Natural Language Processing
509 pages in the complete text edition.
All subject text · MarkdownOriginal content · JSON
M2 · About
- What this NLP guide coversPortion, paper pattern & sources
- Video companions in your NotebookLMPortion, paper pattern & sources
M2 · About Chapter
M2 · Formulas
- The M2 formula sheetAll M2 formulas
- Unigrams, bigrams and trigrams: construct them correctlyN-grams & probability
- Unigram estimation and complete worked examplesN-grams & probability
- Bigram counts: build the table before multiplyingN-grams & probability
- Trigram estimates: two-word history, not two denominatorsN-grams & probability
- Laplace smoothing: why both numerator and denominator changeSmoothing & perplexity
- A complete unseen-bigram question from corpus to final answerSmoothing & perplexity
- Adjusted counts and relative discountingSmoothing & perplexity
- Perplexity: meaning, formula and four complete examplesSmoothing & perplexity
- Log probabilities: replace long products with sumsSmoothing & perplexity
- Markov chains: read a transition table and multiply a pathHMM & decoding
- Estimate transition and emission probabilities from tagged dataHMM & decoding
- HMM path probabilities: three complete numerical examplesHMM & decoding
- POS HMM worked corpus: Mary, Jane, can, will and spotHMM & decoding
- Why exhaustive search grows exponentiallyHMM & decoding
- Viterbi decoding: compare incoming paths, keep the best, trace backHMM & decoding
- A path score is not the probability of the observed sentenceHMM & decoding
- Lesk algorithm, from words to overlap scoresWSD: Lesk & Naive Bayes
- Thesaurus / Walker approach, with three complete examplesWSD: Lesk & Naive Bayes
- Multinomial Naive Bayes: the exact solving methodWSD: Lesk & Naive Bayes
- Mam's bank numerical—every count and every multiplicationWSD: Lesk & Naive Bayes
- More WSD numericals: given likelihoods, unequal priors and repeated cluesWSD: Lesk & Naive Bayes
M2 · Formulas Chapter
M2 · Index
M2 · Mocks
- Paper instructionsM2 Mock A · core calculations
- Q1 (2 marks)M2 Mock A · core calculations
- Q2 (2 marks)M2 Mock A · core calculations
- Q3 (3 marks)M2 Mock A · core calculations
- Q4 (3 marks)M2 Mock A · core calculations
- Q5 (3 marks)M2 Mock A · core calculations
- Paper instructionsM2 Mock B · unseen pairs & decoding
- Q1 (2 marks)M2 Mock B · unseen pairs & decoding
- Q2 (2 marks)M2 Mock B · unseen pairs & decoding
- Q3 (3 marks)M2 Mock B · unseen pairs & decoding
- Q4 (3 marks)M2 Mock B · unseen pairs & decoding
- Q5 (3 marks)M2 Mock B · unseen pairs & decoding
- Paper instructionsM2 Mock C · mixed numerical practice
- Q1 (2 marks)M2 Mock C · mixed numerical practice
- Q2 (2 marks)M2 Mock C · mixed numerical practice
- Q3 (3 marks)M2 Mock C · mixed numerical practice
- Q4 (3 marks)M2 Mock C · mixed numerical practice
- Q5 (3 marks)M2 Mock C · mixed numerical practice
M2 · Mocks Chapter
M2 · Papers
- Q1B · Chunking and chinkingSolved final paper · December 2025
- Q1C · Hyponymy, hypernymy and WSDSolved final paper · December 2025
- Q2B · Laplace-smoothed bigram probabilitySolved final paper · December 2025
- Q4B · HMM emission, transition and sequenceSolved final paper · December 2025
- Q5B · Naïve Bayes word-sense classificationSolved final paper · December 2025
- Q1B · Content and function wordsSolved re-exam · February 2026
- Q2B · Laplace smoothing on a second corpusSolved re-exam · February 2026
- Q3A · Top-down parsingSolved re-exam · February 2026
- Q3B · Lesk word-sense disambiguationSolved re-exam · February 2026
- Q4B · A second HMM sequenceSolved re-exam · February 2026
M2 · Papers Chapter
M2 · Read
- What NLP is doingStart here · language & probability
- Tokens, counts and symbolsStart here · language & probability
- Probability, without skipped stepsStart here · language & probability
- Bayes’ rule: prior, evidence and answerStart here · language & probability
- Small probabilities, logs and calculator useStart here · language & probability
- Words, counts and probabilities: the notation you needN-grams and language models
- Unigrams, bigrams and trigrams: construct them correctlyN-grams and language models
- Collocations and the purpose of a language modelN-grams and language models
- Probability chain rule and the Markov assumptionN-grams and language models
- Unigram estimation and complete worked examplesFrom a corpus to sentence probabilities
- Bigram counts: build the table before multiplyingFrom a corpus to sentence probabilities
- Trigram estimates: two-word history, not two denominatorsFrom a corpus to sentence probabilities
