10/22/13: Annotating social acts: authority claims and alignment moves...; What to do about bad language

Many people remarked on looking at out-of-vocabulary (OOV) rates.  (We later described these as lexical innovations, and then neologisms).

  Q: when is something a true new lexical item, as opposed to a typo?
  - SRILM toolkit suggests using a character-based model when you hit an unrecognized token
  - thank about "successful" innovations.  (see refs on website).  An unusual case: adoption from one lang to another (see the "Beefmoves" paper)



For alignment/authority: for other datasets, are there "cheap" labels?  (E.g. in Thomas/Pang/Lee, we used votes as signals of [coarse-grained] agreement)