Unit 1 - Practice Quiz

1 Which decade is generally considered the starting point of Natural Language Processing, marked by Alan Turing's publication on machine intelligence?

2 What defines the set of structural rules governing the composition of clauses, phrases, and words in any given natural language?

3 What is defined as the study of the internal structure of words and how they are formed?

4 Which level of linguistic analysis focuses specifically on the order of words and how they combine to form sentences?

5 What is the primary focus of semantics in Natural Language Processing?

6 Which of the following describes 'ambiguity' in Natural Language Processing?

7 Which of the following is a common application of NLP?

8 What is the process of breaking down a continuous stream of text into smaller units like words or sentences?

9 What is the primary purpose of stemming in text processing?

10 How does lemmatization primarily differ from stemming?

11 What are stop-words in the context of NLP preprocessing?

12 Why is punctuation handling an important step in text preprocessing?

13 What does OOV stand for in Natural Language Processing?

14 Which of the following is a classic example of text normalization?

15 What major linguistic feature does the basic Bag-of-Words (BoW) model ignore?

16 In the context of n-grams, what is a 'bigram'?

17 What does 'TF' stand for in the TF-IDF representation?

18 What is the main purpose of the Inverse Document Frequency (IDF) component in TF-IDF?

19 What is a 'phoneme' in linguistics?

20 Which famous 1954 experiment provided a highly publicized demonstration of machine translation, translating Russian sentences into English?

21 Early NLP systems like ELIZA (1966) simulated conversation using pattern matching. What was the primary limitation of this approach regarding natural language understanding?

22 If an NLP system uses Context-Free Grammar (CFG) to parse sentences, which of the following phenomena is most challenging for it to handle efficiently?

23 Consider the sentence: 'Can you pass the salt?' While syntactically a question, it functions as a request. Which linguistic level of analysis is required to understand this intended meaning?

24 Analyze the words 'running' and 'runner'. Which of the following statements correctly identifies the morphological processes applied to the root word 'run'?

25 The sentence 'I saw the man with the telescope' exhibits structural ambiguity. Which NLP task is directly responsible for resolving how 'with the telescope' attaches to the rest of the sentence?

26 When building a word sense disambiguation model, you encounter the word 'bank' in the contexts of 'river bank' and 'bank account'. This is an example of dealing with which semantic concept?

27 In the sentence 'John told his father that he was promoted', the pronoun 'he' is ambiguous. Resolving whether 'he' refers to John or his father is an example of which NLP challenge?

28 If you are designing an NLP pipeline for a conversational AI (chatbot) meant to book flights, which sequence of applications is most logical for processing user input?

29 Using Byte Pair Encoding (BPE) for tokenization helps mitigate which specific problem compared to standard word-level tokenization?

30 An algorithm reduces both 'university' and 'universal' to the stem 'univers'. What type of stemming error has occurred, and why is it problematic for downstream tasks?

31 Unlike a standard stemmer, a lemmatizer typically requires which piece of additional information to accurately reduce the word 'saw' to its root form?

32 In which of the following NLP tasks is aggressive stop-word removal most likely to degrade model performance?

33 When designing a regular expression-based tokenizer, dealing with punctuation can be tricky. Which of the following examples best demonstrates why simply splitting text on all punctuation characters is a poor strategy?

34 A model trained on a specific corpus encounters a new word during testing. If the model uses a fixed word-level vocabulary with a single <UNK> token, how does it process the new word?

35 Which of the following scenarios demonstrates a potential negative consequence of applying universal case folding (lowercasing all text) during text normalization?

36 Given a vocabulary of size , you represent a 50-word sentence using a standard Bag-of-Words (BoW) vector. What is the dimension of the resulting vector, and what is its primary characteristic?

37 Consider a sentence containing exactly words. If we extract standard continuous trigrams (n=3) from this sentence, how many trigram tokens will be generated? (Assume )

38 In the TF-IDF formulation, the Inverse Document Frequency (IDF) of a term is often calculated as . If a specific stop-word appears in every single document in a corpus of size , what will its IDF value be?

39 Suppose document A has 100 words, and the word 'AI' appears 5 times. Document B has 500 words, and 'AI' appears 15 times. Using basic Term Frequency (count divided by document length), which document gives higher TF to the word 'AI'?

40 When training an n-gram language model, sparsity becomes a major issue for higher values of . Which technique is commonly applied to handle -grams that appear in the test set but were never seen in the training set?

