Thai Legal Term Correction using Random Forests with Outside-the-sentence Features

We propose a method for finding and correcting misused Thai legal terms in Thai statutory sentences. Our method predicts legal terms using Random Forest classifiers, each of which is optimized for each set of similar legal terms. Each classifier utilizes outsidethe-sentence features, namely, promulgation year, title keywords, and section keywords of statutes, in addition to words adjacent to the targeted legal term. Our experiment shows that our method outperformed not only a Random Forest method without the outside-thesentence features, but also BERT (Bidirectional Encoder Representations from Transformers), a powerful language representation model, in overall accuracy.