NLP ACTUAL EXAM
NLP ACTUAL EXAM 2025/2026 COMPLETE 100 QUESTIONS AND
CORRECT DETAILED ANSWERS (VERIFIED ANSWERS)|ALREADY
GRADED A+||BRAND NEW VERSION!!
When generating text from a language model, which of the following techniques
will likely require the most computational resources and thus be slowest to
generate text?
1 - greedy sampling
2 - impossible to answer this question with the information provided
3 - top-k sampling
4 - beam search
5 - random sampling
4
The fact that the exclamation mark '!' can denote a factorial (4! = 123*4), the
question mark '?' can indicate a missing value (2, 4, ?, 16, 32), and the period '.'
can be a decimal point (4.56), complicates which NLP task?
1 - NONE of the other answers are correct.
2 - Classification
3 - Clustering
4 - Sentence segmentation
5 - POS tagging
6 - Named Entity Extraction
4
The task of determining who or what is being referred to by a pronoun in a
sentence is called:
1 - Named entity recognition
2 - Unit resolutions
3 - Part-of-speech tagging
4 - Co-reference resolution
1|Page
, NLP ACTUAL EXAM
5 - NONE of the other answers are correct
6 - De-pronounification
4
The process of aligning words to a common reference dictionary (e.g. by removing
punctuation), so as to ensure consistent spelling/formatting throughout the
corpus is referred to as:
1 - Word encoding
2 - Word de-punctuation-ification
3 - Word embedding
4 - Word normalisation
5 - Word cleanup
6 - NONE of the other options are correct
4
Which statement about the limitations of Ngram language models is NOT correct:
1 - the expected number of occurrences of an n-gram decreases linearly with the
length of the ngram
2 - the training corpus is never big enough to estimate high-order ngrams well
3 - NONE of the other answers is correct
4 - predictive performance of an ngram model depends greatly on the maximum
ngram length (order of the model)
5 - storing counts for high-order ngrams requires massive amounts of memory
1
GloVE embeddings
1 - represent topics as numerical vectors
2 -NONE of the other answers are correct
3 - represent document as numerical vectors
4 - represent images as numerical vectors
5 - provide a sparse representation
6 - represent words as numerical vectors
7 - provide a low dimensional approximation of a high dimensional representation
2|Page
, NLP ACTUAL EXAM
6
The Mel Spectrogram is just a spectrogram which has:
1 - a logarithmic time axis, logarithmic frequency axis and logarithmic amplitude
scale
2 - a logarithmic time axis, linear frequency axis and linear amplitude scale
3 - a linear time axis, logarithmic frequency axis and logarithmic amplitude scale
4 - None of the other options are correct
5 - a logarithmic time axis, linear frequency axis and logarithmic amplitude scale
6 - a linear time axis, logarithmic frequency axis and linear amplitude scale
7 - a linear time axis, linear frequency axis and logarithmic amplitude scale
8 - a logarithmic time axis, logarithmic frequency axis and linear amplitude scale
3
In order to improve the probability estimates for an n-gram language model we
could:
1 - interpolate the higher-order probability estimates with lower-order estimates
2 - ALL of the other options are correct
3 - smooth all probability estimates by adding a small constant to the word counts
4 - back-off the estimator to use lower-order n-grams whenever counts for higher-
order n-grams are zero
2
A statistical language model computes:
1 - a probability distribution over sequences of words
2 - a probability distribution over languages for a piece of text
3 - NONE of the other options are correct
4 - a probability distribution over emotions (happiness, sadness, anger, etc.)
5 - a probability distribution over grammatical classes (verbs, nouns, etc.)
6 - statistics like the average document length, the average vocabulary size, etc.
1
3|Page
, NLP ACTUAL EXAM
When generating text from a language model with top-k sampling, setting the
value of k to the size of the vocabulary would be equivalent to performing:
1 - random sampling
2 - greedy sampling
3 - t-SNE dimensionality reduction
4 - beam search
5 - NONE of the other answers is correct
1
Which, if any, of the following techniques is NOT used to produce a spectrogram
for analysing audio signals to understand spoken language:
1 - All of the other techniques are used
2 - Fast Fourier Transform (FFT)
3 - Hamming windows on overlapping time-segments
4 - k-Means clustering
5 - Pre-emphasis filtering
4
Machine translation is an example of what type of problem?
1 - classification problem
2 - clustering problem
3 - reinforcement learning problem
4 - regression problem
5 - sequence-to-sequence problem
6 - sequence-labeling problem
7 - NONE of the other answers are correct
8 - masked-language modelling problem
9 - syntactic parsing problem
5
Which of the following prompts to a language model would be considered an
example of one-shot learning?
