Richard N. Aslin, Ph.D. - US grants

A series of experiments are proposed for investigating the manner in which infants and young children perceive snythetic and natural speech sounds as well as nonspeech stimuli that match the complex temporal and frequency relations present in speech. Discrimination and categorization data will be obtained from 6- to 12-months-old infants using an operant headturning procedure. Discrimination, categorization, and labeling data will be obtained from 2- to 4-year-olds using a two-alternative pointing procedure. Studies of infants will focus on the discrimination of foreign speech contrasts, trading relations, categorization of stop consonant place of articulation and vowel category in CV syllables, and the extraction of phonetic features. Studies of young children will focus on the discrimination of foreign speech contrasts, the discrimination and labeling of vowels, vowel normalization and reduction, context effects, trading relations, feature extraction, and the discrimination and labeling of place of articulation in stops, VOT, and TOT. In addition to these studies of speech perception, experiments with 6- to 12-months-olds will address several aspects of developmental psychoacoustics, including thresholds for high frequency tones, frequency discrimination, discrimination of frequency modulation, and psychophysical tuning curves. These psychoacoustic studies will not only establish norms for this age range but also uncover any sensory constraints on speech discrimination. The overall goal of this program of research is to evaluate the mechanisms underlying speech perception in infancy and early childhood. The role of early linguistic experience, particularly in the early stages of language production, will be examined, as well as the onset age for a phonetic mode of analysis.

The purpose of the program of research outlined in this grant proposal is to determine whether human infants in the second six months of life are able to perceptually segment fluent speech into word-length units that could be used to understand their native language. Although many studies in the past 15 years have documented the sophisticated speech discrimination capacities of young infants, those studies have focussed on the infant's perception of isolated speech segments. Studies of speech segmentation based on speech production are limited by the poor articulatory control shown by infants once words are produced and by the fact that perceptual segmentation must precede word production. In contrast to the perception of isolated speech segments, the perception of fluent speech requires on-line processing ant the extraction of discrete acoustic units from a speech waveform characterized by inconsistent acoustic markers for word boundaries. Two questions are critical: (a) can infants extract a familiar acoustic unit from fluent speech and recognize the equivalence of this unit in various contexts, and (b) is this extraction process facilitated by the entry of acoustic units into the lexicon? Three strategies will be employed to study speech segmentation in 6- to 12-month-olds. First, an operant headturning procedure and a visual habituation procedure will be used to assess infants' extraction of word-length units from fluent speech. Variables though to influence segmentation (e.g., speaking rate, intonation and stress, coarticulation of adjacent phonetic segments) will be examined on these segment extraction tasks. Second, the speech INPUT to the infant from the mother will be analyzed in great detail to describe the variability and consistency with which acoustic cues to segmentation are presented to the infant. Short-term "training" studies will assess the role of reference in infants' segmentation of words and pseudowords. Third, infants' preferences for variations in speaking rate and stress patterns will be obtained to determine whether these suprasegmental aspects of fluent speech influence infants' attention. Taken together, these studies will clarify an under-studied but essential aspect of language required for lexical and syntactic development.

Both humans and non-human primates show remarkable learning abilities. However, these abilities are often limited to certain domains, developmental periods, or behavioral contexts. For example, nearly all humans acquire one or more complex linguistic systems-that is, languages -- but not all humans acquire complex musical systems. Similarly, non-human primates are exceptionally adept at learning to forage for and categorize different types of food, but are severely limited in acquiring complex communication systems. Also, both humans and non-human primates appear to learn best in several domains during early periods of development. Thus, learning is nearly always characterized by specializations, rather than by general-purpose mechanisms. Understanding the constraints on learning will contribute to basic research, by accounting for domain- and species-specializations, and to applied research, by refining our understanding of which domains, ages, and contexts are optimal for human
learners.

The goal of the present research project is to explore the ability of human adults, children, infants, and non-human primates (Tamarins) to learn rapid sequential events. A prime example of a rapid sequential event is language, in which sounds are combined to form words, and words are combined to form sentences. Recent findings have demonstrated that human adults and infants can rapidly extract and remember very detailed 'statistics' of linguistic input, such as the frequency and probability that one syllable will follow another. In our proposed research, we will employ miniature artificial 'languages' which simulate some of the structural properties of natural languages, but which can be built with equivalent structures across different domains (speech sequences, tone sequences, visual sequences, motor sequences). At issue is the facility humans and non-human primates show for the extraction of statistical structure from these different learning materials. Are they equally sensitive to the distributions of elements and higher order structure in the materials? Do learning abilities differ across learners of different ages and species, and across different structures and domains?

