Skip to content

Allow training models with non-CGN PoS tags #53

Open

Description

@proycon

Frog is currently quite tied to CGN (as we noticed in #52). I propose adding a parameter style for tagger in frog.cfg to indicate whether a PoS tagset is CGN-like (tagger.style = cgn) , i.e. it uses the HEAD(featurevalue,featurevalue) format and to also allow training models that are not in that style (tagger.style = simple?), in which case the resulting pos tags in the FoLiA would of course have no features at all at just be considered blobs. Additionaly, we could perhaps add another parameter value (tagger.style = cgn-full) for a more verbose CGN-style HEAD(subset=featurevalue,subset=featurevalue) (suggested by @JessedeDoes) which would eliminate a lot of the disambiguation problems we currently face.

Activity

  1. kosloot commented on Jul 5, 2018

    @kosloot
    Collaborator

    Well....
    as #52 is solved for now, I see less urgency. I suggest waiting for a real use-case is best.
    What probably is only needed, is to make sure that Frog can handle 'unstructured' tags, also without brackets and sub information. It probably already can. (or almost)
    A more verbose format looks undesirable to me. More data, more parsing etc.

  2. proycon commented on Jul 5, 2018

    @proycon
    MemberAuthor

    Yes, agreed

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Metadata

Metadata

Assignees

Type

No type

Projects

No projects

    Milestone

    No milestone

    Relationships

    None yet

    Development

    No branches or pull requests

    Issue actions