheart, Patterson, and Marshall (1987). Beginning with the semantic errors, Coltheart (1980a) also extended the list of common properties to 12 (examples of errors are from DE, Patterson & Marcel, 1977): 1. Semantic errors (e.g. BLOWING + “wind,” VIEW -P “scene,” NIGHT + “sleep,” GONE + “lost”). 2. Visual errors (e.g. WHILE -P “white,” SCANDAL ---* “sandals,” POLITE --+ “politics,” BADGE + “bandage”). 3. Function-word substitutions (e.g. WAS --+ “and,” ME + “my,” OFF + “from,” THEY + “the”). 4. Derivational errors (e.g., CLASSIFY + “class,” FACT -+ “facts,” MARRIAGE + “married,” BUY + “bought”). 5. Nonlexical derivation of phonology from print is impossible (e.g. pronouncing nonwords, judging if two nonwords rhyme). 6. Lexical derivation of phonology from print is impaired (e.g. judging if two words rhyme). 7. Words with low imageability/concreteness (e.g. JUSTICE) are harder to read than words with high imageability/concreteness(e.g. Downloaded By: [University of Toronto] At: 19:49 16 February 2010 TABLE). 8. Verbs are harder than adjectives, which are harder than nouns, in reading aloud; 9. Functions words are more difficult than content words in reading aloud. 10. Writing is impaired (spontaneous or to dictation). 11. Auditory-verbal short-term memory is impaired. 12. Whether a word can be read at all depends on its sentence context (e.g. FLY as a noun is easier than FLY as a verb). Given the uniformity of the patients’ symptoms, ‘Coltheart characterised the symptom-complex as a syndrome. DYSLEXIA & CONNECTIONIST NEUROPSYCHOLOGY 379 In the conclusion of their review article, “Deep dyslexia since 1980,” Coltheart et al. (1987) argue that de.ep dyslexia presents cognitive neuro- psychology with a major challenge. They raise two main issues specific to the domain of reading. First, they argue that standard “box-and-arrow” information-processing accounts of deep dyslexia (e.g. Morton & Patter- son, 1980) provide no explanation for the observed combination of symptoms. If a patient makes semantic errors in reading aloud, why are many other types of behaviour virtually always observed? Second, they point out that the standard explanations for semantic errors and for effects of concreteness involve different impairments along the semantic route (Coltheart et al., 1987, pp. 421422). The loss of semantic information for abstract words that explained visual errors in oral reading cannot readily explain semantic errors in oral reading, since semantic errors typically occur on moderately concrete words. The deficit in the semantic routine that gives a pretty account of semantic errors is, rather, an abnormal sloppiness in the procedure of addressing a phono- logical output code from a set of semantic features. Must we now post- ulate several different semantic-routine impairments in deep dyslexia, and if so, why do we not observe patients who have one but not the other: in particular, patients who make semantic errors but do not have difficulty with abstract words? Recently, Hinton and Shallice (1991) have put forward a connectionist approach to deep dyslexia that addresses the first of these points. They reproduced the co-occurrence of semantic, visual, and mixed visual-and- semantic errors by lesioning a connectionist network that develops UMUC- tors for word meanings. Although the success of their simulations is encouraging, there is little understanding of what underlying principles are responsible for them. In this paper, we evaluate and, where possible, improve on the most important design decisions made by Hinton and Shallice. First, we demonstrate the robustness of the account by examining network architectures different from the original model. We also improve Downloaded By: [University of Toronto] At: 19:49 16 February 2010 on the rather arbitrary way that the model realised an explicit response by extending it to generate phonological output from semantics. Next, we evaluate the significance of the particular learning procedure used to train the original model by re-implementing it in a more plausible connectionist formalism. Finally,