Using Automatic Morphological Tools to Process Data from a Learner Corpus of Hungarian
Péter Durst, Martina Katalin Szabó, Veronica Vincze, János Zsibrita
Hungarian language, Natural language processing, Morphological parsing, Automatic error tagging, Learner corpus
The aim of this article is to show how automatic morphological tools originally used to analyze native speaker data can be applied to process data from a learner corpus of Hungarian. We collected written data from 35 students majoring in Hungarian studies at the University of Zagreb, Croatia. The data were analyzed by magyarlanc, a sentence splitter, morphological analyzer, POS-tagger and dependency parser, which found 667 unknown word forms. We investigated the recommendations made by the Hungarian spellchecker hunspell for these unknown words and the correct forms were manually chosen. It was found that if the first suggestion made by hunspell was automatically accepted, an accuracy score of 82% could be attained. We also introduce our automatic error tagger, which makes use of our annotation scheme developed on the basis of the special characteristics of Hungarian morphology and learner language, and which is able to reliably locate and label morphological errors.
Apples - Journal of Applied Language Studies