Question VI hands on PoS

Question VI hands on PoS

by Naomie Mathilde Pont -
Number of replies: 1

Good afternoon,

I was trying to solve the last question of the hands on for this week, but I am struggling to understand why you ticked those answers. Both for parts a and b.

Indeed, when we apply the hypothesis "limited lexical conditioning" and "limited scope for syntactic dependencies (1 neighbor)", we should have that P(t4|t1,t2,t3)=P(t4|t3).

So why is the proposition "the tagging of fast depends on the one of young" ticked ? (where fast has tag t4 and young t1)

Finally, to solve part b, I just don't get at all how we should think.

Thank you so much for your explanations!



In reply to Naomie Mathilde Pont

Re: Question VI hands on PoS

by Martin Rajman -
While it is indeed true that, for an order 1 HMM, we have:
P(t4|t1,t2,t3)=P(t4|t3)
this is by no means sufficient to guarantee that the optimal tagging for "fast" (at position 4) does not depend on the optimal one for "young" (at position 1).

Indeed, as I explained during the review, due to the global nature of the performed optimization, there is an important difference between saying "tag t1 can be removed from the conditioning of t4" and saying "the optimal tagging at position t1 does not depend on the optimal one at position 4".

While the former assertion is indeed a direct consequence of "limited scope for syntactic dependencies (1 neighbor)" hypothesis, the latter can only be guaranteed (for an order k HMM) if there are at least k non ambiguous words between position 1 and position 4.

Therefore, as, in question VI, all words are considered as ambiguous, all 4 answers are correct, while, in question VI, as "birds" (at position 2) is the only non ambiguous word, the optimal taggings are dependent on each other only for the pairs of positions that are on the same side wrt. position 2; thus only answers 1 and 2 are correct.