Annotating learner texts with article error information is a difficult task. To identify which are the main difficulties for annotators, we carry out an annotation experiment in Spanish texts written by Japanese learners. Two expert and two non-expert raters annotate 300 noun phrases containing a definite, indefinite or zero article. We calculate inter-annotator agreement and analyse the sources of disagreement. We find article usage governed by pragmatic factors causes disagreement the most, while lexico-semantic factors are the most reliable. Finally, a learner corpus sample of 30,000 words is annotated with a revised version of the annotation scheme.
Except where otherwise noted, this item's license is described as Attribution-NonCommercial-NoDerivatives 4.0 Internacional