Slide Slam J12 Sandbox Series
Testing Bilinguals for the Known-Word Facilitation Effect in Speech Segmentation
Sejin Lee1, Kristina C. Backer1, Sven L. Mattys2, Heather Bortfeld1; 1University of California, Merced, 2University of York
Introduction: It has been well documented that word knowledge contributes to the segmentation of continuous speech (e.g., McClelland & Elman, 1986; Norris, 1994). Speech segmentation models (i.e., Brent, 1999) demonstrate that identification of known words within a speech stream allows the edges of adjacent new words to be inferred by listeners, thus presenting a means by which new words can be learned without formal instruction about those words. This language learning strategy has been shown to operate in both in adults (Dahan & Brent, 1999; White, Melhorn, & Mattys, 2010) and infants (Bortfeld, et al., 2005; Sandoval & Gómez, 2016). Thus, the nature of the interaction between word knowledge and auditory cues in the incoming signal is the basis of substantial research (see Mattys & Bortfeld, 2016). Recent research pitting top-down (i.e., lexical knowledge) against bottom-up (i.e., statistical cues in the signal) effects demonstrated that prior word knowledge can boost the effect of statistical cues in adult English monolinguals (Palmer et al., 2019). Thus, a known-word facilitation effect aids segmentation of novel artificial speech (see also Poulin-Charronnat et al., 2017). Here, we consider whether bilingual listeners show the same effect. Methods: In the current study, we compared performance by English monolingual (n = 20) and Spanish-English bilingual (n = 22) college students on an artificial speech stream segmentation task. We first documented detailed language background information for all participants. Half of the participants from each language group were familiarized with either a control stream made entirely of novel words (e.g., golatu, pigola, etc.) or a test stream, i.e., the same stream containing occasional instances of a known English word (i.e., “philosophy”). Following the familiarization phase, all participants were tested on their ability to recognize the novel words via a two-alternative forced choice task. In addition to the novel words (strings with 1.00 transitional probabilities across three CV syllables), the recognition task included part-words (i.e., strings straddling portions of novel words) and non-words (i.e., strings not present in the familiarization stream). Results: Using linear mixed-effects modeling, preliminary findings demonstrate that, when collapsing across both monolingual and bilingual groups, participants recognized significantly more novel words (p = .022) from the test stream (M = .75) than from the control stream (M = .61). This known-word facilitation effect was evident in the monolingual group (p = .009) but did not reach significance in the bilingual group (p = .19). Bilingual speakers performed significantly worse overall than monolingual speakers (p = .005; Monolinguals: M = .75; Bilinguals: M = .63). Conclusion: These results confirm the finding that the presence of known words in an otherwise novel and continuous speech stream boosts statistical learning. The reason for the discrepancy between monolinguals and bilinguals remains unclear, but it suggests that language experience affects the utilization of top-down lexical knowledge during artificial speech segmentation. Additional research is needed to delineate the factors and neural mechanisms contributing to the difference between monolinguals and bilinguals in how top-down information impacts performance on artificial speech stream segmentation.