ConlluEditor: a fully graphical editor for Universal dependencies - - PowerPoint PPT Presentation

ConlluEditor: a fully graphical editor for Universal dependencies treebank files

Johannes Heinecke Orange / TGI / Data & IA / DESKIÑ Universal Dependencies Workshop 2019

ConlluEditor Johannes Heinecke

UDW 2019

Movaon Features Demo

2/7 . . . . . . . . . . . . . .

Movaon

Eding CoNLL-U files

readiblity (on one line, sub-trees easy to grasp, all important informaon presented) browsing searching (for forms, lemmas, U/XPOS, deprels, sequence of all of these modifying any CoNLL-U column in graphic mode (including enhanced relaons) trees, not flat graphs abstracon from underlying format git support svg/XƎL

AT

EX export

ConlluEditor Johannes Heinecke

UDW 2019

Movaon Features Demo

3/7 . . . . . . . . . . . . . .

Features I

Eding

full graphical editor for basic and enhanced dependency relaons word edit (form, lemma, UPOS, XPOS, features, misc-column) autocompleon (UPOS, XPOS, deprels, using user provided lists of valid labels) eding multoken words ([1-2] ...) and empty nodes ([5.1] ...) comment eding support for right-to-le scripts like Arabic or Hebrew split and join words (to correct bad tokenizaon), split and join sentences (to modify sentence segmentaon) undo/redo git support (add/commit every n edits, default: 1) export of dependency graphs as .svg-image, XƎL

A

T EX-code (for the kz-dependency package or the deptree.sty XƎL

A

T EX style), sd-parse, CoNLL-U

ConlluEditor Johannes Heinecke

UDW 2019

Movaon Features Demo

4/7 . . . . . . . . . . . . . .

Features II

Searching

regex search funcons (including sequences of tokens, sub-graphs and comments)

Validaon

validaon: indicates undefined UPOS, XPOS, dependency relaons (based on lists given to the server) prohibion of invalid (cyclic) trees normalisaon of token ids (first column, from 1 to n, taking into account multoken words, empty words and heads) validaon with external script (such like UD’s validate.py) on the current sentence limited mul-user support: as long as two users do not edit the same sentence

ConlluEditor Johannes Heinecke

UDW 2019

Movaon Features Demo

5/7 . . . . . . . . . . . . . .

Architecture

CoNLL-U file list of UPOS list(s) of XPOS list(s) of deprels validaon script I I HTTP server (java 8) AJAX Firefox ≥ 63/ Chrome ≥ 60/ Edge ≥ 44 conllueditor (JS) jquery-gui 1.12.1 jquery 3.3.1 bootstrap 4.1.3 gitaa gitbb git

ConlluEditor Johannes Heinecke

UDW 2019

Movaon Features Demo

6/7 . . . . . . . . . . . . . .

Let’s see the real thing

configure validator (in valid.conf): $HOME/UD/tools/validate.py --lang cy --max-err 0 --level 5 {FILE} start server $CE/bin/conlluedit.sh -r \

• -deprels deprel.cy,deprel.ud \
• -UPOS cpos.ud \
• -XPOS xpos.cy \
• -validator valid.conf \

demo.conllu 5555

• pen navigator

— hp:/ /localhost:5555 — demo

ConlluEditor Johannes Heinecke

UDW 2019

Movaon Features Demo

7/7 . . . . . . . . . . . . . .

Thank

thank 1 VERB VBP Mood : Ind T ense : Pres VerbForm : Fin

root you

you 2 PRON PRP Case : Acc Person : 2 PronT ype : Prs

• bj

attention

attention 5 NOUN NN Number : Sing

• b

l : f

• r

for

for 3 ADP IN

case

your

you 4 PRON PRP$ Person : 2 Poss : Yes PronT ype : Prs

nmod:poss

!

! 6 PUNCT , SpaceAfter : No

p u n c t

Thank

1

you

2

attention

5

for

3

your

4

!

6

https://github.com/Orange-OpenSource/conllueditor johannes.heinecke@orange.com