Abstract not found

Despite a century of research, the mechanisms underlying short-term or working memory for serial order remain uncertain. Recent theoretical models have converged on a particular account, based on transient associations between independent item and context representations. In the present article, the authors present an alternative model, according to which sequence information is encoded through sustained patterns of activation within a recurrent neural network architecture. As demonstrated through a series of computer simulations, the model provides a parsimonious account for numerous benchmark characteristics of immediate serial recall, including data that have been considered to preclude the application of recurrent neural networks in this domain. Unlike most competing accounts, the model deals naturally with findings concerning the role of background knowledge in serial recall and makes contact with relevant neuroscientific data. Furthermore, the model gives rise to numerous testable predictions that differentiate it from competing theories. Taken together, the results presented indicate that recurrent neural networks may offer a useful framework for understanding short-term memory for serial order.

Abstract not found

this article may be sent to David Plaut, Mellon Institute 115-- CNBC, Carnegie Mellon University, 4400 Fifth Avenue, Pittsburgh PA 15213--2683; email: plaut@cmu.edu

The lexical semantic system is an important component of human language and cognitive processing. One approach to modeling semantic knowledge makes use of hand-constructed networks or trees of interconnected word senses (Miller, Beckwith, Fellbaum, Gross, & Miller, 1990; Jarmasz & Szpakowicz, 2003). An alternative approach seeks to model word meanings as high-dimensional vectors, which are derived from the cooccurrence of words in unlabeled text corpora (Landauer & Dumais, 1997; Burgess & Lund, 1997a). This paper introduces a new vector-space method for deriving word-meanings from large corpora that was inspired by the HAL and LSA models, but which achieves better and more consistent results in predicting human similarity judgments. We explain the new model, known as COALS, and how it relates to prior methods, and then evaluate the various models on a range of tasks, including a novel set of semantic similarity ratings involving both semantically and morphologically related terms.

The authors examine the implications of research on Chinese for theories of reading and propose the lexical constituency model as a general framework for word reading across writing systems. Word identities are defined by 3 interlinked constituents (orthographic, phonological, and semantic). The implemented model simulates the time course of graphic, phonological, and semantic priming effects, including immediate graphic facilitation followed by graphic inhibition with simultaneous phonological facilitation, a pattern unique to the Chinese writing system. Pseudocharacter primes produced only facilitation, supporting the model’s assumption that lexical thresholds determine phonological and semantic, but not graphic, effects. More generally, both universal reading processes and writing system constraints exist. Although phonology is universal, its activation process depends on how the writing system structures graphic units. The development of models of word reading has been informed primarily by studies of English word identification. This is true for both symbolic models that postulate an internal lexicon and multiple pathways to pronunciation (Besner & Smith, 1992; Coltheart,

The authors investigated spatial, temporal, and attentional manipulations in a short-term repetition priming paradigm. Brief primes produced a strong preference to choose the primed alternative, whereas long primes had the opposite effect. However, a 2nd brief presentation of a long prime produced a preference for the primed word despite the long total prime duration. These surprising results are explained by a computational model that posits the offsetting components of source confusion (prime features are confused with target features) and discounting (evidence from primed features is discounted). The authors obtained compelling evidence for these components by showing how they can cooperate or compete through different manipulations of prime salience. The model allows for dissociations between prime salience and the magnitude of priming, thereby providing a unified account of “subliminal” and “supraliminal” priming.

Abstract: A naturalistic scheme of primitive conceptual representations is proposed using the statistical measure of mutual information. It is argued that a concept represents, not the class of objects that caused its tokening, but the class of objects that is most likely to have caused it (had it been tokened), as specified by the statistical measure of mutual information. This solves the problem of misrepresentation which plagues causal accounts, by taking the representation relation to be determined via ordinal relationships between conditional probabilities. The scheme can deal with statistical biases and does not rely on arbitrary criteria. Implications for the theory of meaning and semantic content are addressed. 1.

Abstract not found

Neuropsychological studies of patients with acquired semantic impairments have yielded two distinct and contrasting patterns of performance in a spoken-word/picture matching task (Warrington & Cipolotti, 1996). Patients labeled access/refractory are strongly influenced by presentation rate, semantic relatedness of distractors, and repetition, yet they seem relatively unaffected by lexical frequency. Degraded-store patients, on the other hand, are strongly affected by lexical frequency but less affected by presentation rate, semantic relatedness, or repetition. Our account of these patterns of performance is based on the distinction between two different types of neurological damage: 1) damage to neuromodulatory systems that function to amplify neural signals while suppressing normal refractory-like effects, and 2) damage to connections between groups of neurons that encode semantic information and are sensitive to frequency/familiarity. We present a connectionist model that learns to map spoken-word input to semantic representations and that incorporates a particular form of neural refractoriness referred to as synaptic depression, as well as a simple form of neuromodulation. We show that the model is capable of accounting for the contrasting pattern of semantic impairment under these two different forms of damage, and further demonstrate how it is capable of handling several documented cases that are exceptions to the basic patterns of impairment. Several predictions and limitations of the current model are discussed.