Multiple node_ids in a single error?

[AngledLuffa]

There are a couple error types (double subj, double obj, etc) which could benefit from remembering multiple error indices. Any issues with make such a PR?

[dan-zeman]

What is error index? The line number + sentence id?

[AngledLuffa]

Sorry if unclear.

Multiple subjects [12, 15] ('مسج' ,'يڪيج') not subtyped as ':outer'.

Currently nodeid == 19

would be nice to also have a field such as childrenid == (12, 15), possibly with a better name

[dan-zeman]

I see. In fact I was thinking that even nodeid is superfluous since the line number already points to the node :-) But I can imagine that other uses of the new Validator object, such as highlighting the error in visualized tree, could benefit from node ids. They are not necessarily children of the main node, and they can be more than two (as in punct-is-nonproj), so it should be a list with a more neutral name (other_node_ids? related_ids? affected_ids?).

And I guess it would be better to do this in the remake branch that @ellepannitto and @harisont created.

[AngledLuffa]

Yes, that last bit makes sense - any sense of a time frame?

[dan-zeman]

a list with a more neutral name (other_node_ids? related_ids? affected_ids?).

Or maybe all the ids including the main one could be in the same list, perhaps with the semantics that the first one is "main".

There is no fixed time frame I think, but it's not going to be rolled out tomorrow.

[harisont]

Sounds good, let me add this to the to-do list in #132!

@AngledLuffa: As Dan says, merging remake into master will take some time, as some more work and testing is needed and all of us have other commitments especially for the rest of this month. If you'd like to help us out with any items (for example implementing this specific feature), though, I'm sure that can speed up the process.

[dan-zeman]

Each Incident object now has an optional list of references to additional nodes or input lines. A reference has filename, lineno, nodeid and sentid. (nodeid could be empty if we were referring to a line that is not a node.) It can also have a comment that explain how the reference relates to the incident.

Currently it is used in a few level 3 tests (multiple subjects, multiple objects, nonprojective punctuation...)

[AngledLuffa]

Neat, thanks. Is it expected that the reference nodeid can be a string? For example, the nmod/obl error has string nodeids:

The parent (node [23] 'جھڙو') is a nominal (and not a predicate), hence the relation should be 'nmod', not 'obl'.
23 <class 'str'>

[AngledLuffa]

I just added a PR which adds the particular feature I had been looking for, using the new References system

[dan-zeman]

Neat, thanks. Is it expected that the reference nodeid can be a string? For example, the nmod/obl error has string nodeids:

The parent (node [23] 'جھڙو') is a nominal (and not a predicate), hence the relation should be 'nmod', not 'obl'.
23 <class 'str'>

It is expected, although definitely not any string. But it cannot be int because it can be "1.1". And it is not float either because 1.1 < 1.10.