In Mandarin, word meanings are differentiated by tones. Therefore, when a Mandarin song is sung according to its musical melody, word meanings could potentially be misunderstood. In this research, we intend to investigate whether or not a singer would adjust the pitch contour so as to best convey word meanings. A Mandarin singing dataset is currently being manually parsed into single words, phonetics of which are manually transcribed (including the tones), and for each word the pitch contour is calculated by the YIN algorithm. Afterwards, the distance between arbitrary pairs of contours can be calculated by a dynamic time warping-based method. By comparing average same-tone distances with the distances calculated without distinguishing the tones, one can measure the extent to which a singer modifies his/her pitch inflection, consciously or not, according to the actual tone of the word. Some mixed results are reported.