Acceptance of evidence-supported hypotheses generates a stronger signal from an underlying functionally-connected network.

Choosing one's preferred hypothesis requires multiple brain regions to work in concert as a functionally connected network. We predicted that a stronger network signal would underlie cognitive coherence between a hypothesis and the available evidence. In order to identify such functionally connected networks in magnetoencephalography (MEG) data, we first localized the generators of changes in oscillatory power within three frequency bands, namely alpha (7-13 Hz), beta (18-24 Hz), and theta (3-7 Hz), with a spatial resolution of 5mm and temporal resolution of 50 ms. We then used principal component analysis (PCA) to identify functionally connected networks reflecting co-varying post-stimulus changes in power. As predicted, PCA revealed a functionally connected network with a stronger signal when the evidence supported accepting the hypothesis being judged. This difference was driven by beta-band power decreases in the left dorsolateral prefrontal cortex (DLPFC), ventromedial prefrontal cortex (VMPFC), posterior cingulate cortex (PCC), and midline occipital cortex.

Functional connectivity reveals load dependent neural systems underlying encoding and maintenance in verbal working memory.

One of the main challenges in working memory research has been to understand the degree of separation and overlap between the neural systems involved in encoding and maintenance. In the current study we used a variable load version of the Sternberg item recognition test (two, four, six, or eight letters) and a functional connectivity method based on constrained principal component analysis to extract load-dependent neural systems underlying encoding and maintenance, and to characterize their anatomical overlap and functional interaction. Based on the pattern of functional connectivity, constrained principal component analysis identified a load-dependent encoding system comprising bilateral occipital (Brodmann's area (BA) 17, 18), bilateral superior parietal (BA 7), bilateral dorsolateral prefrontal (BA 46), and dorsal anterior cingulate (BA 24, 32) regions. For maintenance, in contrast, constrained principal component analysis identified a system that was characterized by both load-dependent increases and decreases in activation. The structures in this system jointly activated by maintenance load involved left posterior parietal (BA 40), left inferior prefrontal (BA 44), left premotor and supplementary motor areas (BA 6), and dorsal cingulate regions (BA 24, 32), while the regions displaying maintenance-load-dependent activity decreases involved bilateral occipital (BA 17, 18), posterior cingulate (BA 23) and rostral anterior cingulate/orbitofrontal (BA 10, 11, 32) regions. The correlation between the encoding and maintenance systems was strong and negative (Pearson's r = -.55), indicting that some regions important for visual processing during encoding displayed reduced activity during maintenance, while subvocal rehearsal and phonological storage regions important for maintenance showed a reduction in activity during encoding. In summary, our analyses suggest that separable and complementary subsystems underlie encoding and maintenance in verbal working memory, and they demonstrate how constrained principal component analysis can be employed to characterize neuronal systems and their functional contributions to higher-level cognition.

The learning of first and second person pronouns in English: network models and analysis.

Although most English-speaking children master the correct use of first and second person pronouns by three years, some children show persistent reversal errors in which they refer to themselves as you and to others as me. Recently, such differences have been attributed to the relative availability of overheard speech during the learning process. The present study tested this proposal with feed-forward neural networks learning these pronouns. Network learning speed and analysis of their knowledge representations confirmed the importance of exposure to shifting reference provided by overheard speech. Errorless pronoun learning was linked to the amount of overheard speech, interactions with a greater number of speakers, and prior knowledge of the basic-level kind PERSON.

Discriminant component pruning. Regularization and interpretation of multi-layered back-propagation networks.

Neural networks are often employed as tools in classification tasks. The use of large networks increases the likelihood of the task's being learned, although it may also lead to increased complexity. Pruning is an effective way of reducing the complexity of large networks. We present discriminant components pruning (DCP), a method of pruning matrices of summed contributions between layers of a neural network. Attempting to interpret the underlying functions learned by the network can be aided by pruning the network. Generalization performance should be maintained at its optimal level following pruning. We demonstrate DCP's effectiveness at maintaining generalization performance, applicability to a wider range of problems, and the usefulness of such pruning for network interpretation. Possible enhancements are discussed for the identification of the optimal reduced rank and inclusion of nonlinear neural activation functions in the pruning algorithm.

Case studies of pronoun development in two hearing-impaired children: normal, delayed or deviant?

This paper is a study of first and second person pronoun development in the spoken language of two young hearing-impaired children. Pronoun development was examined over a period of 11 months, starting at the age of 29 and 28 months, to determine whether the children's acquisition of these pronouns would reflect normal, delayed or deviant patterns of development. Comparison of data from these children with data regarding normally developing children shows the hearing-impaired children's acquisition to be within normal expectations for hearing age and overall linguistic level, and only slightly delayed in terms of chronological age. These results lend support to the view that differences in the hearing-impaired child's language ability are probably the result of a relative lack of auditory and linguistic experience, rather than reorganisation of the hearing-impaired child's psychological and cognitive processing abilities.

Analysis of pronominal errors: a case-study.

The present paper reports a case-study of a normally developing boy who made pronominal errors for about ten months. Comprehension and production of first- and second-person pronouns were longitudinally examined from 1;7 to 2;10 to test three hypotheses concerning pronominal errors: pronominal errors are a result of either (a) semantic confusion, (b) simple imitation, or (c) confusion between self and others. The results showed that the child began using first- and second-person pronouns at about 1;8 and mastered the correct usage by 2;10. Consistent errors for the first- and the second-person pronouns were observed from 1;11 to 2;4, but proportions of errors occurring in his imitative language were low. The comprehension and production data clearly indicated that the child persistently made pronominal errors due to semantic confusion. That is, first-person pronouns referred to a person with whom the child conversed and second-person pronouns referred to himself.

An alternative view of pronominal errors in autistic children.

Persistent pronominal errors in autistic children have been attributed either to a psychosocial deficit or to a linguistic or cognitive deficit. However, recent studies of normal children suggest that the failure to observe pronouns in speech addressed to another person is a major reason children show pronominal errors. The present study investigated if pronominal errors in autistic children can be explained by this alternative hypothesis. Children's attentiveness to the pronoun models and their imitative behaviors were examined under two modeling conditions. The nonaddressee condition provided children with systematic opportunities to observe pronoun models directed to another person as well as those directed to themselves. The addressee condition provided only systematic opportunities to observe pronoun models directed to children. Clear evidence for the alternative hypothesis was obtained for second person pronouns, suggesting that pronominal errors in autistic children can be interpreted within the framework of normal language development.

Regions of Significance in Multiple Regression Analysis.

There has been frequent confusion about the meaning of the various possible tests of significance in multiple regression, and this has led to discussions of "apparent contradictions" in regression. This paper considers the case of two predictor variables, and figures are obtained which show the regions of significance of joint regression coefficients, regression coefficients considered separately, and the multiple correlation. The intersection of these regions of significance and non-significance illustrates how the various "apparent contra- dictions" and anomalies may occur.