- Class-style count-table questions and partial-table trapsFrom a corpus to sentence probabilities
- Perplexity: meaning, formula and four complete examplesPerplexity, log probabilities and smoothing
- Log probabilities: replace long products with sumsPerplexity, log probabilities and smoothing
- Laplace smoothing: why both numerator and denominator changePerplexity, log probabilities and smoothing
- A complete unseen-bigram question from corpus to final answerPerplexity, log probabilities and smoothing
- Adjusted counts and relative discountingPerplexity, log probabilities and smoothing
- Character-model perplexity: keep exact probabilitiesPerplexity, log probabilities and smoothing
- Parts of speech and contextual ambiguityPOS tagging and its learning approaches
- Lexical, rule-based, stochastic, Brill and neural taggersPOS tagging and its learning approaches
- Generative versus discriminative models and Bayes' rulePOS tagging and its learning approaches
- Markov chains: read a transition table and multiply a pathMarkov chains and Hidden Markov Models
- Hidden states, observations, transitions and emissionsMarkov chains and Hidden Markov Models
- Estimate transition and emission probabilities from tagged dataMarkov chains and Hidden Markov Models
- HMM path probabilities: three complete numerical examplesMarkov chains and Hidden Markov Models
- POS HMM worked corpus: Mary, Jane, can, will and spotMarkov chains and Hidden Markov Models
- Why exhaustive search grows exponentiallyChoosing the most likely tag sequence
- Viterbi decoding: compare incoming paths, keep the best, trace backChoosing the most likely tag sequence
- A path score is not the probability of the observed sentenceChoosing the most likely tag sequence
- Chunking groups words; chinking removes words from a candidate chunkChunking, chinking and context-free grammar
- CFG: terminals, nonterminals, production rules and parse treesChunking, chinking and context-free grammar
- Ambiguity and grammar coverageChunking, chinking and context-free grammar
- Recursive-descent top-down parsingTop-down and bottom-up parsing: complete traces
- Three full top-down derivations with parse treesTop-down and bottom-up parsing: complete traces
- Shift-reduce bottom-up parsing: stack, input and actionsTop-down and bottom-up parsing: complete traces
- Three full shift-reduce tracesTop-down and bottom-up parsing: complete traces
- Entity spans, categories and contextNamed Entity Recognition
- Dictionary, rules, machine learning and deep learningNamed Entity Recognition
- Three complete entity-recognition examplesNamed Entity Recognition
- Lexeme, lemma, sense and contextWord meaning, lexical relations and WordNet
- Every lexical relation with examplesWord meaning, lexical relations and WordNet
- WordNet: synsets, glosses and hierarchyWord meaning, lexical relations and WordNet
- What WSD solves and how approaches differWord sense disambiguation: Lesk, Walker and graph methods
- Lesk algorithm, from words to overlap scoresWord sense disambiguation: Lesk, Walker and graph methods
- Thesaurus / Walker approach, with three complete examplesWord sense disambiguation: Lesk, Walker and graph methods
- Random-walk WSD and supervised alternativesWord sense disambiguation: Lesk, Walker and graph methods
- Prior, likelihood and posterior—before the formulaNaive Bayes WSD: probabilities to a complete numerical
- Multinomial Naive Bayes: the exact solving methodNaive Bayes WSD: probabilities to a complete numerical
- Mam's bank numerical—every count and every multiplicationNaive Bayes WSD: probabilities to a complete numerical
- More WSD numericals: given likelihoods, unequal priors and repeated cluesNaive Bayes WSD: probabilities to a complete numerical
M2 · Read Chapter
- Start here · language & probability
- N-grams and language models
- From a corpus to sentence probabilities
- Perplexity, log probabilities and smoothing
- POS tagging and its learning approaches
- Markov chains and Hidden Markov Models
- Choosing the most likely tag sequence
- Chunking, chinking and context-free grammar
- Top-down and bottom-up parsing: complete traces
- Named Entity Recognition
- Word meaning, lexical relations and WordNet
- Word sense disambiguation: Lesk, Walker and graph methods
- Naive Bayes WSD: probabilities to a complete numerical
M2 · Revision
- Last-day reading orderM2 · last-day revision
- Recognise the question and start correctlyM2 · last-day revision
M2 · Revision Chapter
TEE · About
