Natural Language Processing Named Entity Recognition and Information Extraction 3 — Questions and Answers
Question 1: What is the difference between closed-domain and open-domain question answering?
- Closed-domain QA is restricted to a specific topic or dataset, while open-domain QA answers questions about any topic using a large corpus or the web (Correct answer)
- Closed-domain QA uses neural networks while open-domain QA uses rules
- Open-domain QA requires a knowledge graph while closed-domain does not
- Closed-domain QA generates answers while open-domain extracts spans
Correct answer: Closed-domain QA is restricted to a specific topic or dataset, while open-domain QA answers questions about any topic using a large corpus or the web
Closed-domain systems (e.g., medical QA) operate within a bounded knowledge base, while open-domain systems retrieve from broad corpora like Wikipedia to answer any question.
Question 2: What is extractive question answering?
- Selecting a span of text from a given passage as the answer to a question (Correct answer)
- Generating a new answer sentence not present in the source text
- Ranking a list of candidate answers by relevance
- Classifying whether a question is answerable
Correct answer: Selecting a span of text from a given passage as the answer to a question
Extractive QA models (e.g., BERT fine-tuned on SQuAD) predict the start and end token positions of the answer span within the provided context passage.
Question 3: What is template-based information extraction?
- Using predefined patterns or templates to identify and fill structured information slots in text (Correct answer)
- Using neural networks to learn extraction patterns from data
- Generating summaries from structured databases
- Parsing XML or JSON data formats
Correct answer: Using predefined patterns or templates to identify and fill structured information slots in text
Template-based IE uses hand-crafted or learned patterns (e.g., regular expressions, trigger words) to populate structured records from unstructured text.
Question 4: What is entity linking (also called named entity disambiguation)?
- Connecting entity mentions in text to their corresponding entries in a knowledge base such as Wikipedia (Correct answer)
- Tagging words with their entity types using BIO labels
- Resolving pronouns to their antecedents
- Merging duplicate records in a database
Correct answer: Connecting entity mentions in text to their corresponding entries in a knowledge base such as Wikipedia
Entity linking disambiguates mentions (e.g., 'Apple' as fruit vs. company) by mapping them to canonical knowledge base entries using context.
Question 5: What is the BIO tagging scheme's limitation compared to BIOES?
- BIO cannot distinguish single-token entities from multi-token entity starts, while BIOES adds S (Single) and E (End) tags for clarity (Correct answer)
- BIO uses more tags than BIOES
- BIO cannot label the outside of entity spans
- BIO requires a separate model for each entity type
Correct answer: BIO cannot distinguish single-token entities from multi-token entity starts, while BIOES adds S (Single) and E (End) tags for clarity
BIOES extends BIO with E (End of entity) and S (Single-token entity) tags, providing richer boundary signals that can improve NER model performance.
Question 6: What is the purpose of the OntoNotes corpus in NLP research?
- Providing large-scale annotations for multiple tasks including NER, coreference, semantic roles, and parse trees across multiple genres (Correct answer)
- Storing a list of English ontological concepts and their definitions
- Benchmarking machine translation between major world languages
- Hosting pre-trained word embeddings for download
Correct answer: Providing large-scale annotations for multiple tasks including NER, coreference, semantic roles, and parse trees across multiple genres
OntoNotes is a multi-layer annotated corpus used to train and evaluate models on NER, coreference resolution, SRL, and parsing simultaneously.
Natural Language Processing Named Entity Recognition and Information Extraction 3
What is the difference between closed-domain and open-domain question answering?