Self-organising neural networks for adaptive control

Self-organizing neural networks have been implemented in a wide range of application areas such as speech processing, image processing, optimization and robotics. Recent variations to the basic model proposed by the authors enable it to order state space using a subset of the input vector and to apply a local adaptation procedure that does not rely on a predefined test duration limit. Both these variations have been incorporated into a new feature map architecture that forms an integral part of an Hybrid Learning System (HLS) based on a genetic-based classifier system. Problems are represented within HLS as objects characterized by environmental features. Objects controlled by the system have preset targets set against a subset of their features. The system's objective is to achieve these targets by evolving a behavioural repertoire that efficiently explores and exploits the problem environment. Feature maps encode two types of knowledge within HLS — long-term memory traces of useful regularities within the environment and the classifier performance data calibrated against an object's feature states and targets. Self-organization of these networks constitutes non-genetic-based (experience-driven) learning within HLS. This paper presents a description of the HLS architecture and an analysis of the modified feature map implementing associative memory. Initial results are presented that demonstrate the behaviour of the system on a simple control task.

Fundamental deficits of auditory perception in Wernicke’s aphasia

Objective: This work investigates the nature of the comprehension impairment in Wernicke’s aphasia, by examining the relationship between deficits in auditory processing of fundamental, non-verbal acoustic stimuli and auditory comprehension. Wernicke’s aphasia, a condition resulting in severely disrupted auditory comprehension, primarily occurs following a cerebrovascular accident (CVA) to the left temporo-parietal cortex. Whilst damage to posterior superior temporal areas is associated with auditory linguistic comprehension impairments, functional imaging indicates that these areas may not be specific to speech processing but part of a network for generic auditory analysis. Methods: We examined analysis of basic acoustic stimuli in Wernicke’s aphasia participants (n = 10) using auditory stimuli reflective of theories of cortical auditory processing and of speech cues. Auditory spectral, temporal and spectro-temporal analysis was assessed using pure tone frequency discrimination, frequency modulation (FM) detection and the detection of dynamic modulation (DM) in “moving ripple” stimuli. All tasks used criterion-free, adaptive measures of threshold to ensure reliable results at the individual level. Results: Participants with Wernicke’s aphasia showed normal frequency discrimination but significant impairments in FM and DM detection, relative to age- and hearing-matched controls at the group level (n = 10). At the individual level, there was considerable variation in performance, and thresholds for both frequency and dynamic modulation detection correlated significantly with auditory comprehension abilities in the Wernicke’s aphasia participants. Conclusion: These results demonstrate the co-occurrence of a deficit in fundamental auditory processing of temporal and spectrotemporal nonverbal stimuli in Wernicke’s aphasia, which may have a causal contribution to the auditory language comprehension impairment Results are discussed in the context of traditional neuropsychology and current models of cortical auditory processing.

The neural processing of masked speech: Evidence for different mechanisms in the left and right temporal lobes

It has been previously demonstrated that extensive activation in the dorsolateral temporal lobes associated with masking a speech target with a speech masker, consistent with the hypothesis that competition for central auditory processes is an important factor in informational masking. Here, masking from speech and two additional maskers derived from the original speech were investigated. One of these is spectrally rotated speech, which is unintelligible and has a similar (inverted) spectrotemporal profile to speech. The authors also controlled for the possibility of “glimpsing” of the target signal during modulated masking sounds by using speech-modulated noise as a masker in a baseline condition. Functional imaging results reveal that masking speech with speech leads to bilateral superior temporal gyrus (STG) activation relative to a speech-in-noise baseline, while masking speech with spectrally rotated speech leads solely to right STG activation relative to the baseline. This result is discussed in terms of hemispheric asymmetries for speech perception, and interpreted as showing that masking effects can arise through two parallel neural systems, in the left and right temporal lobes. This has implications for the competition for resources caused by speech and rotated speech maskers, and may illuminate some of the mechanisms involved in informational masking.

