UD for Breton 
Tokenisation and Word Segmentation
Tokenisation was originally done using the Apertium morphological analyser for Breton. This joins certain multiword tokens with spaces as single tokens. Where the number of spaces in the original token matches the number of spaces in the multiword token, these are split into separate tokens in UD, where the part of speech of the multiword token is given to the first token, and subsequent tokens are given the part of speech X and attached with the fixed relation.
The most important tokenisation factor is with the words traditionally described as inflected or conjugated prepositions. Here we analyse them as contractions of prepositions and pronouns. For example, dit is tokenised as a multiword token constructed from da “to” and it “you”.
Morphology
Tags
Some comments on various parts of speech:
- X Used for parts of multiword tokens that do not necessary have a part of speech of their own.
- ADP Breton has prepositions. Those words often called inflected or conjugated prepositions are analysed as multiword tokens (see above).
- PART Not used, words described in the literature as verbal particles are tagged as AUX.
Words tagged as AUX:
Copulae:
- bezañ
Auxiliary verbs:
- kaout
- ober
Verbal “particles”:
- a
- en ur
- en em
- e
- o
- na
- ra
Features
Inherent:
Inflectional:
Syntax
The following relation subtypes are used in the Breton data:
- acl:relcl Relative adnominal clause
- aux:pass Auxiliary verb used in the construction of the passive
- flat:name Parts of a multi-word personal name attached to the first part
- nmod:gen Nominal modification with genitive meaning using Celtic-style conjunctive genitive
- nmod:poss Nominal modification with possessive meaning
- nsubj:cop Nominal subject of a non-verbal or copular clause
- obl:agent An oblique introduced with gant “with” that expresses the agent in a passive construction
Treebanks
There is 1 Breton UD treebank: