Language Modeling DATA-SCIENCE-OPTIMIZE Practice Questions (Page 2)

Questions 6–10

1. 6expert · hard

   A team is choosing between an RNN and a transformer for a language model that must process long documents. The transformer achieves lower perplexity but requires significantly more memory. The deployment environment has limited GPU memory. Which model should the team choose?

   • AThe RNN, because transformers are never suitable for long documents
   • BThe transformer, but only if the team reduces the sequence length
   • CThe RNN, because it uses less memory and can still handle long documents
   • DThe transformer, because lower perplexity is the only metric that matters
   Reveal Solution Discussion Select an answer first
2. 7foundation · easy

   Which of the following is a real-world application of language models?

   • ADatabase indexing
   • BImage classification
   • CMachine translation
   • DNetwork routing
   Reveal Solution Discussion Select an answer first
4. 8application · medium

   A startup is developing a voice assistant that must transcribe user speech into text. The system uses an acoustic model and a language model to improve accuracy. Which role does the language model play in this system?

   • AIt filters out background noise from the audio
   • BIt generates the final audio output
   • CIt converts audio signals into phonemes
   • DIt assigns probabilities to word sequences to help disambiguate similar-sounding phrases
   Reveal Solution Discussion Select an answer first
5. 9application · medium

   A developer needs to build a text generation system that produces coherent paragraphs for a chatbot. The system must capture long-range dependencies and handle variable-length input. Which architecture is most suitable?

   • AA bigram n-gram model with Laplace smoothing
   • BA feedforward neural network with a fixed input window
   • CA support vector machine classifier
   • DA transformer-based language model
   Reveal Solution Discussion Select an answer first
6. 10expert · hard

   A data scientist is building a trigram model for a language with a very large vocabulary (1 million words). The model will be used for offline batch processing, so speed is not a concern. The scientist wants to minimize perplexity. Which smoothing technique should be used?

   • ANo smoothing
   • BLaplace smoothing
   • CKneser-Ney smoothing
   • DBackoff to unigrams only
   Reveal Solution Discussion Select an answer first