NAF was developed to fulfill the following requirements:
- NAF should properly represent linguistic information focusing on two kind of linguistic processes (LPs):
- within document processing: LPs whose granularity is the document
- cross document processing, for (event) coreference, etc.
- NAF should be simple
- NAF should work for existing NLP modules, in particular modules developed partners in projects that use NAF, i.e. it should be easy and little afford to adapt existing tools to use NAF.
- All elements in NAF will be identified with URIs (not document/XML-object internal ids)
- NAF should be flexible so that it can contain additional information and alternative representations:
- It should be possible (and preferably easy) to integrate alternative modules (that may be developed by third parties) in the pipeline
- It should be possible to represent other RDF-based layers that link to the URIs used in the SEM annotation layer or to background knowledge