Scribd
Upload
115 views4 pages

NLP Midsem Paper Jan 2024 Regular Exam

The document outlines the mid-semester test details for the M.Tech. in AIML program at Birla Institute of Technology & Science, Pilani, including the course title, exam nature, weightage, duration, and instructions for students. It consists of five questions covering topics such as ambiguity in sentences, neural network architecture for NLP, TF-IDF calculations, skip-gram negative sampling, and the Viterbi algorithm for part-of-speech tagging. Each question has specific tasks and marks assigned, focusing on practical applications of Natural Language Processing concepts.

Uploaded by

Sudip Dey
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
115 views4 pages

NLP Midsem Paper Jan 2024 Regular Exam

The document outlines the mid-semester test details for the M.Tech. in AIML program at Birla Institute of Technology & Science, Pilani, including the course title, exam nature, weightage, duration, and instructions for students. It consists of five questions covering topics such as ambiguity in sentences, neural network architecture for NLP, TF-IDF calculations, skip-gram negative sampling, and the Viterbi algorithm for part-of-speech tagging. Each question has specific tasks and marks assigned, focusing on practical applications of Natural Language Processing concepts.

Uploaded by

Sudip Dey
Copyright
© © All Rights Reserved
We take content rights seriously. If you suspect this is your content, claim it here.
Available Formats
Download as DOCX, PDF, TXT or read online on Scribd
You are on page 1/ 4

Birla Institute of Technology & Science, Pilani

Work Integrated Learning Programmes Division


First Semester 2023-2024
M.Tech. in AIML

Mid-Semester Test
(EC-2 Regular Paper)

Course No. : AIMLCZG530


Course Title : Natural Language Processing
Nature of Exam : Closed Book
Weightage : 30% No. of Pages =3
Duration : 2 Hours No. of Questions = 5
Date of Exam : 21-01-2024_FN
Note to Students:
1. Please follow all the Instructions to Candidates given on the cover page of the answer book.
2. All parts of a question should be answered consecutively. Each answer should start from a fresh page.
3. Assumptions made if any, should be stated clearly at the beginning of your answer.

Question 1. [3+2+2=7 Marks]


A. Find the ambiguity of the below sentences and justify your answer [3 marks]
a) The tank is full of water. I saw a military tank.
b) Before the professor left the stage, the play begins
c) She is looking for a match

B.
i. Given is the following toy corpus. Calculate all the bigram probabilities. [2 marks]
<s> I like apples </s>
<s> Apple is good for health</s>
<s> Apple is in red colour </s>

ii. For above training data in (i), Calculate the probability of below sentence using raw bigram
probabilities and using Laplace smoothing, <s> I am eating apples</s> [2marks]

Question 2. [4+3 = 7 Marks]


A. Study the below neural network designed for learning word embedding along with other NLP
application stated below and answer the following questions.

If the input layer ‘X’ denote the one hot encoding of the vocabulary, “e” is the embedding layer,
“h1”,”h2” are hidden layers and “Y” is the output layer emitting continuous valued output, identify
no more than 3 issues/error in the architecture and suggest modification to suit the below use case
requirement. If there are no corrections required, then mention “No Error” explicitly.
Use Case: Neural network in language modelling for sentence completion
Given a training corpus with below vocabulary each vectorized with four dimensions, and following
test sentence phrase, the neural network, should have predictive ability to identify next best word
to fill in the blank of the test sentence, by analyzing context window with five tokens.
Vocabulary: {he, she, bat, tree, wooden, park, playing, saw, was, a, on, the, with, in, morning,
evening} Test Sentence: “on a morning he was playing in the ________”

B. The number of times each word appears in different documents is given in the table below.
Calculate the TF-IDF value for each term in D1. [1.5 mark]
Find the word embedding for each term using TF-IDF value. Find which words are closest using TF-
IDF word embedding. [0.5 mark]
Which documents are more similar to each other? [0.5 mark]
What is the disadvantage of using the TF-IDF values for the word embedding’s? [0.5 mark]

D1 D2 D3
NLP 10 0 0
is 50 66 89
extremely 20 22 12
interesting 30 32 11
course 20 0 0

Find the word embedding for each term using TF-IDF value.
NLP [0.496,0,0]
is [0,0,0]
extremely [0,0,0]
interesting[0,0,0]
course [0.63,0,0]

Find which words are closest using TF-IDF word embedding. [0.5 mark]
Which documents are more similar to each other? [0.5 mark] D2 and D3
What is the disadvantage of using the TF-IDF values for the word embedding’s? [0.5 mark] -
Sparsity

Question 3. [1.5+3+1.5=6 Marks]


Given a training corpus: “played in the morning”, use the skip-gram negative sampling method and answer
the following: The initial embedding matrix and initial context matrix has dimensions |v| x 3 and is given as
follows:
Note: No need to update or show any weights other than necessary for below questions. Follow only the
approach as discussed in class. i.e., Simplified Skip gram negative sampling with binary classification model.
Round all the calculations to exactly two decimal places.

a) Generate the training dataset for an input target word “played” and context window of 1 next
word and hyper parameter value k=2 for the negative sampling task. Use the information
available in the question.
b) Calculate the error for the above dataset for only the first iteration of skip-gram training, with only
one hidden layer.
c) Explain in no more than 40 words, why skipgram algorithm training was modified from multiclass
to binary classification task.

Question 4. [5 Marks]
Find the appropriate POS tag using statistical model with bigram assumption, for the word “cook” in the
sentence,
“He will cook the food”

Word to Tag combination


1 p(cook/VB) 0.0056
p(cook/
2 NN)
0.0072
3 p(the/DT) 0.014
0.0004
4 P(food/NN)
7

Tag to Tag combination


p(VB/
1 MD)
0.78
p(NN/
2 MD)
0.14
p(DT/ 0.07
3 VB) 5
p(DT/ 0.02
4 NN) 6
Note:
NN- Noun
VB-Verb
MD-Modal
DT-Determiner

Question 5. [5 Marks]
By using Viterbi Algorithm, fill the Viterbi table for the sentence, “He will fight”. The tag transition
probabilities and the word likelihood for this corpus are as follows:

Tag transition probabilities MD NN VB PRP


MD 0.000008 0.31 0.46 0.0056
NN 0.000096 0.209 0.658 0.00068
VB 0.001 0.05 0 0.008
PRP 0.08 0.02 0.001 0.00001
START 0.008 0.000934 0.05677 0.08

h
Word likelihood probabilities e will fight
MD 0 0.8 0
NN 0 0.2 0.4
VB 0 0 0.6
PRP 1 0 0

H FIGH
Viterbi Table E WILL T
NN
VB
MD
PRP
Note:
PRP: PERSONAL PRONOUN
MD:MODAL
VB:VERB BASE FORM
NN:NOUN, SINGULAR OR
MASS

You might also like

pFad - Phonifier reborn

Pfad - The Proxy pFad of © 2024 Garber Painting. All rights reserved.

Note: This service is not intended for secure transactions such as banking, social media, email, or purchasing. Use at your own risk. We assume no liability whatsoever for broken pages.


Alternative Proxies:

Alternative Proxy

pFad Proxy

pFad v3 Proxy

pFad v4 Proxy