Natural Language Processing (NLP) AIE Practice Questions (Page 3)

Questions 11–15

1. 11application · easy

   A data scientist is preparing a dataset of product reviews for a sentiment analysis model. The reviews contain many words like 'running', 'ran', and 'runs'. To reduce the vocabulary size while preserving the core meaning, which technique should be applied?

   • ATokenization
   • BStemming
   • CStop-word removal
   • DPart-of-speech tagging
   Reveal Solution Discussion Select an answer first
2. 12foundation · easy

   Which NLP task is used to determine whether a product review expresses a positive, negative, or neutral opinion?

   • ANamed entity recognition
   • BMachine translation
   • CPart-of-speech tagging
   • DSentiment analysis
   Reveal Solution Discussion Select an answer first
4. 13expert · hard

   A company is developing a multilingual chatbot for customer service. They need to support 10 languages, but they only have high-quality training data for English. They are considering: (1) training a single multilingual neural language model, (2) training separate models for each language, (3) using a translation service to convert all queries to English, (4) using a rule-based system with language-specific rules. Which approach is most likely to provide good performance across all languages with the least development effort?

   • AUse a translation service to convert all queries to English before processing.
   • BUse a rule-based system with language-specific rules.
   • CTrain a single multilingual neural language model on available multilingual data.
   • DTrain separate models for each language using only English data.
   Reveal Solution Discussion Select an answer first
5. 14expert · hard

   A social media monitoring tool analyzes posts for hate speech. The tool incorrectly flags benign posts that use sarcasm or irony. Which NLP challenge is this, and what is a potential mitigation?

   • AContextual ambiguity; use a model trained on sarcasm-annotated data.
   • BTokenization ambiguity; use a different tokenizer.
   • CNamed entity recognition failure; use a larger NER model.
   • DLanguage model overfitting; use a simpler model.
   Reveal Solution Discussion Select an answer first
6. 15expert · hard

   A company is building a language model for a specialized domain (legal documents). They have a large general English corpus and a small legal corpus. Which training strategy is most effective?

   • APre-train on the general corpus and fine-tune on the legal corpus.
   • BUse a rule-based system with hand-crafted legal rules.
   • CTrain only on the general corpus and ignore the legal corpus.
   • DTrain only on the legal corpus from scratch.
   Reveal Solution Discussion Select an answer first