Docs: reparandum for repeated words in poorly edited texts

[lauma]

Is it permited to use role reparandum for poorly proofread texts where verb is ungrammatically repeated, e.g. _yesterday he was was here_. If not, what would be more appropriate?

In Latvian corpus we have a sentence _Zālē bija aizņemtas bija gandrīz visas vietas_ (all seats were taken in the hall) from relatively reputable news source. In this sentence the ungrammaticaly repeated verb "bija" (were) even isn't next to the previous instance of it. Latvian has rather free word order, so either _bija aizņemtas gandrīz visas vietas_ or _aizņemtas bija gandrīz visas vietas_ both are completely fine.

[dan-zeman]

It sounds acceptable to me. But it would be also possible to attach both occurrences to the same parent and use the dep relation for the second copy. I think I have seen something like that in one of the UD treebanks.

[lauma]

Untill now I assumed that dep means incomplete or someway wrong annotation, and, thus, should be used as a last resort.

[dan-zeman]

I would say that this case is a last resort :-) The dep relation definitely should be used as little as possible. But an ungrammatical sentence is something where I would start at least considering dep as a possible solution.

[colinbatchelor]

In the speech subcorpus I'm annotating at the moment I'm using dep where the transcriber has been unable to work out what the speaker said and written [?]. Occasionally I can guess the part of speech from context but it seems safest just to say dep. Is that reasonable?

[dan-zeman]

In situations where you cannot guess a different relation from context, dep sounds reasonable to me.

[amir-zeldes]

@colinbatchelor we do the same for damaged manuscript parts in the Coptic treebank (by convention, dep governed by the root, or local root if there is a projectivity problem)