Overview
ISO 24617-9:2019 - Reference Annotation Framework (RAF) defines a comprehensive, interoperable model for annotating referential phenomena in texts and multimodal interactions. Part of the ISO 24617 semantic annotation series, RAF covers simple anaphora and coreference as well as complex bridging and multimodal reference. The standard specifies an abstract syntax (conceptual model), a recommended concrete XML serialisation (a TEI P5 customisation), and a catalogue of core data categories for referential entities and link structures.
Keywords: ISO 24617-9, RAF, reference annotation, semantic annotation, TEI P5, XML serialisation, coreference, anaphora, multimodal reference.
Key topics and technical requirements
- Meta-model for reference annotation: Separates linguistic marking of referring expressions (markables) from discourse entities (referents), and defines objectal and lexical relations between them.
- Abstract vs concrete syntax: Specifies an abstract syntax (what the annotation means) and prescribes a TEI P5–based XML concrete syntax as the primary interoperable serialisation.
- Entity and link structures: Defines entity structures (referring-expression components and discourse-entity components) and link structures (coreference, bridging, objectal relations).
- Data categories and semantics: Provides normative data categories for referential features (e.g., identity, inclusion, overlap) and guidance on annotation semantics.
- Multimodal anchoring: Supports anchoring referring expressions to communicative segments across modalities (text, gesture, audio).
- Implementation guidance: Includes practical advice on namespaces, feature structures, document architecture, ambiguous/alternative linking, and representing referential chains.
- Normative references: Requires compatibility with TEI P5, XML 1.0 and related ISO metadata standards (e.g., ISO 24622-1).
Practical applications and users
RAF is intended to improve interoperability and reusability of annotated language resources across NLP and language-technology tasks:
- Who uses it
- NLP researchers and corpus linguists building annotated corpora
- Annotation-tool developers and data engineers implementing XML serialisations
- Language resource managers and standards teams ensuring metadata compliance
- Applied ML/NLP teams in NER, coreference resolution, machine translation, question answering, dialogue systems, information retrieval, summarization, and text understanding
- Practical value
- Standardises reference annotation for consistent training data and evaluation
- Enables multimodal annotation (gesture, audio, text) for human–machine dialogue and multimodal ML
- Facilitates interoperability between tools and datasets by leveraging TEI P5 and W3C XML recommendations
Related standards
- ISO 24617 semantic annotation series (other parts)
- TEI P5 Guidelines (customisation target for RAF)
- XML 1.0 (W3C)
- ISO 24622-1 (CMDI metadata guidance)
- ISO 24612 (dialogue act annotation principles)
Using ISO 24617-9:2019 (RAF) helps organisations create robust, interoperable reference-annotated resources that accelerate reproducible research and production-ready NLP systems.
Open preview (PDF)Language resource management — Semantic annotation framework — Part 9: Reference annotation framework (RAF)