Abstract: Using a spelling correction model based on keyboard distance and dependency parsing, the nonword errors generated in the process of text input were solved. The final spelling correction model was constructed by considering three parts: the proximity weight, the dependency weight and the word frequency. The model was validated in the Brown corpus, the Gutenberg corpus and the Inaugural corpus. The experimental results show that the model can be effective for nonword error correction.

Key words: keyboard distance, word distance, dependency parsing,  , SpellKD model, nonword error