Following the trends established in psychology and emerging in L2 research, we explain our support for an Open Science approach in this paper (i.e., developing, analyzing and sharing datasets) as a way to answer controversial and complex questions in applied linguistics. We illustrate this with a focus on a frequently debated question, what underlies individual differences in the dynamic system of post-pubertal L2 speech learning? We provide a detailed description of our dataset which consists of spontaneous speech samples, elicited from 110 late L2 speakers in the UK with diverse linguistic, experiential and sociopsychological backgrounds, rated by ten L1 English listeners for comprehensibility and nativelikeness. We explain how we examined the source of individual differences by linking different levels of L2 speech performance to a range of learner-extrinsic and intrinsic variables related to first language backgrounds, age, experience, motivation, awareness, and attitudes using a series of factor and Bayesian mixed-effects ordinal regression analyses. We conclude with a range of suggestions for the fields of applied linguistics and SLA, including the use of Bayesian methods in analyzing multivariate, multifactorial data of this kind, and advocating for publicly available datasets. In keeping with recommendations for increasing openness of the field, we invite readers to rethink and redo our analyses and interpretations from multiple angles by making our dataset and coding publicly available as part of our 40th anniversary ARAL article.

Recently, scholars have begun to explore the hypothesis that individual differences in domain-general auditory perception, which has been identified as an anchor of L1 acquisition, could explain some variance in postpubertal L2 learners’ segmental and suprasegmental learning in immersive settings. The current study set out to examine the generalizability of the topic to the acquisition of higher-level linguistic production skills—that is the appropriate use of diverse, rich, and abstract vocabulary. The speech of 100 Polish-English bilinguals was elicited using an interview task, submitted to corpus-/rater-based linguistic analyses, and linked to their ability to discriminate sounds based on individual acoustic dimensions (pitch, duration, and amplitude). According to the results, those who attained more advanced L2 lexical proficiency demonstrated not only more relevant experience (extensive immersion and earlier age of arrival), but also more precise auditory perception ability.

Whereas many scholars have emphasized the relative importance of comprehensibility as an ecologically valid goal for L2 speech training, testing, and development, eliciting listeners’ judgments is time-consuming. Following calls for research on more efficient L2 speech rating methods in applied linguistics, and growing attention toward using machine learning on spontaneous unscripted speech in speech engineering, the current study examined the possibility of establishing quick and reliable automated comprehensibility assessments. Orchestrating a set of phonological (maximum posterior probabilities and gaps between L1 and L2 speech), prosodic (pitch and intensity variation), and temporal measures (articulation rate, pause frequency), the regression model significantly predicted how naïve listeners intuitively judged low, mid, high, and nativelike comprehensibility among 100 L1 and L2 speakers’ picture descriptions. The strength of the correlation (r = .823 for machine vs. human ratings) was comparable to naïve listeners’ interrater agreement (r = .760 for humans vs. humans). The findings were successfully replicated when the model was applied to a new dataset of 45 L1 and L2 speakers (r = .827) and tested under a more freely constructed interview task condition (r = .809).