Bird
0
0

You have a bigram model but get zero probability for some word pairs in test data. What is the best way to fix this?

medium📝 Debug Q6 of 15
NLP - Text Generation
You have a bigram model but get zero probability for some word pairs in test data. What is the best way to fix this?
ARemove all zero probability pairs from the test data
BUse only unigram probabilities
CIncrease the size of the test data
DApply smoothing techniques like Laplace smoothing
Step-by-Step Solution
Solution:

  1. Step 1: Identify zero probability problem

    Zero probabilities occur when unseen word pairs appear in test data.

  2. Step 2: Apply smoothing

    Smoothing like Laplace adds small counts to all pairs to avoid zero probabilities.

  3. Final Answer:

    Apply smoothing techniques like Laplace smoothing -> Option D

  4. Quick Check:

    Smoothing fixes zero probs = B [OK]
Quick Trick: Use smoothing to handle unseen pairs [OK]
Common Mistakes:
MISTAKES
  • Removing test data pairs
  • Ignoring zero probabilities
  • Switching to unigram only

Want More Practice?

15+ quiz questions · All difficulty levels · Free

Free Signup - Practice All Questions
More NLP Quizzes
Sentiment Analysis Advanced - Aspect-based sentiment analysis - Quiz 8hard Sentiment Analysis Advanced - Fine-grained sentiment (5-class) - Quiz 9hard Sequence Models for NLP - Why sequence models understand word order - Quiz 3easy Sequence Models for NLP - LSTM for text - Quiz 5medium Sequence Models for NLP - LSTM for text - Quiz 3easy Sequence Models for NLP - Why sequence models understand word order - Quiz 6medium Text Generation - Temperature and sampling - Quiz 8hard Text Generation - Evaluating generated text (BLEU, ROUGE) - Quiz 1easy Text Similarity and Search - Edit distance (Levenshtein) - Quiz 8hard Word Embeddings - Why embeddings capture semantic meaning - Quiz 2easy