N-Grams Smoothing
1. For the following corpus, manually compute the bigram probabilities for all word pairs using Add-One Smoothing. Compare your results with the simulation output.
(eos) I booked a flight (eos) I took a flight (eos)
2. Explain why smoothing is necessary in N-gram models. Provide examples from the experiment where smoothing changes the probability from zero to a non-zero value.
3. Run the simulation with at least two different corpora. Observe and report how the bigram probability distributions change with and without smoothing.
4. Given the following bigram counts, fill in the missing probabilities using Add-One Smoothing:
| Previous Word | Next Word | Count |
|---|---|---|
| I | booked | 3 |
| I | took | 2 |
| booked | a | 3 |
| took | a | 2 |
| a | flight | 5 |
| flight | eos | 5 |
- Calculate ( P(\text{booked} | I) ), ( P(\text{took} | I) ), and ( P(\text{flight} | a) ) using Add-One Smoothing (assume vocabulary size ( V = 6 )).
5. Describe a real-world NLP application where N-gram smoothing would be essential. Explain how zero probabilities could affect the application and how smoothing resolves this.