Audio and text signals reveal learners native languages differently
Acoustic and Lexical Signals for Native Language Identification of Language Learners
Computation and LanguageMachine Learning
Summary
Identifying a person's native language by how they speak a second language can use sounds or words. The authors studied both speech recordings and written transcripts of English language learners to see which works better at different skill levels. They found that listening to the audio works best for beginners, while analyzing the words in transcripts works better for intermediate to advanced learners. The accuracy of text-based identification depends a lot on how much transcript is available, but audio is more stable with less input.
What this means in practice
- •For language assessment developers: Improve tools to detect learners native language from speech or text, adjusting methods by learner proficiency and available data size.
- •For language learning app creators: Integrate audio and transcript analysis to personalize feedback based on native language clues across proficiency levels.
Authors
Ahmet Yavuz Uluslu, Mengjie Qian, Kate Knill, Mark Gales
Abstract
Spoken native language (L1) identification for second language learners has traditionally relied on acoustic attributes of speech. Comparatively little attention has been paid to analysing speech transcripts for L1 traces. In this paper, we present a study of L1 identification using both speech and corresponding transcripts from a spoken learner English corpus. We compare a WavLM-based audio L1 classifier with a zero-shot GPT-4o classifier operating on transcripts and analyse their predictions across proficiency levels using a CEFR-balanced evaluation set. Our experiments show that the audio classifier outperforms the transcript-based approach overall when full transcripts are available, although their relative strengths vary with learner proficiency. Audio performs better at A1, whereas text performs better from A2 to C1 with full transcripts. Transcript-based accuracy generally increases with proficiency, while acoustic performance remains comparatively stable. Restricting transcript length substantially reduces text-based accuracy, whereas audio is less sensitive to input length and performs relatively well with limited input. These findings provide insights into L1 identification from acoustic and textual representations of learner speech and how performance varies with proficiency and the amount of available language material.