(SEM VII) THEORY EXAMINATION 2024-25 TEXT ANALYSIS

TEXT ANALYSIS (KDS073) – COMPLETE SOLVED PAPER

Time: 3 Hours  Max Marks: 100
Instructions: Attempt all Sections

SECTION A (2 × 10 = 20 Marks)

Attempt all questions in brief

a) How does context influence error detection?

Context helps determine whether a word fits semantically and syntactically within a sentence. Even if a word is spelled correctly, context can reveal errors (e.g., “I went to see the sea” vs “see” instead of “sea”).

b) Backward algorithm in HMM for PoS tagging

The backward algorithm computes the probability of future observations given a current state. It works by recursively summing probabilities from the end of the sentence to the current position.

c) Unification of feature structures (number & gender)

Example:

NP: [number = singular, gender = masculine]

Verb: [number = singular]
Unification succeeds because the number feature agrees, ensuring grammatical correctness.

d) Limitations of CFGs in modeling syntax           Cannot handle long-distance dependencies

Poor handling of agreement features                   Limited semantic representation

Inefficient for ambiguous sentences

e) Dictionary-based vs distributional word similarity

f) Selectional restrictions                                      Semantic constraints on how words can combine.
Example: “The stone ate food” violates selectional restrictions.

g) Classification of speech sounds

Speech sounds are classified as:                               Vowels / Consonants

Voiced / Unvoiced                                                     Place and manner of articulation

h) Effect of vocal tract shape on speech spectrum

Vocal tract shape determines formant frequencies, influencing vowel quality and timbre of speech sounds.

i) Viterbi algorithm

A dynamic programming algorithm used to find the most probable sequence of hidden states in HMMs, commonly applied in PoS tagging and speech recognition.

j) LPC vs PLP coefficients

SECTION B (10 × 3 = 30 Marks)

Attempt any three

a) FSA for regular expression (ab)*c                           States loop over ab

Final state reached on c                                              Accepts strings like c, abc, ababc

b) Ambiguity in sentence using dependency grammar

Sentence: “The dog saw the man with the telescope”

Ambiguity:                                                                   Man has telescope

Dog used telescope                                                     Resolution: Use semantic constraints or context to determine attachment.

c) Syntax-driven semantic analysis                               Syntax guides semantic interpretation.
Sentence: “John gave Mary a book”                              Agent: John

Recipient: Mary                                                             Theme: book

d) Filter-bank vs LPC methods

e) Likelihood distortions in speech recognition

Occurs due to noise or channel mismatch.
Example: Background noise alters acoustic likelihoods, causing misrecognition.

SECTION C (10 × 5 = 50 Marks)

Attempt one from each question

Q3(a) Minimum Edit Distance

Transform “intention” → “execution”                           Operations:

Substitution                                                                      Insertion

Deletion

Minimum edit distance = 5                                            (Using dynamic programming alignment)

Q3(b) Interpolation vs Backoff smoothing

Q4(a) Treebanks in NLP

Treebanks are annotated syntactic corpora used to train parsers.
Example: Penn Treebank provides parsed sentence structures.

Q4(b) CYK parsing algorithm

Bottom-up parsing                                                              Uses dynamic programming

Works with CFG in Chomsky Normal Form                         Example sentence: “He saw a cat”

Q5(a) Supervised WSD

Steps:                                                                                    Collect labeled data

Extract contextual features                                                   Train classifier

Predict sense                                                                         Example: “bank” → river bank or financial bank

Q6(a) Log-spectral distance

Measures difference between two spectra:

D=1N∑(log⁡S1−log⁡S2)2D = \sqrt{\frac{1}{N}\sum (\log S_1 - \log S_2)^2}D=N1​∑(logS1​−logS2​)2​

Used in speech quality analysis.

Q7(b) Role of HMMs in speech recognition

HMMs model:                                                                       Hidden phoneme states

Observable acoustic signals                                                  Forward algorithm: Computes likelihood
Backward algorithm: Computes future probabilities