Jenny R. Saffran - US grants

Statistical learning is a promising addition to the language learning arsenal, given the massive amount of statistical information in linguistic input. However, the fact that such mechanisms exist raises several pressing problems. In particular, different types of statistical learning mechanisms are required for the acquisition of different aspects of language. What remains unknown is how these mechanisms interact to derive linguistic structure, given multiple possible levels of analysis. A related issue concerns the domain-specificity of statistical language learning processes.

The following objectives will be addressed, using previously developed laboratory learning paradigms which permit careful manipulation of the input and detailed assessment of what participants are able to learn: (1) to determine whether learners can track the multiple levels of statistical regularity present in linguistic input, (2) to reveal constraints on the particular types of computations performed by infant, child, and adult language learners, (3) to determine whether the same types of regularities are detected in linguistic and non-linguistic learning tasks. The answers to these questions will inform an emerging theoretical framework, constrained statistical learning, intended to elucidate the study of language acquisition and other pressing issues in human learning and development. The proposed teaching objectives will serve to extend the excitement of the psychology laboratory to the classroom by training undergraduate Research and Teaching Fellows to bridge the gap between research and teaching by leading discussion sections, by facilitating the participation of students in a large lecture class in running small-scale research projects, and by encouraging scientific writing in the psychology classroom.

DESCRIPTION (provided by applicant): First language acquisition is a hallmark of normative human development. A substantial body of research suggests that language learning is facilitated by the ability to track statistical regularities in linguistic input. Research during the current project period clearly demonstrates that infants are powerful statistical learners. However, the relevance of these abilities to the learning problems presented in infants' linguistic environments remains poorly understood. The research proposed in this application will test specific hypotheses concerning infant statistical language learning, focusing on how infants make use of statistical information. Specific Aim One is to determine which surface statistics infants can use given natural language input. Specific Aim Two is to determine how the statistics of sound sequences in real speech influence word learning. Specific Aim Three is to determine whether infants' on-line language processing is affected by statistical information. The results of the research proposed in this competing continuation application will promote positive developmental outcomes by expanding our understanding of the learning mechanisms underlying normative development. Individuals who are less facile at statistical learning may be at risk for developmental language disorders. Subsequent research will use the outcome of these studies to motivate investigations including populations of young children at risk for atypical language development.

DESCRIPTION (provided by applicant): Language and communication deficits are one of the hallmarks of autism spectrum disorders (ASD). Language delay is often the first concern of parents, and early language abilities are predictive of broader outcomes for children on the autism spectrum. Recent studies have noted that comprehension is relatively more affected than production in young children with ASD, yet little research has focused on this aspect of early language functioning. Contemporary versions of the weak central coherence account suggest that autism is characterized by a cognitive style in which there is a bias towards local or low-level perceptual processing rather than global, integrative processing. This theoretical account yields specific hypotheses regarding early phonological and semantic processing which will be tested in this project. The proposed research will employ implicit eye-gaze paradigms developed to assess real-time comprehension (lexical representations) in typically developing infants and toddlers. A series of five studies are planned to address the following three specific aims: 1) to investigate the hypothesis that toddlers with ASD display a bias towards enhanced perceptual processing in lexical tasks compared to typically developing peers; 2) to examine the claim that toddlers with ASD exhibit poorer contextual integration on lexical tasks than typical controls; and 3) to determine whether early phonological and semantic processing abilities predict later language outcomes. A total of 150 toddlers will participate, 75 with ASD and 75 typically developing toddlers matched on parent-reported vocabulary comprehension and sex. A looking-while-listening paradigm will be used in Studies 1-4. Study 5 will involve a 1-year follow-up assessment to examine predictors of language outcomes as well as re-administration of one of the comprehension processing tasks. This research is significant in that it will provide a comprehensive characterization of early lexical processing in toddlers with ASD and provide the first systematic evaluation of whether the weak central coherence account can help explain the early comprehension deficits in this group. Findings will lead to better understanding of the autism phenotype(s) and have direct implications for developing early interventions.

