Automatic Spelling Diagnostic

• Applicants: Professorin Kay Margarethe Berkling, Ph.D.; Professorin Dr. Johanna Fay; Dr.-Ing. Sebastian Stüker; Professor Dr. Alexander Waibel
• Subject Area: General and Comparative Linguistics, Experimental Linguistics, Typology, Non-European Languages
• Term: from 2013 to 2015
• Project identifier: Deutsche Forschungsgemeinschaft (DFG) - Project number 223983369

The goal of this research project is the development of an instrument that allows for the automatic analysis of spelling errors in freely composed texts, that are available in digital form. For this purpose this project is a collaboration of researchers from the field of language didactics and automatic natural language processing.The innovative approach pursued in this project is based on the analysis of the spelling of students that takes into account the pronunciation of the spelling in addition to the normally used, misspelled grapheme sequence. The approach is based on the assumption that students, that learn to write, work closely with the pronunciation of the word sequence that they want to spell.Tools and techniques from the area of automatic speech recognition and speech synthesis allow to compare the text achieved by the students with an orthographically correct version of the text, that was automatically reconstructed with the help of confusion probabilities and language model probabilities. Spelling errors can then be automatically annotated and classified. This process is based on the classification of spelling errors by Fay that has been specifically developed for the use with freely composed text. Our tool will be evaluated with respect to commonly used quality measures.The interdisciplinary approach to our research opens new possibilities in the field of spelling didactics:- For research in the field of spelling, our instrument, for the first time, allows to analyze very large amounts of text corpora from students.- For the individual diagnostic, while teaching spelling, our instruments offers the pre-requisites for a continuous, individual error profile of the students, from which individual training measures can be derived.

• DFG Programme: Research Grants