The issue of processing and understanding spontaneous speech interests speech engineers and linguists, because spontaneous speech is the most natural and the most frequent form of language used for communication. However, the lack of well-defined databases of spontaneous speech prohibits the scale and the depth of spontaneous speech research to a great extent. By using the methodology of corpus linguistics, this paper shows that linguistic theories can be examined, approved or disproved by quantitative empirical analyses. The studies introduced in this paper contain a pioneering work on corpus-based analyses of syllable contractions in spontaneous Mandarin by construing spontaneous Mandarin database incorporated with linguistic considerations into the spontaneous speech phenomena. Corpus-based results presented in this paper not only confirm characteristics of syllable contraction with respect to Chinese phonology, but also provide new insights into lexical factors that influence the production of syllable contractions.