Can ASR systems help characterize the acquisitional stage of L2 Polish learners?
Résumé
Following the exponential developments in artificial intelligence for language data analysis, we are interested in the link between the performance of a speech recognition system and learners' acquisitional stage. We explore how metrics for evaluating ASR systems for automatic transcription tasks can provide information for characterizing learners' acquisitional stage (El Ayari & Watorek 2024). Following on from our previous study on French acquisition, this study is conducted on Polish acquisition, and specifically on data from the VILLA project (Dimroth et al. 2013). This corpora document early stages of Polish acquisition under controlled conditions, with beginner learners from different mother tongues: Dutch, English, French, German and Italian. The participants were exposed to fourteen hours of Polish; the tasks performed tracked their acquisitional itineraries.
Polish is a language that is at the same time scarcely taught and little studied from the point of view of second-language acquisition. We cross-reference the results obtained by the automatic transcription system with existing manual transcriptions of audio tasks. Contrary to our first study, in which all learners shared the same first language (LANGSNAP corpus), the analysis of the VILLA corpus enables us to observe and measure the potential impact of the mother tongue on interlanguage (Corder 1980) produced by learners (Watorek 2020), on both lexical and grammatical criteria.