This thesis studies learning mechanisms and cognitive biases present from birth involved in language acquisition, in particular in speech segmentation and the extraction of linguistic regularities. Due to the sequential nature of speech, uncovering language structure is closely related with how infants segment speech. We investigated infant abilities to track distributional properties on the stimuli, and the role of prosodic cues and of memory constraints. In two experiments we investigated neonates’ capacities to segment and extract words from continuous speech by using fNIRS. Experiment 1 demonstrates that neonates can segment and extract words from continuous speech based on distributional cues alone; whereas Experiment 2 shows that newborns can extract words when they are marked only by prosodic contours. Additionally we implemented a method for the study of the dynamics of the functional connectivity of the neonatal brain during speech segmentation tasks. We identified stable and reproducible functional networks with small-world properties that were task independent. Moreover, we observed periods of high global and low global connectivity, which remarkably, were task dependent, with stronger values when neonates listen to speech with structure. In another set of experiments we studied memory constraints on the encoding of six-syllabic words in newborns using fNIRS. Experiment 4 demonstrates that the edge syllables of a sequence are better encoded, and Experiment 5 goes beyond by showing that a subtle pause enhances the encoding of intermediate syllables, which evidences the role of prosodic cues in speech processing. A final group of experiments explore how information is encoded when it is presented continuously across different modalities; specifically if an abstract encoding of the sequences’ constituents is generated. Experiments 6-9 suggest that adults form an abstract representation of words based on the position of the syllables, but only in the speech modality. In Experiments 10 and 11 we used pupillometry to test the same in 5-month-old infants. Nevertheless results were not conclusive, we did not find evidence of an abstract encoding.
