Plaut, D. C., McClelland, J. L., Seidenberg, M., & Patterson, K. E. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56 --115.  Home/Search   Document Details and Download   Summary   Related Articles   Check

....University of California, San Diego, 1999 Professor Garrison W. Cottrell, Chair As is true for almost all computational modeling of higher brain functions, debate currently rages over models which address the ability to read. Three current models (the single route connectionist system of (Plaut et al. 1996), the Dual Route Cascade of (Coltheart et al. 1993) and the connectionist Dual Process model of (Zorzi et al. 1998b) are examined in relation to various human reading data. An apparently serial effect seen in irregular words, where early irregularities cause more slowing in pronunciation time ....

....to various human reading data. An apparently serial effect seen in irregular words, where early irregularities cause more slowing in pronunciation time than late irregularities, is reanalyzed with respect to enemy friend ratios, and the results of subject experiments are presented. The work of (Plaut et al. 1996) is extended to handle multisyllabic words, modeling the effect seen in the human studies, and thus showing that the position of irregularity effect is not a serial phenomena after all. xv Chapter I Introduction POLONIUS: What do you read, My Lord HAMLET: Words, words, words. ....

[Article contains additional citation context not shown here]

Plaut, D., McClelland, J., Seidenberg, M., and Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103(1):56--115.

.... that distinguishes their own Dual Route Cascaded (DRC) model (Coltheart, Curtis, Atkins Haller, 1993; Coltheart, Langdon Haller, 1996; Coltheart Rastle, 1994; Rastle Coltheart, 1999b; Rastle Coltheart, 1998) from its major competitors, the Triangle connectionist model of reading (Plaut, McClelland, Seidenberg Patterson, 1996), the Multiple Levels Model (Norris, 1994) and the Connectionist Dual Process Model (Zorzi, Houghton Butterworth, 1998) Rastle and Coltheart point out that the DRC model alone possesses a serial mechanism to process the mapping between the orthographic and phonological form of words; the other ....

Plaut, D.C., McClelland, J.L., Seidenberg, M.S. & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains.

....are graphemically, morphologically, or semantically related ( Taft, 1991 ] for a review) The author is supported by a Medical Research Council studentship. Substantial progress has been made modelling the form based lexical representations in the light of graphemic or phonological similarity [ Plaut et al. 1994 ] but there is currently no principled measure of semantic similarity. Word meaning is much more difficult to quantify. The network described here is an attempt to address this problem. It is inspired by two relatively independent approaches to semantic representation from cognitive psychology ....

D. C. Plaut, J. L. McClelland, M. S. Seidenberg, and K. E. Patterson. Understanding normal and impaired word-reading: Computational principles in quasi-regular domains. Technical report, Carnegie Mellon University, 1994.

....in the retrieval of that trace. These networks have also been used to solve optimization problems and soft constraint satisfaction problems (Hopfield and Tank, 1986; Rumelhart et al. 1986b) They have played central roles in cognitive models of lexical access (Kawamoto, 1993) reading aloud (Plaut et al. 1996), visual object recognition (Mjolsness, 1991) and consciousness (Mathis and Mozer, 1995) In addition to their useful functional properties, attractor networks exhibit interesting dynamic behavior as activation states unfold in time. This allows attractor network models to make predictions about ....

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., and Patterson, K. (1996). Understanding normal and impaired word reading: Computational 26 principles in quasi-regular domains. Psychological Review, 103(1):56-- 115.

....representing semantic information must now be addressed if any headway is to be made. Although semantics obviously plays a very important role in language, cognitive models concerned with language have either not attempted to implement this component [2, 20] or implemented it only on a smallscale [3, 4, 6, 7, 16, 17, 18]. If the experimental results from tasks such as reading and lexical decision are to be simulated, there must be serious investigations into how semantics can be represented on a large scale, e.g. for thousands of words. Recently, work has begun on using large corpora to extract semantic ....

