A friend of yours had a peek at this final exam a priori and she sent you a brief text (encoded for privacy) to help you allot appropriate time needed for preparation. Your decision will be based on the top 2 decoded output, so you choose to apply 'beam search' setting beam width, K = 2. You have the following information: Given Vocab, V = {trivial, hard, super, }, a decoder σ, K = 2, and the beam table 1 at time t = 1. k blk (k) P b 1 [, super] 0.7 2 [, trivial] 0.3 1007 Table 1: Here, b is the k-th path sequence, and P (b()) is the probability of the path sequence at time t. Additionally, you know the following bigram probabilities: P(|hard) = P( \trivial) = 1 (d) aboodiboP(|super) = 0 P(trivial super) = 0.6 P(hard super) = 0.4 (2) (4) Recall that a hypothesized path is marked completed when an end-of-sequence token ([/s]) is produced. The stopping criteria of your beam search is: i) cutoff time step t = 3 or, ii) cutoff completed number of hypotheses (n = 2) has been reached. la) Write down the values of (6,11,622)), i.e. the top two path sequences decoded at the completion of your beam search at time t. 1b) Repeat the above answer when the bigram probability of observing either trivial, hard following super is equal to a fair coin toss (i.e. P(trivial|super) P(hard super) = 0.5). How would you make a (time allotment) decision now? boodilsdil naixem eda entrad = neve the

A friend of yours had a peek at this final exam a priori and she sent you a brief text (encoded for privacy) to help you allot appropriate time needed for preparation. Your decision will be based on the top 2 decoded output, so you choose to apply 'beam search' setting beam width, K = 2. You have the following information: Given Vocab, V = {trivial, hard, super, }, a decoder σ, K = 2, and the beam table 1 at time t = 1. k blk (k) P b 1 [, super] 0.7 2 [, trivial] 0.3 1007 Table 1: Here, b is the k-th path sequence, and P (b()) is the probability of the path sequence at time t. Additionally, you know the following bigram probabilities: P(|hard) = P( \trivial) = 1 (d) aboodiboP(|super) = 0 P(trivial super) = 0.6 P(hard super) = 0.4 (2) (4) Recall that a hypothesized path is marked completed when an end-of-sequence token ([/s]) is produced. The stopping criteria of your beam search is: i) cutoff time step t = 3 or, ii) cutoff completed number of hypotheses (n = 2) has been reached. la) Write down the values of (6,11,622)), i.e. the top two path sequences decoded at the completion of your beam search at time t. 1b) Repeat the above answer when the bigram probability of observing either trivial, hard following super is equal to a fair coin toss (i.e. P(trivial|super) P(hard super) = 0.5). How would you make a (time allotment) decision now? boodilsdil naixem eda entrad = neve the