Natural behavioral patterns of speech recognition error recovery

Due to faster entry speed and lower attention demand, speech recognition is widely used in the multimodal text entry system. But error recovery is a big challenge to its usability for the accuracy of speech recognition is not good enough. However, few studies focused on the question how users recovered the errors in a multimodal entry system, especially in realistic scenarios. The present study investigated the natural behavioral patterns of speech recognition error recovery both in laboratory and field conditions. And further to examine whether there was any difference between these conditions. Results indicated that the handwriting modality was more frequently adopted in both laboratory and filed conditions. Similar behavioral patterns were found in these two conditions. And further, the usage of modality to recover error showed obvious individual variety in the laboratory condition, which was classified into three preference types. However, it was not apparently varied among individuals in the field condition. Furthermore, the entry modality had little effect on the choice of recovery modality. Finally, based on the findings in the present study, several applicable implications were provided.