Prosody and fluency of Finland Swedish as a second language : Investigating global parameters for automated speaking assessment

This study investigates prosody and fluency of Finland Swedish as a second language (L2). The main objective is to investigate global measures of prosody and fluency as predictors of overall oral proficiency, fluency, and pronunciation ratings. We analyzed parameters related to temporal fluency, timing (based on syllable durations), and f0 change from spontaneous speech produced by 30 native and 235 non-native speakers of Finland Swedish representing proficiency levels from beginner to intermediate. We used pairwise comparisons to investigate the differences between native speech (L1) and L2 samples from different proficiency levels. To study the predictability of ratings with acoustic parameters, we fitted a multiple linear regression model for each assessed dimension of L2 skills. The comparison of L1 and L2 samples as well as L2 samples with different proficiency and fluency levels showed clear differences in f0 change and fluency parameters. Standard deviation of syllable durations also showed differences with respect to L2 learners’ fluency level. The results for multiple linear regression models, however, indicate contribution of rate-normalized standard deviation of syllable duration to fluency ratings, alongside traditionally used fluency parameters. As for proficiency ratings, f0 slope complemented fluency parameters in the prediction model. The predictive power of the parameters varied depending on the assessed dimension of L2 skills. This study provides new information on the prosodic features of Finland Swedish as a second language and suggests new research on the assessment of non-dominant varieties of pluricentric languages. The results support previous findings on the importance of speed and pausing measures in predicting oral L2 skills. However, further investigation of language-specific f0 and timing parameters as part of automated or computer-assisted speaking assessment is called for.peerReviewe