Simultaneous Speech Segmentation and Cross-Situational Statistical Learning in Monolinguals, Bilinguals, and Multilinguals

Statistical learning (SL) is a mechanism that learners use to segment words from continuous speech and map them to their correct referents through the computation of co-occurrence probabilities (Saffran et al., 1996a,1996b; Yu & Smith, 2007). So far, most SL studies have investigated segmentation and mapping using separate tasks (Graf-Estes et al., 2007). Therefore, very little is known about how these two processes interact, which is crucial to understand real world language learning. Moreover, nothing is known about how knowledge of an additional language influences this interaction. In this study, we exposed monolinguals, bilinguals, trilinguals, and quadrilinguals to a joint SL task allowing the tracking of both syllable co-occurrences and word-object co-occurrences at the same time. Two familiarization phases were used, during which participants simultaneously heard two phrases of an artificial language and saw two unique objects on the screen across multiple trials. In the test phase, participants were tested on their speech segmentation and mapping abilities. Our results indicated that although all groups showed different learning trajectories and strategies, they all succeeded at learning label words and mapping them to their correct referents after a second exposure. We also found that quadrilinguals outperformed both bilinguals and monolinguals in segmenting label words. Our preliminary findings suggest that, regardless of language experience, learners are capable of overcoming the cognitive load and the complexity of computing simultaneously available segmentation and mapping statistics, provided they are given sufficient exposure. In addition, knowledge of four languages could enhance the ability to detect word boundaries using statistical information