....values and assess how good the best resultant semantic representations really are. 3 Implementation of the Semantic Component of Reading and Lexical Decision Models Psychological models of reading and lexical decision have been implemented using neural networks with varying degrees of success [2, 3, 4, 7, 16, 17, 18, 20]. A major problem has been in implementing the semantic component of such models since there is no established theory of what should be represented or how. Modellers have tended towards using simple notions of semantic micro feature representations as a practical way of implementing the lexical ....

Plaut DC, McClelland JL, Seidenberg MS & Patterson KE. Understanding Normal and Impaired Word Reading: Computational Principles in Quasi-Regular Domains. Psychological Review 1996; 103:56-115

.... 1991; Wolpert, 1992; Vapnik Chervonenkis, 1971) Psychologically, generalization is important for modeling human nonword reading performance (e.g. the fact that people generally pronounce nonwords like nust according to the regularities of the English language; Seidenberg McClelland, 1989; Plaut, McClelland, Seidenberg, Patterson, 1996), and more generally for understanding how human and animal cognition can exhibit flexibility in ever changing environments. Thus, the fact that interactivity impairs generalization, often to a very significant degree, is problematic for any attempt to develop a biologically, computationally, and ....

....interactive networks, the simulations in this paper provide more direct evidence that conjunctive attractor dynamics are the source of the problem. However, some other reports in the literature have demonstrated successful generalization in interactive networks (e.g. Plaut McClelland, 1993; Plaut et al. 1996). Indeed, Plaut and colleagues attributed the generalization success of their networks to their ability to form componential attractors, where attractor dynamics operated within, but not between, the representations of the compositional features of letters phonemes of words (just like the ....

[Article contains additional citation context not shown here]

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. E. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56--115.

....Our models provide a principled and parsimonious account. 1 Introduction Connectionist modelling has been a fruitful approach to understanding normal and impaired cognition, as demonstrated by Plaut, McClelland, Seidenberg and Patterson s (1996) modelling of normal and impaired word recognition [1]. Insofar as distributed representations are a critical aspect of such models, they instantiate an important anatomical fact about the brain: the brain itself seems to rely on distributed representations. However, such models often go no further in representing anatomical reality; thus, for ....

Plaut, D.C., McClelland, J.L., Seidenberg, M.S. & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56--115.

....(e.g. Cottrell and Small, 1984) iii) Sometimes, spurious attractive states arose which corresponded to no interpretation (e.g. Cottrell and Small, 1984) In Section 3.3. 2, we show that certain spurious states may provide a plausible model of parsing of an ungrammatical string (cf. Plaut, McClelland, Seidenberg, and Patterson, 1996). iv) Syntactic and semantic information simultaneously constrained parsing (e.g. Cottrell Small, 1983) The development of the backpropagation algorithm (Rumelhart et al. 1986) led to a new class of learning based connectionist parsing models. Currently, the most successful models are ....

Plaut, D.C., McClelland, J.L., Seidenberg, M.S., & Patterson, K.E. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56--115.

....simulations show how such an approach can account for many aspects of reading (and related tasks) that are not possible without the influence of the semantic system. Further possible refinements of this approach are discussed. Introduction Recently, improved versions (Bullinaria, 1994a,b; Plaut et al. 1994) of earlier connectionist models of reading (Sejnowski Rosenberg, 1987; Seidenberg McClelland, 1989) have begun to achieve a performance on their training data and an ability to generalize to unseen words and non words that is comparable to human subjects. These models also exhibit some ....

....clusters (12) vowel clusters (10) and offset phoneme clusters (12) and the orthography consisted of units for the corresponding three sets of letter clusters (12, 9 and 13) plus two units to code for the presence or absence of a final e . This constitutes a highly simplified version of the Plaut et al. 1994) representation, that (after removing homographs and homophones) allows the representation of 513 real monosyllabic words from the standard Seidenberg McClelland (1989) corpus. Rather than attempting to generate a realistic semantic representation for these words, we simply assigned each a ....

[Article contains additional citation context not shown here]

Plaut, D.C., McClelland, J.L., Seidenberg, M.S. & Patterson, K.E. (1994). Understanding Normal and Impaired Word Reading: Computational Principles in Quasi-Regular Domains. Psychological Review, to appear.