- Latest and older papers: exact question mapAll recovered TEE papers · unit, topic and marks map
- What this NLP guide coversPortion, paper pattern & sources
- Video companions in your NotebookLMPortion, paper pattern & sources
- Add-k: preserve a normalized vocabulary rowOptional extensions — after the TEE questions
- Forward HMM: sum paths and normalize posteriorsOptional extensions — after the TEE questions
TEE · About Chapter
TEE · Formulas
- Core formula sheetCore formulas
- Unigrams, bigrams and trigrams: construct them correctlyN-grams & probability
- Unigram estimation and complete worked examplesN-grams & probability
- Bigram counts: build the table before multiplyingN-grams & probability
- Trigram estimates: two-word history, not two denominatorsN-grams & probability
- Laplace smoothing: why both numerator and denominator changeSmoothing & perplexity
- A complete unseen-bigram question from corpus to final answerSmoothing & perplexity
- Adjusted counts and relative discountingSmoothing & perplexity
- Perplexity: meaning, formula and four complete examplesSmoothing & perplexity
- Log probabilities: replace long products with sumsSmoothing & perplexity
- Markov chains: read a transition table and multiply a pathHMM & decoding
- Estimate transition and emission probabilities from tagged dataHMM & decoding
- HMM path probabilities: three complete numerical examplesHMM & decoding
- POS HMM worked corpus: Mary, Jane, can, will and spotHMM & decoding
- Why exhaustive search grows exponentiallyHMM & decoding
- Viterbi decoding: compare incoming paths, keep the best, trace backHMM & decoding
- A path score is not the probability of the observed sentenceHMM & decoding
- Lesk algorithm, from words to overlap scoresWSD: Lesk & Naive Bayes
- Thesaurus / Walker approach, with three complete examplesWSD: Lesk & Naive Bayes
- Multinomial Naive Bayes: the exact solving methodWSD: Lesk & Naive Bayes
- Mam's bank numerical—every count and every multiplicationWSD: Lesk & Naive Bayes
- More WSD numericals: given likelihoods, unequal priors and repeated cluesWSD: Lesk & Naive Bayes
- Label and one-hot encoding: three complete examplesRepresentation & similarity
- Bag of words and bag of n-gramsRepresentation & similarity
- TF–IDF from the exact slide conventionRepresentation & similarity
- Cosine: products, lengths and divisionRepresentation & similarity
- Jaccard: intersection, union and distanceRepresentation & similarity
- Manhattan, Euclidean and Minkowski: three complete comparisonsRepresentation & similarity
- Distributional hypothesis and a co-occurrence matrixRepresentation & similarity
- Word2Vec: CBOW and skip-gram from training pairsRepresentation & similarity
- Word analogies and sentence embeddingsRepresentation & similarity
- The GloVe objective and numerical interpretationRepresentation & similarity
- Text classification and its full pipelineApplication calculations
- Precision, recall, F1 and accuracyApplication calculations
- Spelling correction and edit distanceApplication calculations
- Information retrievalApplication calculations
- N-gram interpolation: weighted evidence from several historiesFaculty topics: interpolation and Stupid Backoff
- Backoff and the important score-versus-probability distinctionFaculty topics: interpolation and Stupid Backoff
- Prepare a grammar for CYK / CKYFaculty topic: CNF and CYK / CKY parsing
- CYK / CKY: complete charts, backpointers and ambiguityFaculty topic: CNF and CYK / CKY parsing
- TEE formula sheet: vectors and applicationsTEE · additional formulas
TEE · Formulas Chapter
TEE · Index
TEE · Mocks
- Write the whole paper under exam conditionsTimed practice · Solved final paper · December 2025
- Q1A · NLU and NLGTimed practice · Solved final paper · December 2025
- Q1B · Chunking and chinkingTimed practice · Solved final paper · December 2025
- Q1C · Hyponymy, hypernymy and WSDTimed practice · Solved final paper · December 2025
- Q1D · Pragmatics and discourseTimed practice · Solved final paper · December 2025
- Q2A · Morpheme, lexeme and NERTimed practice · Solved final paper · December 2025
- Q2B · Laplace-smoothed bigram probabilityTimed practice · Solved final paper · December 2025
- Q3A · Ambiguity and bottom-up parsingTimed practice · Solved final paper · December 2025
- Q3B · Co-occurrence matrix and cosineTimed practice · Solved final paper · December 2025
- Q4A · Three preprocessing techniquesTimed practice · Solved final paper · December 2025
- Q4B · HMM emission, transition and sequenceTimed practice · Solved final paper · December 2025
