Sounds like ah, aha, aw, eh, er, erm, hmm, huh, mm, oh, ooh, oops, phew, uh, uh-huh, uh-uh, um and yup are ubiquitous in informal spoken English. In human dialogs, these sounds are important in conversation control and for conveying attitudes. Spoken dialog systems make use of these sounds to achieve various effects such as concern, dissatisfaction, disfluency, etc., which are unavailable for single lexical sound. Ward (2005) has labeled all sounds which are not laughter and not words as non-lexical items. His study has initiated a series of systematic studies of non-lexical utterances, and has provided us with a fresh quarry of future study on the topic. In this thesis, the author first presents a critical review of the study of non-lexical utterances carried out by Professor Nigel Ward (2005). Possible limitations in issues like corpus construction, term definitions, categorizational standards and meaning measurement in Ward’s work are examined and discussed. Based on this critical analysis, the present author attempts to redefine non-lexical utterances and to distinguish them from semi-lexical utterances, which have traditionally been classified as discourse markers and studied as exclamations and interjections.With the literature review and theoretical discussions of the relevant issues in the field, a 263,173-word corpus is constructed for the study of semi-lexical utterance oh, and data for analysis come from the text file of the subtitles of a set of popular American TV series. The empirical study is mainly based on the statistics made by the Concordance and ConCapp software, while the WASP software is also utilized in the analytic process of the aural stimuli and this method may serve as a basic platform for future researchers.Based on an analysis of the data, collocations of semi-lexical utterance oh are classified into five types in an order of occurrence. Distinctive paralinguistic features including pitch, loudness and rate are also examined in the interpretation of oh in specific context. In contrast to the compositional hypotheses put forward by Ward (2005), a cognitive contextual hypothesis is proposed in the interpretation of semi-lexical utterance, which holds that the meaning of semi-lexical utterance is context-dependent, and it should be identified through a cognitive approach. The findings of the study have important research and pedagogical implications for the study of lexicography, and strategies in spoken English.