.... introducing our model, we present three existing examples of models that impose different restrictions on knowledge representation and processing: i) a dual route model (Coltheart, 1978) in which lexical knowledge and spelling to phonology rules are strictly separated, ii) a single route model (Plaut et al. 1996), in which knowledge is represented and processed in a single system, allowing for any spelling phonology correspondence ranging from rules to lexical knowledge, and (iii) a multiple level processing model (Norris, 1993) which allows for predefined levels of correspondences between spelling and ....

....regular pronunciations can be correctly converted from text to speech via the rule based route as well as via the lexical route, whereas irregular words can only be converted to speech correctly via the lexical route. Single route theory (see, e.g. Glushko, 1979; Seidenberg McClelland, 1989; Plaut et al. 1996) states that skilled reading aloud is accomplished by a single mechanism, in which a less strict distinction exists between different spelling phonology correspondence levels. Glushko (1979) introduces the idea of a similarity matching component, that converts spelling strings to their ....

[Article contains additional citation context not shown here]

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., and Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56--115.

....the combined model of normal processing and damage in aphasia. One could even imagine using simulation results to provide insight into the breakdown occurring in specific patients. Examples of recent computational investigations of low level language processing include the word reading models of Plaut, McClelland, Seidenberg, and Patterson (1996) and Shallice, Glasspool, and Houghton (1995) and the word production models of Levelt, Roelofs, and Meyer (1999) and Dell, Schwartz, Martin, Saffran, and Gagnon (1997) In practice, a computational model is often constructed with the aim of testing claims about one or two specific theoretical ....

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103 (1), 56--115.

....models, these spurious states were considered a liability because some parsable sentences got stuck in them. Below, in Section 3.3. 2, we show that certain spurious states are an asset in that they provide a plausible model of what happens when one attempts to parse an ungrammatical string (cf. Plaut et al. 1996); iv) Syntactic and semantic information were used simultaneously to constrain the parse (e.g. Cottrell Small, 1983; Cottrell, 1985) The development of the backpropagation algorithm for learning (Rumelhart et al. 1986) and its promotion as a useful tool in psychological modeling (Rumelhart ....

Plaut, D.C., McClelland, J.L., Seidenberg, M.S., & Patterson, K.E. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56--115.

....processes and so on. We find other examples also outside of cognitive modeling, such as modeling economic processes, physical phenomena, etc. A widely used practice in connectionist natural language modeling is localistic and handcrafted feature based encoding [Seidenberg, 1989] Elman 1990] [Plaut, 1996], Henderson, 1998] which restricts the capacity of the processing system. It would be preferable that those representations evolve in the course of experiencing the language in its external sequential form, which is in accordance with our capacity to learn any language without any prior ....

....of representations of the processed sequences. This question is especially apparent in natural language modeling. In earlier connectionist models, the lexemes were represented in a static manner with some artificial and not always effective encoding schemes (e.g. in [Seidenberg, 1989] and [Plaut, 1996], and sentences consisted of some artificial and very limited in number words [Elman, 1990] Miikkulainen, 1991] Tabor, 1997] When modeling some other problems, for example learning lexical phonotactics [Stoianov, 1998] Stoianov, 1999] or learning the mapping from orthographic to phonetic ....

Plaut, D. C., McClelland, J., Seidenberg, M., Patterson, K., Understanding normal and impaired word reading: computational principles in quasi-regular domains, Psychological Review, 103, 56, 1996.

....mapping from written to spoken forms in natural language. Connectionist models, if successfully trained on this problem and if their performance correlates to human performance in reading, can supply a framework for lexical processing. For example, Seidenberg McClelland (1989, hence SM89) and Plaut et al. (1996) suggest that a singleroute distributed process performs this transformation, as opposed to the symbolic dual route model which claims that the reader must also have a lexical route that handles exceptions, or irregular words(Coltheart 1980, 1993) The dual route model has a connectionist ....

