The influence of morphological regularities on the dynamics of a connectionist network.

The effects of morphological regularities on the behavior of connectionist networks were studied by training identical networks on orthographic-semantic mappings that either contained such regularities or did not. Morphological regularities had a substantial impact on both the amount of training needed to learn a mapping and the number of words that could be included in the training set. A variety of analyses demonstrated how morphological regularities structure the organization and componentiality of a network's internal representations.

Implicit memory for phonological processes in visual stem completion.

Four experiments were conducted investigating the role of phonology in repetition priming. Experiment 1 used a cross-modal priming paradigm in which participants made semantic judgments about spoken words and then performed a visual stem completion task. In Experiments 2-4, both the primes and the test stems were presented visually. The results of the first three experiments revealed that priming transfers across interpretations of a homophone. That is, seeing or hearing week primes both week and weak. The results of Experiment 4 showed that homophone priming cannot be attributed to the orthographic similarity of homophonic words. Together, these results indicate that repetition priming on a visual word completion task includes a phonological component.

Attentional modulation of the phonetic significance of acoustic cues.

Four experiments addressing the role of attention in phonetic perception are reported. The first experiment shows that the relative importance of two cues to the voicing distinction changes when subjects must perform an arithmetic distractor task at the same time as identifying a speech stimulus. The contribution of voice onset time to phonetic labeling decreases when subjects are distracted, while that of FO onset frequency increases. The second experiment shows a similar pattern for two cues to the distinction between the vowels /i/ (as in "beat") and /I/ (as in "bit"). Under low attention conditions, formant pattern has a smaller effect on phonetic labeling while vowel duration has a larger effect. Together these experiments indicate that careful attention to speech perception is necessary for strong acoustic cues (voice-onset time and formant patterns) to achieve their full impact on phonetic labeling, while weaker acoustic cues (FO onset frequency and vowel duration) achieve their full impact on phonetic labeling without close attention. Experiment 3 shows that this pattern is obtained when the distractor task places little demand on verbal short-term memory. Experiment 4 provides a data set for testing formal models of the role of attention in speech perception. Attention is shown to influence the signal-to-noise ratio in the phonetic encoding of acoustic cues; the sustained phonetic contribution of weak cues without close attention stems from reduced competition from strong cues. This principle is instantiated in a network model in which the role of attention is to reduce noise in the phonetic encoding of acoustic cues. Implications of this work for understanding speech perception and general theories of the role of attention in perception are discussed.

On the locus of redundancy effects in a letter-detection task.

It takes longer to detect a target if it is the initial letter of a nonredundant letter string such as BNHTW than if it is the initial letter of a redundant letter string such as BBBBB (Johnson, 1986b; Johnson & Blum, 1988). The results of the present study reveal that the redundancy effect also occurs for mixed-case letters strings (e.g., BbbBb) and digit strings (e.g., 22222). In addition, these results suggest that the cause of the redundancy effect is not related to the visual properties of the noninitial letters per se but is instead related to the presence of the target in noninitial positions. Together, these results rule out a number of hypotheses about the cause of the redundancy effect and suggest that the locus of this effect is in processes involved in response selection and/or the read-out of information from the perceptual system.

Similarity effects in word and pseudoword repetition priming.

Repetition priming refers to the facilitation in the visual identification of a stimulus produced by a recent encounter with that stimulus. In the paradigm used here, subjects performed a naming task in which a sequence of primes was presented; then they performed a tachistoscopic identification task in which the stimuli that were presented varied in their similarity to the primes. The results indicated that repetition priming facilitated the identification of repeated words and pronounceable nonwords that were similar, but not identical, to the recently encountered primes. Moreover, the number of presentations of the primes influenced the amount of facilitation for repeated words but, in contrast to previous findings, had no effect on the amount of facilitation for repeated nonwords or similar words and nonwords. The results are interpreted within a connectionist model of visual word identification.

Why are "What" and "Where" Processed by Separate Cortical Visual Systems? A Computational Investigation.

In the primate visual system, the identification of objects and the processing of spatial information are accomplished by different cortical pathways. The computational properties of this "two-systems" design were explored by constructing simplifying connectionist models. The models were designed to simultaneously classify and locate shapes that could appear in multiple positions in a matrix, and the ease of forming representations of the two kinds of information was measured. Some networks were designed so that all hidden nodes projected to all output nodes, whereas others had the hidden nodes split into two groups, with some projecting to the output nodes that registered shape identity and the remainder projecting to the output nodes that registered location. The simulations revealed that splitting processing into separate streams for identifying and locating a shape led to better performance only under some circumstances. Provided that enough computational resources were available in both streams, split networks were able to develop more efficient internal representations, as revealed by detailed analyses of the patterns of weights between connections.