Abstract: Objective To explore the mechanism of speech production in nonfluent aphasia. Methods A series of very familiar idioms were presented to nonfluent aphasic patients using DMDX software on a computer screen under 3 different conditions. Group 1 was asked to read the last 2 words of each idiom without voice prompt. Group 2 was asked to read the last 2 words of each idiom with voice prompt.Group 3 was asked to read the last 2 words with voice prompt under the condition of 2 matching forms. The idioms in group 1 and group 2 were entirely same but the sequences were random. The number of errors was compared. Results The scores of 3 groups were (8.93±8.04),(13.63±6.26), and (11.63±6.77) respectively. That the 3 levels of single factor analysis of variance showed: main effect F(2,28)=10.125, P<0.001 showed that there was a statistically significant difference between 3 groups. That paired t-test between Group 1 and Group 2 was t=3.511, P<0.01 suggested that the scores improved with voice prompt. That paired t-test between Group 1 and Group 3 was t=2.526, P<0.05 suggested that the scores significantly improved with voice prompt. That paired t-test between Group 2 and Group 3 was t=3.325, P<0.01 suggested that Group 2 was better than Group 3 under the condition of voice prompt. Conclusion Voice prompt can improve nonfluent aphasic patients on spoken output capacity for familiar idioms, and the patients may accompany disorder in speech coding for the selective extraction.

Key words: nonfluent aphasia, idioms, speech production