Thanks for your responses and the references. >From your responses and the papers of Lucia Specia et al., my main take away >so far is that a simple and unique feature won't provide a Confidence Measure. >When using a bunch of them, one need to consider that they are often strongly >correlated. Getting a reasonably good confidence score seems doable, which is >a good thing, but requires significant efforts to get there (manually >annotated sentences, a training process, optimizations). Let's see if we have >the courage and the time to tackle this soon!
François _______________________________________________ Moses-support mailing list [email protected] http://mailman.mit.edu/mailman/listinfo/moses-support
