[SRILM User List] Some Problems about the generation of mesh using lattice-tool

Andreas Stolcke stolcke at speech.sri.com
Thu Dec 17 18:19:46 PST 2009


On 12/17/2009 5:20 PM, wei chen wrote:
> Hi all,
> I am using lattice-tool to transform lattice into mesh, and I know
> that the process that
> * Let N := set of all lattice nodes
> * while N is not empty
> * n = node with highest posterior in N
> * p = path through n with highest probability
> * p' = portion of p containing nodes in N
> * align p' to sausage (*)
> * N = N - nodes in p'
> but how to determine the align number of the sausage ? just using the
> alignments of the best sentence including the node with highest
> posterior probablility?
You start out with as many alignment positions as there are words in the
first lattice path chosen.
Each additional path being aligned might increase the number of
positions as a result of an "insertion" into the alignment.

> And during the generation, some warning like "Lattice::alignLattice:
> warning: failed to align 1 word(s), max posterior = 1.43348e-06"
> happens, could you please tell me why it happens?Thanks a lot!
Sometimes you are trying to align a partial path (p') in which the end
node is already aligned to a position that comes *before* the alignment
position for the start node. Empirically, the posterior probabilities
associated with such paths are very small (as in your example), so you
can consider them as noise in your lattice (and ignore them).

Andreas

> Best wishes,
> Wei Chen
>
>
> _______________________________________________
> SRILM-User site list
> SRILM-User at speech.sri.com
> http://www.speech.sri.com/mailman/listinfo/srilm-user

-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://www.speech.sri.com/pipermail/srilm-user/attachments/20091217/4b084260/attachment.html>


More information about the SRILM-User mailing list