The neural processing of masked speech: evidence for different mechanisms in the left and right temporal lobes

It has been previously demonstrated that extensive activation in the dorsolateral temporal lobes associated with masking a speech target with a speech masker, consistent with the hypothesis that competition for central auditory processes is an important factor in informational masking. Here, masking from speech and two additional maskers derived from the original speech were investigated. One of these is spectrally rotated speech, which is unintelligible and has a similar (inverted) spectrotemporal profile to speech. The authors also controlled for the possibility of "glimpsing" of the target signal during modulated masking sounds by using speech-modulated noise as a masker in a baseline condition. Functional imaging results reveal that masking speech with speech leads to bilateral superior temporal gyrus (STG) activation relative to a speech-in-noise baseline, while masking speech with spectrally rotated speech leads solely to right STG activation relative to the baseline. This result is discussed in terms of hemispheric asymmetries for speech perception, and interpreted as showing that masking effects can arise through two parallel neural systems, in the left and right temporal lobes. This has implications for the competition for resources caused by speech and rotated speech maskers, and may illuminate some of the mechanisms involved in informational masking.

From dichotic listening to the irrelevant sound effect: a behavioural and neuroimaging analysis of the processing of unattended speech

The assumption that ignoring irrelevant sound in a serial recall situation is identical to ignoring a non-target channel in dichotic listening is challenged. Dichotic listening is open to moderating effects of working memory capacity (Conway et al., 2001) whereas irrelevant sound effects (ISE) are not (Beaman, 2004). A right ear processing bias is apparent in dichotic listening, whereas the bias is to the left ear in the ISE (Hadlington et al., 2004). Positron emission tomography (PET) imaging data (Scott et al., 2004, submitted) show bilateral activation of the superior temporal gyrus (STG) in the presence of intelligible, but ignored, background speech and right hemisphere activation of the STG in the presence of unintelligible background speech. It is suggested that the right STG may be involved in the ISE and a particularly strong left ear effect might occur because of the contralateral connections in audition. It is further suggested that left STG activity is associated with dichotic listening effects and may be influenced by working memory span capacity. The relationship of this functional and neuroanatomical model to known neural correlates of working memory is considered.

Word frequency and bigram frequency effects on linguistic processing and speech motor performance in individuals with aphasia and normal speakers

Models of normal word production are well specified about the effects of frequency of linguistic stimuli on lexical access, but are less clear regarding the same effects on later stages of word production, particularly word articulation. In aphasia, this lack of specificity of down-stream frequency effects is even more noticeable because there is relatively limited amount of data on the time course of frequency effects for this population. This study begins to fill this gap by comparing the effects of variation of word frequency (lexical, whole word) and bigram frequency (sub-lexical, within word) on word production abilities in ten normal speakers and eight mild–moderate individuals with aphasia. In an immediate repetition paradigm, participants repeated single monosyllabic words in which word frequency (high or low) was crossed with bigram frequency (high or low). Indices for mapping the time course for these effects included reaction time (RT) for linguistic processing and motor preparation, and word duration (WD) for speech motor performance (word articulation time). The results indicated that individuals with aphasia had significantly longer RT and WD compared to normal speakers. RT showed a significant main effect only for word frequency (i.e., high-frequency words had shorter RT). WD showed significant main effects of word and bigram frequency; however, contrary to our expectations, high-frequency items had longer WD. Further investigation of WD revealed that independent of the influence of word and bigram frequency, vowel type (tense or lax) had the expected effect on WD. Moreover, individuals with aphasia differed from control speakers in their ability to implement tense vowel duration, even though they could produce an appropriate distinction between tense and lax vowels. The results highlight the importance of using temporal measures to identify subtle deficits in linguistic and speech motor processing in aphasia, the crucial role of phonetic characteristics of stimuli set in studying speech production and the need for the language production models to account more explicitly for word articulation.

