saving /home/johannes/GitHub/UD_Turkish-PUD/tr_pud-ud-test.conllu sen… #1
Conversation
…tence: 892 (/127.0.0.1:35690)
| @@ -12166,7 +12166,7 @@ | |||
| # sent_id = w01072079 | |||
| # text = İmparator Caracalla 3. yüzyılda kısa bir süre geçerli olan yeni bir bölünme uygulamıştır. | |||
| # text_en = By the 3rd century the emperor Caracalla made a new division which lasted only a short time. | |||
| 1 İmparator İmparator NOUN NN Number=Sing 13 nsubj _ _ | |||
| 1 İmparator İmparator NOUN NN Number=Sing 13 nsubj:outer _ _ | |||
There was a problem hiding this comment.
Why is this labeled as outer subject? Isn't the actual error elsewhere? Is the clause kısa bir süre geçerli olan yeni bir bölünme really subject? Isn't it rather an object clause?
There was a problem hiding this comment.
Revisiting it, I think you're right, "Imperator" should be just nsubj and "bölümne" definitely not csubj (objd or rather ccomp?)
| @@ -3857,7 +3857,7 @@ | |||
| # text = Seçim bölgesi, seçmenlerin yüzde 62'si AB'den ayrılmayı destekleyen Kuzey Kesteven konsey bölgesindedir. | |||
| # text_en = The constituency is in the council area of North Kesteven, where 62% of voters backed leaving the EU. | |||
| 1 Seçim seçim NOUN NN Number=Sing 2 nmod:poss _ _ | |||
| 2 bölgesi bölge NOUN NN Case=Nom|Number=Sing|Number[psor]=Sing|Person[psor]=3 13 nsubj _ SpaceAfter=No | |||
| 2 bölgesi bölge NOUN NN Case=Nom|Number=Sing|Number[psor]=Sing|Person[psor]=3 13 nsubj:outer _ SpaceAfter=No | |||
There was a problem hiding this comment.
Why is this labeled as outer subject? Shouldn't the other subject, seçmenlerin yüzde 62'si, be rather attached to a nested clause?
There was a problem hiding this comment.
Hm, I took the definition word by word: nsubj:outer specifies a nominal subject of a copular clause whose predicate is itself a clause, and concluded that Seçim bölgesi is the subject (but not nominal, in fact). Do you see it the other way round (62'si is nsubj:outer)?
There was a problem hiding this comment.
62'si here is the subject of the destekleyen which is a participle.
On a general note: the annotations in this treebank is really in a bad shape. There has been some fixes from BOUN people a few years ago, but as far as I know they were not applied. I think we should re-activate that effort and apply their changes before diverging from the base a lot.
There was a problem hiding this comment.
I see, I wanted only to correct the things detected by the automatic validator to be sure that the PUD treebanks won't be excluded from the future versions, because I think the PUds are really useful from a typological point of view. But I agree a new effort may be necessary to revisit things ....
There was a problem hiding this comment.
I agree. It would be a pity if we drop PUD treebanks. I'd be happy to have a look at the Turkish one - at least to get it validate by next data freeze, but problems are beyond the issues of validation. I hope we can also improve the actual annotations. I will start contacting the BOUN people. Maybe we can factor their changes in before working on the validation issues.
| 4 gerçekleşen gerçekleş VERB VB Aspect=Perf|Mood=Ind|Number=Sing|Tense=Pres|VerbForm=Part 6 nmod:poss _ _ | ||
| 5 lerin _ X GW Case=Gen|Number=Plur|Person=3 4 goeswith _ _ | ||
| 4 gerçekleşen gerçekleş VERB VB Aspect=Perf|Case=Gen|Mood=Ind|Number=Plur|Tense=Pres|Typo=Yes|VerbForm=Part 6 nmod:poss _ _ | ||
| 5 lerin _ X GW _ 4 goeswith _ _ |
There was a problem hiding this comment.
This fix is uncontroversial, but it is actually odd for a PUD treebank to have Typo=Yes and/or goeswith at all. This is not a naturally occurring text. It has been translated into Turkish specifically for the purpose of UD annotation, so there shouldn't be any errors in the translation.
There was a problem hiding this comment.
This is a mistake during automatic segmentation during the annotation process. I do not think the original text would have this word split into two. Correct solution would be merging these tokens. I'd annotate as:
gerçekleşenlerin gerçekleş VERB _ Case=Gen|Number=Plur|Tense=Pres|VerbForm=Part 5 acl _ _
But it seems in PUD participles and verbal nouns are all split. (Also, nomd(:poss) is also the default for clausal modifiers that behave like nouns as the one above.)
There was a problem hiding this comment.
I didnt't realise that PUD translations are "error free". In these case I agree that we should retokenise it into gerçekleşenlerin. What deprel do you prefer, acl or rather nmod?
There was a problem hiding this comment.
I didnt't realise that PUD translations are "error free". In these case I agree that we should retokenise it into gerçekleşenlerin. What deprel do you prefer,
aclor rathernmod?
From the global UD perspective I would say that if it is correctly tagged as VERB, then it heads a clause, meaning that it can be acl but not nmod or nmod:poss. But I haven't checked what's usually done in the Turkish treebanks.
There was a problem hiding this comment.
This is one of the tricky cases, and inconsistently annotated in different treebanks. These clauses behave like nouns (as in here, participating in a typical genitive - possessive construction). However, UD does not have dependency type for 'a nominal clause modifier of a noun'. Some treebanks use acl, and others use nmod(:poss). I am more inclined for acl (the type of modification can be inferred from the morphological features), but I do understand the others wanting to treat these clauses like nouns.
This is corrected version of PR #1, fixing only two validation errors.
three minor validation errors corrected