PoS tagging review question not done this morning

PoS tagging review question not done this morning

by Jean-Cédric Chappelier -
Number of replies: 5

Here is the question we didn't have time to address this morning:

Consider the following lexicon excerpt, where D, N, P, and V are the tags associated with the entries (D stands for determiner, N for noun, P for pronoun, and V for verb):
cat: N; V
running: N; V
saw: N; V
the: D
you: P
and consider the following sentence:
the cat you saw running

What is the condition to be verified by the parameters of the order-1 HMM model (using the provided lexicon excerpt) for the word “cat” to be tagged as a noun in the above sentence?

In reply to Jean-Cédric Chappelier

Re: PoS tagging review question not done this morning

by Ilija Gjorgjiev -

Is quiz 2 going to cover only the topics: "Out of Vocabulary Forms; Spelling Error Correction" and  "Morphology" or it is going to cover also the topics from the previous quiz?

Thank you,

Ilija

In reply to Ilija Gjorgjiev

Re: PoS tagging review question not done this morning

by Jean-Cédric Chappelier -
Well... For sure the main topic will be on "Out of Vocabulary Forms; Spelling Error Correction" and  on "Morphology"; but we do not expect you to forget all what you have learned so far ;-)

Those quizzes are not certification evaluations ("summative assesments") but are feedback evaluations ("formative assessments"). So, on one hand it'll be contradictory to tell you "forget everything before", on the other hand it'll be a bit waste of time to provide again feedback on something on which we already did...
(réponse de Normand)
In reply to Jean-Cédric Chappelier

Re: PoS tagging review question not done this morning

by Mike Yan Michelis -

As the HMM can be split into two parts at the entry "you", I thought maybe the condition could be:

p(N|D) * p("cat"|N) * p(P|N) > p(V|D) * p("cat"|V) * p(P|V) 

As this would be what is selected in the "max" during the Viterbi algorithm. Would this be correct?

In reply to Mike Yan Michelis

Re: PoS tagging review question not done this morning

by Jean-Cédric Chappelier -

Exactly!

maybe one minor remark: it's not "the HMM can be split into two parts" but "the lattice", i.e. the unfolding of the HMM paths for this very sentence.