Automatic Preposition Errors Correction Using Inductive Learning

In this paper, we describe a system for correcting English preposition errors automatically. Non-native English writers often make these errors. Our system uses rules extracted automatically based on preposition context features, such as preceding and following nouns. Additional rules are generated recursively from the extracted rules using inductive learning. Our system achieves 82% accuracy and 32% coverage, which are competitive with other systems. Apart from the performance, it has an advantage of being more understandable while investigating why a given preposition was erroneous. This is because we use rules and they give this advantage over maximum entropy approaches.