Randy Diehl, Ph.D. - US grants

The goal of this proposal is to examine the development of complex auditory information processing in infants. Two sets of experiments are planned: one dealing with the perception of rhythmic structure and the other with the perception of speech sounds. The first set of experiments is intended to delineate the precise conditions that give rise to rhythm discrimination and the formation of rhythmic categories and to test competing hypotheses about how rhythm is perceptually organized. In the second set of experiments, we will study speech sound discrimination and category formation. A main issue here is whether, and at what age, infants display the same kinds of phonetic trading relations and context effects as are found for adults. In addition, both sets of experiments are designed to test, in the auditory domain, a developmental theory of information processing that is based on earlier work on infant visual perception. In all experiments we will use a visual habituation procedure to assess auditory discrimination and categorization. The principal advantage of this procedure over other techniques is that it is suitable for a wide range of age groups, and it will therefore enable us to monitor developmental changes in auditory information processing.

The long-term objectives of the research are to develop our knowledge of how humans discriminate and recognize speech sounds and to apply this knowledge toward the explanation of specific phonetic regularities. The proposed studies will build on previous work in our laboratory to yield a more complete account of auditory factors in speech perception and of listener-oriented factors in speech production. In general, a phonological contrast is realized by a covarying set of phonetic properties, and this covariation is often quite uniform across languages. Typically, phoneticians have attempted to explain phonetic covariation in terms of putative physical constraints on speech production. Correspondingly, theorists within the field of speech perception have often assumed that phonetic trading relations (where the perception of one acoustic-phonetic dimension is affected by the setting of another acoustic parameter) result primarily from the listener's implicit knowledge of speech production constraints. An alternative hypothesis is that many significant types of phonetic covariations reflect a strategy of speech communities to enhance the auditory distinctiveness of phonological contrasts, and relatedly, that many phonetic trading relations are rooted in general auditory factors that are not specific to speech. To evaluate this alternative hypothesis, a variety of perceptual and auditory modeling experiments will be conducted, each designed to factor out the possible perceptual contribution of speech-relevant knowledge. The experiment will employ the following methods, among others: (1) Speech/nonspeech comparisons. Categorization performance on speech stimuli varying along certain acoustic dimensions are compared to performance on nonspeech stimuli that are acoustically analogous in certain relevant respects. (2) Cross-native-language comparisons. Certain phonetic dimensions are relevant in some languages but not others. It is possible to factor out the contribution of language-relevant experience by comparing categorization performance on such a phonetic dimension between a group of subjects whose native language exploits the dimension and a group of subjects for whom the dimension is linguistically irrelevant. (3) Auditory modeling. A realistic model of mammalian auditory-nerve response is used to predict performance on several types of speech perceptual task (e.g., discrimination, categorization and auditory distance judgements). These converging methods will yield important information about how the auditory representation of speech sounds constrains speech perceptual performance. The results will also permit a detailed evaluation of listener-oriented accounts of phonetic covariation. Finally, the research will provide a foundation for future work on how various forms of auditory pathology affect speech perception.

Research studies have shown that the two hemispheres of the brain have different functions. For example, the left hemisphere has been argued to specialize language processing while artistic/creative abilities are specialized by the right hemisphere. This dissertation will evaluate two possible accounts for hemispheric specialization of pitch processing. The first account is a functional account which states that hemispheric specialization of pitch processing is determined by the function of pitch contrasts. For example, pitch contrasts that are more linguistic (e.g. linguistic tone, stress, and intonation) are processed by the left hemisphere, and pitch contrasts that are less linguistic (e.g. music) are processed by the right hemisphere. The second account is an acoustic account which states that acoustic correlates determine hemispheric specialization, and therefore, all processing of pitch, regardless of function, is carried out by one hemisphere of the brain. Thus, pitch used in both intonation and music might be processed by the right hemisphere or the left hemisphere.

Three sets of experiments involving the perception of linguistic tone (or linguistic pitch contrast) will be conducted in order to evaluate these two accounts. Linguistic tone is defined as pitch pattern that is used to signal meaning at the word level (e.g. in Chinese, /ma/ spoken with a high pitch means 'mother'; the same syllable means 'a horse' when spoken with a falling pitch). The first experiment will use brain-damaged individuals as subjects; patients with either left-hemisphere or right-hemisphere damage will be asked to identify linguistic tones. The second experiment is a dichotic listening experiment which will present linguistic tones to only one ear of normal-hearing subjects and assess which ear/hemisphere combination better detects these tones. The third experiment will use neuroimaging technique to examine directly which cerebral hemisphere is active when normal-hearing subjects are listening to linguistic tones.

Most languages of the world (although not English) utilize linguistic tones. A better understanding of how the brain processes linguistic tones will not only help us to understand language processing and the universality of language, but may also help us to develop therapeutic techniques to help rehabilitate the oral communication skills of many brain damaged patients.

