1
A language model assigns
Correct Answer
a probability to a sequence of words
☑ MCQ PRACTICE
Natural Language Processing Unit 5
Practice objective questions for quick revision and examination preparation. Try answering each question before revealing the answer.
Natural Language Processing - Unit-5
2
A bigram model predicts a word based on
Correct Answer
the previous one word
3
A trigram model predicts a word based on
Correct Answer
the previous two words
4
The Markov assumption in n-gram models states that
Correct Answer
the next word depends only on a limited history
5
N-gram probabilities are usually estimated from
Correct Answer
relative frequency counts in a training corpus
6
Perplexity is used to evaluate language models; a model with
Correct Answer
lower perplexity is better
7
Perplexity is related to cross-entropy H as
Correct Answer
2 raised to the power H
8
The data sparsity problem in n-gram models means that
Correct Answer
many valid n-grams never occur in the training data
9
Smoothing is used to
Correct Answer
give some probability mass to unseen n-grams
10
Add-one (Laplace) smoothing
Correct Answer
adds one to every n-gram count
11
Good–Turing smoothing re-estimates counts using
Correct Answer
the number of n-grams that occur a given number of times
12
Kneser–Ney smoothing is based on
Correct Answer
continuation probability, how many different contexts a word follows
13
In backoff, if a higher-order n-gram is not observed, the model
Correct Answer
falls back to a lower-order n-gram
14
Interpolation combines
Correct Answer
estimates from different n-gram orders using weights
15
Class-based language models
Correct Answer
group words into classes to reduce data sparsity
16
Variable-length n-gram models
Correct Answer
vary the length of the context depending on the data
17
LDA, used in Bayesian topic-based language models, stands for
Correct Answer
Latent Dirichlet Allocation
18
Language model adaptation is the process of
Correct Answer
tuning a model to a new domain, topic or speaker
19
A cache language model
Correct Answer
increases the probability of recently used words
20
Cross-lingual language modeling
Correct Answer
uses data or models from other languages to help the target language
21
Words that are not in the training vocabulary are called
Correct Answer
out-of-vocabulary (OOV) words
22
Bayesian parameter estimation for language models makes use of
Correct Answer
a prior distribution over the model parameters
23
Which of the following is an application of language models?
Correct Answer
Speech recognition and machine translation
24
Multilingual language modeling aims to
Correct Answer
build language models that can handle several languages
25
An n-gram model with a larger n generally
Correct Answer
needs more training data and suffers more from sparsity
Fill in the Blanks
26
A bigram model conditions on the previous __________ word(s).
Correct Answer
one
27
A trigram model conditions on the previous __________ words.
Correct Answer
two
28
__________ is the standard measure for evaluating language models; lower is better.
Correct Answer
Perplexity
29
Perplexity equals 2 raised to the power of the __________ entropy.
Correct Answer
cross
30
__________ smoothing adds one to every n-gram count.
Correct Answer
Laplace
31
Kneser–__________ smoothing is based on continuation probability.
Correct Answer
Ney
32
In __________–Turing smoothing, counts are re-estimated using counts of counts.
Correct Answer
Good
33
Falling back to a lower-order n-gram when the higher-order one is unseen is called __________.
Correct Answer
backoff
34
Combining estimates of different n-gram orders with weights is called __________.
Correct Answer
interpolation
35
LDA stands for Latent __________ Allocation.
Correct Answer
Dirichlet
36
__________-based language models group words into classes.
Correct Answer
Class
37
Words that are not in the training vocabulary are called out-of-__________ (OOV) words.
Correct Answer
vocabulary
38
The Markov assumption says the next word depends only on a limited __________.
Correct Answer
history
39
Language model adaptation tunes a model to a new __________.
Correct Answer
domain
40
A __________ language model increases the probability of recently used words.
Correct Answer
cache