--- title: "The ISLE Corpus of Non-Native Spoken English — Menzel 2000" description: "Building pronunciation tutors that give learners specific corrective feedback requires recorded learner… 5 references, cited by 11 papers here." authors: ["Wolfgang Menzel", "Eric Atwell", "Patrizia Bonaventura", "Daniel Herron", "Peter Howarth", "Rachel Morton", "Clive Souter"] year: 2000 venue: "Proceedings of the Second International Conference on Language Resources and Evaluation (LREC'00)" source: "https://aclanthology.org/L00-1234/" licence: "CC BY-NC 4.0 (ELRA, LREC proceedings)" licence_url: "https://creativecommons.org/licenses/by-nc/4.0/" summary_licence: "Elamur Cosmic License" canonical: "https://research.elamur.ai/free/en/papers/menzel-2000-isle-corpus-non-native-spoken/" --- [Elamur Research](https://md.research.elamur.ai/free/en/) · [this page in HTML](https://research.elamur.ai/free/en/papers/menzel-2000-isle-corpus-non-native-spoken/) [home](https://md.research.elamur.ai/free/en/) / [papers](https://md.research.elamur.ai/free/en/papers/) / [2000](https://md.research.elamur.ai/free/en/papers/2000/) / The ISLE Corpus of Non-Native Spoken English # The ISLE Corpus of Non-Native Spoken English **Authors:** [Wolfgang Menzel](https://md.research.elamur.ai/free/en/authors/wolfgang-menzel/), [Eric Atwell](https://md.research.elamur.ai/free/en/authors/eric-atwell/), Patrizia Bonaventura, Daniel Herron, Peter Howarth, Rachel Morton and [Clive Souter](https://md.research.elamur.ai/free/en/authors/clive-souter/) · **Year:** 2000 · **Venue:** [Proceedings of the Second International Conference on Language Resources and Evaluation (LREC'00)](https://md.research.elamur.ai/free/en/venues/lrec/) ([LREC 2000](https://md.research.elamur.ai/free/en/venues/lrec/2000/)) · **Publisher:** [European Language Resources Association (ELRA)](https://md.research.elamur.ai/free/en/publishers/european-language-resources-association-elra/) · **Licence:** [CC BY-NC 4.0 (ELRA, LREC proceedings)](https://creativecommons.org/licenses/by-nc/4.0/) **Links:** [Publisher's page](https://aclanthology.org/L00-1234/) · [PDF](https://aclanthology.org/L00-1234.pdf) · [OpenAlex](https://openalex.org/W1753200545) **Citations:** 52 citations (OpenAlex, 2026-10-14) · [11](https://md.research.elamur.ai/free/en/papers/menzel-2000-isle-corpus-non-native-spoken/cited-by/) in our library, of the 149,439 papers whose bibliographies we read ## Summary Building pronunciation tutors that give learners specific corrective feedback requires recorded learner speech whose errors are marked in detail. Few such resources existed. The ISLE project collected about 18 hours of read English from 23 Italian and 23 German intermediate learners. The material combines an autobiographical Everest text, used to test speaker adaptation, with 164 short phrases written around known trouble spots: problem phones, weak forms, tricky stress and consonant clusters. The annotation pipeline was partly automatic. A British English HMM recogniser produced word alignments, and canonical phones and primary stress came from dictionary look-up. Trained linguists then edited only the places where the speaker departed from that reference, marking phone deletions, insertions and substitutions plus stress shifts. Statistics from the corpus confirmed that learners usually replace difficult English sounds with the closest native ones or drop them, and that short function words produce many errors. The corpus served to train and evaluate error localisation, phone-error diagnosis and stress detection. It is distributed via ELDA. ## Key points - **Speakers and recordings:** 46 speakers (23 Italian, 23 German), with a few extra non-native and native British speakers for comparison. Recordings are 16 kHz, 16-bit WAV, about 18 hours in total, drawn from 250 utterances. - **Material in seven blocks (A–G):** - Blocks A–C: roughly 1300 words (82 sentences) of the Everest text, used only for adaptation and given only word-level error notes. - Blocks D–G: minimal pairs, item selection, reading and