universaldependencies / cairo Goto Github PK

We have used brat's universal dependencies configuration (slightly adjusted with some renamed/added dependency relations) in brat v1.3 to annotate dependency relations and POS-tags in a small dataset.

After acquiring the brat-standoff-formatted .ann files, I ran the brat2conllu.pl script on them, but the resulting CoNLL-U formatted data lacks dependency structure and dependency relation labels.

Example

Plain Text (example.txt)

in a bowl .

Brat Standoff Annotation (example.ann)

T1.1	ADP 0 2	in
R1.1-1	case Arg1:T1.3 Arg2:T1.1
#1.1	AnnotatorNotes T1.1	LEMMA=in POSTAG=IN
T1.2	DET 3 4	a
A1.2-1	Definite T1.2 Ind
A1.2-2	PronType T1.2 Art
R1.2-1	det Arg1:T1.3 Arg2:T1.2
#1.2	AnnotatorNotes T1.2	LEMMA=a POSTAG=DT
T1.3	NOUN 5 9	bowl
A1.3-1	Number T1.3 Sing
#1.3	AnnotatorNotes T1.3	LEMMA=bowl POSTAG=NN
T1.4	PUNCT 10 11	.
R1.4-1	punct Arg1:T1.3 Arg2:T1.4
#1.4	AnnotatorNotes T1.4	LEMMA=. POSTAG=.

brat2conllu.pl Output (example.conllu)

Running perl brat2conllu.pl example.txt example.ann > example.conllu yields:

# sent_id = s1
# text = in a bowl .
1	in	_	ADP	_	_	0	_	_	Offset=0-2
2	a	_	DET	_	Definite=Ind|PronType=Art	0	_	_	Offset=3-4
3	bowl	_	NOUN	_	Number=Sing	0	_	_	Offset=5-9
4	.	_	PUNCT	_	_	0	_	_	Offset=10-11

• Expected: the CoNLL-U output contains the correct dependency labels and structure.
• Actual: as you can see, all words have 0 as their head (i.e. all words are roots) and the dependency labels are completely left out.

Question: what are we doing wrong? Does the script expect another version of the brat standoff format?

Hi.

The documentation for brat2conllu.pl shows as an example to use ru.txt and ru.brat as input. From brat I can only export ".ann" files, but I do not have any .brat files. Art those the same? I renamed my .ann files to .brat and it seems to work.

Thanks!

Hi,

What license are the sentences released under? I wanted to add some of them to Tatoeba, but can only do so if they are released with a compatible license such as CC BY or CC0.

CC0 would be ideal, as it's compatible with the terms of the Mozilla Common Voice project.