Welcome to the DEL Lab!

child draw a picture of brain
 

 

We study Development, Education, and Learning. 

We aim to advance theories of human cognition, inform education, and improve child development.

 

 

Our Team

What we do

Join us

The structures in the input learning data

The input learning data in the environment presents many structures; human learning can be characterized as the acquisition of these structures. These structures are learnable based on experiential learning and repeated exposures to them, including the input from social partners (e.g., parents, teachers, peers). Our past research has examined how domains—such as multi-digit number symbols and maps—with overlapping relational structures are learned by young children during casual play activities with a social partner. Our future plans include delineating the structures in language input—using data from naturalist setting and machine learning methods—and specifying how they shape language acquisition. Understanding the structure in the input learning data and their effects on learning are critical for addressing inequalities in the early learning environment and for creating healthy learning environments for all children. 

Relevant publications: 

Intake to the learning system revealed by eye-tracking

Input is not the same as intake; to understand what information is picked up by learners, we have to know where they attend to and the sequence through which they do so.  Eye-tracking allows us to study the source of intake data and the temporal dynamics through which they are entered, processed, and integrated into learners’ memory. Our past research has utilized both stationary and head-mounted eye-trackers to examine the processes underlying adults’ and children’s visual attention. Current projects use head-mounted eye-trackers to study the processes through which preschool and kindergarten children attend, represent, and learn from formal instruction on place value (the above video clip shows an example). One of the ultimate goals is to advance a process-driven understanding of effective learning and education.

Relevant publications: 

Bridging early learning and later achievement

Learning and cognition build in time, with earlier experiences preparing and setting the stage for later learning. Thus, early learning environments may create hidden competencies and hidden deficits that have far-reaching consequences for later formal learning and success in life. We seek to characterize early everyday learning environments and to uncover the bridge between this early learning and later achievements. Going beyond correlations, we want to understand the developmental pathways and mechanisms that lead to successful and unsuccessful learning. We have discovered, for example, that prior to formal schooling, many children are building piecemeal knowledge about the structures of multi-digit number names and their written forms, knowledge that may play a causal role in the acquisition of place value and calculation skills later in school. 

Relevant publications: 

Machine Learning as simulations of learning mechanisms

Many theories in cognitive development are based on broad, verbal descriptions, leaving the details of learning mechanisms unspecified. It is hard to verify and compare these mechanisms because of the lack of specificity. Computational models offer one solution to this problem by clearly specifying the process of learning. We use machine learning models 1) as a pattern extraction device to study the structure in the input learning data, 2) to simulate proposed learning mechanisms for human learning, 3) test different theoretical predictions, and 4) test the efficacies of different instructional methods. Our past research has examined the structure in children’s early multi-digit number knowledge and has shown how a deep learning model (LSTM) may capture children’s learning of multi-digit number symbols. Current and future projects aim to further examine how different statistical structures of the input data impact learning, the sources of cross-cultural differences in early knowledge, and how the arrangement of teaching materials impacts learning.

Relevant publications: 

Mastering human-invented symbol systems

Much of human learning is about acquiring and mastering a variety of human-invented symbol systems, from spoken language to written text, number, equations, to computer icons, graphs, gestures, even cultural conventions and practices. Most of our research is dedicated to understanding how children (and adults) learn these symbol systems and become competent users of them. Our past research has studied various cognitive processes—e.g., attention, perceptual, memory, action—that support the acquisition of these symbol systems. But these systems serve more than communicative functions, they are tools for thoughts—transforming the ways that we perceive, reason, and act upon the world. Our future projects aim to further examine this relationship, particularly with respect to how the process of acquiring symbol systems in fact trains and entrenches various cognitive processes. 

Relevant publications: 

Translational studies in schools

Well-controlled experiments in the laboratory are critical for examining causal mechanisms of human learning. But to test their scalability and potential interactions with other cognitive processes during learning, laboratory research must connect to and be informed by learning in real life. In collaboration with researchers from human development, learning science, and cognitive science, we are able to work directly with schools and teachers in addressing students’ difficulties in learning. We are looking for more opportunities to work with educational organizations (e.g., schools, daycares, museums). Our larger goal is to understand learning and cognition in the wild—where children succeed or fail—and to use experimental methods (including clean laboratory experiments) and computational models to better understand in-the-world learning.

Relevant publications: 

  • Yuan, L., Prather, R., Mix, K.S., & Smith, L.B. (under review). It's not symbol-grounding; it's relational mapping: when and how physical representations benefit the learning of symbolic numbers.
  • Yuan, L., Johns, E., Mix, K. S., & Smith, L. B. (in prep). Multiple pathways to place value knowledge but prior symbol knowledge matters.