Modeling infant speech sound discrimination using simple associative networks

Infants' responses in speech sound discrimination tasks can be nonmonotonic over time. Stager and Werker (1997) reported such data in a bimodal habituation task. In this task, 8-month-old infants were capable of discriminations that involved minimal contrast pairs, whereas 14-month-old infants were not. It was argued that the older infants' attenuated performance was linked to their processing of the stimuli for meaning. The authors suggested that these data are diagnostic of a qualitative shift in infant cognition. We describe an associative connectionist model showing a similar decrement in discrimination without any qualitative shift in processing. The model suggests that responses to phonemic contrasts may be a nonmonotonic function of experience with language. The implications of this idea are discussed. The model also provides a formal framework for studying habituation-dishabituation behaviors in infancy.

Steady-spectrum contexts and perceptual compensation for reverberation in speech identification

Perceptual compensation for reverberation was measured by embedding test words in contexts that were either spoken phrases or processed versions of this speech. The processing gave steady-spectrum contexts with no changes in the shape of the short-term spectral envelope over time, but with fluctuations in the temporal envelope. Test words were from a continuum between "sir" and "stir." When the amount of reverberation in test words was increased, to a level above the amount in the context, they sounded more like "sir." However, when the amount of reverberation in the context was also increased, to the level present in the test word, there was perceptual compensation in some conditions so that test words sounded more like "stir" again. Experiments here found compensation with speech contexts and with some steady-spectrum contexts, indicating that fluctuations in the context's temporal envelope can be sufficient for compensation. Other results suggest that the effectiveness of speech contexts is partly due to the narrow-band "frequency-channels" of the auditory periphery, where temporal-envelope fluctuations can be more pronounced than they are in the sound's broadband temporal envelope. Further results indicate that for compensation to influence speech, the context needs to be in a broad range of frequency channels. (c) 2007 Acoustical Society of America.

Processing of wh-questions in a case of posterior cortical atrophy

Posterior cortical atrophy (PCA) is a type of dementia that is characterized by visuo-spatial and memory deficits, dyslexia and dysgraphia, relatively early onset and preserved insight. Language deficits have been reported in some cases of PCA. Using an off-line grammaticality judgement task, processing of wh-questions is investigated in a case of PCA. Other aspects of auditory language are also reported. It is shown that processing of wh-questions is influenced by syntactic structure, a novel finding in this condition. The results are discussed with reference to accounts of wh-questions in aphasia. An uneven profile of other language abilities is reported with deficits in digit span (forward, backward), story retelling ability, comparative questions but intact abilities in following commands, repetition, concept definition, generative naming and discourse comprehension.

On the nature and cause of Specific Language Impairment: a view from sentence processing and infant research

This paper addresses the nature and cause of Specific Language Impairment (SLI) by reviewing recent research in sentence processing of children with SLI compared to typically developing (TD) children and research in infant speech perception. These studies have revealed that children with SLI are sensitive to syntactic, semantic, and real-world information, but do not show sensitivity to grammatical morphemes with low phonetic saliency, and they show longer reaction times than age-matched controls. TD children from the age of 4 show trace reactivation, but some children with SLI fail to show this effect, which resembles the pattern of adults and TD children with low working memory. Finally, findings from the German Language Development (GLAD) Project have revealed that a group of children at risk for SLI had a history of an auditory delay and impaired processing of prosodic information in the first months of their life, which is not detectable later in life. Although this is a single project that needs to be replicated with a larger group of children, it provides preliminary support for accounts of SLI which make an explicit link between an early deficit in the processing of phonology and later language deficits, and the Computational Complexity Hypothesis that argues that the language deficit in children with SLI lies in difficulties integrating different types of information at the interfaces.

Constancy in the perception of speech when the level of room-reflections varies

A speech message played several metres from the listener in a room is usually heard to have much the same phonetic content as it does when played nearby, even though the different amounts of reflected sound make the temporal envelopes of these signals very different. To study this ‘constancy’ effect, listeners heard speech messages and speech-like sounds comprising 8 auditory-filter shaped noise-bands that had temporal envelopes corresponding to those in these filters when the speech message is played. The ‘contexts’ were “next you’ll get _to click on”, into which a “sir” or “stir” test word was inserted. These test words were from an 11-step continuum, formed by amplitude modulation. Listeners identified the test words appropriately, even in the 8-band conditions where the speech had a ‘robotic’ quality. Constancy was assessed by comparing the influence of room reflections on the test word across conditions where the context had either the same level of room reflections (i.e. from the same, far distance), or where it had a much lower level (i.e. from nearby). Constancy effects were obtained with both the natural- and the 8-band speech. Results are considered in terms of the degree of ‘matching’ between the context’s and test-word’s bands.