....this mapping. Since the first successful connectionist system that transformed text to phonemes the NETtalk (Sejnowski Rosenberg 1987) a number of other connectionist architectures that model the human reading process have been suggested. Among these, Seidenberg McClelland (1989) and Plaut, McClelland, Seidenberg Patterson (1996), have been influential on connectionist NLP research (e.g. Milostan Cottrell 1998; Harm 1998) A connectionist model that takes the opposite side of the single route dual route controversy was proposed by Zorzi (1998) who suggests that the MLP might benefit from an extra set of ....

[Article contains additional citation context not shown here]

Plaut, D.C., J McClelland, M Seidenberg & K Patterson (1996). Understanding Normal and Impaired Word Reading: Computational Principles in Quasi-Regular Domains. Psychological Review, 103, pp.56-115.

....explain the human neural system as well as higher order cognitive functions, including language processing. As artificial connectionist models prove successful with the problems they have been designed for, it is challenging to seek better approximation of human cognitive functions. For example, Plaut et al. 1996) claim to model effects observed in human lexical processing. In this comprehensive study, performance analyses of Multilayered Perceptron and Attractor Neural Networks were presented. Both models were trained on mapping from orthography to phonology and employed static lexical representations. ....

.... the output domain, which is a distributed representation of phonemes (possible successors) There is a number of NN models; most of them are designed for static data processing, and many connectionist language implementations employ such static 4 Ivelin Stoianov and John Nerbonne models, e.g. Plaut et al. 1996). Language takes place in time, however. We produce and hear sequences of sounds, which we may represent statically, but at least spoken language is always expressed sequentially. Therefore, an NN model for language should have an internal dynamics allowing reaction that depends on the past input ....

[Article contains additional citation context not shown here]

Plaut, D., McClelland, J., Seidenberg, M. and Patterson, K.(1996), Understanding normal and impaired word reading: Computational principles in quasiregular domains, Psychological Review 103, 56--115.

....not correspond to any of the normal inflection types when confronted with words that look like the irregulars it has learned may be seen as analogous to this difficulty experienced by humans. The unclassified responses may potentially be eliminated by using a phonological attractor at the output (Plaut, McClelland, Seidenberg Patterson, 1996). This is a direction for future research. Conclusion Selective attention was shown to be a powerful aid to learning in neural networks. Both types of selective attention networks mastered the training set completely, while the network without selective attention did not. The mechanism of ....

Plaut, D.C., McClelland, J.L., Seidenberg, M.S., & Patterson, K.E. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56115.

....of language is represented in the mind of the language user. SIMULATING GRAMMATICALITY JUDGMENTS 6 Grammaticality Judgments The approach that we have briefly summarized is beginning to be applied to a range of questions about acquisition, processing, and breakdown following brain injury (Plaut, McClelland, Seidenberg, Patterson, 1995; MacDonald et al. 1994; Seidenberg, 1997) Here we want to return to the concept of grammaticality and to the task of making grammaticality judgments, both of which are central to the standard approach. We have suggested that knowledge of language is not a set of rules for generating sentences ....

Plaut, D., McClelland, J., Seidenberg, M., & Patterson, K. (1995). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review.

....knowledge being Figure 3: Sentence Gestalt Network Model MAKE mAk Phonology Context Meaning Orthography Figure 4: Linguistic Processing Framework otherwise coded into the system by the others. 4. 5 Generalization and Novelty Although the linguistic processing model developed by Plaut et al. [15] focuses mainly on learning to read (bold connections in Figure 4) the system they have developed demonstrates some interesting behavior which may be applicable to modeling fast mapping. Plaut and his co authors develop a recurrent network which learns to map orthography, the printed letters of a ....

....a network similar to that of St. John and McClelland be used to determine which input features deserve the most attention. These feature vectors may then be used as the basis for a system similar to the DISCERN model. Finally, we propose a recurrent system similar to the one used by Plaut et al. [15] to represent the long term memory. This type of system provides the necessary generalization and ability to represent novel inputs which is necessary for the representation of memory. In this model, the sentence gestalt network attention mechanism would steer the focus of the network to ....

