The interfaces for chunking are explained in:
Chunks only get scores in the confidence chunking interface,
but not in the first-best or n-best interfaces.
Chunkings themselves only get scores when run from
the n-best interface, not from the first-best or
confidence chunking interface.
Unfortunately, there's no way to return a per-chunk
score from a first-best chunker because there's no
model of what that probability should be.
What you could do is run the confidence-based-chunker
and only take the chunks above 0.5 confidence, which
are guaranteed to be consistent with each other (that
is, no overlaps), and get per-chunk scores.
- Bob Carpenter
On 12/4/2010 8:31 PM, Yogesh wrote:
> I am using CharLmHmmChunker for tagging. The score for Chunking that I get
> is Infinity.
> Is there some way to get the actual probabilities?