The Effects of Instruction in Connected Speech Processes on L2 Speech Perception: Evidence from Bayesian Mixed-Effects Models

Connected speech processes (CSPs) are changes in pronunciation that occur when words are spoken in fluent speech, as opposed to being articulated carefully in isolation. For example, the pronunciations of find and out in a sentence like ‘let’s find out’ may approximate fine doubt due to resyllabification. While L1 listeners typically accommodate such phonological alterations with ease, CSPs often pose perceptual challenges for L2 listeners.

Building on a preceding diagnostic study, the present intervention targeted six CSP categories identified as particularly challenging for L2 listeners: Deletion of Onset, Glide Insertion, Reduction of Stop, CC-V Linking, Palatalisation, Regressive Assimilation. 66 Mandarin-speaking university students from Taiwan with upper-intermediate to advanced English proficiency participated in the study. Participants were randomly assigned to one of three instructional conditions: (1) explicit rule-based instruction; (2) inductive noticing-based instruction; (3) a control without direct pronunciation instruction. Speech perception was assessed before and after the intervention using measures of intelligibility (recognition of CSP-affected words) and comprehensibility (perceived ease of understanding).

Data were analysed using Bayesian mixed-effects models and Bayes factors. Overall, participants demonstrated gains in intelligibility but an unexpected decline in comprehensibility following the intervention. Despite the observed main effects of time, there was no robust evidence for Group-by-Time interactions for either measure. Bayes factors further indicated that this pattern reflected data ambiguity rather than support for a null effect. The talk will focus on methodological rationales and provide preliminary data simulation results informed by the study.

Teams link: https://teams.microsoft.com/meet/3551416611792?p=n8sUC57wg70xw24DRq