DMRX creates <realpred> nodes without required pos field

[goodmami]

Jacy has a predicate like _te_adjunct_rel, which PyDelphin converts to the following realpred:

<realpred lemma="te" sense="adjunct" />

Then dmrx._decode_pred returns None for elem.get('pos'), and Pred.realpred trips on the non-string value of the pos field.

See: http://lists.delph-in.net/archives/developers/2017/002631.html

[goodmami]

Strict fix:

• throw error when Pred.stringpred() has no pos
• throw (more informative) error when Pred.realpred() has no pos
• (maybe unnecessary) throw error in dmrx._encode_node() if pos is None
• (maybe unnecessary) throw error in dmrx._decode_pred() if pos is None

Robust fix:

• create pos="" attribute in dmrx.encode_node() when pos is None
• create default pos (x? empty string?) in dmrx.decode_pred() when pos is None
• refactor Pred class to allow, e.g., empty strings for pos

[fcbond]

Maybe the default pos should be 'u' unknown?

…

On Wed, Dec 20, 2017 at 2:44 AM, Michael Wayne Goodman < ***@***.***> wrote: Strict fix: - throw error when Pred.stringpred() has no pos - throw (more informative) error when Pred.realpred() has no pos - (maybe unnecessary) throw error in dmrx._encode_node() if pos is None - (maybe unnecessary) throw error in dmrx._decode_pred() if pos is None Robust fix: - create pos="" attribute in dmrx.encode_node() when pos is None - create default pos (x? empty string?) in dmrx.decode_pred() when pos is None - refactor Pred class to allow, e.g., empty strings for pos — You are receiving this because you are subscribed to this thread. Reply to this email directly, view it on GitHub <#129 (comment)>, or mute the thread <https://github.com/notifications/unsubscribe-auth/ABD8xuc2NXL0R7geiBZMvGvGXjdcQiCVks5tCHSRgaJpZM4RHzzw> .

-- Francis Bond <http://www3.ntu.edu.sg/home/fcbond/> Division of Linguistics and Multilingual Studies Nanyang Technological University

[goodmami]

Oh I thought x was the catch-all pos. But u works, too.

But I want to avoid doing something quietly that would yield a different MRS on round-tripping. E.g., if PyDelphin reads in _te_adjunct_rel, then outputs _te_u_adjunct_rel, it has changed the semantics for the purpose of preventing a PyDelphin exception.

I would rather throw an error (grammar devs need to get with the times, man), make normalization (i.e. inserting the 'u' pos) an explicit operation by the user (e.g. x = mrs.normalize(simplemrs.loads_one('...'))), or make it robust in a way that is equivalent on round-trip (e.g. empty string pos, then make sure PyDelphin writes _te_adjunct_rel and not _te__adjunct_rel)