Overview
ISO 24612:2012 - Language resource management - Linguistic annotation framework (LAF) - defines a standard framework for representing linguistic annotations of primary language data (text, speech, audio, image, video). LAF provides an abstract data model and an XML serialization (pivot format) so annotations expressed in one representation can be mapped and exchanged with another. The standard targets interoperable, machine-processable annotation of corpora and multimedia language resources.
Key topics and technical requirements
- Abstract data model: annotations are represented as a directed graph (nodes and edges) where nodes may reference regions of primary data and carry annotation content.
- Nodes, edges, feature structures: nodes and edges can be associated with feature structures describing linguistic properties (e.g., morphosyntactic tags), enabling multi-layered annotation.
- Anchors and regions: primary data locations are addressed by anchors (medium-dependent: character offsets, timestamps, coordinates). Regions are ordered lists of anchors; non-contiguous regions and segmentation conventions are supported.
- Stand-off annotation architecture: annotations are serialized in documents separate from the primary data, allowing multiple annotation layers to reference the same primary source.
- XML pivot format and headers: LAF specifies an XML serialization as a pivot format plus headers (resource header, primary data document header, annotation document headers) that define media types, anchor types, file types, annotation declarations and provenance.
- Primary data handling: primary data is treated as read-only to preserve references. Text primary data defaults to UTF-8/UTF-16; XML primary data can be referenced with XPath 2.0 expressions.
- Conformance: an annotation scheme must be isomorphic to the LAF data model (i.e., renderable/mappable to the graph-based model).
Applications and who uses it
LAF is practical for:
- NLP engineers and researchers building and sharing annotated corpora (POS, syntax, discourse, multimodal annotations).
- Corpus developers and linguists standardizing annotation schemes across projects.
- Speech and multimedia annotation teams annotating audio/video with timestamps and spatial anchors.
- Tool developers and integrators who need a pivot format to map between diverse annotation tools and formats.
- Digital archivists and data managers ensuring long-term interoperability and reproducibility of annotated language resources.
Benefits include improved interoperability, consistent metadata and provenance, easier format mapping, and support for layered and multimodal annotations.
Related standards
- ISO 12620 (data categories for annotation content) and other ISO/TC 37 standards are referenced for content category standardization. LAF defines the representational framework; content categories are provided by complementary standards.