UD-treebank-Romanian

UD_Romanian_SMSL (Romanian Social Media Sexist Language)

Summary

• Genre: Social Media
• Corpus size: 210 samples
• The Romanian Social Media Sexist Language UD treebank (called UD_RoSMSL) is a reference treebank in UD format for Romanian sexist language.

Tasks

• Annotation of 210 samples (source: git: CoRoSeOf - paper: CoRoSeOf - An Annotated Corpus of Romanian Sexist and Offensive Tweets)
• Added 2 new deprel labels non-existing in the Romanian language, i.e, vocative:mention and discourse:emo, which results in Syntax errors: 182(*** FAILED *** with 182 errors)
• Without the new labels validation is successful 🎉
• TBA the new 2 Unknown DEPREL label in the documentation upon submission:
• Unknown DEPREL label: 'vocative:mention'
• Unknown DEPREL label: 'discourse:emo'
• Updated README.md on: https://github.com/UniversalDependencies/UD_Romanian-TueCL and committed it on dev
• documentation added by @coltekin: two documentation pages:
• https://universaldependencies.org/ro/dep/discourse-emo
• https://universaldependencies.org/ro/dep/vocative-mention

Statistics of UD_RoSMSL

• Tree count: 210

POS TAGS

• Number of unique POS tags: 17
• Unique POS tags: ['ADJ', 'ADP', 'ADV', 'AUX', 'CCONJ', 'DET', 'INTJ', 'NOUN', 'NUM', 'PART', 'PRON', 'PROPN', 'PUNCT', 'SCONJ', 'SYM', 'VERB', 'X']

Relations

• Number of unique dependency relations: 47 (+ 2 new ) out of 52 relations currently permitted in Romanian language
• • Unique deprels: ['acl', 'advcl', 'advcl:tcl', 'advmod', 'advmod:tmod', 'amod', 'appos', 'aux', 'aux:pass', 'case', 'cc', 'cc:preconj', 'ccomp', 'ccomp:pmod', 'compound', 'conj', 'cop', 'csubj', 'dep', 'det', 'discourse', 'discourse:emo', 'expl', 'expl:pass', 'expl:poss', 'expl:pv', 'fixed', 'flat', 'goeswith', 'iobj', 'list', 'mark', 'nmod', 'nsubj', 'nsubj:pass', 'nummod', 'obj', 'obl', 'obl:agent', 'obl:pmod', 'obl:tmod', 'orphan', 'parataxis', 'punct', 'reparandum', 'root', 'vocative', 'vocative:mention', 'xcomp']

Tokenization and Word Segmentation

• This corpus contains 210 sentences
• Total number of tokens: 4417
• Total number of unique tokens: 1692

Mispellings

• Total number of misspelled words: 451 (documented in the misc column with the CorrectForm)

Data Splits

The treebank will be divided into three parts:

• 12.5% for the test set (ro-ud-smsl-test.conllu)
• 12.5% for the development set (ro-ud-smsl-dev.conllu)
• 75% for the training set (ro-ud-smsl-train.conllu)

The split will be done done randomly. The corpus has been divided into documents containing instances of sexist language found in social media, from Twitter.

Language Family

• Language: Romanian (code: ro)

• Language family: Indo-European

                - Italic
                  - Latino-Faliscan
                        - Romance
                          - Eastern Romance
                            - Daco-Romance
                                - Romanian
  

©