Expressive interjection (EI) is defined as non-lexical speech sound which indicates the speaker's cognitive/affective state changes. It is a type of vocal affect burst, i.e., brief and sudden nonverbal expressions that are produced spontaneously and unconsciously. Although EI as a social signal is assumed to play an important part in speech communication, very little is known about its linguistic, paralinguistic, and pragmatic nature. The goal of this study is to unveil the structure and functions of vocal affect bursts in human interactions. This paper focuses on the surface structure of EIs, which is an indispensable foundation for further analysis and modeling. Based on linguistic/acoustic analyses of a natural, spontaneous dialog corpus, the distinctiveness of EI was revealed as: (1) less variation in transcribed expression, (2) may have very short duration, and (3) higher F0 and intensity. In addition, it was revealed that an apparent correlation between formant frequencies and perceived paralinguistic information was observed only for EIs with the vowel /a/, which suggests that the vowel /a/ as an EI can accommodate richer paralinguistic information than other vowels.