CS 563: Natural Language Processing
Assignment-1: Part-of-Speech Tagging
Deadline: 15 February 2024
● Markings will be based on the correctness and soundness of the outputs.
● Marks will be deducted in case of plagiarism.
● Proper indentation and appropriate comments (if necessary) are mandatory.
● Use of frameworks like scikit-learn etc is not allowed.
● All benchmarks(accuracy etc), answers to questions and supporting examples
should be added in a separate file with the name ‘report’.
● All code needs to be submitted in ‘.py’ format. Even if you code it in ‘.IPYNB’
format, download it in ‘.py’ format and then submit
● You should zip all the required files and name the zip file as:
○ <roll_no>_assignment_<#>.zip, eg. 1501cs11_assignment_01.zip.
● Upload your assignment ( the zip file ) in the following link:
○ CS563-NLP-2024-Assignments
Problem Statement:
● The assignment targets to implement Hidden Markov Model (HMM) to perform
Part-of-Speech (PoS) tagging task
Implementation:
HMM based Model:
● HMM Parameter Estimation
○ Input: Annotated tagged dataset
○ Output: HMM parameters
○ Procedure:
■ Step1: Find states.
■ Step2: Calculate Start probability (π).
■ Step3: Calculate transition probability (A)
■ Step4: Calculate emission probability (B)
● Features for HMM:
○ Train two HMM models based on:
■ First order markov assumption (Bigram) where current word PoS
tag is based on the previous and current words
■ Second order markov assumption (Trigram) where current word
PoS tag is based on the previous two words along with the current
word
Testing:
● After calculating all these parameters use these parameters to tag the test input
sequence using the Viterbi algorithm
Dataset:
● Dataset consists of sentences and each word is tagged with its corresponding
PoS tag
● Brown dataset: Brown_train.txt
● Format of dataset:
○ Each line contains <Word/Tag> (word followed by ‘/’ and tag)
○ Sentences are separated by a new line
Documents to submit:
● Model code
● Perform 5 fold cross-validation on the Training dataset and report both average &
individual fold results (Accuracy, Precision, Recall and F-Score).
● Create a confusion matrix using Python Library.
● Briefly discuss Bigram vs Trigram assumption while training HMMs.
● With some examples (good pairs and bad pairs) why the model is confused and
when it is giving correct results. Analyze and Explain the reason behind it.
● Also, Implement a RNN based model for this task and compare the result of both
RNN and HMM.
● Discuss which model is better? With some justification and analysis when RNN is
better than HMM and when HMM is better than RNN and when both fail and
why?
● Write a report (doc or pdf format) on how you are solving the problems as well as
all the results including model architecture (if any).
For any queries regarding this assignment, contact:
Ramakrishna Appicharla (ramakrishnaappicharla@gmail.com),
Arpan Phukan (arpanphukan@gmail.com),
Sandeep Kumar (sandeep.kumar82945@gmail.com), and,
Aizan Zafar (aizanzafar@gmail.com)