Plaut, D. C., McClelland, J. L., Seidenberg, M. S. & Patterson, K. E. (1994). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Submitted to Psychological Review.

....to phonology. In a complete model of reading, message understanding must play a role, but many psycholinguistic phenomena can be explained in the context of this simple mapping task. A difficulty in learning this mapping is that in a language such as English, the mapping is quasiregular (Plaut et al. 1996); there are a wide range of exceptions to the general rules. As with nearly all psychological phenomena, more frequent stimuli are processed faster, leading to shorter naming latencies. The regularity of mapping interacts with this variable, a robust finding that is well explained by connectionist ....

....irregularities will not be as affected by conflicting information. This result is validated by simulations with the one syllable DRC model (Coltheart and Rastle, 1994) Several connectionist systems have been developed to model the orthography to phonology process (Seidenberg and McClelland, 1989; Plaut et al. 1996). These connectionist models provide evidence that the task, with accompanying phenomena, can be learned through a single mechanism. In particular, Plaut et al. henceforth PMSP) develop a recurrent network which duplicates the naming latencies appropriate to their data set, consisting of ....

Plaut, D., McClelland, J., Seidenberg, M., and Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103(1):56--115.

....that response times to words in a dense semantic neighborhood will be faster than words which have few semantically similar neighbors in the language. This is conceptually consistent with the neighborhood effect seen in the mapping from orthography to phonology [Seidenberg McClelland, 1989, Plaut et al. 1996] in that patterns with many neighbors are faster in both pathways, but since there is no regularity in the random mapping used here, it is clear that the cause of this effect is different than that of previous experiments. We also report a rather distressing finding. Reaction time in this model ....

....recurrent networks which are trained to settle to a stable output. Using attractor network models, A number of experiments have been reported which lend credence to the idea that many of the effects seen during lexical access with human subjects are naturally modeled using attractor networks. Plaut et al. 1996) showed that the regularity effect demonstrated by [Seidenberg McClelland, 1989] holds for attractor networks as well as for feed forward networks. Plaut (1995) demonstrated semantic and associative priming in an attractor network model which implemented a random mapping whose intent was to ....

[Article contains additional citation context not shown here]

Plaut, D. C., McClelland, J. L., Seidenberg, M. S. & Patterson, K. E. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103(1):56--115.

....English orthography and phonology to avoid a completely position specific representation. Their model learns to read non words very well, but it does so by building in a lot of knowledge into the representation, rather than having the network learn this knowledge. One could plausibly assume (cf. Plaut et al. 1996) that some of this knowledge is acquired prior to reading acquisition; that is, children normally know how to pronounce words (i.e. talk) before they start learning to read. This idea is explored by Harm, Altmann and Seidenberg (1994) who showed that pretraining a network on phonology can help ....

....are damaged but the exceptions are preserved does not occur. It remains possible that some realistic single route model of reading, incorporating factors which have been claimed to be important to connectionist accounts of reading such as word frequency and phonological consistency effects (cf. Plaut et al. 1996) might give rise to the relevant double dissociation. However, Bullinaria and Chater s results indicate that modeling phonological dyslexia is potentially a major challenge for any single route connectionist model of reading. Single and dual route theorists argue about whether non word and ....

[Article contains additional citation context not shown here]

Plaut, D., McClelland, J.L., Seidenberg, M.S., & Patterson, K.E. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56--115.

....of noise to the connection strengths or the destruction of connections or units, is thought to resemble the symptoms of brain damaged patients. Simulations of disorder reported in the literature generally focus upon modelling language breakdown, including reading disorders, such as dyslexia (Plaut et al. 1994) and spoken disorders, for example aphasic naming (Dell et al. 1995) The systematic lesioning of a modular architecture in which destruction of connection or the modification of parameters in more than one component module may contribute to the modelling of a range of disorders has also been ....

