Single-modal spatial representations ask how much of a scene’s spatial structure can be recovered from one modality alone, without recourse to cross-modal correspondence. The question matters because every multimodal system is ultimately built from unimodal encoders, and the spatial content a fused representation can carry is bounded by what each modality already captures on its own.
What separates the modalities is not whether they carry spatial information but how overtly they carry it. Vision presents geometry almost directly, arranging appearance and shape over a spatial grid or the volume a scene occupies. Language and sound hold the same structure only obliquely, the one as symbolic relations among named entities, the other as the physical trace that propagation leaves on a signal. Others still are tied to a particular sensor or task, trading generality for fidelity to the data they were built around. This chapter proceeds along that gradient, from the most explicit geometric encodings of space toward the most implicit, examining each modality on its own terms before the question of how they combine is taken up.