description exercises with a target perplexity of 6–10, fully phone- and stress-annotated. - **Annotation format and effort:** - Labels use Entropic's UK English phone set rather than IPA. - Deletion is written as `0`, insertion as a hyphen attached to a neighbouring label, substitution as a plain label swap. - Six annotators needed about 5–6 hours per speaker, roughly 300 hours overall. - **Two error measures:** the paper ranks phones and words by error sensitivity (how often the unit is wrong) and by error productivity (its share of all errors). - For Italians, `uh` was wrong 47.2% of the time, usually realised as `uw`. - Schwa insertion made up about 6.7% of all Italian errors. - **Error types and cost:** - Stress errors largely matched predictions. - Word-level errors were unsystematic. - Phone errors mixed L1 interference with idiosyncratic slips. - Annotation cost was cited at up to one dollar per phone. ## Interesting ideas - Annotators did not transcribe from scratch; they corrected only where reality differed from an automatically generated canonical reference, which cut effort and made the deviations explicit. - Ranking units by both sensitivity and productivity separates rare-but-fragile items from frequent items that dominate the error mass. - Annotators were told never to move unit boundaries, even wrong ones, so the word, phone and stress tiers stayed aligned; boundary problems were kept apart from label problems. - Annotators were asked to use a coarse inventory that matched the recogniser rather than the finest one they could produce, keeping the labels usable by the system, with an escape hatch for clear L1 sounds. - Exercises were designed as low-perplexity, constrained-choice tasks so that recognition stays reliable enough for diagnosis. ## For Mova How these ideas could serve Mova, our small open language model — its files, stages and tests: - **I1** (learn): Treat the agreed silver tree from witness parsers as the canonical layer and have LLM teacher tiers or humans record only edits (relabel, reattach), so each correction becomes a reusable repair-rule example. - **I2** (learn): Report parser errors per deprel and per lemma under both measures, so effort goes to function words or relations that cause most failures, not only to those with the worst rate. - **I3** (tokens): When repairing tags or lemmas, freeze token boundaries and log tokenization faults on a separate channel, so downstream UD and frame layers keep stable indices. - **I4** (tree): Normalise treebanks from different UD dialects onto one coarse label set that Mova actually consumes, while keeping an "outside inventory" tag for constructions that don't map cleanly. - **I5** (answer): For multiple-choice school tests, restrict interpretation to the given options and check each against the text world, which keeps the answer path short and explainable. *Our summary of the paper, in our own words — not its text. The work belongs to its authors; read it at the links above.* ## Institutions - [University of Leeds](https://md.research.elamur.ai/free/en/universities/university-of-leeds/) — University, Leeds, United Kingdom · 42 works · 1,823 citations - [Universität Hamburg](https://md.research.elamur.ai/free/en/universities/universitat-hamburg/) — University, Hamburg, Germany · 81 works · 5,549 citations 📚 Shelves (6): [🖥️ Computer Science](https://md.research.elamur.ai/free/en/shelves/computer-science/) · 18,614 works · [🤖 Artificial Intelligence](https://md.research.elamur.ai/free/en/shelves/computer-science/artificial-intelligence/) · 17,965 works · [🎭 Arts and Humanities](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/) · 17,450 works · [🗣️ Language and Linguistics](https://md.research.elamur.ai/free/en/shelves/arts-and-humanities/language-and-linguistics/) · 17,296 works · [💭 Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/) · 1,649 works · [💭 Experimental and Cognitive Psychology](https://md.research.elamur.ai/free/en/shelves/psychology/experimental-and-cognitive-psychology/) · 930 works 🧭 Topics (1): [English](https://md.research.elamur.ai/free/en/topics/english/) · 2,379 works ## Cite Cite as: Wolfgang Menzel, Eric Atwell, Patrizia Bonaventura, Daniel Herron, Peter Howarth, Rachel Morton and Clive Souter. 