The Development of Melodic Representations at Early Age: Towards a Computational Model.

T his research work concerns the early development of music perception and cognition, particularly, the melodic representations at an early age, utilizing a computational modeling perspective. This way, we propose an extensive review of the existent literature intended to contextualize the problem into both approaches of experimental psychology and computational modeling, and also a conceptualization of a developmental model gathering the information previously collected. To achieve this, the work was divided into more specific steps, such as: (1) Collect relevant experimental studies; (2) Extract data from the experimental studies that can give clues on which behaviors develop at specific times related to specific stimulus in the course of infant development between prenatal and postnatal stages; (3) Build a developmental timetable of auditory processes with the collected data; (4) Use the previous collected information to simulate experimental studies on the appearance of music cognition in the first year of life, using Artificial Neural Networks.

The literature review focused mainly in experimental studies coming from developmental psychology with infants in the first years of life, and then compared them with similar studies done with adults. More over, infants’ lack of cultural exposure and learning (contrarily to that of adults) could provide clues to whether certain features of music cognition are innate or not.

The collected data was analyzed in a developmental perspective that is reflected in a constructed timetable. In this timetable it is possible to observe the evolution of the representations in music cognition.

This data collected was then used to simulate the experimental studies. The focus of the simulations was on classification problems, aiming to reproduce infants’ discrimination tasks. The simulations had a cognitive basis on perception of melodic contour, musical scales and pitch. The simulations of the experiments were done using Artificial Neural Networks, with different representations for the stimuli. These simulations produced significant classification results, validating neural networks as an appropriate methodology for reproducing infants’ discrimination tasks and encouraged further research towards a computational developmental model.