DESCRIPTION: Focusing on vowel sounds from a wide variety of languages, we use a combination of acoustic measurements of natural tokens, computer simulation of preferred vowel systems, and perceptual and statistical pattern recognition experiments to test predictions of the Theory of Adaptive Dispersion (TAD) and several of its corollaries (namely, the Size Principle, the Hyper and Hypo Theory, and the Auditory Enhancement Hypothesis). TAD states that the aim of talkers is to make phonological contrasts sufficiently distinctive to promote linguistic comprehension by the listener while minimizing the articulatory effort needed to achieve this degree of distinctiveness. Among the predictions tested are these: (1) As vowel inventories increase in size, there will tend to be a greater use of various contrast-enhancing properties, including vowel inherent spectral change, an expanded vowel space, and a greater use of distinctive vowel length differences. (2) For a given level of redundancy in the communication setting, alternative sources of contrast enhancement will tend to vary inversely across dialects of the same language. (3) Contrast enhancement will be more in evidence when a vowel is part of a word that represents new rather than given information. (4) Computer simulations of preferred vowel inventories based on TAD will tend to be more accurate when the vowel tokens are dynamically specified. Because TAD and its corollaries are primarily aimed at explaining regularities of speech production rather than of speech perception, they must be supplemented by specific hypotheses that predict how speech sounds are categorized by listeners. Two such hypotheses that will be tested are: the Tonotopic Distance Hypothesis and the Linear Boundary Hypothesis. The first of these states that perceived boundary locations among vowel categories are determined by the distances between adjacent spectral peaks but that the weight of each distance cue is inversely related to the size of the distance. The second hypothesis states that perceived category boundaries between vowel stimuli tend to be linear when formant frequencies are scaled in auditorily motivated units such as Bark. These perceptual hypotheses will be extensively tested both within and across languages.

With National Science Foundation support, Dr. Lori Holt, Dr. Andrew Lotto, and Dr. Randy Diehl will collaboratively investigate the basic perceptual and cognitive processes that underlie complex auditory categorization, such as phonetic categorization.

One of the goals of the project is to develop efficient and robust L2 training techniques. In order to accomplish this, it is important to understand how the auditory system deals with variability in the input to create useful categories. With prior NSF support, the PIs have developed a methodology for training listeners to categorize novel non-speech sounds that have some of the same complexity as speech sounds. This approach allows the researcher to have complete control of listeners' experience and provides a view of category learning as it is happening. In this project, this methodology will be applied specifically to study the problem of feature determination and weighting in categorization. Results from these studies will be analyzed within a Bayesian statistical decision framework, which provides a means to compute how an "ideal observer" would perform in the task. Thus, it is possible to determine if listeners are performing optimally and, if not, discover what biases they exhibit in weighting strategies.

Based on the results of these methodologies, training regimens for L2 category learning will be created and tested by training adult Japanese listeners on English /l/-/r/. In addition to the broader impact derived from a better understanding of auditory categorization and improved L2 training techniques, this project will develop techniques that may be useful for therapy with new users of hearing aids or cochlear implants. This program of research also provides a unique training opportunity for students. The three laboratories involved are connected to world-class experimental groups with different emphases (perceptual systems, cognitive neuroscience, audition and communication disorders). This inter-disciplinary setting will provide substantial interdisciplinary interaction for students.

[unreadable] DESCRIPTION (provided by applicant): Concepts from ideal observer theory will be used to study auditory category learning and speech perception. Ideal observers provide a benchmark for assessing human performance on perceptual tasks and may serve as an appropriate starting point for developing models of actual performance. In the first set of experiments, listeners will learn to categorize novel non-speech stimuli. Prior probabilities of the categories and degree of category overlap will be varied to assess listeners' sensitivity to the distributional properties of the stimuli. The second set of experiments will test whether selective adaptation and contrast effects in speech and non- speech perception can be modeled as changes in the listeners' prior probability and stimulus likelihood distributions. The third set of experiments will extend recent pilot work in our laboratory aimed at developing ideal classifiers for naturally produced speech sounds. The results of this pilot work suggest that listeners adopt a strategy of optimizing categorization performance with respect to the natural distributional properties of phonemes in their language. This hypothesis will be comprehensively tested in these experiments. The fourth set of experiments will examine how observers weight and integrate multiple sources of evidence in making phoneme category judgments. The aim is to learn how closely listeners approximate optimal informational weightings in such tasks when different stimulus properties are degraded in either predictable or unpredictable ways. In the fifth set of experiments, natural phoneme distributions will be used to estimate optimal trajectories between a set of native language (L1) phoneme contrasts and a competing set of contrasts in a second language (L2). Perception data of L2 learners at varying stages of proficiency will be compared to the estimated optimal trajectories. The techniques developed and tested in the proposed studies can serve as important tools for the study of speech perception in listeners with varying degrees of hearing loss, including those with cochlear implants. In particular, the derivation of ideal classifiers whose auditory input is restricted to mimic that of particular hearing-impaired individuals may allow investigators to distinguish between limits on performance due to sensory factors and those attributable to non-optimal categorization strategies. [unreadable] [unreadable] [unreadable]