The Directorate of Social, Behavioral and Economic Sciences offers postdoctoral research fellowships to provide opportunities for recent doctoral graduates to obtain additional training, to gain research experience under the sponsorship of established scientists, and to broaden their scientific horizons beyond their undergraduate and graduate training. Postdoctoral fellowships are further designed to assist new scientists to direct their research efforts across traditional disciplinary lines and to avail themselves of unique research resources, sites, and facilities, including at foreign locations. This postdoctoral fellowship trains a young scientist exploring
the early process of language learning by studying how young infants discover words in spoken language. This project also assesses how language experience affects the word discovery process by studying bilingual-learning infants and adults. By including infants, results from the project further knowledge about how learners acquire linguistic regularities from spoken language and provide insights into the mechanisms underlying early language development, both of which inform efforts to understand language impairments. Additionally, by including bilingual and adult populations, the project highlights how bilingualism changes the process of language acquisition, particularly early on in development. Beyond the scientific contributions, this fellowship also 1) supports the mentorship of undergraduate students from underrepresented backgrounds, 2) engages the local community, including populations typically underrepresented in research, through direct participation in the research and through presentation of the results by the fellow, and 3) supports the fellow?s outreach plan to broaden the participation of students from diverse backgrounds in the sciences. Finally, this fellowship supports the training and development of the fellow, a Mexican-American woman and first generation college graduate that holds a personal commitment to advancing underrepresented populations in STEM.

Acquiring the words of the native language(s) typically begins by discovering them in speech. This process starts in the first year of life and is critical for language development. However, identifying the sounds that compose the words in spoken language is not trivial for the young child. The patterns that define words in natural languages are complex: they consist of multiple features and vary in their degree of reliability. Additionally, spoken words rarely occur in isolation, or with pauses to signal where they start and end. How do infants discover the structure of words given the complexities of spoken language? How does this process change as learners gain experience with language? To address these questions, this research manipulates the complexity of the language input, together with the language experience of the learner, to understand how infants discover words in speech. The studies in this project use an artificial language learning approach to study these questions. Study 1 examines how monolingual English learning infants (8 to 10 months of age) acquire a novel, complex word pattern embedded in fluent speech. Studies 2 and 3 assess how language experience affects learning by testing bilingual-learning infants (Study 2), and adults (Study 3). These studies outline how typically developing infants acquire linguistic patterns from spoken language, which is a necessary endeavor for understanding early language development. This project also informs efforts to understand communication and language disorders, and has the potential to be used in the future as a framework to study children with language and developmental delays.

Abstract Language delay is often one of the first concerns of parents of toddlers with autism spectrum disorder (ASD), and early language abilities predict broader outcomes for children on the autism spectrum. Yet, mechanisms underlying language deficits in children with ASD remain unspecified. One prominent component of linguistic behavior is the use of predictions or expectations during learning and processing. Several researchers have recently posited prediction-deficit accounts of ASD. The basic assumption of the prediction accounts is that information is processed by making predictions and testing violations against expectations (prediction errors). Flexible (neurotypical) brains attribute differential weights to prediction errors to determine when new learning is appropriate, while individuals with ASD are thought to assign disproportionate weight to prediction errors. These prediction deficits are hypothesized to lead to higher levels of perceived novelty, resulting in `hyperplasticity' of learning based on the most recent input. The proposed project will be the first study to examine whether language deficits in young children with ASD are linked to atypical domain-general prediction processes. Seven studies are proposed to address the following Specific Aims: (1) to determine the ability of toddlers with ASD to generate predictions compared to typically developing (TD) peers; (2) to establish whether toddlers with ASD exhibit hyperplasticity of learning relative to TD peers; (3) to examine the extent to which certain child characteristics predict individual variability in predictive behavior and hyperplasticity in toddlers with ASD; and (4) establish whether predictive behavior or hyperplasticity of learning on verbal/visual tasks predicts vocabulary and/or syntactic ability one year later in young children with ASD. The sample will be comprised of 75 ASD toddlers (three cohorts of 25 toddlers) and 75 TD controls distribution-matched on cognition (raw scores), SES, and sex. In this project we will utilize anticipatory eye movements (AEMs) and looking-while-listening (LWL) methods to investigate prediction in visual and verbal tasks in toddlers with and without ASD. Studies 1 (auditory) and 2 (visual) examine prediction as indexed by AEMs under conditions in which probabilities of the occurrence of events are relatively stable. Studies 3 (verbal) and 4 (nonverbal) will investigate hyperplasticity of learning in AEM tasks in which probabilities of the events change and predictions must be updated. Study 5 will explore hyperplasticity within the context of novel word learning using a LWL task. Study 6 will examine the relationship between child characteristics (cognition, language, autism symptom severity) and prediction. In Study 7, performance on Studies 1-5 will be used to predict language abilities in the same sample of toddlers one year later (from 18-35 mo to 30-47 mo) using both standardized tasks and an online incremental language processing task. All studies were developed to have minimal task demands in order to include toddlers with ASD across a wide range of functioning. Findings from this research will provide important new insights into ASD phenotypes and mechanisms underlying language development that could open new pathways for intervention.