David A. Medler

Neural correlates of sensory and decision processes in auditory object identification.

Physiological studies of auditory perception have not yet clearly distinguished sensory from decision processes. In this experiment, human participants identified speech sounds masked by varying levels of noise while blood oxygenation signals in the brain were recorded with functional magnetic resonance imaging (fMRI). Accuracy and response time were used to characterize the behavior of sensory and decision components of this perceptual system. Oxygenation signals in a cortical subregion just anterior and lateral to primary auditory cortex predicted accuracy of sound identification, whereas signals in an inferior frontal region predicted response time. Our findings provide neurophysiological evidence for a functional distinction between sensory and decision mechanisms underlying auditory object identification. The present results also indicate a link between inferior frontal lobe activation and response-selection processes during auditory perception tasks.

Distinct Brain Systems for Processing Concrete and Abstract Concepts

Behavioral and neurophysiological effects of word imageability and concreteness remain a topic of central interest in cognitive neuroscience and could provide essential clues for understanding how the brain processes conceptual knowledge. We examined these effects using event-related functional magnetic resonance imaging while participants identified concrete and abstract words. Relative to nonwords, concrete and abstract words both activated a left-lateralized network of multimodal association areas previously linked with verbal semantic processing. Areas in the left lateral temporal lobe were equally activated by both word types, whereas bilateral regions including the angular gyrus and the dorsal prefrontal cortex were more strongly engaged by concrete words. Relative to concrete words, abstract words activated left inferior frontal regions previously linked with phonological and verbal working memory processes. The results show overlapping but partly distinct neural systems for processing concrete and abstract concepts, with greater involvement of bilateral association areas during concrete word processing, and processing of abstract concepts almost exclusively by the left hemisphere.

Syntactic and Semantic Modulation of Neural Activity during Auditory Sentence Comprehension

In previous functional neuroimaging studies, left anterior temporal and temporal-parietal areas responded more strongly to sentences than to randomly ordered lists of words. The smaller response for word lists could be explained by either (1) less activation of syntactic processes due to the absence of syntactic structure in the random word lists or (2) less activation of semantic processes resulting from failure to combine the content words into a global meaning. To test these two explanations, we conducted a functional magnetic resonance imaging study in which word order and combinatorial word meaning were independently manipulated during auditory comprehension. Subjects heard six different stimuli: normal sentences, semantically incongruent sentences in which content words were randomly replaced with other content words, pseudoword sentences, and versions of these three sentence types in which word order was randomized to remove syntactic structure. Effects of syntactic structure (greater activation to sentences than to word lists) were observed in the left anterior superior temporal sulcus and left angular gyrus. Semantic effects (greater activation to semantically congruent stimuli than either incongruent or pseudoword stimuli) were seen in widespread, bilateral temporal lobe areas and the angular gyrus. Of the two regions that responded to syntactic structure, the angular gyrus showed a greater response to semantic structure, suggesting that reduced activation for word lists in this area is related to a disruption in semantic processing. The anterior temporal lobe, on the other hand, was relatively insensitive to manipulations of semantic structure, suggesting that syntactic information plays a greater role in driving activation in this area.

Modulation of the semantic system by word imageability

A prevailing neurobiological theory of semantic memory proposes that part of our knowledge about concrete, highly imageable concepts is stored in the form of sensory-motor representations. While this theory predicts differential activation of the semantic system by concrete and abstract words, previous functional imaging studies employing this contrast have provided relatively little supporting evidence. We acquired event-related functional magnetic resonance imaging (fMRI) data while participants performed a semantic similarity judgment task on a large number of concrete and abstract noun triads. Task difficulty was manipulated by varying the degree to which the words in the triad were similar in meaning. Concrete nouns, relative to abstract nouns, produced greater activation in a bilateral network of multimodal and heteromodal association areas, including ventral and medial temporal, posterior-inferior parietal, dorsal prefrontal, and posterior cingulate cortex. In contrast, abstract nouns produced greater activation almost exclusively in the left hemisphere in superior temporal and inferior frontal cortex. Increasing task difficulty modulated activation mainly in attention, working memory, and response monitoring systems, with almost no effect on areas that were modulated by imageability. These data provide critical support for the hypothesis that concrete, imageable concepts activate perceptually based representations not available to abstract concepts. In contrast, processing abstract concepts makes greater demands on left perisylvian phonological and lexical retrieval systems. The findings are compatible with dual coding theory and less consistent with single-code models of conceptual representation. The lack of overlap between imageability and task difficulty effects suggests that once the neural representation of a concept is activated, further maintenance and manipulation of that information in working memory does not further increase neural activation in the conceptual store.