- Q5A · TF–IDF matrixTimed practice · Solved final paper · December 2025
- Q5B · Naïve Bayes word-sense classificationTimed practice · Solved final paper · December 2025
- Q6A · Coreference, pronouns and Word2VecTimed practice · Solved final paper · December 2025
- Q6B · Hobbs: Ravi and SureshTimed practice · Solved final paper · December 2025
- Q7A · Text classification pipelineTimed practice · Solved final paper · December 2025
- Q7B · Corpora and Jaccard similarityTimed practice · Solved final paper · December 2025
- Write the whole paper under exam conditionsTimed practice · Solved re-exam · February 2026
- Q1A · Generic NLP pipelineTimed practice · Solved re-exam · February 2026
- Q1B · Content and function wordsTimed practice · Solved re-exam · February 2026
- Q1C · Word2Vec and GloVeTimed practice · Solved re-exam · February 2026
- Q1D · Anaphora, cataphora and coreferenceTimed practice · Solved re-exam · February 2026
- Q2A · NLP tasks and syntax versus semanticsTimed practice · Solved re-exam · February 2026
- Q2B · Laplace smoothing on a second corpusTimed practice · Solved re-exam · February 2026
- Q3A · Top-down parsingTimed practice · Solved re-exam · February 2026
- Q3B · Lesk word-sense disambiguationTimed practice · Solved re-exam · February 2026
- Q4A · Bag of Words and three cosine similaritiesTimed practice · Solved re-exam · February 2026
- Q4B · A second HMM sequenceTimed practice · Solved re-exam · February 2026
- Q5A · Why TF–IDF is usefulTimed practice · Solved re-exam · February 2026
- Q5B · Sentence-window vectors and cosineTimed practice · Solved re-exam · February 2026
- Q6A · Pronoun approaches and lexical relationsTimed practice · Solved re-exam · February 2026
- Q6B · Hobbs: Ramesh and ShyamTimed practice · Solved re-exam · February 2026
- Q7A · Information retrievalTimed practice · Solved re-exam · February 2026
- Q7B · Gutenberg corpus and normalizationTimed practice · Solved re-exam · February 2026
TEE · Mocks Chapter
TEE · Papers
- Q1A · NLU and NLGSolved final paper · December 2025
- Q1B · Chunking and chinkingSolved final paper · December 2025
- Q1C · Hyponymy, hypernymy and WSDSolved final paper · December 2025
- Q1D · Pragmatics and discourseSolved final paper · December 2025
- Q2A · Morpheme, lexeme and NERSolved final paper · December 2025
- Q2B · Laplace-smoothed bigram probabilitySolved final paper · December 2025
- Q3A · Ambiguity and bottom-up parsingSolved final paper · December 2025
- Q3B · Co-occurrence matrix and cosineSolved final paper · December 2025
- Q4A · Three preprocessing techniquesSolved final paper · December 2025
- Q4B · HMM emission, transition and sequenceSolved final paper · December 2025
- Q5A · TF–IDF matrixSolved final paper · December 2025
- Q5B · Naïve Bayes word-sense classificationSolved final paper · December 2025
- Q6A · Coreference, pronouns and Word2VecSolved final paper · December 2025
- Q6B · Hobbs: Ravi and SureshSolved final paper · December 2025
- Q7A · Text classification pipelineSolved final paper · December 2025
- Q7B · Corpora and Jaccard similaritySolved final paper · December 2025
- Q1A · Generic NLP pipelineSolved re-exam · February 2026
- Q1B · Content and function wordsSolved re-exam · February 2026
- Q1C · Word2Vec and GloVeSolved re-exam · February 2026
- Q1D · Anaphora, cataphora and coreferenceSolved re-exam · February 2026
- Q2A · NLP tasks and syntax versus semanticsSolved re-exam · February 2026
- Q2B · Laplace smoothing on a second corpusSolved re-exam · February 2026
- Q3A · Top-down parsingSolved re-exam · February 2026
- Q3B · Lesk word-sense disambiguationSolved re-exam · February 2026
- Q4A · Bag of Words and three cosine similaritiesSolved re-exam · February 2026
- Q4B · A second HMM sequenceSolved re-exam · February 2026
- Q5A · Why TF–IDF is usefulSolved re-exam · February 2026
- Q5B · Sentence-window vectors and cosineSolved re-exam · February 2026
- Q6A · Pronoun approaches and lexical relationsSolved re-exam · February 2026
- Q6B · Hobbs: Ramesh and ShyamSolved re-exam · February 2026
- Q7A · Information retrievalSolved re-exam · February 2026
- Q7B · Gutenberg corpus and normalizationSolved re-exam · February 2026
- Q1a · BoW and one-hot encoding2025-02-21 · re-exam · recovered AI paper
- Q1b · Lesk WSD2025-02-21 · re-exam · recovered AI paper
- Q1c · Syntax parsing approaches2025-02-21 · re-exam · recovered AI paper
- Q1d · NLU versus NLG and applications2025-02-21 · re-exam · recovered AI paper
- Q2a · TF–IDF cosine after a fixed stoplist2025-02-21 · re-exam · recovered AI paper
