Subject Code: KAI052
Subject Name: Natural Language Processing
Course: B.Tech (Semester V)
Maximum Marks: 100
Duration: 3 Hours
Exam Year: 2023–24
Sections: A, B, and C
SECTION A – Short Answer Questions (2 × 10 = 20 Marks)
All questions are compulsory.
How has NLP evolved over time?
Explain the challenges associated with language modeling in NLP.
Discuss strategies for handling ambiguity in parsing.
How is Dynamic Programming employed in parsing algorithms?
What are the limitations of supervised approaches in handling Word Sense Disambiguation (WSD)?
How do semantic attachments help disambiguate word senses?
Applications of filter bank methods in speech signal processing.
How do filter banks contribute to speech analysis?
Describe the role of Perceptual Linear Prediction (PLP).
Explain how feature extraction helps understand speech patterns.
SECTION B – Medium-Length Questions (10 × 3 = 30 Marks)
Attempt any three of the following:
Provide an overview of Hidden Markov Models (HMM) and Maximum Entropy Models (MEM) for word-level analysis — their strengths and weaknesses.
Discuss Probabilistic CYK parsing and Probabilistic Lexicalized CFGs, explaining how they improve traditional parsing.
Compare First-Order Logic and Propositional Logic, highlighting expressiveness and relational representation.
Analyze challenges in representing and classifying speech sounds and their impact on speech recognition systems.
Explain the significance of Likelihood Distortions in speech analysis and how they affect model assessment.
SECTION C – Long/Analytical Questions (10 × 5 = 50 Marks)
Attempt one part from each question.
Q3. Word-Level Analysis
a. Explain Minimum Edit Distance and its role in word-level analysis.
OR
b. Discuss challenges in evaluating N-grams and the role of smoothing techniques.
Q4. Grammar and Parsing
a. Compare Dependency Grammar with Phrase Structure Grammar — key differences in syntactic representation.
OR
b. Distinguish Shallow Parsing from Deep Parsing — advantages and limitations.
Q5. Word Similarity and Disambiguation
a. Explain how thesaurus-based and distributional methods measure word similarity — discuss strengths and weaknesses.
OR
b. Compare WSD techniques using dictionaries and thesauri — how these lexical resources disambiguate word senses.
Q6. Speech Processing
a. Explore acoustic phonetics and how speech acoustics explain perceptual differences between sounds.
OR
b. Explain Linear Predictive Coding (LPC) — its modeling process and advantages in speech synthesis.
Q7. Speech Analysis and HMM Evaluation
a. Explain time alignment in speech analysis using Dynamic Time Warping (DTW) and multiple alignment paths.
OR
b. Describe evaluation of HMMs — Optimal State Sequence and Viterbi Search Algorithm for most likely path estimation.
Key Topics to Study
Evolution of NLP: From rule-based to deep learning models.
Parsing: CYK, dependency parsing, and probabilistic approaches.
Language Modeling: N-grams, smoothing, and neural models.
Word Sense Disambiguation: Supervised vs unsupervised techniques, semantic roles.
Speech Processing: Filter banks, LPC, PLP, and DTW.
Machine Learning in NLP: HMMs, MEMs, evaluation with Viterbi algorithm.
Logic & Representation: FOL vs Propositional Logic for meaning representation.
Word Similarity: Thesaurus-based, distributional semantics, embeddings.
Study Tips
Revise algorithms and their mathematical basis: HMMs, CYK parsing, DTW, LPC.
Understand architecture flow: Speech input → feature extraction → phoneme recognition → semantic interpretation.
Prepare comparisons: (e.g., Shallow vs Deep parsing, FOL vs PL).
Focus on probabilistic methods: smoothing, likelihood distortion, expectation maximization.
Link theory with applications: speech assistants, machine translation, POS tagging, chatbots.
