Assume that you have collected the data in the following tables, and assume that all other observed counts are 0. In the bigram table, rows represent w-1, columns represent w eg, C(computer, keyboard) = 2. C(x,-1,20) computer keyboard monitor store computer keyboard monitor store 0 0 Table 1: Bigram frequency. Rows = w-1, columns = w. 10 3 | computer keyboard monitor store Table 2: Unigram frequency. Consider the following sentence fragment S: "I shopped at the computer You need to determine whether the sentence is more likely to end with computer store" or "computer monitor." () Compute the raw bigram probabilities for the candidate words {store, monitor) to complete the sentence S. i.e. P(store computer) and P(monitor computer). IN one word more likely than the other, and if so which one? 2 pts (b) Compute the kneser-Ney smoothed bigram probability of the candidate words {store, monitor to complete the sentence. Use d = 0.5 as the discount term. Is one word more likely than the other, and if so which one? If the result has changed, why do you think it changed? 5 pts 2 of 6 Homework 3: Language modeling Deadline: Feb 17th, 3:00pm CS 4650/7650 Natural Language Understanding (c) Change the discount term to d = 0.1 and re-compute the Kneser-Ney smoothed bigram probability of the candidate words {store, monitor} to complete the sentence. Is one word more likely than the other, and if so which one? If the result has changed, why do you think it changed? (3 pts] Assume that you have collected the data in the following tables, and assume that all other observed counts are 0. In the bigram table, rows represent w-1, columns represent w eg, C(computer, keyboard) = 2. C(x,-1,20) computer keyboard monitor store computer keyboard monitor store 0 0 Table 1: Bigram frequency. Rows = w-1, columns = w. 10 3 | computer keyboard monitor store Table 2: Unigram frequency. Consider the following sentence fragment S: "I shopped at the computer You need to determine whether the sentence is more likely to end with computer store" or "computer monitor." () Compute the raw bigram probabilities for the candidate words {store, monitor) to complete the sentence S. i.e. P(store computer) and P(monitor computer). IN one word more likely than the other, and if so which one? 2 pts (b) Compute the kneser-Ney smoothed bigram probability of the candidate words {store, monitor to complete the sentence. Use d = 0.5 as the discount term. Is one word more likely than the other, and if so which one? If the result has changed, why do you think it changed? 5 pts 2 of 6 Homework 3: Language modeling Deadline: Feb 17th, 3:00pm CS 4650/7650 Natural Language Understanding (c) Change the discount term to d = 0.1 and re-compute the Kneser-Ney smoothed bigram probability of the candidate words {store, monitor} to complete the sentence. Is one word more likely than the other, and if so which one? If the result has changed, why do you think it changed? (3 pts]