- Q2b · Weather Naive Bayes2025-02-21 · re-exam · recovered AI paper
- Q3a · Coreference and anaphora2025-02-21 · re-exam · recovered AI paper
- Q3b · Cohesion, coherence and discourse markers2025-02-21 · re-exam · recovered AI paper
- Q3c · Question answering system2025-02-21 · re-exam · recovered AI paper
- Q4a · NLP pipeline2025-02-21 · re-exam · recovered AI paper
- Q4b · Morphology and syntax2025-02-21 · re-exam · recovered AI paper
- Q4c · Most probable bigram continuations2025-02-21 · re-exam · recovered AI paper
- Q5a · Preprocess the NLP passage2025-02-21 · re-exam · recovered AI paper
- Q5b · Stemming and lemmatization tools2025-02-21 · re-exam · recovered AI paper
- Q5c · Hobbs on John and Mary2025-02-21 · re-exam · recovered AI paper
- Q6a · Corpora and acquisition2025-02-21 · re-exam · recovered AI paper
- Q6b · HMM components and estimates2025-02-21 · re-exam · recovered AI paper
- Q7a · Word classes, chunking and chinking2025-02-21 · re-exam · recovered AI paper
- Q7b · Named entity recognition2025-02-21 · re-exam · recovered AI paper
- Q7c · Co-occurrence with inconsistent requested comparisons2025-02-21 · re-exam · recovered AI paper
- Q1a · TF–IDF in retrieval2024-12-03 · final · recovered AI paper
- Q1b · NER and approaches2024-12-03 · final · recovered AI paper
- Q1c · Skip-gram and CBOW2024-12-03 · final · recovered AI paper
- Q1d · Pragmatics and discourse2024-12-03 · final · recovered AI paper
- Q2a · Six preprocessing operations2024-12-03 · final · recovered AI paper
- Q2b · Lexical similarity and cosine2024-12-03 · final · recovered AI paper
- Q2c · Animal categorical Naive Bayes2024-12-03 · final · recovered AI paper
- Q3a · Hobbs and ambiguous professor–student reference2024-12-03 · final · recovered AI paper
- Q3b · Classification and sentiment in industry2024-12-03 · final · recovered AI paper
- Q4a · Top-down syntax tree2024-12-03 · final · recovered AI paper
- Q4b · Data acquisition and four augmentation methods2024-12-03 · final · recovered AI paper
- Q5a · Feature engineering, BoW and Euclidean2024-12-03 · final · recovered AI paper
- Q5b · WSD approaches and uses2024-12-03 · final · recovered AI paper
- Q5c · Discourse analysis of a conversation2024-12-03 · final · recovered AI paper
- Q6a · Henry bigram continuations2024-12-03 · final · recovered AI paper
- Q6b · Five linguistic levels2024-12-03 · final · recovered AI paper
- Q6c · NLU and NLG working together2024-12-03 · final · recovered AI paper
- Q7a · POS, chunks and boundary-aware HMM2024-12-03 · final · recovered AI paper
- Q7b · Co-occurrence, dot products and plot2024-12-03 · final · recovered AI paper
- Q1a · Anaphora versus cataphora2024-02-12 · re-exam · recovered AI paper
- Q1b · Five corpora2024-02-12 · re-exam · recovered AI paper
- Q1c · Linguistic foundations2024-02-12 · re-exam · recovered AI paper
- Q1d · Five regex operators2024-02-12 · re-exam · recovered AI paper
- Q2a · Wordlist puzzle and lexical resource2024-02-12 · re-exam · recovered AI paper
- Q2b · NLP applications and limitations2024-02-12 · re-exam · recovered AI paper
- Q3a · Top-down restaurant parse2024-02-12 · re-exam · recovered AI paper
- Q3b · Preprocessing pipeline2024-02-12 · re-exam · recovered AI paper
- Q4a · Embeddings and Skip-gram2024-02-12 · re-exam · recovered AI paper
- Q4b · POS, NER, chunking and TTR2024-02-12 · re-exam · recovered AI paper
- Q5a · Lexical and vector semantics2024-02-12 · re-exam · recovered AI paper
- Q5b · Classification label structures2024-02-12 · re-exam · recovered AI paper
- Q6a · TF–IDF in document representation2024-02-12 · re-exam · recovered AI paper
- Q6b · HMM tables and path2024-02-12 · re-exam · recovered AI paper
- Q7a · Healthcare and finance applications2024-02-12 · re-exam · recovered AI paper
- Q7b · Naive Bayes sentiment2024-02-12 · re-exam · recovered AI paper
- Q1a · WSD and pronoun resolution2023-12-01 · final · recovered AI paper
- Q1b · WordNet and its hierarchy2023-12-01 · final · recovered AI paper
- Q1c · Sentiment feature extraction2023-12-01 · final · recovered AI paper
- Q1d · Binary, multiclass and multilabel2023-12-01 · final · recovered AI paper
- Q2a · Text corpus structure and applications2023-12-01 · final · recovered AI paper
- Q2b · Lesk versus Walker and bank disambiguation2023-12-01 · final · recovered AI paper
- Q3a · Hobbs algorithm and limitations2023-12-01 · final · recovered AI paper
- Q3b · Stemming and lemmatization2023-12-01 · final · recovered AI paper
- Q4a · Lexical and vector semantics2023-12-01 · final · recovered AI paper
