Search PubMed⌕ Search

Biomedical subjects

J L Elman

Publications and source records attributed to J L Elman.

11 recordsLinked to original sources
PubMed1995

Learning and morphological change.

An account is offered to change over time in English verb morphology, based on a connectionist approach to how morphological knowledge is acquired and used. A technique is first described that was developed for modeling historical change in connectionist networks, and that technique is applied to model English verb inflection as it developed from the highly complex past tense system of Old English towards that of the modern language, with one predominant "regular" inflection and a small number of irregular forms. The model relies on the fact that certain input-output mappings are easier than others to learn in a connectionist network. Highly frequent patterns, or those that share phonological regularities with a number of others, are learned more quickly and with lower error than low-frequency, highly irregular patterns. A network is taught a data set representative of the verb classes of Old English, but learning is stopped before reaching asymptote, and the output of this network is used as the teacher of a new net. As a result, the errors in the first network were passed on to become part of the data set of the second. Those patterns that are hardest to learn led to the most errors, and over time are "regularized" to fit a more dominant pattern. The results of the networks simulations were highly consistent with the major historical developments. These results are predicted from well-understood aspects of network dynamics, which therefore provide a rationale for the shape of the attested changes.

Humans↗
PubMed1993

Learning and development in neural networks: the importance of starting small.

It is a striking fact that in humans the greatest learning occurs precisely at that point in time--childhood--when the most dramatic maturational changes also occur. This report describes possible synergistic interactions between maturational change and the ability to learn a complex domain (language), as investigated in connectionist networks. The networks are trained to process complex sentences involving relative clauses, number agreement, and several types of verb argument structure. Training fails in the case of networks which are fully formed and 'adultlike' in their capacity. Training succeeds only when networks begin with limited working memory and gradually 'mature' to the adult state. This result suggests that rather than being a limitation, developmental restrictions on resources may constitute a necessary prerequisite for mastering certain complex domains. Specifically, successful learning may depend on starting small.

Child↗
PubMed1988

Learning the hidden structure of speech.

In the work described here, the backpropagation neural network learning procedure is applied to the analysis and recognition of speech. This procedure takes a set of input/output pattern pairs and attempts to learn their functional relationship; it develops the necessary representational features during the course of learning. A series of computer simulation studies was carried out to assess the ability of these networks to accurately label sounds, to learn to recognize sounds without labels, and to learn feature representations of continuous speech. These studies demonstrated that the networks can learn to label presegmented test tokens with accuracies of up to 95%. Networks trained on segmented sounds using a strategy that requires no external labels were able to recognize and delineate sounds in continuous speech. These networks developed rich internal representations that included units which corresponded to such traditional distinctions as vowels and consonants, as well as units that were sensitive to novel and nonstandard features. Networks trained on a large corpus of unsegmented, continuous speech without labels also developed interesting feature representations, which may be useful in both segmentation and label learning. The results of these studies, while preliminary, demonstrate that backpropagation learning can be used with complex, natural data to identify a feature structure that can serve as the basis for both analysis and nontrivial pattern recognition.

Computer Simulation↗
PubMed1985

Interference between phonemes during phoneme monitoring: evidence for an interactive activation model of speech perception.

We explore the recent finding (Newman & Dell, 1978) that the time needed to detect a target phoneme in a phoneme monitoring task is increased when the preceding word contains a phoneme similar to the target. Normal adult native speakers of English monitored auditorily presented sentences and responded as quickly as possible whenever they detected a specified phoneme. We found that preceding word-initial phonemes, despite being processed more quickly, increased the response latency to the following target phoneme more than did preceding word-medial phonemes. There was also an increase in response latency even when the subject could be highly certain that the similar preceding phoneme was not an instance of the target phoneme. We argue that the interference effects are due to fundamental characteristics of perceptual processing and that more time is needed to categorize the target phoneme. We present a computer simulation using an interactive activation model of speech perception to demonstrate the plausibility of our explanation.

Adolescent↗
PubMed1983

Oral vs. manual tapping with delayed auditory feedback as measures of cerebral dominance.

Two sets of findings have been reported which make contradictory claims about the cerebral lateralization of mechanisms for coordinating auditory feedback with motor control. One difficulty in evaluating the two sets of claims is that they result from experiments involving methodologies which are different enough to make direct comparisons difficult. The present study reports results of an experiment using a technique which partially bridges the gap between two of these paradigms. This technique involves examining the effect of delayed auditory feedback (DAF) on performance of both an oral and a manual tapping task. Ear of delivery of DAF, subject language background, and nature of the auditory stimulus were the experimental variables. The findings support the claim that the left hemisphere is differentially involved in the use of auditory feedback to control oral, but not manual, motor activity. No evidence was found to indicate that this functional cerebral asymmetry is stimulus- or language-dependent.

Acoustic Stimulation↗
PubMed1979

Perceptual origins of the phoneme boundary effect and selective adaptation to speech: a signal detection theory analysis.

Differences in discriminability of stimuli near phoneme boundaries and findings from selective adaptation have been used to argue for the existence of neurophysiological mechanisms--feature detectors--which mediate the perception of speech and speechlike sounds. A detection theory model was used in order to discover whether or not the phoneme boundary effect and the shift in phoneme boundary after adaptation might rather be attributable to changes in response bias. This model was applied in the analysis of phoneme identifications of three sets of stimuli before and after adaptation. While the origins for the phoneme boundary effect appear to lie below the level of response bias, findings suggest that identification changes after adaptation may be due solely to shifts in criterion, rather than changes at the sensory level.

Adult↗
PubMed1978

Contrast effects on stop consonant identification.

Changes in the identification of speech sounds following selective adaptation are usually attributed to a reduction in sensitivity of auditory feature detectors. An alternative explanation of these effects is based on the notion of response contrast. In several experiments, subjects identified the initial segment of synthetic consonant-vowel syllables as either the voiced stop [b] or the voiceless stop [ph]. Each test syllable had a value of voice onset time (VOT) that placed it near the English voiced-voiceless boundary. When the test syllables were preceded by a single clear [b] (VOT = -100 msec), subjects tended to identify them as [ph], whereas when they were preceded by an unambiguous [ph] (VOT = 100 msec), the syllables were predominantly labeled [b]. This contrast effect occurred even when the contextual stimuli were velar and the test stimuli were bilabial, which suggests a featural rather than a phonemic basis for the effect. To discount the possibility that these might be instances of single-trial sensory adaptation, we conducted a similar experiment in which the contextual stimuli followed the test items. Reliable contrast effects were still obtained. In view of these results, it appears likely that response contrast accounts for at least some component of the adaptation effects reported in the literature.

Humans↗