David Schlangen : Home Page > minutes2007_10_01
- present: Michaela, David
- incremental parsing:
- Amit's parser still rather mysterious. Incremental input of
words and output of tree at each word now works, but parser lags
behind one word. This is because of peculiarities of algorithm.
- other options:
- Bikel parser. Written in Java. Not incremental at the moment.
- openCCG. Written in Java. Not incremental at the moment.
---> for now:
- we'll try to get some simple features out of Amit's
parser, probably just by parsing its output (+ hopefully
getting at its open predictions). But minimise further
commitment to this parser.
- get interface to ASR going -- this will be needed later
anyways, regardless of which parser is used.
- other fun things:
- word-based models of utterance lenghts, e.g. n-grams on words
or POS tags
- look at length distributions of instances of different dialogue
acts (using SWBD DA annotation). Are there DAs that have a
typical lenght? (Small standard deviation.)
- given DA history, can we predict current DA, and via this
information, predict / constrain the prediction of the length
of the current utterance.
das, 10/01/07 02:30 (GMT)
Keyword: inpro,
meetings,
minutesAdd a new page under this one
