Title: The Emergence of Compositionality: From Implicit to Explicit Representations of Relational Structure
Abstract: A full account of mature adult cognition requires understanding its developmental origins. One fundamental human cognitive capacity is compositionality—the ability to use a limited set of components to generate an unlimited array of compositions through principled recombination. Language, mathematics, chemistry, and music are all examples. A perennial question is how learners, exposed primarily to complex compositions (e.g., sentences composed of words in specific orders; multi-digit numbers composed of digits with different place values), come to learn both the components and the principles that govern their combination (e.g., syntactic structure). Focusing on the crucial domain of multi-digit symbolic numbers, I will illustrate (1) how preschool children develop an initial, implicit grasp of multi-digit numbers by mapping spoken number names to their written forms through statistical learning—a process that can be simulated with general neural network models; (2) how this early implicit understanding lays the groundwork for later explicit comprehension of base-ten principles, as revealed through latent variable modeling and network analysis; and (3) how implicit knowledge alone is insufficient, as shown in multi-day training experiments: acquiring explicit relational understanding is supported by actively aligning relational structures across language, perception, and action through a dynamic, constructive learning process.
Jillian Sifuentes
Email