In this paper we provide a framework for shallow dialog annotations (SDA), and for their use in the context of the processing and retrieval of multimodal meeting recordings. The SDA model groups the following elements: dialog segmentation into utterances and episodes, detection of dialog acts and adjacency pairs, and detection of referring expressions and coreference links, including references to documents. An instantiated XML annotation model based on boundaries, labels and links, is provided. The use of SDA data in a meeting retrieval interface is also described.