from what i remember, it extrapolates from known instances to unseen ones using a number of metrics; the least number of deletions/insertions/subs, searching for the phrase that requires least transformation to match etc. i'm sure he must be using some decent smoothing techniques too, as he has worked for a while with Ney, who is famous for his work on the subject.
if you need more details check his papers. http://www.isi.edu/~och/
Re:South Park episode display classic irony
on
Isn't It Ironic?
·
· Score: 1
well, the irony (or commentary ?) in this may be the fact that during the colonization of america more natives were killed by diseases brought by the europeans than by combat.
the europeans would rub blankets on people with smallpox and present them as gifts to their hosts.
while it is true that bayesian filtering is not heuristic, it isn't all that sophisticated either. there are better statistical modelling approaches, but i guess the bayesian will still give quite good results.
Slashdot Mirror
from what i remember, it extrapolates from known instances to unseen ones using a number of metrics; the least number of deletions/insertions/subs, searching for the phrase that requires least transformation to match etc. i'm sure he must be using some decent smoothing techniques too, as he has worked for a while with Ney, who is famous for his work on the subject.
if you need more details check his papers. http://www.isi.edu/~och/
well, the irony (or commentary ?) in this may be the fact that during the colonization of america more natives were killed by diseases brought by the europeans than by combat.
the europeans would rub blankets on people with smallpox and present them as gifts to their hosts.
most audio equipment of today wouldn't reproduce it either - amplifiers etc. have filters with cutoff at 20-20kHz, the human hearing range.
it seems to me that the beatles' masterplan of dog-torture never worked out.
while it is true that bayesian filtering is not heuristic, it isn't all that sophisticated either. there are better statistical modelling approaches, but i guess the bayesian will still give quite good results.