(SEM VIII) THEORY EXAMINATION 2023-24 NATURAL LANGUAGE PROCESSING

SECTION A

(2 × 10 = 20 | Very Short Answers)

a. Explain Language Modelling
Language modelling is the process of assigning probabilities to sequences of words to predict the likelihood of a sentence in a language.

b. Discuss pragmatic analysis
Pragmatic analysis deals with understanding the meaning of sentences based on context, speaker intention, and real-world knowledge.

c. Concept of knowledge
Knowledge refers to organized information, facts, rules, and relationships that help a system understand and reason about language.

d. Knowledge representation
Knowledge representation is a method of structuring information so that a computer system can use it for reasoning and inference.

e. Stochastic Part-of-Speech tagging
Stochastic POS tagging assigns tags to words based on probability models such as Hidden Markov Models.

f. Concept of Parsing
Parsing is the process of analyzing sentence structure according to grammatical rules to determine syntactic relationships.

g. Frequency and Amplitude
Frequency refers to the number of occurrences of a linguistic feature, while amplitude represents the strength or magnitude of a signal component.

h. Auxiliary verb with example
Auxiliary verbs help main verbs form tense or mood.
Example: She is reading a book.

i. Dependency tags with example
Dependency tags show relationships between words.
Example: In “She eats food”, eats is the root, she is subject, food is object.

j. TF and IDF
TF measures term frequency in a document, while IDF measures how rare a term is across documents.

SECTION B

(Attempt any THREE | 3 × 10 = 30 Marks)

2(a) Evaluating Language Understanding Systems

Language understanding systems are evaluated based on accuracy, precision, recall, robustness, efficiency, and scalability. Evaluation techniques include test corpora, benchmark datasets, human judgment, and task-based performance such as question answering or translation quality.

2(b) Knowledge representation using Semantic Networks and Production Rules

Semantic networks represent knowledge using nodes and links, where nodes represent concepts and links represent relationships.

Production rules use IF–THEN statements to encode logic.
Example:
IF weather is rainy THEN carry umbrella.

2(c) Feature Systems and Augmented Grammar

Feature systems add attributes like number, gender, tense to grammar rules for better parsing accuracy.
Augmented grammars enhance context-free grammar by including semantic and syntactic constraints to reduce ambiguity.

2(d) Handling questions in Context-Free Grammar

Questions are handled by transforming sentence structures using grammar rules.
Example:
Statement: You are reading a book.                      Question: Are you reading a book?
CFG rearranges subject and auxiliary verb to generate questions.

2(e) Viterbi Search Algorithm

The Viterbi algorithm finds the most probable sequence of hidden states in stochastic models like HMMs.
It uses dynamic programming to compute optimal paths efficiently, commonly used in POS tagging and speech recognition.

SECTION C

3(a) Probabilistic Context-Free Grammars (PCFGs)

PCFGs extend CFGs by assigning probabilities to grammar rules.
They help resolve ambiguity by selecting the most probable parse tree.
PCFGs are widely used in syntactic parsing and NLP applications.

3(b) Steps in Natural Language Understanding

Steps include lexical analysis, syntactic analysis, semantic analysis, discourse analysis, and pragmatic analysis.
These stages transform raw text into meaningful representations for reasoning.

4(a) Transition Network Grammars & Top-Down Chart Parsing

Transition Network Grammars use state diagrams to represent grammar rules.
Top-down chart parsing starts from the start symbol and predicts sentence structure using charts to avoid redundancy.

4(b) Types of knowledge

Types include:                                                                       Linguistic knowledge

World knowledge                                                                 Pragmatic knowledge

Discourse knowledge                                                           Domain knowledge

5(a) Dependency parsing & Shift-Reduce parsing

Dependency parsing identifies relationships between words.
Shift-reduce parsing uses a stack and input buffer, applying shift and reduce actions to build dependency structures.

5(b) Transition Network Grammar & Recursive Transition Network

Transition Network Grammar represents grammar using state machines.
Recursive Transition Networks allow recursive rules for complex sentence structures.
(The given NP grammar can be drawn using state transitions from ART/DET → ADJ → N.)

6(a) Encoding uncertainty & Shift-Reduce parsing diagram

Encoding uncertainty represents ambiguity using probabilities.
Shift-reduce parsing builds parse trees using stack-based operations guided by grammar rules.

6(b) Database interface in NLP

A database interface allows users to query databases using natural language.
It converts natural language queries into structured SQL queries.

7(a) Probabilistic Context-Free Grammar (detailed)

PCFG assigns probabilities to production rules and selects the most likely parse tree.
It improves parsing accuracy in ambiguous sentences.

7(b) Hidden Markov Model & Baum-Welch Algorithm

HMM is a probabilistic model with hidden states and observable outputs.
Baum-Welch algorithm estimates HMM parameters using expectation-maximization.
Implementation issues include data sparsity, computational complexity, and convergence problems.