2000. The ISLE Corpus of Non-Native Spoken English. Proceedings of the Second International Conference on Language Resources and Evaluation (LREC'00). The paper itself, not this page. BibTeX from the library's record of the work. ```bibtex @inproceedings{menzel2000isle, title = {{The ISLE Corpus of Non-Native Spoken English}}, author = {Wolfgang Menzel, Eric Atwell, Patrizia Bonaventura, Daniel Herron, Peter Howarth, Rachel Morton and Clive Souter}, booktitle = {Proceedings of the Second International Conference on Language Resources and Evaluation (LREC'00)}, year = {2000}, doi = {10.63317/2h2y5rayo2bb}, url = {https://doi.org/10.63317/2h2y5rayo2bb} } ``` - [← previous Experiments on Sentence Boundary Detection](https://md.research.elamur.ai/free/en/papers/stevenson-2000-experiments-sentence-boundary-detection/) - [next → A Unified Statistical Model for the Identification of English BaseNP](https://md.research.elamur.ai/free/en/papers/xun-2000-unified-statistical-model-identification-english/) - [↑ back to papers of 2000](https://md.research.elamur.ai/free/en/papers/2000/) ## References 5 references, 0 of them with a card of ours. ↗ leads to the work, «find» searches for one we have no id for yet. 1. de Cock, S., Granger, S. , Petch-Tyson, S. 1997. The Louvain International Database of Spoken English Interlanguage (LINDSEI) Project, rapport interne, Centre for English Corpus Linguistics, Université catholique de Louvain, Ehzani, F. Knodt, E. 1998. Speech Technology in Computer-Aided Language Learning: Strengths and Limitations of a New CALL Paradigm. in: Language Learning & Technology, vol. 2, no. 1, p. 45-60. [find](https://scholar.google.com/scholar?q=de+Cock%2C+S%2E%2C+Granger%2C+S%2E+%2C+Petch%2DTyson%2C+S%2E+1997%2E+The+Louvain+International+Database+of+Spoken+English+Interlanguage+%28LINDSEI%29+Project%2C+rapport+interne%2C+Centre+for+English+Corpus+Linguistics%2C+Universit) 2. Eisen, B., Tillmann, H. G., Draxler, Ch. 1992. Consistency of Judgements in Manual Labelling of Phonetic Segments: The Distinction between Clear and Unclear Cases, Proc. Int. Conf. On Spoken Language Processing ICSLP’92, p. 871-874. [find](https://scholar.google.com/scholar?q=Eisen%2C+B%2E%2C+Tillmann%2C+H%2E+G%2E%2C+Draxler%2C+Ch%2E+1992%2E+Consistency+of+Judgements+in+Manual+Labelling+of+Phonetic+Segments%3A+The+Distinction+between+Clear+and+Unclear+Cases%2C+Proc%2E+Int%2E+Conf%2E+On+Spoken+Language+) 3. Herron, D., Menzel, W., Atwell, E., Bisiani, R., Daneluzzi, F., Morton, R., Schmidt, J. A. (1999) Automatic Localization and Diagnosis of Pronunciation Errors for Second Language Learners of English. Proc. 6th European Conference on Speech Communication and Technology, Eurospeech '99, Budapest, vol. 2, p.855-858. [find](https://scholar.google.com/scholar?q=Herron%2C+D%2E%2C+Menzel%2C+W%2E%2C+Atwell%2C+E%2E%2C+Bisiani%2C+R%2E%2C+Daneluzzi%2C+F%2E%2C+Morton%2C+R%2E%2C+Schmidt%2C+J%2E+A%2E+%281999%29+Automatic+Localization+and+Diagnosis+of+Pronunciation+Errors+for+Second+Language+Learners+of+English%2E+) 4. Hunt, J. 1996. The Ascent of Everest. Stuttgart: Ernst Klett Verlag, English Readers Series Young, S., Kershaw, D. Odell, J., Ollason, D., Valtchev, V., Woodland, P. 1999. The HTK Book 2.2, Entropic, Cambridge. [find](https://scholar.google.com/scholar?q=Hunt%2C+J%2E+1996%2E+The+Ascent+of+Everest%2E+Stuttgart%3A+Ernst+Klett+Verlag%2C+English+Readers+Series+Young%2C+S%2E%2C+Kershaw%2C+D%2E+Odell%2C+J%2E%2C+Ollason%2C+D%2E%2C+Valtchev%2C+V%2E%2C+Woodland%2C+P%2E+1999%2E+The+HTK+Book+2%2E2%2C+Entropic%2C+) 5. Power, K., Morton, R., Matheson, C., Ollason, D. 1996. The graphvite Book 1.1, Entropic, Cambridge. [find](https://scholar.google.com/scholar?q=Power%2C+K%2E%2C+Morton%2C+R%2E%2C+Matheson%2C+C%2E%2C+Ollason%2C+D%2E+1996%2E+The+graphvite+Book+1%2E1%2C+Entropic%2C+Cambridge%2E) *Updated 2 November 2026*