Plaut, D. C., McClelland, J. L., Seidenberg, M. S. & Patterson, K.E. (1994) Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Carnegie Mellon University, Pittsburgh, PA: Technical Report PDP.CNS.94.5.

....of a naming response is as follows. A representation of pronunciation is built up over time, based on the results of processing at one or more other levels of representation (e.g. lexical, semantic, orthographic, and syntactic knowledge; Coltheart, Curtis, Atkins, Haller, 1993; Kawamoto, 1988; Plaut, McClelland, Seidenberg, Patterson, 1996). When the pronunciation is resolved according to some criterion of completeness, the response is initiated. Often, the exact nature of the criterion is left unexplained, but a common assumption is that a response is initiated as soon as an entire pronunciation is complete by some criterion (but ....

.... the criterion is left unexplained, but a common assumption is that a response is initiated as soon as an entire pronunciation is complete by some criterion (but see Kawamoto, Kello, Jones, Bame, 1998) For example, activation or saturation thresholds have been used (e.g. Coltheart et al. 1993; Plaut et al. 1996). One reason why issues of response generation are often neglected is what Bock (1996) has termed the mind in the mouth assumption. She argued that researchers often implicitly assume that articulation provides a relatively direct reflection of cognitive processing, but the link from cognition ....

[Article contains additional citation context not shown here]

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains.

....three connectionist models of word reading. The models represent simplified versions of two alternate hypotheses concerning the architecture of the word reading system. Simulations 1 and 2 implemented versions of the triangle framework (Harm and Seidenberg, 1999; Seidenberg and McClelland, 1989; Plaut et al. 1996), in which orthography is mapped onto phonology both directly and via semantics. Simulation 3 implemented a novel architecture in which orthography is mapped directly onto internal representations that mediate the mapping between semantics and phonology. All three simulations showed that ....

....the word reading system as a whole. In Kello and Plaut (2000) we made an initial foray into this issue by examining the effect of a time criterion in two competing models of word reading: the dual route cascaded (DRC) model (Coltheart et al. 1993) and the distributed attractor model reported in Plaut et al. 1996). Both simulations failed to account for the result that, as tempo increased, the number of spelling sound errors remained constant whereas the occurrence of other types of errors increased. In the simulations, the occurrence of all errors, including spelling sound errors, increased as the time ....

[Article contains additional citation context not shown here]

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains.

....controlled, but contrary to the earlier results, significant effects of frequency were observed with AoA controlled. The fact that AoA affects performance independent of frequency seems to present a challenge for models of skilled word recognition (e.g. Coltheart, Curtis, Atkins, Haller, 1993; Plaut, McClelland, Seidenberg, Patterson, 1996; Seidenberg McClelland, 1989) that do not explicitly take this factor into account. In this article we provide a critique of AoA both as a variable in behavioral experiments and AGE OF ACQUISITION EFFECTS 3 as a concept relevant to theories of word recognition. We argue that previous studies ....

....ranging from initial acquisition to skilled performance. Connectionist modeling Connectionist models employing distributed representations and gradual learning from experience provide a theoretical framework for examining effects of frequency of exposure over time (e.g. Harm Seidenberg, 1999; Plaut et al. 1996; Seidenberg McClelland, 1989) Such models illustrate three points relevant to the AoA debate. First, both when a word is learned and how often it is used are intrinsically related insofar as both affect the weights that govern network performance. Second, how quickly a word is learned depends ....

[Article contains additional citation context not shown here]

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. E. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56-115.

....taken place in the domain of English word reading. An early connectionist model (Seidenberg McClelland, 1989) provided a good account of word reading but was poor at pronouncing word like pseudowords (e.g. MAVE; Besner, Twilley, McCann, Seergobin, 1990) A more recent series of simulations (Plaut, McClelland, Seidenberg, Patterson, 1996) showed that the limitations of this preliminary model stemmed from the model s use of poorly structured orthographic and phonological representations. By contrast, networks with more appropriate representations were able to learn to pronounce both regular and exception words, and yet al..so ....

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56-115.