- Q4b · Three-document TF–IDF2023-12-01 · final · recovered AI paper
- Q5a · Word embeddings and CBOW2023-12-01 · final · recovered AI paper
- Q5b · HMM tables and given path2023-12-01 · final · recovered AI paper
- Q6a · Text classification pipeline2023-12-01 · final · recovered AI paper
- Q6b · VSM, TF and IDF in retrieval2023-12-01 · final · recovered AI paper
- Q1a · Normalization changes vocabulary2023-01-30 · re-exam · recovered AI paper
- Q1b · Good corpus characteristics2023-01-30 · re-exam · recovered AI paper
- Q1c · Word2Vec versus Doc2Vec2023-01-30 · re-exam · recovered AI paper
- Q1d · Ambiguity in NLP2023-01-30 · re-exam · recovered AI paper
- Q2a · Normalize the Tanisha–Simran passage2023-01-30 · re-exam · recovered AI paper
- Q2b · Text vectorization alternatives2023-01-30 · re-exam · recovered AI paper
- Q3a · Semantics, pragmatics and two ambiguities2023-01-30 · re-exam · recovered AI paper
- Q3b · Naive Bayes text classifier2023-01-30 · re-exam · recovered AI paper
- Q4a · Jaccard, cosine and Euclidean without preprocessing2023-01-30 · re-exam · recovered AI paper
- Q4b · CBOW and Skip-gram2023-01-30 · re-exam · recovered AI paper
- Q5a · Password regex2023-01-30 · re-exam · recovered AI paper
- Q5b · Two NLP applications2023-01-30 · re-exam · recovered AI paper
- Q6a · Mention-pair model2023-01-30 · re-exam · recovered AI paper
- Q6b · Classification evaluation including ROC/AUC2023-01-30 · re-exam · recovered AI paper
- Q7a · Organization and facility mentions2023-01-30 · re-exam · recovered AI paper
- Q7b · Tagging and two complete parse trees2023-01-30 · re-exam · recovered AI paper
- Q1a · Need and difficulties of NLP2022-11-22 · final · recovered AI paper
- Q1b · Levels of language analysis2022-11-22 · final · recovered AI paper
- Q1c · Medical chatbot2022-11-22 · final · recovered AI paper
- Q1d · Sentiment with logistic regression2022-11-22 · final · recovered AI paper
- Q2a · CBOW and Skip-gram architecture2022-11-22 · final · recovered AI paper
- Q2b · Seven regex constructions2022-11-22 · final · recovered AI paper
- Q3a · Text preprocessing2022-11-22 · final · recovered AI paper
- Q3b · Penn Treebank tagging2022-11-22 · final · recovered AI paper
- Q4a · Mention-pair coreference2022-11-22 · final · recovered AI paper
- Q4b · Text classification pipeline2022-11-22 · final · recovered AI paper
- Q5a · Corpora and structure2022-11-22 · final · recovered AI paper
- Q5b · Multinomial Naive Bayes2022-11-22 · final · recovered AI paper
- Q6a · GloVe2022-11-22 · final · recovered AI paper
- Q6b · BoW and TF–IDF movie corpus2022-11-22 · final · recovered AI paper
- Q7a · Information extraction, chunking and NER2022-11-22 · final · recovered AI paper
- Q7b · Distributional hypothesis and cosine2022-11-22 · final · recovered AI paper
TEE · Papers Chapter
- Solved final paper · December 2025
- Solved re-exam · February 2026
- 2025-02-21 · re-exam · recovered AI paper
- 2024-12-03 · final · recovered AI paper
- 2024-02-12 · re-exam · recovered AI paper
- 2023-12-01 · final · recovered AI paper
- 2023-01-30 · re-exam · recovered AI paper
- 2022-11-22 · final · recovered AI paper
TEE · Read
- What NLP is doingStart here · language & probability
- Tokens, counts and symbolsStart here · language & probability
- Probability, without skipped stepsStart here · language & probability
- Bayes’ rule: prior, evidence and answerStart here · language & probability
- Small probabilities, logs and calculator useStart here · language & probability
- What natural language processing actually doesNLP foundations: language, tasks and the processing pipeline
- Building blocks and levels of linguistic analysisNLP foundations: language, tasks and the processing pipeline
- Rule-based, statistical and deep-learning approachesNLP foundations: language, tasks and the processing pipeline
- NLU versus NLG, explained through one assistantNLP foundations: language, tasks and the processing pipeline
- Generic NLP pipeline, evaluation and deploymentNLP foundations: language, tasks and the processing pipeline
- Ambiguity and other challenges with precise examplesNLP foundations: language, tasks and the processing pipeline
- Tokenization: sentence, word, character and subwordText preprocessing, normalization and lexical diversity
- Cleaning, case folding, punctuation and stopwordsText preprocessing, normalization and lexical diversity
- Stemming versus lemmatizationText preprocessing, normalization and lexical diversity
- Type–token ratio: three fully worked sumsText preprocessing, normalization and lexical diversity
