Producing an annotated corpus with automatic spelling correction

• Autores: Michael Flor, Yoko Futagi
• Localización: Twenty years of learner corpus research: looking back, moving ahead / Sylviane Granger (ed. lit.), Gaëtanelle Gilquin (ed. lit.), Fanny Meunier (ed. lit.), 2013, ISBN 978-2-87558-199-0, págs. 139-154
• Idioma: inglés
• Texto completo no disponible (Saber más ...)
• Resumen

  • This paper describes ConSpel, a software system for automatic detection and correction of non-word misspellings. We also present an ongoing research project for constructing an ETS (Educational Testing Service) Spelling Corpus. The corpus consists of essays written by native and non-native speakers of English to the writing prompts of TOEFL® and GRE® tests. Essays are annotated for misspellings by trained annotators, using a semi-automated methodology. An evaluation of the ConSpel system was conducted, using the data from the completed phase of the annotation project. The ConSpel system achieves above 95% accuracy in error detection. The evaluation also indicates that an advanced correction algorithm, which takes into account the local context of misspellings, achieves correction accuracy of 77% and consistently outperforms a baseline context-blind approach.