41 Which of the following best characterizes the primary historical consequence of the 1966 ALPAC (Automatic Language Processing Advisory Committee) report on the early trajectory of Natural Language Processing?

42 Natural languages exhibit phenomena such as center embedding (e.g., "The mouse the cat the dog bit chased ran away"). In the context of the Chomsky hierarchy, what does the theoretical existence of arbitrary-depth center embedding imply about human language?

43 Consider the morphological transformation from the root word "compute" to the word "computational". Which sequence of morphological processes accurately characterizes this transition?

44 The sentence "I saw the man with the telescope" exhibits severe structural ambiguity. In a Probabilistic Context-Free Grammar (PCFG), how is the correct parse tree mathematically resolved during syntactic inference?

45 The Principle of Compositionality asserts that the meaning of a complex expression is determined by the meanings of its constituent expressions and the rules used to combine them. Which of the following linguistic phenomena breaks this principle, requiring specialized representation in vector space models?

46 The Winograd Schema Challenge (e.g., "The trophy didn't fit into the brown suitcase because it was too large.") is designed to test machine intelligence. What specific NLP challenge does this schema fundamentally target?

47 When applying Byte-Pair Encoding (BPE) tokenization to a highly agglutinative language like Turkish or Finnish, what is the primary mathematical and practical advantage over strict word-level tokenization in deep learning models?

48 Which of the following scenarios is a canonical example of "over-stemming" (a false positive reduction) resulting from the heuristic rules of algorithms like the Porter Stemmer?

49 Unlike algorithmic stemming, dictionary-based lemmatization relies heavily on accurate Part-of-Speech (POS) tagging. Given a standard WordNet Lemmatizer, what is the expected outcome if the word "saw" in the sentence "He used a saw" is incorrectly tagged as a verb prior to lemmatization?

50 While traditionally beneficial for Information Retrieval, aggressive stop-word removal severely degrades performance in certain modern NLP tasks. In which of the following tasks would stripping stop-words (e.g., "the", "is", "not") be most disastrous for model inference, and why?

51 In clinical NLP, uniformly stripping all punctuation can lead to catastrophic semantic loss. Which of the following exemplifies a structural edge case where a punctuation mark functions as a critical semantic token rather than merely a syntactic boundary?

52 The FastText model fundamentally improves upon the standard Word2Vec architecture primarily in its handling of Out-Of-Vocabulary (OOV) words. What is the mathematical mechanism by which FastText accomplishes this?

53 When preprocessing noisy multilingual text, what is the specific role of Unicode Normalization Form KC (NFKC) compared to NFC, and why is this critical for NLP normalization?

54 Given two documents represented as standard Bag-of-Words (BoW) frequency vectors and , the raw dot product inherently scales with document length, skewing similarity metrics. Which normalization mathematically transforms these vectors such that their dot product strictly equals their angular cosine similarity?

55 In an -gram language model with a vocabulary size , the transition from bigrams to trigrams drastically exacerbates data sparsity. Which advanced smoothing technique addresses this by utilizing absolute discounting combined with an interpolation weight that falls back to lower-order models based on the diversity of contexts a word appears in?

56 In the standard TF-IDF formulation, Inverse Document Frequency is often calculated as . If a term appears in every single document in a training corpus of size , what is the mathematical consequence on its TF-IDF weight, and how is this edge case typically mitigated in libraries like scikit-learn?

57 Zipf's Law states that the frequency of a word in a natural language corpus is inversely proportional to its rank , mathematically expressed as . In the context of NLP model training, what critical structural challenge does this empirical distribution formalize?

58 Consider a long document where the word "excellent" is repeated 100 times. Using standard Term Frequency (TF), the weight is $100$, which can disproportionately dominate the document vector. To dampen this effect, sublinear TF scaling is often applied. Which of the following formulas represents the standard sublinear scaling for a raw term frequency ?

59 While Named Entity Recognition (NER) is traditionally framed as a token-level sequence labeling task using architectures like BiLSTM-CRF, modern generative models occasionally reframe NER as a sequence-to-sequence generation task. What is the primary analytical trade-off (drawback) of this generative approach compared to strict token classification?

60 The perplexity of a test sequence under a bigram language model is defined as . If the un-smoothed language model assigns a probability of $0$ to a single unseen bigram in the test sequence, what mathematically happens to the perplexity, and what does this signify?