An embodiment perspective on the acquisition of speech perception

Integrating unsupervised and reinforcement learning in human categorical perception: A computational model

Categorical perception identifies a tuning of human perceptual systems that can occur during the execution of a categorisation task. Despite the fact that experimental studies and computational models suggest that this tuning is influenced by task-independent effects (e.g., based on Hebbian and unsupervised learning, UL) and task-dependent effects (e.g., based on reward signals and reinforcement learning, RL), no model studies the UL/RL interaction during the emergence of categorical perception. Here we have investigated the effects of this interaction, proposing a system-level neuro-inspired computational architecture in which a perceptual component integrates UL and RL processes. The model has been tested with a categorisation task and the results show that a balanced mix of unsupervised and reinforcement learning leads to the emergence of a suitable categorical perception and the best performance in the task. Indeed, an excessive unsupervised learning contribution tends to not identify task-relevant features while an excessive reinforcement learning contribution tends to initially learn slowly and then to reach sub-optimal performance. These results are consistent with the experimental evidence regarding categorical activations of extrastriate cortices in healthy conditions. Finally, the results produced by the two extreme cases of our model can explain the existence of several factors that may lead to sensory alterations in autistic people.

Speech Production From a Developmental Perspective

Purpose Current approaches to speech production aim to explain adult behavior and so make assumptions that, when taken to their logical conclusion, fail to adequately account for development. This failure is problematic if adult behavior can be understood to emerge from the developmental process. This problem motivates the proposal of a developmentally sensitive theory of speech production. The working hypothesis, which structures the theory, is that feedforward representations and processes mature earlier than central feedback control processes in speech production. Method Theoretical assumptions that underpin the 2 major approaches to adult speech production are reviewed. Strengths and weaknesses are evaluated with respect to developmental patterns. A developmental approach is then pursued. The strengths of existing theories are borrowed, and the ideas are resynthesized under the working hypothesis. The speech production process is then reimagined in developmental stages, with each stage building on the previous one. Conclusion The resulting theory proposes that speech production relies on conceptually linked representations that are information-reduced holistic perceptual and motoric forms, constituting the phonological aspect of a system that is acquired with the lexicon. These forms are referred to as exemplars and schemas, respectively. When a particular exemplar and schema are activated with the selection of a particular lexical concept, their forms are used to define unique trajectories through an endogenous perceptual-motor space that guides implementation. This space is not linguistic, reflecting its origin in the prespeech period. Central feedback control over production emerges with failures in communication and the development of a self-concept.

The role of geminates in infants' early word production and word-form recognition

Infants learning languages with long consonants, or geminates, have been found to 'overselect' and 'overproduce' these consonants in early words and also to commonly omit the word-initial consonant. A production study with thirty Italian children recorded at 1;3 and 1;9 strongly confirmed both of these tendencies. To test the hypothesis that it is the salience of the medial geminate that detracts attention from the initial consonant we conducted three experiments with 11-month-old Italian infants. We first established baseline word-form recognition for untrained familiar trochaic disyllables and then tested for word-form recognition, separately for words with geminates and singletons, after changing the initial consonant to create nonwords from both familiar and rare forms. Familiar words with geminates were recognized despite the change, words with singletons were not. The findings indicate that a feature occurring later in the word affects initial consonant production and perception, which supports the whole-word phonology model.

Learning words and learning sounds: Advances in language development

Phonological development is sometimes seen as a process of learning sounds, or forming phonological categories, and then combining sounds to build words, with the evidence taken largely from studies demonstrating 'perceptual narrowing' in infant speech perception over the first year of life. In contrast, studies of early word production have long provided evidence that holistic word learning may precede the formation of phonological categories. In that account, children begin by matching their existing vocal patterns to adult words, with knowledge of the phonological system emerging from the network of related word forms. Here I review evidence from production and then consider how the implicit and explicit learning mechanisms assumed by the complementary memory systems model might be understood as reconciling the two approaches.

Frame dominance in infants with hearing loss

PURPOSE

According to the frames then content (f/c) hypothesis (P. F. MacNeilage & B. L. Davis, 1990), the internal structure of syllables with consonant plus vowel structure (CV) during canonical babbling is determined primarily by production system properties related to rhythmic mandibular oscillations (motor frames). The purpose of this study was to evaluate whether auditory sensitivity affects the internal organization of CV syllables in infants identified in the 1st year of life with hearing loss.

METHOD

CV co-occurrence patterns were analyzed for 13 infants with varying degrees of hearing sensitivity (normal hearing [n = 4], mild-to-moderately severe hearing loss (n = 6), and severe-to-profound hearing loss [n = 3]). Consonants and vowels within CV syllables were grouped according to place of articulation. Thus, an inventory of CV syllables with labial, coronal, and dorsal consonant onsets was created.

RESULTS

77% of predicted frames were confirmed above chance. Additionally, there was no association between pure-tone average and any CV co-occurrence. Finally, co-occurrences that were not predicted by the hypothesis were statistically confirmed in very few instances.

CONCLUSIONS

Auditory sensitivity may not influence intrasyllabic organization within CV syllables once infants begin canonical babbling, as the co-occurrences observed are primarily those predicted by the f/c hypothesis.