- Vocabulary and vectors—what each coordinate meansText to numbers: label encoding, one-hot, BoW and TF–IDF
- Label and one-hot encoding: three complete examplesText to numbers: label encoding, one-hot, BoW and TF–IDF
- Bag of words and bag of n-gramsText to numbers: label encoding, one-hot, BoW and TF–IDF
- TF–IDF from the exact slide conventionText to numbers: label encoding, one-hot, BoW and TF–IDF
- Combined slide numerical: TF–IDF followed by cosineText to numbers: label encoding, one-hot, BoW and TF–IDF
- Cosine: products, lengths and divisionSimilarity and distance: every numerical step
- Jaccard: intersection, union and distanceSimilarity and distance: every numerical step
- Manhattan, Euclidean and Minkowski: three complete comparisonsSimilarity and distance: every numerical step
- Choose the representation before the metricSimilarity and distance: every numerical step
- Corpus structure and annotationsCorpora, lexical resources and data acquisition
- Gutenberg, web/chat, Brown, Reuters and InauguralCorpora, lexical resources and data acquisition
- Raw text, tokens and sentences are different viewsCorpora, lexical resources and data acquisition
- Wordlists, pronunciation and semantic resourcesCorpora, lexical resources and data acquisition
- Acquiring data and preserving labels during augmentationCorpora, lexical resources and data acquisition
- How corpus evidence powers later unitsCorpora, lexical resources and data acquisition
- Characters, classes, anchors and repetitionRegular expressions: read, build and test text patterns
- Search, match, fullmatch, findall, substitution and flagsRegular expressions: read, build and test text patterns
- Six full regex questions and solutionsRegular expressions: read, build and test text patterns
- Pattern reference for the handout's remaining tasksRegular expressions: read, build and test text patterns
- Words, counts and probabilities: the notation you needN-grams and language models
- Unigrams, bigrams and trigrams: construct them correctlyN-grams and language models
- Collocations and the purpose of a language modelN-grams and language models
- Probability chain rule and the Markov assumptionN-grams and language models
- Unigram estimation and complete worked examplesFrom a corpus to sentence probabilities
- Bigram counts: build the table before multiplyingFrom a corpus to sentence probabilities
- Trigram estimates: two-word history, not two denominatorsFrom a corpus to sentence probabilities
- Class-style count-table questions and partial-table trapsFrom a corpus to sentence probabilities
- Perplexity: meaning, formula and four complete examplesPerplexity, log probabilities and smoothing
- Log probabilities: replace long products with sumsPerplexity, log probabilities and smoothing
- Laplace smoothing: why both numerator and denominator changePerplexity, log probabilities and smoothing
- A complete unseen-bigram question from corpus to final answerPerplexity, log probabilities and smoothing
- Adjusted counts and relative discountingPerplexity, log probabilities and smoothing
- Character-model perplexity: keep exact probabilitiesPerplexity, log probabilities and smoothing
- N-gram interpolation: weighted evidence from several historiesFaculty topics: interpolation and Stupid Backoff
- Backoff and the important score-versus-probability distinctionFaculty topics: interpolation and Stupid Backoff
- Parts of speech and contextual ambiguityPOS tagging and its learning approaches
- Lexical, rule-based, stochastic, Brill and neural taggersPOS tagging and its learning approaches
- Generative versus discriminative models and Bayes' rulePOS tagging and its learning approaches
- Markov chains: read a transition table and multiply a pathMarkov chains and Hidden Markov Models
- Hidden states, observations, transitions and emissionsMarkov chains and Hidden Markov Models
- Estimate transition and emission probabilities from tagged dataMarkov chains and Hidden Markov Models
- HMM path probabilities: three complete numerical examplesMarkov chains and Hidden Markov Models
- POS HMM worked corpus: Mary, Jane, can, will and spotMarkov chains and Hidden Markov Models
- Why exhaustive search grows exponentiallyChoosing the most likely tag sequence
- Viterbi decoding: compare incoming paths, keep the best, trace backChoosing the most likely tag sequence
- A path score is not the probability of the observed sentenceChoosing the most likely tag sequence