In addition to behavioral experiments with humans and non-human primates, a series of computational studies will allow us to investigate the formal properties of learning mechanisms, in order to ask what architectural and neural differences might underlie such differences in learning abilities. What kinds of computational architectures can learn the types of regularities and patterns that human infants learn? Is the inability of adults or non-human primates to learn some types of complex sequential events due to the absence of a learning device specialized for that domain, or could small differences in computation and/or memory lead to large differences in learning outcome?

The goal of the project is to reveal how normal human infants, during the second six months of postnatal development, acquire the sound- structures that will become words in their native language. This process of early world learning, which occurs before infants begin to produce words in their speech, must involve the segmentation of stretches of fluent adult speech that correspond to words. Infants presumably use both acoustic cues, such as pauses and prosody (e.g., pitch and stress), and distributional cues, such as the statistical patterning of sequences of sounds, to solve the word-segmentation task. Using a preferential listening technique, preceded by a familiarization phase, 8-month-olds will be tested for their ability to segment multi-syllabic word-like units from artificial language corpora. These corpora will be brief (2-4 minutes) and will be created by a speech synthesizer to control for the presence (or absence) of acoustic cues to word boundaries. The proposed experiments will examine the relative importance of acoustic and statistical cues to work boundaries, the temporal ordering of statistical cues, the limitations on which statistical cues can be used, and the robustness of statistical cues in long-term memory. These studies on word segmentation, therefore, will not only provide important information about a fundamental aspect of early language acquisition, but they will also serve as a model system for the examination of other aspects of statistical language learning.

DESCRIPTION (provided by applicant): The goal of this program of research is to determine how the developing human infant forms representations of the visual world. The focus of the research is on a class of powerful learning mechanisms that have been shown by the Principal Investigator and his colleagues to rapidly extract from sequences of auditory stimuli the statistical properties that form coherent units (e.g., words and melodies). Visual statistical learning will be studied to determine whether a similarly powerful set of mechanisms is present in a modality other than audition. Infants ranging in age from 3- to 12-months of age, as well as adults, will be tested on a variety of statistical learning tasks in which small visual shapes are arranged into scenes. At issue is how infants and adults learn that some of these shapes appear together (co-occur) across many different scenes, forming the basic building blocks for representing those scenes in memory. Four different techniques will be used with infants. The primary technique involves the repeated presentation of a sequence of 16-28 different scenes composed of 3-6 different shapes. After a decline in looking time (habituation) to these displays, infants will be presented with test displays containing coherent (high statistical relatedness) and incoherent (unrelated) shapes that were embedded in the scenes. The other three techniques involve forced-choice preferential looking, automated corneal reflection eye-tracking, and anticipatory eye-movements to learned categories. These techniques will be used to determine whether newly learned features activate attention in cluttered scenes, how these features are learned when low-level properties of the scenes compete for attention, how variations over time in the input statistics affect the accuracy of feature learning, and how other perceptual constraints affect feature learning. A key hypothesis of the Pl's statistical approach will be tested -- that learners represent the largest coherent unit in a complex array of elements, rather than also representing all of the embedded elements that are redundant with this larger unit. Taken together, these proposed studies will reveal how infants and adults learn new information from complex visual scenes and represent that information in a computationally efficient manner. Failure to learn efficiently could lead to deficits in the early phases of learning in infancy and negatively affect the formation of higher level categories.

0931687
Berger

This project aims to improve the use of near-infrared light for studying infants' brain activity. Although current optical methods of monitoring brain activity in infants show promise, many poorly-corrected variables (physical, physiological, and anatomical) still corrupt the data from too many infant participants. Through the research team's combination of expertise in biophotonics, brain and cognitive sciences, and pediatrics, the sources of noise will be reduced to levels where meaningful scans can be performed on babies routinely. The method used will be near-infrared spectroscopy (NIRS), which measures changes in the absorption of light by blood in the scalp and brain. The instrumentation will include dedicated "scalp-only" channels whose source-detector separations (~5 mm) are too short to be sensitive to brain signals. When one of these channels' readout is subtracted from standard channels (with separations of 20-35 mm), much physiological noise is cancelled, and stimulus related activations become easier to observe. The researchers' home-built corrected-near infrared-spectroscopy (C-NIRS) system for adults will be restructured for infant studies and enhanced with more sources and detectors, enabling measurements from a stimulated region and concurrent monitoring of a "control" region. An instrument of this scale designed with dedicated near-channel noise-suppression will be the first of its kind. In parallel, a more flexible silicone-based headband will be developed to fit a range of head shapes, be comfortable and safe for infants, and reduce the amplitude of motion artifacts in the data. The new instrument will be tested on a cohort of 60 healthy, alert infant participants (age 6-9 mos.), along with a commercial near-infrared instrument that lacks correction channels.
Measurement protocols will focus on well-studied visual and auditory stimulation paradigms. The reductions in physiological noise and motion artifacts should yield a higher "hit rate" of meaningful data runs using the C-NIRS instrument than for traditional NIRS, thereby establishing C-NIRS as a valuable tool in the arsenal of the cognitive scientist studying infant development.