Time course of semantic processes during sentence comprehension: an fMRI study.

The ability to create new meanings from combinations of words is one important function of the language system. We investigated the neural correlates of combinatorial semantic processing using fMRI. During scanning, participants performed a rating task on auditory word or pseudoword strings that differed in the presence of combinatorial and word-level semantic information. Stimuli included normal sentences comprised of thematically related words that could be readily combined to produce a more complex meaning, semantically incongruent sentences in which content words were randomly replaced with other content words, pseudoword sentences, and versions of these three sentence types in which syntactic structure was removed by randomly re-ordering the words. Several regions showed greater BOLD signal for stimuli with words than for those with pseudowords, including the left angular gyrus, left superior temporal sulcus, and left inferior frontal gyrus, suggesting that these areas are involved in semantic access at the single word level. In the angular and inferior frontal gyri these differences emerged early in the course of the hemodynamic response. An effect of combinatorial semantic structure was observed in the left angular gyrus and left lateral temporal lobe, which showed greater activation for normal compared to semantically incongruent sentences. These effects appeared later in the time course of the hemodynamic response, beginning after the entire stimulus had been presented. The data indicate a complex spatiotemporal pattern of activity associated with computation of word and sentence-level semantic information, and suggest a particular role for the left angular gyrus in processing overall sentence meaning.

Attentional and linguistic interactions in speech perception.

The role of attention in speech comprehension is not well understood. We used fMRI to study the neural correlates of auditory word, pseudoword, and nonspeech (spectrally rotated speech) perception during a bimodal (auditory, visual) selective attention task. In three conditions, Attend Auditory (ignore visual), Ignore Auditory (attend visual), and Visual (no auditory stimulation), 28 subjects performed a one-back matching task in the assigned attended modality. The visual task, attending to rapidly presented Japanese characters, was designed to be highly demanding in order to prevent attention to the simultaneously presented auditory stimuli. Regardless of stimulus type, attention to the auditory channel enhanced activation by the auditory stimuli (Attend Auditory>Ignore Auditory) in bilateral posterior superior temporal regions and left inferior frontal cortex. Across attentional conditions, there were main effects of speech processing (word+pseudoword>rotated speech) in left orbitofrontal cortex and several posterior right hemisphere regions, though these areas also showed strong interactions with attention (larger speech effects in the Attend Auditory than in the Ignore Auditory condition) and no significant speech effects in the Ignore Auditory condition. Several other regions, including the postcentral gyri, left supramarginal gyrus, and temporal lobes bilaterally, showed similar interactions due to the presence of speech effects only in the Attend Auditory condition. Main effects of lexicality (word>pseudoword) were isolated to a small region of the left lateral prefrontal cortex. Examination of this region showed significant word>pseudoword activation only in the Attend Auditory condition. Several other brain regions, including left ventromedial frontal lobe, left dorsal prefrontal cortex, and left middle temporal gyrus, showed Attention x Lexicality interactions due to the presence of lexical activation only in the Attend Auditory condition. These results support a model in which neutral speech presented in an unattended sensory channel undergoes relatively little processing beyond the early perceptual level. Specifically, processing of phonetic and lexical-semantic information appears to be very limited in such circumstances, consistent with prior behavioral studies.