- Chunking groups words; chinking removes words from a candidate chunkChunking, chinking and context-free grammar
- CFG: terminals, nonterminals, production rules and parse treesChunking, chinking and context-free grammar
- Ambiguity and grammar coverageChunking, chinking and context-free grammar
- Recursive-descent top-down parsingTop-down and bottom-up parsing: complete traces
- Three full top-down derivations with parse treesTop-down and bottom-up parsing: complete traces
- Shift-reduce bottom-up parsing: stack, input and actionsTop-down and bottom-up parsing: complete traces
- Three full shift-reduce tracesTop-down and bottom-up parsing: complete traces
- Entity spans, categories and contextNamed Entity Recognition
- Dictionary, rules, machine learning and deep learningNamed Entity Recognition
- Three complete entity-recognition examplesNamed Entity Recognition
- Prepare a grammar for CYK / CKYFaculty topic: CNF and CYK / CKY parsing
- CYK / CKY: complete charts, backpointers and ambiguityFaculty topic: CNF and CYK / CKY parsing
- Lexeme, lemma, sense and contextWord meaning, lexical relations and WordNet
- Every lexical relation with examplesWord meaning, lexical relations and WordNet
- WordNet: synsets, glosses and hierarchyWord meaning, lexical relations and WordNet
- What WSD solves and how approaches differWord sense disambiguation: Lesk, Walker and graph methods
- Lesk algorithm, from words to overlap scoresWord sense disambiguation: Lesk, Walker and graph methods
- Thesaurus / Walker approach, with three complete examplesWord sense disambiguation: Lesk, Walker and graph methods
- Random-walk WSD and supervised alternativesWord sense disambiguation: Lesk, Walker and graph methods
- Prior, likelihood and posterior—before the formulaNaive Bayes WSD: probabilities to a complete numerical
- Multinomial Naive Bayes: the exact solving methodNaive Bayes WSD: probabilities to a complete numerical
- Mam's bank numerical—every count and every multiplicationNaive Bayes WSD: probabilities to a complete numerical
- More WSD numericals: given likelihoods, unequal priors and repeated cluesNaive Bayes WSD: probabilities to a complete numerical
- Distributional hypothesis and a co-occurrence matrixVector semantics and word embeddings
- Sparse, dense, distributional and distributedVector semantics and word embeddings
- Word2Vec: CBOW and skip-gram from training pairsVector semantics and word embeddings
- Word analogies and sentence embeddingsVector semantics and word embeddings
- Why GloVe uses global statisticsGloVe: global co-occurrence and weighted error
- The GloVe objective and numerical interpretationGloVe: global co-occurrence and weighted error
- Reference, pragmatics and discourseUnit 6 · reference & discourse
- Pronoun resolution: methods and examplesUnit 6 · reference & discourse
- Hobbs’ algorithm, step by stepUnit 6 · reference & discourse
- Text classification and its full pipelineUnit 7 · applications
- Sentiment analysis: meaning, methods, exampleUnit 7 · applications
- Precision, recall, F1 and accuracyUnit 7 · applications
- Spelling correction and edit distanceUnit 7 · applications
- Machine translationUnit 7 · applications
- Question answeringUnit 7 · applications
- Information extractionUnit 7 · applications
- Information retrievalUnit 7 · applications
TEE · Read Chapter
- Start here · language & probability
- NLP foundations: language, tasks and the processing pipeline
- Text preprocessing, normalization and lexical diversity
- Text to numbers: label encoding, one-hot, BoW and TF–IDF
- Similarity and distance: every numerical step
- Corpora, lexical resources and data acquisition
- Regular expressions: read, build and test text patterns
- N-grams and language models
- From a corpus to sentence probabilities
- Perplexity, log probabilities and smoothing
- Faculty topics: interpolation and Stupid Backoff
- POS tagging and its learning approaches
- Markov chains and Hidden Markov Models
- Choosing the most likely tag sequence
- Chunking, chinking and context-free grammar
- Top-down and bottom-up parsing: complete traces
- Named Entity Recognition
- Faculty topic: CNF and CYK / CKY parsing
- Word meaning, lexical relations and WordNet
- Word sense disambiguation: Lesk, Walker and graph methods
- Naive Bayes WSD: probabilities to a complete numerical
- Vector semantics and word embeddings
- GloVe: global co-occurrence and weighted error
- Unit 6 · reference & discourse
- Unit 7 · applications
TEE · Revision
- All seven units: final revisionTEE · full-course revision
- Numerical answer frameworkTEE · full-course revision