....with domain characteristics. It is a program of scientific research to discover what the principles and domain characteristics are and how they give rise to different types of representations. As a starting place in the discovery of the relevant principles, we have suggested (McClelland, 1993; Plaut, McClelland, Seidenberg, Patterson, 1996) that the principles include the following: that the activations and connection weights that support representation and processing are graded in nature; that processing is intrinsically gradual, stochastic, and interactive; and that mechanisms underlying processing adapt to task constraints. ....

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains.

.... notion that language knowledge takes the form of rules (although such rules may be expressed in terms of connections between localist connectionist units; see, e.g. Norris, 1994; Reggia, Marsland, Berndt, 1988) The third response to the challenge, adopted by distributed connectionist theories (Plaut, McClelland, Seidenberg, Patterson, 1996; Seidenberg McClelland, 1989; Van Orden, Pennington, Stone, 1990; Van Orden Goldinger, 1994) and elaborated in the current chapter, is more radical. It eschews the notion that the knowledge supporting online language performance takes the form of explicit rules. Of course, such performance ....

....McCann, Seergobin, 1990) and at lexical decision under some conditions (Besner et al. 1990; Fera Besner, 1992) Thus, the model failed to refute traditional claims that localist, word specific representations and separate mechanisms are necessary to account for skilled reading. More recently, Plaut, McClelland, Seidenberg, and Patterson (1996, also see Plaut McClelland, 1993; Seidenberg, Plaut, Petersen, McClelland, McRae, 1994) have shown that the limitations of the Seidenberg and McClelland model in pronouncing nonwords stems not from any general limitation in the abilities of connectionist networks in quasi regular domains (as ....

[Article contains additional citation context not shown here]

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains.

....by the removal of some proportion of the units and or connections in certain regions of the model. Perhaps the most widely investigated class of disorders concern selective impairments in reading the acquired dyslexias (Hinton Shallice, 1991; Mozer Behrmann, 1990; Plaut Shallice, 1993; Plaut, McClelland, Seidenberg, Patterson, 1996). The current work extends the relevance of connectionist modeling in cognitive neuropsychology by demonstrating that the same computational principles which are effective for understanding normal cognitive processing, and the effects of brain damage, can also provide insight into the nature of ....

....provides a brief overview of findings from empirical studies attempting to remediate the reading impairments of acquired dyslexic patients. Following this, two computational simulations are presented, both involving networks that that are trained to derive the meanings of written words (see Plaut, 1996, for more details) The first demonstrated that, in retraining a network after damage, the degree of relearning and generalization depended on the location of the lesion. The results have interesting implications for understanding the nature and variability of recovery in patients. In the second ....

[Article contains additional citation context not shown here]

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56--115.

....viewed as no coincidental association that all reported patients with marked surface dyslexia have also been profoundly anomic, suggesting reduced semantic phonological activation. The chapter summarizes the simulation of surface dyslexia in the computational model of reading developed by Plaut, McClelland, Seidenberg, and Patterson (1996), and presents new data from three surface alexic patients. The graded consistency effects in the patients reading performance are more compatible with the distributed connectionist framework than with dual route models maintaining a strict dichotomy between regular and exception words. 1 ....

....had demonstrated the adequacy of a single mechanism for reading both exception words and nonwords, because the original simulation achieved notably less success than most human readers do in generalizing its knowledge to the pronunciation of nonwords. In the most recent phase of this debate, Plaut, McClelland, Seidenberg and Patterson (1996) presented four new simulations of the O P computation in English. As one principal development on their predecessor (Seidenberg McClelland, 1989) the networks in the Plaut et al. model employed orthographic and phonological representations designed to capture more successfully the similarities ....

[Article contains additional citation context not shown here]

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56--115.

....included training patterns consisting of each single grapheme and the corresponding phoneme. These were included because children are taught these correspondences in the process of learning to read, although equivalent networks not trained on these correspondences exhibit equivalent behavior (see Plaut, McClelland, Seidenberg, Patterson, 1994). Furthermore, rather than present each word with a probability proportional to its frequency of occurrence (Kucera Francis, 1967) and update the weights immediately, we accumulated the error derivatives for the training cases, each weighted by its frequency, before changing the weights. This ....

