Research
The way we learn changes through development. As our brains develop, the set of cognitive mechanisms supporting learning grows and we gain different and typically more advanced assets. The main goal of my research is to understand how humans learn and how they represent what they have learned. I study developmental changes in learning and knowledge representation, primarily gaining insights from behavioral experiments with young children and adults, with the aid of computational modelling, neurophysiological measures and work with nonhuman animals.
One may form semantic link between a pear and a quince based on similarity in their features, or based on similarity in language contexts they occur in (see Figure 1). The formation of semantic links between the concepts similar in meaning is of critical importance for conceptual development as it allows for generalization of knowledge about familiar to novel items. My work investigates development of learning mechanisms needed for acquisition of these semantic links from the two main sources of information: 1) regularities in the visual input and 2) the statistical structure of language.
The first line of research looks at how different attentional mechanisms, one available early (distributed attention) and the other available later in development (selective attention), affect what object features we remember and use in future generalization. The second line of research focuses on the development of abilities to form semantic (associative and taxonomic) links from co-occurrence regularities in language.
Visual category learning
To understand how developmental differences in attention optimization affect learning and generalization, I study categories with exceptions. I found that the effect that information that violates our expectations (exceptions) has on generalization, changes with the development of selective attention. Young, 4-year-old learners, struggle to ignore irrelevant features and focus on the relevant ones, which results in their generalization being driven by both, features consistent with the salient regularity and ones that violate it. With the development of selective attention, participants start to generalize only from regular category members (broader concept) to exceptions, but not the vice versa. In other words, with the development of selective attention, the representation of categories with exceptions seems to shift to a hierarchical one, in which exceptions are a sub-set of regular category members (for more details, see Savic & Slotusky, 2019; Savic, Blanco, & Sloutsky, 2018, 2019).
My ongoing and future work in this area will focus on (a) developmental differences in integration of novel information (i.e. When do category boundaries change?), (b) testing limitations of computational models to account for developmental differences and hierarchical representations (collaboration with Nate Blanco and Vladimir Sloutsky), and (c) understanding neural underpinnings of these representations by conducting research with nonhuman animals (collaboration with Vladimir Sloutsky and Ed Wasserman).
Lexico-semantic development fostered by statistical regularities in language
Words tend to co-occur in language in predictable ways. For example, pear and quince reliably co-occur with juicy, whereas happy and sad reliably co-occur with feel. I study how these co-occurrence regularities in language can foster both: (1) early emerging associative links, between the concepts whose labels directly co-occur (juicy - pear), and (2) building on associative, foster later developing taxonomic links, between the concepts whose labels share patterns of co-occurrence (pear and quince both reliably co-occur with juicy, eat, or brandy, see Figure 1).
This line of research resulted in a set of novel paradigms, which we successfully use in the lab to track formation of novel and familiar semantic links from infancy to adulthood. I found substantial evidence that mere exposure to co-occurrence regularities can foster formation of both associative and taxonomic links, but that learning mechanisms supporting formation of taxonomic links (e.g. information integration across the episodes of learning) show protracted development. These patterns parallel my findings on development changes in semantic organization of familiar concepts (see Savic, Savic, & Kovic, 2017; Savic, Unger, & Sloutsky, 2018; Unger, Savic, & Sloutsky, 2020; Savic, Thierry, & Kovic, 2020).
These findings motivated development of two novel research projects I will focus on in the future and offered me an opportunity to apply for grant support in a role of a Co-PI (National Science Foundation - pending application; R01: National Institutes of Health - to be submitted in June 2020). The main goals of these projects are to investigate how development of abilities to learn from statistical regularities in the linguistic input may support semantic development and reading comprehension, as well as whether the developmental changes in learning mechanism can be linked to changes in neural connectivity (collaboration with Vladimir Sloutsky, Layla Unger and Zeynep Saygin).