Bài kiểm tra cơ bản về NLP

Câu trả lời và giải thích xuất hiện khi bạn làm bài kiểm tra.

1. An application must assign one category to each whole customer message. Which NLP output type fits that task?
   • A document-level classification label
   • A sequence of new translated sentences
   • Character spans marking names only
   • A transcript of audio that was never supplied
2. In the guide’s sentence “Mina at Northstar Labs said the delayed launch was disappointing,” what could an entity-recognition task mark?
   • The overall sentiment as a single document label
   • Mina as a person and Northstar Labs as an organization
   • Một bản dịch sang tiếng Pháp
   • A summary claiming why the launch was delayed
3. Why might a tokenizer split a written word into more than one token?
   • Token boundaries always match syllables exactly
   • It must first translate every input
   • Token units depend on the tokenizer and can include word pieces
   • Every tokenizer removes all punctuation before processing
4. For “the delayed launch was disappointing,” which output best matches a sentiment task?
   • A span marking “launch” as an organization
   • Bản dịch sang ngôn ngữ khác
   • A list of the sentence’s tokens only
   • A label describing the expressed reaction as negative
5. A text classifier fails on sarcasm and negation common in real user input. What should the team do?
   • Build evaluation examples reflecting language conditions the application encounters
   • Remove context so every phrase is isolated
   • Assume tidy benchmark accuracy covers every style
   • Ask the model to label its own errors without checking
6. Why keep an original passage beside information extracted by an NLP system?
   • To guarantee the extraction is correct without review
   • So a reader can confirm the extracted result in context
   • To convert named entities into new model parameters
   • To remove the need to evaluate the application
7. A long document is silently truncated before processing. What is the main risk named by the guide?
   • The model will automatically recover every omitted paragraph
   • The text becomes a structured database
   • Important sections may be removed from the input
   • Truncation changes the author of the document
8. A team has a narrow text-routing task. What comparison does the guide recommend considering?
   • Only the largest language model available
   • A model that ignores the task definition
   • A model output without any evaluation examples
   • A simple rule or keyword baseline alongside more complex systems