Kaingang/Universal Dependencies

kgp.annotated2.ud.conllu and kgp.annotated2.nomorph.conllu are the files we chose to train the parser on (not kgp.annotated.ud.conllu or kgp.annotated.nomorph.conllu).

Evaluation

Number of forms, UAS, LAS scores for kgp.annotated2.ud.conllu

$ udpipe --accuracy --parse kgp.withmorph.udpipe ../kgp.annotated2.ud.conllu
Loading UDPipe model: done.
Parsing from gold tokenization with gold tags - forms: 160, UAS: 70.62%, LAS: 55.00%

• Number of sentences in corpus: 18

Number of forms, UAS, LAS scores for kgp.annotated2.nomorph.conllu

$ udpipe --accuracy --parse kgp.nomorph.udpipe ../kgp.annotated2.nomorph.conllu
Loading UDPipe model: done.
Parsing from gold tokenization with gold tags - forms: 160, UAS: 70.62%, LAS: 55.00%

• Number of sentences in corpus: 18

Dependency

(for each relation: a description of the relation, noting various ways it might be used in the language; and two examples of the relation from your corpus, preferably illustrating what you described.)

Five dependency relations that we used at least twice in our annotation

nsubj

We used the NSUBJ relation to relate subjects to non-verbal predicates. Where Portuguese and English use a copula "to be" verb, Kaingang instead uses a particular aspect marker "vẽ".

"Ka fej kusũg vẽ."

tree flower red ASP

"The tree's flower is red"

In the above example, we would connect the nonverbal predicate kusũg to its NSUBJ dependent: fej