In this thesis, an interactional perspective will be taken as the main spirit to explore the functionality of conversational CS. The research will be based on the TV discussion programmes collected from Taiwan. It is found that the loci of CS are tightly correlated with its surrounding contexts such as discourse, code and participants to arrive at locally-situated meanings. Based on such findings, I have proposed three types of CS in terms of its interaction with the contextual configurations: discourse-related CS, code-related CS and participant-related CS. Four basic functions of CS can be inferred from the above findings; they are amplification, contrasting, shifting, framing and differentiation. It is argued that these functions may derive from the features of CS itself: the contrastive codes, the act of switching and the act of marking. An attempt is also made to re-examine intrasentential CS from functional and facilitative perspectives. It is argued that CS is employed by speakers to highlight the implicational meanings of an utterance, a global connection to negate with a prior discussion, or the underlying contradiction between social and self expectations. Besides, the switch sites where CS occurs within a sentence are actually very- flexible, depending on the purposes and communicative effects intended to achieve in each interactive exchange. Reiteration, lexical triggering and force of contrast can best account for such a facilitation process. CS may be creatively deployed by speakers to highlight the pragmatic function/meaning of dui/tioh, tags and metalanguaging phrases, which meanwhile weaken their inherent referential meanings. From the way how CS arrives at locally-situated meanings in conversational exchanges, speakers' intentionality of language alternation can thus be detected. By means of CS, speakers intend to contextualise the upcoming speech activity by relating the current talk to the prior talk or knowledge; in so doing, the interpretations of an utterance or a stretch of talk can thus be constrained and ambiguity can also be avoided. Speakers also intend to solve a potential or an emergent problem caused by either turn-taking rules or personal confrontation, to signal the marginality of metacognitive activities from the main discourse, or to enhance communicative efficiency in internal structuring such as narrative, argumentation and side remarks. Hearers tend to display their willingness for cooperation and participation in the proceeding talk by complying with the code choice made by the current speaker. At last, by repositioning the role of CS in interaction, a procedural model regarding the production and interpretation of CS is then preliminarily proposed.