BERTScore and Semantic Evaluation Quiz

Answers and explanations appear as you take the quiz.

1. What core weakness of BLEU and ROUGE does BERTScore address?
   • They penalize valid paraphrases that use different words
   • They run too slowly on long documents
   • They require human annotators for every score
   • They only work on translation, not summarization
2. How does BERTScore decide that two tokens are similar?
   • By counting shared characters
   • By cosine similarity of their contextual embeddings
   • By checking if they appear in the same dictionary entry
   • By exact string matching
3. What does it mean that BERTScore embeddings are 'contextual'?
   • They are computed only once per vocabulary word
   • They ignore the surrounding words entirely
   • The same word gets different vectors depending on its sentence
   • They are based on word frequency counts
4. Which three values does BERTScore typically report?
   • BLEU, ROUGE, and METEOR
   • Accuracy, recall, and perplexity
   • Bias, variance, and loss
   • Precision, recall, and F1
5. What is the purpose of optional IDF weighting in BERTScore?
   • To downweight very common, less informative words
   • To speed up embedding computation
   • To convert scores into percentages
   • To translate text before scoring
6. Why is baseline rescaling sometimes applied to BERTScore?
   • To remove the need for a reference text
   • To make raw scores spread across a more interpretable range
   • To convert it into BLEU for comparison
   • To eliminate the embedding model entirely
7. In BERTScore, which direction corresponds to recall?
   • For each candidate token, find its best-matching reference token
   • Compare sentence lengths only
   • For each reference token, find its best-matching candidate token
   • Count exact matching n-grams
8. Why can IDF weighting change BERTScore’s token contribution?
   • It replaces contextual embeddings with word counts
   • It removes the reference text
   • It forces every token to have identical similarity
   • Common words receive less weight than more informative words