Hi,
I have a question for the following question. Since the question asks about all possible ways we can compute the prob. I think 3-gram can be reduced to 2-gram by marginalization. So the marked choice might also be valid? Since other correct options also consider marginalization sometimes.
in the conditioning. (marginalization deals with the part BEFORE the conditioning, not after)Sorry for not explaining clearer. I simply mean here in the slide, since the 3 gram parameters P(aaa)...P(zzz) already give us the ability to compute parameters of 2-gram parameters P(aa)...P(zz) by marginalization. So it is possible to compute the probability in a 2-gram fashion when we have the parameters of 3-grams. It's also possible to rewrite the 2-gram form to the 3-gram form by just adding more conditions too. Since the question states: select all ways to compute the max likelihood. Here I am suggesting a way to do so.
Sorry but I still think you misunderstand my point. I mean it is possible to compute the prob in 2-gram fashion in 3-gram model setting because we can first reduce the 3-gram model to 2-gram by marginalization. and the question did not constrain the way to compute this prob.
Follow the pulses of EPFL on social networks
© 2023 EPFL, all rights reserved