tanto/como, both/and, etc

[AngledLuffa]

Maybe my first attempt at figuring out como was too broad, didn't get any engagement at any rate, so let's try a smaller version. What to do with both X and Y, tanto X como Y, and similar constructions?

both X and Y in English GUM/EWT is tagged CCONJ/CCONJ

tanto / quanto in it_isdt: ADV / ADV

tanto / como in pt_bosque: ADV / ADV

atât / cât in ro_simonero: ADV / ADV

tant / com in ca_ancora: ADV / SCONJ

which brings us to Spanish tanto/como, naturally not consistent between AnCora and GSD

in es_ancora: ADV / SCONJ is typical

in es_gsd: PRON/CCONJ, ADV/CCONJ, ADV/ADV, ADV/ADP, PRON/ADP

out of these Romance languages, seems ADV/SCONJ is unique to the AnCora datasets, and SCONJ doesn't seem right since Y is often a noun. So it might be better to switch the two AnCora datasets and GSD to ADV/ADV. Still, I'd be okay with using ADV/SCONJ in ES just to make the general annotation of como easier.

[AngledLuffa]

@dan-zeman thoughts? ADV/ADV or ADV/SCONJ in Spanish & Catalan?

[nschneid]

Some discussion in UniversalDependencies/UD_English-EWT#212

[dan-zeman]

I think that for como the two main options are SCONJ or relative ADV (which was until now tagged as PRON in AnCora, and still is e.g. in one of the Galician treebanks). The SCONJ is probably grammaticalized usage of the original adverb and it is more frequent, but they are not always easy to distinguish. The adverbial interpretation makes sense if it can be turned into a question (replacing relative como with interrogative cómo).

Despite the fact that tanto ... como is functionally coordinating construction, I prefer not to further complicate things by labeling these instances of como as CCONJ. Instead, I would stick with SCONJ as in AnCora.

As for tanto, I don't see it as a conjunction (which it wouldn't be outside tanto ... como, unlike como). It is a demonstrative (ADV or DET or PRON, not sure how much we want to distinguish these from each other).

[AngledLuffa]

Sounds good. Since we're mostly following what AnCora already does, there isn't much to do there (a couple apparent stragglers changed). The GSD changelist is substantially larger:

UniversalDependencies/UD_Spanish-GSD#95

I tried to separate it into logically different changes. Not 100% sure about the head of tanto in the final change. Any thoughts? Look good?

[dan-zeman]

Any thoughts? Look good?

Not yet. Thoughts posted at the PR.

[AngledLuffa]

I don't see the thoughts yet - do you mean they will be later?

[dan-zeman]

I added two comments to the code there. They are labeled as part of a review that is "Pending". Let me see if I can close the review somehow so the comments become visible to other people than me (assuming this is the problem).

[AngledLuffa]

Whatever you just did, I can see it now!

[AngledLuffa]

Alright, I think we're agreed now that tanto X como Y is improved. Probably we should take another pass over the datasets and look at patterns for tanto como X and try to improve that some. It is inconsistent whether head(X) == tanto or if head(tanto) == head(X)