NEW
Font size
WorksheetsNatural Language Processing Quiz
Total questions: 30
Worksheet time: 15mins
Which of the following best describes the main challenge in Natural Language Understanding (NLU)?
Converting speech to text
POS tagging
Tokenizing sentences
Mapping words to structured representations capturing meaning
Which of the following is not a core phase of Natural Language Processing?
Morphological Analysis
Phoneme segmentation
Syntax Analysis
Pragmatic Analysis
The Noisy Channel Model in NLP is primarily used in:
POS tagging
Word sense disambiguation
Spelling correction and speech recognition
Dependency parsing
The main assumption in the Noisy Channel Model is:
Noise can be removed completely by probabilistic inference
Observed data is generated by corruption of a true intended message
Every channel introduces equal noise
Noise is language independent
Which part-of-speech category contributes most to the syntactic structure of English sentences?
Nouns and verbs
Adjectives and adverbs
Prepositions and conjunctions
Pronouns and determiners
A grammar that generates exactly the sentences of a language is known as:
Descriptive grammar
Formal grammar
Prescriptive grammar
Cognitive grammar
Which of the following is a disadvantage of simple N-gram models?
They are non-parametric
They cannot be smoothed
They are not probabilistic
They suffer from data sparsity and lack of generalization
The purpose of smoothing in language models is to:
Increase accuracy by removing stop words
Reassign probabilities to unseen n-grams
Reduce model size
Improve parsing efficiency
Perplexity of a language model is:
The log of word likelihood
The inverse probability of the test set normalized by word count
The measure of ambiguity
The ratio of observed to expected frequencies
Which of the following is not true for neural language models?
They map words to continuous vector spaces
They can capture long-range dependencies
They require fewer parameters than N-grams
They can generalize across similar contexts
A key limitation of feed-forward neural language models is:
They cannot model word sequences of variable length
They cannot process bigrams
They cannot use embeddings
They are unsupervised
Which neural architecture most effectively replaces the N-gram assumption?
RNN-based models
CNN-based models
Naïve Bayes models
CRF-based models
POS tagging involves assigning:
Lemmas to words
Grammatical categories to words
Parse trees to sentences
Dependency arcs between words
Rule-based POS tagging relies primarily on:
Statistical probabilities
Lexical rules and contextual constraints
Machine learning
Embedding spaces
The Hidden Markov Model assumes that:
Words depend on their POS tags
Tags depend only on the previous tag
Words depend on previous words
Tags depend on all previous tags
In an HMM-based POS tagger, emission probabilities represent:
Transition between tags
Probability of word given a tag
Probability of tag given word
Probability of a sequence
A common issue in neural POS tagging is:
Sparse matrix representation
Overfitting on rare words
Lack of parallelism
Fixed sequence length
The major advantage of neural POS tagging over rule-based systems is:
Reduced interpretability
Automatic feature learning
Requirement of large rule sets
Dependency on lexicons
Bottom-up parsing differs from top-down parsing by:
Constructing parse tree from input words upward
Using probabilistic inference
Needing grammar probabilities
Being deterministic
The CKY algorithm is applicable only to grammars in:
Regular form
Chomsky Normal Form
Dependency form
Context-sensitive form
A “treebank” is used for:
Storing lexical data
Annotated corpora with syntactic parse trees
Neural embeddings
Sentence segmentation
The sum of probabilities of all production rules expanding a non-terminal in PCFG must be:
Zero
Less than one
Exactly one
Greater than one
Probabilistic CKY parsing extends CKY by:
Adding backtracking
Multiplying probabilities along parse paths
Removing ambiguity
Adding syntax trees manually
Which parsing approach is most suitable for ambiguous sentences?
Deterministic parsing
Probabilistic parsing
Transition-based parsing
Left-corner parsing
Which of the following is not a limitation of CFGs?
Inability to capture long-distance dependencies
Inability to model hierarchical structure
Inability to represent recursion
Inability to handle ambiguity
Vector semantics is based on which linguistic hypothesis?
Rule-based grammar
Distributional hypothesis
Dependency grammar
Chomsky's hierarchy
In vector space models, similarity between words is most commonly measured using:
Euclidean distance
Cosine similarity
Jaccard coefficient
Pearson correlation
Singular Value Decomposition (SVD) is used in LSA to:
Reduce dimensionality of term-document matrices
Increase vocabulary size
Normalize probabilities
Create new features manually
The concept of “word sense” refers to:
POS category
Context-dependent meaning variations of a word
Morphological form
Speech phoneme
Dense vector embeddings are preferred over one-hot vectors because they:
Are binary
Capture similarity and reduce dimensionality
Are sparse
Represent words independently
