Non-intrusive error-correction of text input chords: a language model approach

An error correction strategy that can be applied to existing chord based text entry systems is proposed. The strategy is capable of correcting 99.1% of all single bit errors (insertions and deletions) and 98.1% of certain double bit errors (substitutions) occurring in words. The strategy compares the entered words against entries in a reference wordlist. A genetic algorithm is used to search for the close-to-optimal chord-to-character mapping in terms of single bit errors, certain double bit errors and mean fingers per chord (effort). The paper also explores the magnitude of the error-correcting degradation that occurs with multiple bit errors per word. The chord-to-character mappings proposed is capable of correcting 0.6% more single bit errors than the classic micro writer design, and one design is presented that result in about 17% fewer finger movements.