dimsum16 / dimsum-data Goto Github PK

$ fgrep 'AUX' dimsum16.train | fgrep 'v.stative' | wc -l
     342
$ fgrep 'AUX' dimsum16.train | fgrep -v 'v.stative' | wc -l
    2352
$ fgrep 'v.stative' dimsum16.train | fgrep -v 'AUX' | wc -l
    3089

E.g., the following is a copula, so should be VERB rather than AUX:

1       Is      is      AUX     O                       v.stative       lowlands-4
2       your    your    DET     O                               lowlands-4
3       cat     cat     NOUN    O                       n.animal        lowlands-4
4       safe    safe    ADJ     O                               lowlands-4
5       from    from    ADP     O                               lowlands-4
6       disease disease NOUN    O                       n.state lowlands-4

Whereas the following is an auxiliary, so there should be no supersense:

1       Were    were    AUX     O                       v.stative       lowlands-32
2       ordered ordered VERB    O                       v.communication lowlands-32
3       to      to      ADP     O                               lowlands-32
4       chant   chant   VERB    O                       v.communication lowlands-32
5       praise  praise  NOUN    O                       n.communication lowlands-32

What were annotators told for the CPH supersense-labeled tweets? I need to give my annotator a policy for various tweet-centric phenomena.

Differences I am noticing between Lowlands and Ritter datasets:

• Lowlands consists mostly of promotional material/shares of other content. Most tweets are short, headline-like, and end with a URL. Ritter data is more conversational.

• URLs, usernames, numbers, etc. are obscured in Lowlands data, but not Ritter data.

• In Lowlands data, numbers at the beginning of a noun phrase are joined to form an MWE.

• In Lowlands data, usernames are supersense-tagged as PERSON and URLs as COMMUNICATION. They are not supersense-tagged in Ritter. I would argue that it's probably not the best use of annotator time to assign labels that can be applied deterministically, nor is it interesting to give credit for such labels in our evaluation.

• Some Ritter tokens are inexplicably missing supersense labels. E.g. in ritter-train.tsv:

  follow  NN  O
  back    NN  O
  

With the CMU Twitter dataset (Tweebank), some kinds of MWEs are already annotated. It would be nice to either (i) choose supersenses subject to the annotated MWE segmentation, or (ii) choose the segmentation and supersenses such that the annotated MWEs, at minimum, are present (additional MWEs may be predicted).

These occur between sentences lowlands-73 and lowlands-76 (are 74 and 75 supposed to be there?), as well as before the start of the Reviews data. I removed them manually in 07db7a2 but just wanted to check that these aren't due to a preprocessing bug.

The lexical head field is 0 in the Reviews data and blank in the rest of the data.