Slide Slam C18 Sandbox Series
Statistical learning modulation through the variation of stimulus rhythmic structure.
Ireri Gomez Varela1, M. Florencia Assaneo1; 1Instituto de Neurobiología, Universidad Nacional Autónoma de México UNAM.
Statistical learning is the ability to extract regularities present in the environment. In the context of speech, statistical learning is generally regarded as one of the cognitive processes that supports early lexical and morphological rules acquisition. This phenomenon has been widely studied in the past decades by means of the Statistical Word Form Learning (SWFL) protocol, where participants are exposed to a learning phase where four tri-syllabic pseudo-words are presented in a continuous stream and learning is assessed post exposure. One of the stable features of studies employing this protocol is the isochronous nature of the stimulus, i.e., each syllable in the stream has the same duration. Although it has been shown that natural speech possesses some temporal regularity, with a mean syllabic rate across languages between 2 to 8 Hz, it is far from being perfectly rhythmic. In order to assess the ecological validity of SWFL, therefore, it is crucial to explore its robustness against variations in the temporal structure of the stimulus. In this work, we study how disrupting the rhythmic characteristics of the auditory stimulus modulates SWL performance. One of the requirements for detecting variations in SWFL performance is to design stimuli with a high learning rate. Thus, our first goal was to generate and validate 4 different isochronous pseudo-languages optimized to evaluate Spanish speaking participants. To construct the 4 tri-syllabic pseudo-words comprising each pseudo-language, we controlled the syllables' position frequency according to the statistical properties of the Spanish language using the Syllabarium database. For the stimuli synthesis, the two methods typically used in the literature were also compared: (i) generating a co-articulated stream using Mbrola against (ii) concatenating isolated synthesized syllables. Significant perceptual differences were observed, favoring Mbrola, which was then selected to construct the final stimuli set. Finally, we obtained four pseudo-languages with significantly high learning rates -approximately 75% of correct answers. Each of the pseudo-languages was then resynthesized in a non-synchronous version: the duration of each syllable composing the continuous flow of pseudo-words was randomly selected from a flat distribution of syllabic frequencies between 2 and 8 Hz. The main experimental design consists of the presentation of four blocks; each with one of the previously constructed pseudo-language presented either synchronously (all syllables last 250 ms) or non-synchronously. Preliminary data shows learning significantly above chance for both conditions, with an enhancement in the synchronous one. Further work will assess the functional relationship between learning performance and stepwise decrements in the temporal regularities of the stimulus as well as with different values of syllable rate for isochronous stimuli.