Autonomic communication in a computing system analyzes the individual system element as it is affected by and affects other elements. In the human-machine communication aspect, the autonomic system performs its services autonomously, adjusting the behaviour of its services to suit what the user might request implicitly or explicitly. The pervasive multimodal multimedia (MM) computing system aims at realizing anytime, anywhere computing. Its autonomic communication includes the protocols that selects, on behalf of the user, the modalities and media devices that are appropriate to the given interaction context (IC). The modalities are the modes of interaction (i.e. for data input and output) between the user and the computer while the media devices are the physical devices that are used to support the chosen modalities. IC, itself, is the combined user, environment, and system contexts. In this paper, we present the autonomic communication protocols involved in the detection of IC and the MM computing system’s corresponding adaptation. The technical challenges involved in formulating this system’s infrastructure include, among others: (1) the establishment of relationship between IC and its suitable modalities, and the quantification of this suitability, (2) the classification of media devices and its relationship to modality, (3) the modeling of IC and its incremental definition, and (4) the establishment of the system’s fault-tolerance mechanism concerning failed or missing media device. The heart of this paradigm’s design is the machine learning’s knowledge acquisition and the use of the layered virtual machine for definition and detection of IC.