On the adequacy of L2 pronunciation feedback from automatic speech recognition: A focus on Google Translate
Bibliographic record
Abstract
This study investigates automatic speech recognition (ASR) in Google Translate as a source for L2 pronunciation feedback. To be effective, ASR should transcribe learner errors accurately and perform equally well on male and female voices, avoiding gender bias. We assess Google Translate on three Quebec francophone (QF) segmental errors in English: th-substitution ( think → [t] ink ); h-deletion ( happy → appy ); and h-epenthesis ( ice → [h] ice ). Eight QFs (4F/4M) recorded 120 sentences with and without an error on the final item (e.g., I don’t know who to *tank/thank ). Errors were equally divided between real word output (* tank ) and nonword output (e.g., My sister is afraid of *tunder ). We anticipate real word errors, corresponding to entries in the Google Translate lexicon, will be accurately transcribed, whereas nonwords, by definition absent from the lexicon, should be erroneously matched to similar-sounding real words (i.e., the intended output “thunder”), constituting misleading feedback. Forthcoming data analyses will determine the relative contribution of error type, real/nonword output, and gender to final-word transcription and feedback accuracy. Preliminary findings suggest a hierarchy of accuracy (h-deletion, h-epenthesis ˃th-substitution) specific to real-word output. Indeed, ASR shows a clear inability to flag nonword errors. A gender bias effect is not apparent; in fact, ASR generally transcribed the sentences recorded by females more accurately. Mistranscriptions unrelated to final items have yet to be examined. Our presentation will address the implications of our findings for L2 teachers/learners and for developers seeking to design ASR specifically for L2 uses.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.012 | 0.060 |
| Meta-epidemiology (narrow) | 0.001 | 0.000 |
| Meta-epidemiology (broad) | 0.001 | 0.000 |
| Bibliometrics | 0.002 | 0.001 |
| Science and technology studies | 0.001 | 0.002 |
| Scholarly communication | 0.003 | 0.003 |
| Open science | 0.001 | 0.002 |
| Research integrity | 0.001 | 0.001 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".