Recognizing Article Errors Based on the Three Head Words

In this paper, we present a statistical model for detecting article errors, which Japanese learners of English often make in English writing. The model detects article errors based on probabilities estimated looking at the three head words the head verb, the preposition, and the head noun in a corpus. Unfortunately, simply looking at the three head words causes the data sparseness problem. To abate the problem, we apply the backed-off estimate to estimate the probabilities. Experiments show that the performance (F-measure=0.70) of the model is better than that of other methods. Apart from the performance, the model has two advantages: (i) Rules for detecting article errors are automatically generated as conditional probabilities once a corpus is given. (ii) Recall and precision rates of the model are adjustable.