[ Skip to the content ]

Institute of Formal and Applied Linguistics Wiki

[ Back to the navigation ]

You are here: start » user » zeman » treebanks » ro

This is an old revision of the document!

Table of Contents

Romanian (ro)

Resurse pentru Gramaticile de Dependenta (RGD)

Versions

Obtaining and License

The syntactically annotated Romanian texts are available at http://www.phobos.ro/roric/texts/xml/. This is a tcsh script that will download the corpus: 

#!/bin/tcsh -f
foreach i ( t1 t2 t3 t4 t5 t6 t7 t8 t9 t10 tp1 tp2 tp3 tp4 tp5 tp6 tp7 tp8 tp9 tp10 tp11 tp12 tp13 tp14 tp15 tp16 tp17 tp18 tp19 tp20 tp21 tp22 tp23 tp24 tp25 tp26 tp27 tp28 tp29 tp30 tp31 tp32 tp33 tp34 tp35 tp36 tp37 tp38 tp39 tp40 tp41 tp42 tp43 tp44 tp45 tp46 tp47 tp48 tp49 tp50 tp51 tp52 tp53 tp54 tp55 tp56 tp57 tp58 tp59 tp60 tp61 tp62 tp63 tp64 tp65 tp66 tp67 tp68 tp69 tp70 tp71 tp72 tp73 tp74 tp75 tp76 tp77 tp78 tp79 tp80 tp81 tp82 )
  wget http://www.phobos.ro/roric/texts/xml/$i.xml
end
wget http://www.phobos.ro/roric/texts/xml/dga.dtd

Licensing terms are unknown. I tend to interpret their “public availability” in the following way:

The texts were prepared by members of RORIC-LING, Faculty of Mathematics and Computer Science (Facultatea de Matematica şi Informatica), University of Bucharest (Universitatea din Bucureşti), Str. Academiei nr. 14, sector 1, RO-010014, Bucureşti, Romania.

References

Domain

Newspaper.

Size

The corpus contains 36150 tokens in 4042 clauses, yielding 8.94 tokens per clause on average. There is no official training-test data split. We use the files t1.xml – t10.xml (2640 tokens / 266 clauses) for testing and the rest (33510 tokens / 3776 clauses) for training of our HamleDT experiments.

Inside

The corpus contains texts from Portugal and Brazil. The texts were automatically parsed using the PALAVRAS parser (Bick 2000: Eckhard Bick. The Parsing System “Palavras”: Automatic Grammatical Analysis of Portuguese in a Constraint Grammar Framework. Dr.phil. thesis. Aarhus University. Aarhus, Denmark: Aarhus University Press. November 2000.) and revised by linguists (the Bosque part, referred here, was totally revised; the other parts of the Floresta sintáctica project were either partially or not at all revised).

Morphological annotation includes lemmas. In the CoNLL version, the original Floresta tags were converted to fit the CPOS, POS and FEAT columns of the CoNLL format. Use DZ Interset to inspect the CoNLL tagset.

Multi-word expressions have been concatenated into one token, using underscore as the joining character (e.g. “7_e_Meio”, “Hillary_Clinton”).

Sample

The first two sentences of the CoNLL 2006 training data:

1 Um um art art <arti>|M|S 2 >N _ _
2 revivalismo revivalismo n n M|S 0 UTT _ _
3 refrescante refrescante adj adj M|S 2 N< _ _
1 O o art art <artd>|M|S 2 >N _ _
2 7_e_Meio 7_e_Meio prop prop M|S 3 SUBJ _ _
3 é ser v v-fin PR|3S|IND 0 STA _ _
4 um um art art <arti>|M|S 5 >N _ _
5 ex-libris ex-libris n n M|P 3 SC _ _
6 de de prp prp <sam-> 5 N< _ _
7 a o art art <-sam>|<artd>|S 8 >N _ _
8 noite noite n n F|S 6 P< _ _
9 algarvia algarvio adj adj F|S 8 N< _ _
10 . . punc punc _ 3 PUNC _ _

The first two sentences of the CoNLL 2006 test data:

1 É é adv adv <foc> 9 FOC _ _
2 por por prp prp _ 9 ADVL _ _
3 isso isso pron pron-indp <dem>|M|S 2 P< _ _
4 que que adv adv <foc> 9 FOC _ _
5 , , punc punc _ 6 PUNC _ _
6 explica explicar v v-fin PR|3S|IND 0 STA _ _
7 , , punc punc _ 6 PUNC _ _
8 não não adv adv _ 9 ADVL _ _
9 tem ter v v-fin PR|3S|IND 6 ACC _ _
10 pena pena n n F|S 9 ACC _ _
11 de de prp prp _ 10 N< _ _
12 Hillary_Clinton Hillary_Clinton prop prop F|S 11 P< _ _
13 . . punc punc _ 6 PUNC _ _
1 « « punc punc _ 8 PUNC _ _
2 Eles ele pron pron-pers M|3P|NOM 8 SUBJ _ _
3 [ [ punc punc _ 8 PUNC _ _
4 Hillary Hillary prop prop F|S 9 APP _ _
5 e e conj conj-c <co-app> 4 CO _ _
6 Bill_Clinton Bill_Clinton prop prop M|S 4 CJT _ _
7 ] ] punc punc _ 8 PUNC _ _
8 podem poder v v-fin PR|3P|IND 0 QUE _ _
9 ter ter v v-inf _ 8 MV _ _
10 alguma algum pron pron-det <quant>|F|S 11 >N _ _
11 espécie espécie n n F|S 9 ACC _ _
12 de de prp prp _ 11 N< _ _
13 acordo acordo n n M|S 12 P< _ _
14 e e conj conj-c <co-vfin>|<co-fmc> 8 CO _ _
15 quem quem pron pron-indp <interr>|M/F|P 16 SC _ _
16 somos ser v v-fin PR|1P|IND 8 CJT _ _
17 nós nós pron pron-pers M/F|1P|NOM 16 SUBJ _ _
18 para para prp prp _ 16 ADVL _ _
19 dizer dizer v v-inf _ 18 P< _ _
20 se se conj conj-s _ 21 SUB _ _
21 é ser v v-fin PR|3S|IND 19 ACC _ _
22 bom bom adj adj M|S 21 SC _ _
23 ou ou conj conj-c <co-sc> 22 CO _ _
24 mau mau adj adj M|S 22 CJT _ _
25 ? ? punc punc _ 8 PUNC _ _

Parsing

Bosque is a mildly nonprojective treebank. 2778 of the 212,545 tokens in the CoNLL 2006 version are attached nonprojectively (1.31%).

The results of the CoNLL 2006 shared task are available online. They have been published in (Buchholz and Marsi, 2006). The evaluation procedure was non-standard because it excluded punctuation tokens. These are the best results for Portuguese:

Parser (Authors) LAS UAS
MST (McDonald et al.) 86.82 91.36
Malt (Nivre et al.) 87.60 91.22
Nara (Yuchang Cheng) 85.07 90.30

[ Back to the navigation ] [ Back to the content ]