On 11.08.2016, at 22:56, Marshall Schor <[email protected]> wrote:
> 
> I guess we've always (probably incorrectly) thought that those use-cases which
> wanted unique, fixed identifiers for every feature structure would just add 
> them
> as a feature, with whatever kind of id was wanted...

I didn't want to do that because then the annotation editor would have dictated
parts of the type system. The FS addresses serve excellently as out-of-band IDs
(i.e. isolated from the type system).

We've also started adding ID features to some types (e.g. Token, Sentence),
but these represent *externally pre-existing* (in-band) IDs that we obtain e.g.
from a corpus file.

Cheers,

-- Richard

Reply via email to