....that two separate systems are needed to read pronounceable nonwords and exception words. The present article addresses only one of their arguments, and so we cannot claim to have produced a single route system that provides a complete account of reading of regular and exception words (although see Plaut et al. 1994, for additional simulations and discussion) However, we have addressed what we take to be the most central argument against a single mechanism account. Coltheart and colleagues state in their article that dual route theorists claim that there cannot be a single procedure that can correctly ....

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1994). Understanding normal and impaired word reading: Computational principles in quasi-regular domains (Technical Report PDP.CNS.94.5). Pittsburgh, PA: Department of Psychology, Carnegie Mellon University.

....the initiation of a naming response is as follows. A representation of pronunciation is built up over time, based on the results of processing at one or more other levels of representation (e.g. lexical, semantic, orthographic, and syntactic knowledge; Coltheart et al. 1993; Kawamoto, 1993; Plaut et al. 1996). When the pronunciation is resolved according to some criterion of completeness, the response is initiated. Often, the exact nature of the criterion is left unexplained, but a common assumption is that a response is initiated as soon as an entire pronunciation is complete by some criterion (e.g. ....

....to some criterion of completeness, the response is initiated. Often, the exact nature of the criterion is left unexplained, but a common assumption is that a response is initiated as soon as an entire pronunciation is complete by some criterion (e.g. an activation or saturation threshold; Plaut et al. 1996; Coltheart et al. 1993; for an alternate view, see Kawamoto, Kello, Jones, Bame, 1998) One reason why issues of response generation are often neglected is what Bock (1996) has termed the mind in the mouth assumption. She argued that researchers often implicitly assume that articulation ....

[Article contains additional citation context not shown here]

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains.

.... most theories ascribe the central characteristic of surface dyslexia regularization of low frequency exception words to the operation of the phonological route, either in normal operation when isolated from semantics (Coltheart et al. 1993; Plaut, Behrmann, Patterson, McClelland, 1993; Plaut et al. 1994), or after partial damage (Patterson et al. 1990; Shallice McCarthy, 1985) For this reason, it would be inappropriate to use an implementation of an isolated semantic route to model the pattern of errors in surface dyslexia, and its change over the course of rehabilitation. Plaut: Relearning ....

Plaut: Relearning After Damage in Connectionist Networks 38 Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. E. (1994). Understanding normal and impaired word reading: Computational principles in quasi-regular domains (Technical Report PDP.CNS.94.5). Pittsburgh, PA: Department of Psychology, Carnegie Mellon University.

....fundamental properties with them. As a result, it is natural to damage a PDP system to varying degrees. Moreover, such systems have been shown to be sensitive to relative degrees of systematicity within a single task, both in terms of rate of acquisition and in terms of the effects of damage (Plaut, McClelland, Seidenberg, Patterson, 1996; Seidenberg McClelland, 1989) In this paper, we explore whether optic aphasia can be accounted for by the effects of damage to a PDP network in which multiple inputoutput mappings of varying systematicity are mediated by the same internal (semantic) representations. Simulation 1: Basic ....

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56--115.

No context found.

Plaut, D. C., McClelland, J. L., Seidenberg, M., & Patterson, K. E. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103, 56 --115.

No context found.

Plaut DC, McClelland JL, Seidenberg MS, Patterson K. Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review 1996; 103: 56-115.

No context found.

. Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103:56--

No context found.

D. C. Plaut, J. L. McClelland, M. S. Seidenberg, and K Patterson. Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103(1):56--115, 1996.

No context found.

Plaut, D. C., McClelland, J. L., Seidenberg, M. S., & Patterson, K. (1996). Understanding normal and impaired word reading: Computational principles in quasi-regular domains. Psychological Review, 103(1), 56-115.

Online articles have much greater impact   More about CiteSeer.IST   Add search form to your site   Submit documents   Feedback  

CiteSeer.IST - Copyright Penn State and NEC