Auditory sensitivity and the prelinguistic vocalizations of early-amplified infants

PURPOSE

Vocalization development has not been studied thoroughly in infants with early-identified hearing loss who receive hearing aids in the 1st year of life. This study sought to evaluate the relationship between auditory sensitivity and prelinguistic vocalization patterns in infants during the babbling stage.

METHOD

Spontaneous vocalizations of 15 early-identified infants with varying degrees of hearing sensitivity, from normal to profound hearing loss, were audiotaped and perceptually transcribed. Associations between the infant's unaided pure-tone average and the following vocalizations were explored: canonical babbling ratio, percentage of utterances containing canonical syllables, canonical syllable shapes, number of syllable sequences, and consonant-onset patterns in canonical syllables.

RESULTS

Hearing sensitivity was significantly associated with the percentage of utterances containing canonical syllables, the vocalization types used in utterances, and canonical syllable shapes used by the infants.

CONCLUSIONS

Auditory sensitivity contributes significantly to the emergence of babbling patterns. In addition, there is a need for continued study of the vocalizations of infants with milder forms of hearing loss, because in this study, their vocalizations were highly variable despite having received early amplification.

Developmental dyslexia, learning and the cerebellum

Theoretical frameworks for dyslexia must explain how the well-established phonological deficits and the literacy deficits arise. Our longstanding research programme has led to a distinctive 'twin level' framework that proposes, first, that the core deficits are well described in terms of poor skill automaticity. Second, these 'cognitive level' symptoms are attributed to abnormal cerebellar function--a 'brain-level' analysis. The evidence includes data from behavioural, imaging, neuroanatomical and learning studies. The frame-work leads to an 'ontogenetic' analysis that links cerebellar deficit at birth, via problems in articulation and working memory, to the known phonological, speed and literacy difficulties. Differences in locus of cerebellar impairment, experience and/or links to other brain regions may account for subtypes of dyslexia and possibly other developmental disorders. The automaticity/ cerebellar deficit framework provides an explicit demonstration that it is possible to explain motor, speed and phonological deficits within a unified account, integrating previously opposed approaches.

The neuroanatomical and functional organization of speech perception

A striking property of speech perception is its resilience in the face of acoustic variability (among speech sounds produced by different speakers at different times, for example). The robustness of speech perception might, in part, result from multiple, complementary representations of the input, which operate in both acoustic-phonetic feature-based and articulatory-gestural domains. Recent studies of the anatomical and functional organization of the non-human primate auditory cortical system point to multiple, parallel, hierarchically organized processing pathways that involve the temporal, parietal and frontal cortices. Functional neuroimaging evidence indicates that a similar organization might underlie speech perception in humans. These parallel, hierarchical processing 'streams', both within and across hemispheres, might operate on distinguishable, complementary types of representations and subserve complementary types of processing. Two long-opposing views of speech perception have posited a basis either in acoustic feature processing or in gestural motor processing; the view put forward here might help reconcile these positions.

Early sound patterns in the speech of two Brazilian Portuguese speakers

Sound patterns in the speech of two Brazilian-Portuguese speaking children are compared with early production patterns in English-learning children as well as English and Brazilian-Portuguese (BP) characteristics. The relationship between production system effects and ambient language influences in the acquisition of early sound patterns is of primary interest, as English and BP are characterized by differing phonological systems. Results emphasize the primacy of production system effects in early acquisition, although even the earliest word forms show evidence of perceptual effects from the ambient language in both BP children. Use of labials and coronals and low and midfront vowels in simple syllable shapes is consistent with acquisition data for this period across languages. However, potential ambient language influences include higher frequencies of dorsals, use of multisyllabic words, and different phone types in syllable-offset position. These results suggest that to fully understand early acquisition of sound systems one must account for both production system effects and perceptual effects from the ambient language.

Acquisition of serial complexity in speech production: a comparison of phonetic and phonological approaches to first word production

Comparison was made between performance-based and competence-based approaches to the understanding of first word production. The performance-related frame/content approach is representative of the biological/functional perspective of phonetics in seeking explanations based on motor, perceptual and cognitive aspects of speech actions. From this perspective, intrasyllabic consonant-vowel (CV) co-occurrence patterns and intersyllabic sequence patterns are viewed as reflective of biomechanical constraints emerging from mandibular oscillation cycles. A labial-coronal sequence effect involved, in addition, the problem of interfacing the lexicon with the motor system, as well as the additional problem of initiation of movement complexes. Competence-based approaches to acquisition are within the generative phonological tradition; involving an initial assumption of innate, speech-specific mental structures. While various current phonological approaches to acquisition involve consideration of sequence effects and intrasyllabic patterns, they do not adequately establish the proposed mental entities in infants of this age, and are nonexplanatory in the sense of not considering the causes of the structures and constraints that they posit.

Whole-Word Phonology and Templates

Advances in psycholinguistics have identified cognitive mechanisms that may account for the phenomena of whole-word phonology and phonological templates in normally developing children. Deficits in these same mechanisms may also account for certain types of disordered phonologies. In this paper, these cognitive mechanisms are described, strategies for identifying whole-word phonological patterns in normal and disordered phonologies are proposed, and intervention strategies that draw on these same mechanisms as a way to overcome their inappropriate persistence are recommended.