Understanding what infants understand about objects and words that they encounter in the world has been an important goal in developmental science, but the field understands relatively little about how infants perform either of these two tasks. Several neuroimaging methods have been used to determine how adult brains recognize familiar objects and words, but most of these methods are not suitable for use with infants. The goal of this research project is to deploy two neuroimaging methods that are amenable for use with infants, as a novel way to gain insights into the fundamental brain mechanisms that enable object recognition and word understanding in 3- to 12-month-old infants. One technique, electroencephalography (EEG) involves measuring electrical activity generated by the brain from sensors on the scalp. The other, functional near-infrared spectroscopy (fNIRS), shines near-infrared light through the skull and measures how it is absorbed by the brain at each location as an index of how active that part of the brain is. Recording both these measures while infants watch and listen to stimuli will provide important insights into how the infant brain processes this information.

EEG has been in use with infants for many years, whereas functional near-infrared spectroscopy (fNIRS) is a relatively newer non-invasive neuroimaging technique ideally suited for use with infants. fNIRS, like fMRI, provides a signature of metabolic activity in localized regions of the brain but is more suitable for infants because it delivers light to the scalp via a tight-fitting cap. In this project, EEG and fNIRS will be used to measure electrical and metabolic activity in the brain as infants watch objects of various categories such as vehicles or furniture, or listen to words that they know or do not know (or nonsense words). The key analysis tool is a suite of machine-learning algorithms from the field of computer science that take the EEG or fNIRS signals to determine which components of these signals best predict, on a trial by trial basis, what the infant was seeing or hearing.

The outcome of the proposed research will localize in the brain where encoding of visual objects and spoken words takes place and over what time period after stimulus onset that processing occurs. These are fundamental aspects of object and language processing have eluded study in the human infant because of methodological challenges. Establishing protocols for analyzing data generated by these two methods will contribute to analytic techniques available for future infant brain research. Identifying normative properties of these processing mechanisms in the infant brain will set the stage for future research investigating how the developing brain is affected by variations in early experience, by compensation after injury, and by a variety of genetic anomalies.

Project Summary Word recognition is crucial not only for comprehending spoken language but for mapping spoken words onto text in reading. Individuals with language and reading deficits (e.g., Specific Language Impairment, Dyslexia, Autism, which together affect up to 16% of children) have been shown to have deficits in word recognition, making it crucial to understand this process. A hallmark of word recognition is that listeners activate neural representations of multiple candidate words that are consistent with the early acoustic input, and these candidates compete for recognition as they unfold in real-time. The overall goal of this proposal is to capitalize on recent developments in multivariate and machine- learning techniques for analyzing signals obtained from the human brain to measure the real-time unfolding of spoken word recognition. Although these techniques have been most widely used with fMRI data, we propose to extend them to EEG data because EEG is easily used with children and clinical populations, and provides access to the time-course of word recognition, thereby revealing underlying cognitive mechanisms of word recognition, such as lexical competition. Our preliminary findings using this EEG-based paradigm have demonstrated that we can decode the recognition of a specific word (among a set of 8-12 alternatives) at each msec time-step after stimulus onset. The method is sensitive to partial activation of competing words that share some phonological features with the target word, thereby revealing the dynamics of lexical competition as the word-recognition system settles on the final target. Our objectives are to conduct a series of small-scale experiments that achieve three aims. First, we develop and optimize the method with adults (e.g., the experimental procedure and computational implementation). Second, we validate the method with adults by measuring its test/re-test reliability, comparing its estimates of word recognition with traditional behavioral paradigms, and examining how lexical status, and semantic and orthographic expectations shape lexical competition revealed by the EEG measure. This will yield a new, non-invasive, and highly reliable method suitable for assessing spoken word recognition in adults, children, and special populations. Third, we will preliminarily extend the method to children to pave the way for future developmental studies. Taken together, accomplishing these three aims would provide an innovative and powerful tool for assessing a crucial component of language processing in a wide variety of typical and atypical populations.