We present a model for reference resolution that attempts to overcome the standard resolution mechanisms based on co-reference and to account for entities not explicitly mentioned. Both of these characteristics are necessary in order to deal with other-expression in a multimodal corpus of spoken dialogues. Inspired by principles of Cognitive Grammar, the fundamental assumption of the model is that all reference is accomplished via access to and restructuring of domains of reference. The application of this idea to the interpretation of other-expressions leads to several predictions which are validated by a corpus study.