Skip to content
Permalink
master
Switch branches/tags
Go to file
 
 
Cannot retrieve contributors at this time
115 lines (98 sloc) 6.34 KB

Summary

MUDT (Maltese Universal Dependencies Treebank) is a manually annotated treebank of Maltese, a Semitic language of Malta descended from North African Arabic with a significant amount of Italo-Romance influence. MUDT was designed as a balanced corpus with four major genres (see Splitting below) represented roughly equally.

Introduction

Origin

This treebank is the product of the PhD thesis Constituent order in Maltese: A quantitative analysis by Slavomír Čéplö. The text (see References) contains a detailed description of the annotation decisions and composition of the treebank. The treebank was originally produced in accordance with UDv1, this version is brought up to the UDv2.5 standard.

Splitting

MUDT contains 2074 sentences and 44,162 tokens (both defined orthographically) in the following text types:

Text type Subtype Sentence count
newspaper news 239
op-eds 240
Subtotal 479
quasi-spoken newspaper interviews 280
parliament: debates and Q&A 294
Subtotal 574
fiction short stories 246
novel chapters 251
Subtotal 497
non-fiction humanities 249
science, encyclopedic and instructional 275
Subtotal 524
Total 2074

The annotated sentences have been manually split into train, test and dev sets as follows:

File Sentence count Token count
mt_mudt-ud-train.conllu 1123 22880
mt_mudt-ud-test.conllu 518 11073
mt_mudt-ud-dev.conllu 433 10209

Conversion

The original treebank was annotated according to UDv1, this version was automatically converted to UDv2 as follows:

  • UPOS were added by automatic conversion from XPOS using a conversion table
  • UDv1-specific relations where only the label was changed were automatically converted to UDv2 relations (e.g. dobj > obj or nmod:advmod > obl)
  • for all *_DEF and DEF tokens, as well as for b', f', m', s' and t', SpaceAfter=No was added automatically
  • Udapi was used to convert those relations which involved changes in dependency relations (e.g. conj or flat) to UDv2, as well as to add SpaceAfter=No to punctuation
  • all ToDo items were checked manually and fixed wherever necessary
  • all validation errors were fixed manually

Updates

Update to UDv2.6

  • fixed a projective punctuation error

Update to UDv2.5

  • fixed errors related to ellipsis

Update to UDv2.4

With the introduction of stricter checks, the following updates to UPOS, XPOS and relations were implemented:

  • NEW RELATION: XPOS NEG (ma) is now advmod:neg with UPOS ADV; same goes for XPOS FOC lanqas when negating a verbal clause
  • NEW RELATION: in non-copular verbless clauses (Čéplö 2018: 104-108), XPOS KIEN that takes the place of the expletive pronoun is now attached as cop:expl
  • XPOS GEN_PRON (possessive pronouns tiegħi, tiegħek, tiegħu etc.) is now attached as nmod:poss (previously case)
  • XPOS PRON_PERS_NEG (negated personal pronouns which function as negators in copular and verbal clauses) is now aux:neg with UPOS AUX
  • in the same vein, interrogative pronouns hux, hix etc. now have UPOS AUX as well
  • all tokens that are used as auxiliaries (XPOS KIEN, *għad and tantx) now have UPOS AUX
  • XPOS PRON_INT when used as advmod now have UPOS ADV
  • XPOS PRON_INT kemm and kif that function as coordinating conjunctions (kemm ... kif, kif ukoll) now have UPOS CCONJ
  • XPOS ADJ ċertu/-a/i and iktar modifying nouns are now attached as amod (previously det)
  • XPOS NUM_WHD when modifying nouns is now attached as nummod (previously det)
  • numerals like "miljun" attached as nummod now have UPOS NUM (previously NOUN)
  • kemm-il is now attached as det (previously nummod)
  • currency symbols and mathematical symbols now have UPOS SYM (previously PUNCT)

General updates:

  • XPOS PRON_PERG_NEG previously attached as cop are now attached as aux:neg as per the MUDT guidelines

To do (status as of 2019-10-28)

  • doublecheck SpaceAfter=No in general, for ' and " and in particular
  • doublecheck the attachment of ADPs to compounds
  • doublecheck nmods attached to a non-verbal predicate
  • doublecheck flats with X_ENG
  • doublecheck "ċertu/-a/i" and "iktar" and other candidates for determiners

Changelog

  • 2019-10-28
    • Updated to UDv2.5, fixed a number of errors related to ellipsis, updated README
  • 2018-10-04
    • Removed the old version of the Maltese UD treebank, prepared the repository and updated README
  • 2018-10-07
    • Added the files, updated README
  • 2018-10-11
    • Added the files converted to UDv2 (see above), updated README
  • 2018-11-08
    • Changed the train-test-dev split per official guidelines, updated README accordingly; fixed tokenization errors
  • 2019-05-01
    • Updated to UDv2.4, updated README

References

(Čéplö 2018) Slavomír Čéplö. (2018) Constituent order in Maltese: A quantitative analysis. Prague: Charles University.

=== Machine-readable metadata (DO NOT REMOVE!) ================================
Data available since: UD v2.3
License: CC BY-SA 4.0
Includes text: yes
Genre: news legal nonfiction fiction wiki
Lemmas: not available
UPOS: converted from manual
XPOS: manual native
Features: not available
Relations: manual native
Contributors: Čéplö, Slavomír; Zeman, Daniel
Contributing: here
Contact: bulbul@bulbul.sk
===============================================================================