Speech perception and phonological short-term memory capacity in language impairment: preliminary evidence from adolescents with specific language impairment (SLI) and autism spectrum disorders (ASD)

Background: The cognitive bases of language impairment in specific language impairment (SLI) and autism spectrum disorders (ASD) were investigated in a novel non-word comparison task which manipulated phonological short-term memory (PSTM) and speech perception, both implicated in poor non-word repetition. Aims: This study aimed to investigate the contributions of PSTM and speech perception in non-word processing and whether individuals with SLI and ASD plus language impairment (ALI) show similar or different patterns of deficit in these cognitive processes. Method & Procedures: Three groups of adolescents (aged 14–17 years), 14 with SLI, 16 with ALI, and 17 age and non-verbal IQ matched typically developing (TD) controls, made speeded discriminations between non-word pairs. Stimuli varied in PSTM load (two- or four-syllables) and speech perception load (mismatches on a word-initial or word-medial segment). Outcomes & Results: Reaction times showed effects of both non-word length and mismatch position and these factors interacted: four-syllable and word-initial mismatch stimuli resulted in the slowest decisions. Individuals with language impairment showed the same pattern of performance as those with typical development in the reaction time data. A marginal interaction between group and item length was driven by the SLI and ALI groups being less accurate with long items than short ones, a difference not found in the TD group. Conclusions & Implications: Non-word discrimination suggests that there are similarities and differences between adolescents with SLI and ALI and their TD peers. Reaction times appear to be affected by increasing PSTM and speech perception loads in a similar way. However, there was some, albeit weaker, evidence that adolescents with SLI and ALI are less accurate than TD individuals, with both showing an effect of PSTM load. This may indicate, at some level, the processing substrate supporting both PSTM and speech perception is intact in adolescents with SLI and ALI, but also in both there may be impaired access to PSTM resources.

Clustered functional MRI of overt speech production

To investigate the neural network of overt speech production, eventrelated fMRI was performed in 9 young healthy adult volunteers. A clustered image acquisition technique was chosen to minimize speechrelated movement artifacts. Functional images were acquired during the production of oral movements and of speech of increasing complexity (isolated vowel as well as monosyllabic and trisyllabic utterances). This imaging technique and behavioral task enabled depiction of the articulo-phonologic network of speech production from the supplementary motor area at the cranial end to the red nucleus at the caudal end. Speaking a single vowel and performing simple oral movements involved very similar activation of the corticaland subcortical motor systems. More complex, polysyllabic utterances were associated with additional activation in the bilateral cerebellum,reflecting increased demand on speech motor control, and additional activation in the bilateral temporal cortex, reflecting the stronger involvement of phonologic processing.

Temporal-envelope constancy of speech in rooms and the perceptual weighting of frequency bands

Three experiments measured constancy in speech perception, using natural-speech messages or noise-band vocoder versions of them. The eight vocoder-bands had equally log-spaced center-frequencies and the shapes of corresponding “auditory” filters. Consequently, the bands had the temporal envelopes that arise in these auditory filters when the speech is played. The “sir” or “stir” test-words were distinguished by degrees of amplitude modulation, and played in the context; “next you’ll get _ to click on.” Listeners identified test-words appropriately, even in the vocoder conditions where the speech had a “noise-like” quality. Constancy was assessed by comparing the identification of test-words with low or high levels of room reflections across conditions where the context had either a low or a high level of reflections. Constancy was obtained with both the natural and the vocoded speech, indicating that the effect arises through temporal-envelope processing. Two further experiments assessed perceptual weighting of the different bands, both in the test word and in the context. The resulting weighting functions both increase monotonically with frequency, following the spectral characteristics of the test-word’s [s]. It is suggested that these two weighting functions are similar because they both come about through the perceptual grouping of the test-word’s bands.