Natural Language Processing (ELEDSE8A) — DU B. Sc. (Hons) Electronics Previous Year Question Papers

: Natural Language Processing Maximulh Marks Instruction for Candidates Q1 (a) (b) (c) (d) (e) (I) Identify and briefly explain any three major challenges faced in Natural Language Processing. Differentiate between Morphology and Lexical semantics in the context of word forms. Define Information Retrieval. How does it serve as a key application of NLP? What are stop words? Give 5 examples What is Tokenization? Give example of sentence tokenization. Differentiate Term Frequency vs Document Frequency. (3) Q2 (a) Discuss the various phases of Natural Language processing (such as (6) Moaphological, Syntactic, Semantic, and Pragmatic analysis). Using the s.entence ''7lJze o/d 77€cz77 reczc7 cz boofr, " illustrate how the output of the syntactic phase acts as the input for the semantic phase. (b) Formulate the Trigram Language .Model. Assume a training corpus (6) yields the following specific frequency counts: C({I,love,NLP})=80 C({I, love, coding})= 120 Based on this data, calculate the trigram probabilities P({NL,P} I {1, 1ove}) and P({coding} I {1, 1ove}). Explain the Markov assumption utilized in this calculation. (c) What does it mean to conduct an empirical comparison of smoothing (6) techniques? Discuss how applications of Language Modelling evaluate the effectiveness of these different smoothing algorithms. Q3 (a) Describe the application of the Markov Model and the Maximum (6) Entropy model in the .context of part of Speech Tagging. How do these models use probability to resolve tagging ambiguities? (b) Elaborate on the architectures of continuous Bag-of-Words (CBOW) (6) and skip-gram models. Detail how these models generate dense vector representations for words. Q4 (c) In a skip-gram model with a window size of2, identify the training pairs (input, output) generated for the target word "NLP" in the sentence: "The field of NLP is rapidly developing". Show the mathematical objective function that the mo.del tries to maxirize during training. (a) Discuss the process of Natural Language Generation and its reliance on (6) Lexical Semantics. (b) Differentiate between Knowledge-Based WSD (using lexical resources) (6) and Supervised WSD (using annotated corpora) (.c) Describe the Maximuln Entropy model for pos tagging. Discuss how it (6) differs from a standard Markov Model in terms of the features it considers for classification. a) Q5 (a) Explain the difference between Extractive and Abstractive Text (6) Summarization with one example each. Define Tern Frequehcy (TF). Calculate TF for word "cat" in document (c) Explain Document Frequency Weighting with formula. Why do (6) common words get zero weight? (6) "the cat the cat dog" showing all steps. (b) Q6 (a) What is Named Entity Recognition (NER)? List any fournamed entity (6) Explain stemming. Apply stemmi.ng to words: "running", "connection", (c) What is sentiment Analysis? Classify sentiment of these 3 reviews using simple keyword method. Show steps. "Excellent product, very happy! "