Various techniques, such as acoustic model adaptation and pronunciation adaptation, have been reported to improve the recognition of nonnative or accented speech. In this paper, we propose to analyze the regular pairs of the pronunciation variation of the nonnative Mandarin speech spoken by Dai, Lisu and Naxi speakers from Yunnan. According typical pronunciation variations of these 3 accents, the more than one pronunciation for a part of words (i.e. tonal syllables or characters) have been inserted in the standard Mandarin pronunciation dictionary. The experiments show that an improvement is reached with the new dictionary and a simple 2-gram language model for all kinds of nonnative speakers.