Annotation overview

Gold standard annotation


  1. Export raw ATF from the CDLI
    Manually export ATF of texts for annotation from the CDLI database. This can be done by doing a search at and clicking on “download” after the results appear.

  2. Convert texts to pseudo-CoNLL for morphological annotation
    Use the Python script X to prepare a tokenized and CoNLL-style version of the textual information to facilitate morphological annotation.

  3. Morphological annotation
    Manually annotate morphology and subject, direct and indirect object dependencies in the files. (See below for more details.)

  4. Conversion to full annotations set in CoNLL-U Format
    Using the X Python script, convert the new annotations to the fuller version that will be used by subsequent processes.

  5. Conversion to Brat standoff
    Use the CoNLL-U to Brat standoff converter to prepare data for syntactic annotations using the Brat editor.

  6. Syntax annotation
    Manual annotation of syntax using the Brat interface

CoNLL-like and CoNLL-U formats

CoNLL-U format information:
CoNLL-U implementation based on:

Original CoNLL-U syntax calls for such structure marking:

# newdoc id = mf920901-001
# newpar id = mf920901-001-p1
# sent_id = mf920901-001-p1s1A
# text = Slovenská ústava: pro i proti
# text_en = Slovak constitution: pros and cons

We adapt this heading information as follows:

  • In a comment line above the textual information, the text id must be mentioned, eg: # newdoc id = P653433

In CoNLL-U, blank lines are used to separate sentences. Since our documents are generally considered to be a sentence, a blank line will generally appear between texts and sometimes rarely inside a text to separate full sentences.

CoNLL-U Fields

Original CoNLL-U Fields

ConLL-U fields description based on the Universal Dependencies website

  • ID: Word index, integer starting at 1 for each new sentence; may be a range for multiword tokens; may be a decimal number for empty nodes.
  • FORM: Word form or punctuation symbol.
  • LEMMA: Lemma or stem of word form.
  • UPOSTAG: Universal Dependencies (UD) part-of-speech tag.
  • XPOSTAG: Language-specific part-of-speech tag; underscore if not available.
  • FEATS: List of morphological features from the universal feature inventory or from a defined language-specific extension; underscore if not available.
  • HEAD: Head of the current word, which is either a value of ID or zero (0).
  • DEPREL: Universal dependency relation to the HEAD (root iff HEAD = 0) or a defined language-specific subtype of one.
  • DEPS: Enhanced dependency graph in the form of a list of head-deprel pairs.
  • MISC: Any other annotation.

MTAAC CoNLL-like fields for annotation

  • ID: all information about the surface, column, line and token (o.col1.1.1; o.1.1 if there is no column). Only the column number is optional.
  • FROM: token from text, ATF transliteration
  • SEGM: normalized form of the token
  • XPOSTAG: ORACC ETCSRI morphological tags based on the segmentation and using POS tag or named entity tag instead of “STEM” for the stem (eg.: GN.ABL)
  • HEAD: id of token that is the verb for which this token is a subject of object
  • DEPREL: relationship with verb as subject, direct object or indirect object (nsbj/dobj/iobj)
  • MISC: semantic role of this word eg. “seller”

MTAAC CoNLL-U fields with processed data

  • LEMMA: Lemma to which the token should be associated
  • UPOSTAG: Universal dependencies part of speech tag, based on a mapping between the ETCSRI POS and the UD POS
  • FEATS: Unimorph tags, in order of morpheme appearance
  • DEPS: will not be used at this time


  • Morphological annotations are added to the CoNLL file manually using any plain text editor or a spreadsheet program.

  • Syntax annotations can be added manually in the CoNLL file or using the Brat interface. We have a development Brat server up at

Brat website:
Another dependency annotation tool: UD Annotatrix

Morphological annotation example

Automated annotation workflow

Émilie Pagé-Perron