Research
Mechanisms of Language Learning
The questions I address in my work are anchored in the perspective that infants’ ability to track statistical regularities, such as how often a perceived unit of the environment (e.g., a sound or object) occurs and how these units co-occur, is a key mechanism by which they come to know a language. Statistical regularities are pervasive in our environments and tracking them is crucial to adaptive behavior in humans and nonhumans alike. For example, statistical learning contributes to our ability to identify objects, and to predict their trajectory when they temporarily go out of view. A key question that I address in my work is how can tracking simple statistics, something even rats can do, result in learning a language. My work shows that considering how learning unfolds over time and experience can help answer this question. For example, infants readily learn simple, “local” dependencies, but struggle to learn “long-distance” dependencies (or LDDs). Although this initially led to speculation about whether representing LDDs has an innate basis, I found that infants successfully detect LDDs in an artificial language if they are first given experience with adjacent dependencies (Lany & Gómez, 2008; Psychological Science). Another question is how infants learn the meanings of words, often needing just one encounter with a word to correctly determine what it means. My work shows that infants’ grounding in simple local dependencies can lead to developing powerful intuitions that guide word learning (Lany & Saffran, 2010, Psychological Science; Lany & Saffran, 2011, Developmental Science). Consider the clear difference in the suggested meaning of “dax” in the sentence “I want a dax” vs. “I want to dax”. Using artificial language materials, I showed that tracking the overlap in statistical and referential properties of words spurs the development of these important intuitions.
These studies provided important initial evidence that tracking statistical structure provides an important foundation for language development. Tightly controlled laboratory studies, however elegantly designed, are limited in what they can say about language learning “in the wild”, and researchers have rightfully questioned whether they tap the same mechanisms that support real-world language learning. My recent findings are among the first to show that infants under 2 who are better able to learn statistical regularities in such laboratory tasks are more likely to combine words in their own speech and have larger vocabularies (Lany, 2014,; Lany & Saffran, 2011; Lany & Shoaib, 2020), and that infants’ experience with their native language supports the development of statistical learning ability in the first place (Lany & Shoaib, 2020). We have also shown that developments in language proficiency lead to changes in how 2-to-3-year-olds utilize statistics relevant for segmentation in the service of word learning (Shoaib, Wang, Hay, & Lany, 2018; Lany, Karaman, & Hay, under revision).
Moreover, my work has shown that statistical learning ability is related not just to vocabulary size, but is also related to real-time language processing. Attaining a reasonable understanding of speech in real-time is a major developmental achievement in its own right, and infants who are best at real-time language processing also have more accelerated language-learning trajectories. I showed that learning statistical regularities promotes real-time language comprehension by making speech more predictable. By 15 months of age, infants who perform better on two classic statistical-learning tasks, one testing the ability to use statistical regularities to find words in fluent speech, and the other testing the ability to learn simple local dependencies, are also faster to comprehend simple sentences in their native language (Lany et al., 2018). Furthermore, infants with better real-time comprehension are better able to learn the meanings of novel words presented in English sentence frames, especially when they need to use statistical structure within those sentences to do so (Lany, 2018,). These findings clearly reveal a powerful synergy between statistical learning and real-time processing, and suggest that individual differences in statistical learning ability matter for attaining well-recognized language milestones.
In a new line of work, I am investigating the role of cross-modal statistical regularities generated by caregivers as they talk to very young infants in early language development. Specifically, caregivers often shake or loom an object towards an infant as they say its name. I recently showed that that such word-object synchrony influences object encoding (Lany et al., 2022a), and can lead an auditory signal to affect object categorization. These special effects of words on object processing are likely to be a precursor to word learning itself, and they may contribute to unifying word and referents in symbolic relations (Lany et al., 2022b).
In a new line of work, I am investigating the role of cross-modal statistical regularities generated by caregivers as they talk to very young infants in early language development. Specifically, caregivers often shake or loom an object towards an infant as they say its name. I recently showed that that such word-object synchrony influences object encoding (Lany et al., 2002a, b), and can lead an auditory signal to affect object categorization. These special effects of words on object processing are likely to be a precursor to word learning itself, and they may contribute to unifying word and referents in symbolic relations.
Research grants
The Origins of Infant Word Learning
LEVERHULME TRUST (UK)
March 2024 - February 2028
CAREER: Discovering the Underpinnings of Statistical Language Learning in Infants
NATIONAL SCIENCE FOUNDATION (USA)
March 2014 - February 2020