Attentional Modulation in the Detection of Irrelevant Deviance: A Simultaneous ERP/fMRI Study

Little is known about the neural mechanisms that control attentional modulation of deviance detection in the auditory modality. In this study, we manipulated the difficulty of a primary task to test the relation between task difficulty and the detection of infrequent, task-irrelevant deviant (D) tones (1300 Hz) presented among repetitive standard (S) tones (1000 Hz). Simultaneous functional magnetic resonance imaging (fMRI)/event-related potentials (ERPs) were recorded from 21 subjects performing a two-alternative forced-choice duration discrimination task (short and long tones of equal probability). The duration of the short tone was always 50 msec. The duration of the long tone was 100 msec in the easy task and 60 msec in the difficult task. As expected, response accuracy decreased and response time (RT) increased in the difficult compared with the easy task. Performance was also poorer for D than for S tones, indicating distraction by task-irrelevant frequency information on trials involving D tones. In the difficult task, an amplitude increase was observed in the difference waves for N1 and P3a, ERP components associated with increased attention to deviant sounds. The mismatch negativity (MMN) response, associated with passive deviant detection, was larger in the easy task, demonstrating the susceptibility of this component to attentional manipulations. The fMRI contrast D > S in the difficult task revealed activation on the right superior temporal gyrus (STG) and extending ventrally into the superior temporal sulcus, suggesting this regions involvement in involuntary attention shifting toward unattended, infrequent sounds. Conversely, passive deviance detection, as reflected by the MMN, was associated with more dorsal activation on the STG. These results are consistent with the view that the dorsal STG region is responsive to mismatches between the memory trace of the standard and the incoming deviant sound, whereas the ventral STG region is activated by involuntary shifts of attention to task-irrelevant auditory features.

PDP networks can provide models that are not mere implementations of classical theories

Abstract There is widespread belief that connectionist networks are dramatically different from classical or symbolic models. However, connectionists rarely test this belief by interpreting the internal structure of their nets. A new approach to interpreting networks was recently introduced by Berkeley et al. (1995). The current paper examines two implications of applying this method: (1) that the internal structure of a connectionist network can have a very classical appearance, and (2) that this interpretation can provide a cognitive theory that cannot be dismissed as a mere implementation.

Using Extra Output Learning to Insert a Symbolic Theory into a Connectionist Network

This paper examines whether a classical model could be translated into a PDP network using a standard connectionist training technique called extra output learning. In Study 1, standard machine learning techniques were used to create a decision tree that could be used to classify 8124 different mushrooms as being edible or poisonous on the basis of 21 different Features (Schlimmer, 1987). In Study 2, extra output learning was used to insert this decision tree into a PDP network being trained on the identical problem. An interpretation of the trained network revealed a perfect mapping from its internal structure to the decision tree, representing a precise translation of the classical theory to the connectionist model. In Study 3, a second network was trained on the mushroom problem without using extra output learning. An interpretation of this second network revealed a different algorithm for solving the mushroom problem, demonstrating that the Study 2 network was indeed a proper theory translation.

Training redundant artificial neural networks: imposing biology on technology

One biological principle that is often overlooked in the design of artificial neural networks (ANNs) is redundancy. Redundancy is the replication of processes within the brain. This paper examines the effects of redundancy on learning in ANNs when given either a function-approximation task or a pattern-classification task. The function-approximation task simulated a robotic arm reaching toward an object in two-dimensional space, and the pattern-classification task was detecting parity. Results indicated that redundant ANNs learned the pattern-classification problem much faster, and converge on a solution 100% of the time, whereas standard ANNs sometimes failed to learn the problem. Furthermore, when overall network error is considered, redundant ANNs were significantly more accurate than standard ANNs in performing the function-approximation task. These results are discussed in terms of the relevance of redundancy to the performance of ANNs in general, and the relevance of redundancy in biological systems in particular.