1 - "apple => fruit, broccoli => vegetable, eggplant =>"
4|Page
NLP ACTUAL EXAM 2025/2026 COMPLETE 100 QUESTIONS AND
CORRECT DETAILED ANSWERS (VERIFIED ANSWERS)|ALREADY
GRADED A+||BRAND NEW VERSION!!
When generating text from a language model, which of the following techniques
will likely require the most computational resources and thus be slowest to
generate text?
1 - greedy sampling
2 - impossible to answer this question with the information provided
3 - top-k sampling
4 - beam search
5 - random sampling
4
The fact that the exclamation mark '!' can denote a factorial (4! = 123*4), the
question mark '?' can indicate a missing value (2, 4, ?, 16, 32), and the period '.'
can be a decimal point (4.56), complicates which NLP task?
1 - NONE of the other answers are correct.
2 - Classification
3 - Clustering
4 - Sentence segmentation
5 - POS tagging
6 - Named Entity Extraction
4
The task of determining who or what is being referred to by a pronoun in a
sentence is called:
1 - Named entity recognition
2 - Unit resolutions
3 - Part-of-speech tagging
4 - Co-reference resolution
1|Page
, NLP ACTUAL EXAM
5 - NONE of the other answers are correct
6 - De-pronounification
4
The process of aligning words to a common reference dictionary (e.g. by removing
punctuation), so as to ensure consistent spelling/formatting throughout the
corpus is referred to as:
1 - Word encoding
2 - Word de-punctuation-ification
3 - Word embedding
4 - Word normalisation
5 - Word cleanup
6 - NONE of the other options are correct
4
Which statement about the limitations of Ngram language models is NOT correct:
1 - the expected number of occurrences of an n-gram decreases linearly with the
length of the ngram
2 - the training corpus is never big enough to estimate high-order ngrams well
3 - NONE of the other answers is correct
4 - predictive performance of an ngram model depends greatly on the maximum
ngram length (order of the model)
5 - storing counts for high-order ngrams requires massive amounts of memory
1
GloVE embeddings
1 - represent topics as numerical vectors
2 -NONE of the other answers are correct
3 - represent document as numerical vectors
4 - represent images as numerical vectors
5 - provide a sparse representation
6 - represent words as numerical vectors
7 - provide a low dimensional approximation of a high dimensional representation
2|Page
, NLP ACTUAL EXAM
6
The Mel Spectrogram is just a spectrogram which has:
1 - a logarithmic time axis, logarithmic frequency axis and logarithmic amplitude
scale
2 - a logarithmic time axis, linear frequency axis and linear amplitude scale
3 - a linear time axis, logarithmic frequency axis and logarithmic amplitude scale
4 - None of the other options are correct
5 - a logarithmic time axis, linear frequency axis and logarithmic amplitude scale
6 - a linear time axis, logarithmic frequency axis and linear amplitude scale
7 - a linear time axis, linear frequency axis and logarithmic amplitude scale
8 - a logarithmic time axis, logarithmic frequency axis and linear amplitude scale
3
In order to improve the probability estimates for an n-gram language model we
could:
1 - interpolate the higher-order probability estimates with lower-order estimates
2 - ALL of the other options are correct
3 - smooth all probability estimates by adding a small constant to the word counts
4 - back-off the estimator to use lower-order n-grams whenever counts for higher-
order n-grams are zero
2
A statistical language model computes:
1 - a probability distribution over sequences of words
2 - a probability distribution over languages for a piece of text
3 - NONE of the other options are correct
4 - a probability distribution over emotions (happiness, sadness, anger, etc.)
5 - a probability distribution over grammatical classes (verbs, nouns, etc.)
6 - statistics like the average document length, the average vocabulary size, etc.
1
3|Page
, NLP ACTUAL EXAM
When generating text from a language model with top-k sampling, setting the
value of k to the size of the vocabulary would be equivalent to performing:
1 - random sampling
2 - greedy sampling
3 - t-SNE dimensionality reduction
4 - beam search
5 - NONE of the other answers is correct
1
Which, if any, of the following techniques is NOT used to produce a spectrogram
for analysing audio signals to understand spoken language:
1 - All of the other techniques are used
2 - Fast Fourier Transform (FFT)
3 - Hamming windows on overlapping time-segments
4 - k-Means clustering
5 - Pre-emphasis filtering
4
Machine translation is an example of what type of problem?
1 - classification problem
2 - clustering problem
3 - reinforcement learning problem
4 - regression problem
5 - sequence-to-sequence problem
6 - sequence-labeling problem
7 - NONE of the other answers are correct
8 - masked-language modelling problem
9 - syntactic parsing problem
5
Which of the following prompts to a language model would be considered an
example of one-shot learning?
1 - "apple => fruit, broccoli => vegetable, eggplant =>"
4|Page
