Text Embeddings Quiz

Answers and explanations appear as you take the quiz.

1. What does a text embedding supply for a language comparison task?
   • A required spoken audio recording
   • A verbatim dictionary entry for every coordinate
   • A numerical vector representation of text
   • Only a grammar-repair instruction
2. In a representation trained to capture synonym similarity, what relationship is desirable for happy and joyful?
   • A relatively high similarity under the chosen scoring method
   • A mandatory identical vector for every word
   • A guarantee they are synonyms in every possible context
   • A requirement that neither word can be encoded
3. What was a key limitation of early word embeddings like Word2Vec and GloVe?
   • They could not be stored on a computer
   • They only worked for images
   • They required no training data
   • Each word had a single fixed vector regardless of context, so 'bank' meant the same thing everywhere
4. How do modern contextual embeddings improve on static word embeddings?
   • They remove all numbers from the representation
   • They give a word a different vector depending on its surrounding sentence
   • They only embed the first letter of each word
   • They make every word's vector identical
5. Which documented default similarity method does Sentence Transformers use for embedding comparison?
   • Cosine similarity
   • Alphabetical order
   • Document creation date
   • Character count alone
6. What role can compatible text vectors play in a dense-retrieval RAG pipeline?
   • Guaranteeing every generated claim is correct
   • Eliminating the source corpus
   • Ranking candidate passages before they are supplied as context
   • Training a new language model on every query
7. Which pooling strategy is the stated default in the original Sentence-BERT paper?
   • A mandatory maximum over every possible vocabulary item
   • Mean pooling of output token vectors
   • Concatenation of all corpus documents
   • Randomly selecting an unchanged input byte
8. Which Sentence-BERT design allows stored sentences to be encoded before a new comparison sentence arrives?
   • Mandatory joint processing of every possible pair
   • Rebuilding encoder weights for each candidate
   • Encoding only the final comparison score
   • Independent sentence encoding through its siamese architecture