Awesome

Summary

The Galician-TreeGal is a treebank for Galician developed at LyS Group (Universidade da Coruña) and at CiTIUS (Universidade de Santiago de Compostela).

Introduction

The resource derives from a subset (called xeral) of the XIADA corpus (v2.6), created at the Centro Ramón Piñeiro para a Investigación en Humanidades (http://corpus.cirp.es/xiada/).

All the information except the syntactic one was semi-automatically converted to UD from the original resource. The dependency labels were assigned using cross-lingual parsing techniques, and then manually corrected by a linguist (see the references for more information). At the end of this process, several corrections were carried out in order to agree with the UD guidelines.

Galician-TreeGal v0.42 contains 1000 sentences of the xeral corpus (~25k tokens), and it is divided 60-40 splits (train-test).

Differences from the generic Galician guidelines

Morphology

Features

• The Galician-TreeGal treebank uses the following morphological features:
  • Nouns have inherent or inflected Gender (common, feminine, or masculine) and Number (singular or plural).
  • Verbs can have four different VerbForm features (finite, gerund, participle, or infinitive):
    • Forms in participle inflect for Gender (feminine or masculine) and for Number (singular or plural).
    • Both finite and infinitive verb form may inflect for Number (singular or plural) and for Person (1, 2, 3).
    • Finite forms also inflect for Tense (present, future, past, imperfect, and pluperfect) and Mood (indicative, subjuntive, imperative, and conditional).
  • Adjectives agree with nouns with respect to the features Gender and Number. In addition, some adjectives inflect for Degree (comparative, superlative).
  • Negative adverbs have negative Polarity in Galician-TreeGal. Also, relative and interrogative adverbs are labeled with Rel and Int (PronType), respectively.
  • Few adpositions (e.g., sen) and conjunctions (nin) are also labeled with negative polarity.
  • PronType is used to differentiate the following pronouns and determiners: Demonstrative, indefinite, relative, interrogative, personal, and article. Personal (Prs) stands for personal and for possessive pronouns and determiners (Poss disambiguates between these two types). PRON Articles (Art) are those determiners acting as pronouns (heads) when a noun is elided (e.g., "a [elided noun] de Abel")).
  • Determiners and pronouns inflect for Number (singular or plural) and for Gender (feminine, masculine, common, and neutral).
  • Possessive determiners and pronouns (Poss) also have a language-specific feature which encodes the number of the possessor: Number[psor] (singular or plural).
  • Clitic pronouns inflect for Case with three values (nominative, accusative, dative).
  • Determiners agree with nouns in the same way as adjectives.

For more information, see Garcia, Marcos (2016), Universal Dependencies Guidelines for the Galician-TreeGal Treebank (note that this document follows old UD guidelines).

Syntax

• Galician-TreeGal uses four dependency relation subtypes:
  • aux:pass for passive auxiliaries.
  • flat:name for exocentric complex names.
  • flat:foreign for foreign complex names with different structures.
  • nsubj:pass for nominal subjects of passive verbs.

Acknowledgments

• Garcia, Marcos, 2016. Universal Dependencies Guidelines for the Galician-TreeGal Treebank. Technical Report. LyS Group, Universidade da Coruña.

• Garcia, Marcos, Carlos Gómez-Rodríguez and Miguel A. Alonso, 2018. New treebank or repurposed? On the feasibility of cross-lingual parsing of Romance languages with Universal Dependencies. Natural Language Engineering, 24(1): 91-122.

• Rojo, Guillermo, Marisol López Martínez, Eva Domínguez Noya and Fco. Mario Barcala, 2015. Corpus de adestramento do Etiquetador/Lematizador do Galego Actual (XIADA), v2.6. Centro Ramón Piñeiro para a Investigación en Humanidades.

Stats

• Tree count: 1000
• Word count: 25548
• Token count: 23478
• Dep. relations: 37 of which 4 language specific
• POS tags: 16
• Category=value feature pairs: 44

Issues

• There are a few tokens with blank spaces (telephone numbers: "xxx xxx xxx").
• The morphological features of some tokens (analyzed as part of proper nouns in the original corpus) are not available.
• The dependency labels need to be reviewed by another expert to calculate inter-annotator agreement.

Changelog

• 2024-04-21 v0.45:

  • Correction of capitalized lemmas.
  • Correction of UPOS and XPOS annotation in PROPN.
  • Corrections of PROPN > NOUN annotations.
  • Minor corrections of 'nmod' > 'nummod' dependencies.
  • Minor corrections of 'flat', 'appos' and 'compound' dependencies.

• 2019-03-26 v0.42:

  • Correction of AUX verbs and some others POS-tags.
  • Adaptation of copulative analysis to the latest UD guidelines.
  • Minor review of other cases.

• 2018-03-21 v0.41:

  • Few features and POS corrections.
  • 'orphan' dependency correction.

• 2017-06-02 v0.4:

  • POS-tags, morphological features and dependency relations adapted to UDv2.
  • Multiword expressions expanded (previously as single tokens).
  • Complex proper nouns expanded (previously as single tokens).
  • General review.
  • Added original text.