KAF is a layered annotation format, based on XML. The annotation is stand-off, meaning that the original source document remains unchanged and is kept read-only. KAF provides annotation layers for basic natural language processing and is open to extensions with other annotation layers needed by specific applications, which may be standardized later on. If a process adds information which cannot be held by existing layers, a layer of annotation is added. Layers may be linked by means of references from one layer to items in another (lower level) layer. Any previous layers remain intact and can still be used by other processes.

KAF as developed within the EU Knowledge Yielding Ontologies for Transition-based Organization (KYOTO) project. It is compatible with LAF (ISO Linguistic Annotation Framework) but imposes a more specific standardization of the annotation format itself. KAF can be seen as a multi-layer format for text annotation: the first two layers, explicitly dedicated to representing morphosyntactic and syntactic information, are inspired by MAF and SynAF and are implemented “over” the semantic layer. For semantic annotation, the ISO community provides SemAF which is especially dedicated to the representation of events and time. KAF layers are to be seen as dialects of the ISO standards, yet maintaining (different degrees of) mappability to them. The KYOTO dialects do not corrupt the compliance with ISO standards and their underlying philosophy; instead, they are in line with the strategy in ISO which provides high-level models (meta-models) able to be adapted, tailored and implemented according to specific needs.