Getting machines to handle language, which turned out to be most of the problem.
12 topics · 15 curated works
No prior grounding assumed.
Opinion Mining and Sentiment Analysis
Bo Pang & Lillian Lee · 2008
Surveys sentiment classification and opinion mining a decade into the field, organising the accumulated techniques by the linguistic phenomena —…
+1 more at this level
Assumes you know the vocabulary.
Neural Machine Translation of Rare Words with Subword Units
Rico Sennrich, Barry Haddow & Alexandra Birch · 2016
Segments words into frequency-ranked byte-pair-encoded subwords so a fixed-size vocabulary can still compose rare and unseen words instead of mapping…
+6 more at this level
Primary sources and full treatments.
Speech and Language Processing
Daniel Jurafsky & James H. Martin · 2000
The standard NLP text across three decades of the field changing underneath it, which makes its revisions a history of the subject.
+5 more at this level
12 of 15 works
Bo Pang & Lillian Lee
Surveys sentiment classification and opinion mining a decade into the field, organising the accumulated techniques by the linguistic phenomena — negation, comparison, subjectivity — that make bag-of-words classifiers fail.
link checked 17 Sept 2026Steven Bird, Ewan Klein & Edward Loper
Teaches language processing by making the reader build the pieces — tokenisers, taggers, parsers, classifiers — rather than by presenting the theory and its results.
link checked 17 Sept 2026Rico Sennrich, Barry Haddow & Alexandra Birch
Segments words into frequency-ranked byte-pair-encoded subwords so a fixed-size vocabulary can still compose rare and unseen words instead of mapping them to an unknown token.
link checked 17 Sept 2026Pranav Rajpurkar, Jian Zhang, Konstantin Lopyrev & Percy Liang
Argues reading comprehension should be benchmarked by requiring a system to extract an exact answer span from a passage, and builds a 100,000-question crowd-sourced dataset to force that comparison.
link checked 17 Sept 2026Guillaume Lample et al.
Replaces hand-built gazetteers and orthographic features with a bidirectional LSTM-CRF learned end to end from labelled text and unsupervised word embeddings, matching feature-heavy NER systems with almost none of the engineering.
link checked 17 Sept 2026Abigail See, Peter J. Liu & Christopher D. Manning
Combines a sequence-to-sequence summariser with a copy mechanism that can point at source words directly, fixing the factual-inaccuracy and repetition failures that pure abstraction produced.
link checked 17 Sept 2026Alex Wang et al.
Argues that no single task measures general language understanding and assembles nine tasks into one benchmark with a shared leaderboard, which then became the standard a model had to clear before claiming general competence.
link checked 17 Sept 2026Dan Hendrycks et al.
Tests language models on multiple-choice questions spanning 57 academic and professional subjects, from elementary mathematics to law, to show that fluency on text benchmarks does not imply broad factual knowledge.
link checked 17 Sept 2026Jason Wei et al.
Shows that prompting a large language model with a few worked reasoning steps, rather than just answers, substantially improves its performance on arithmetic, commonsense and symbolic reasoning tasks it otherwise gets wrong.
link checked 17 Sept 2026Daniel Jurafsky & James H. Martin
The standard NLP text across three decades of the field changing underneath it, which makes its revisions a history of the subject.
link checked 17 Sept 2026Brown, Della Pietra, Della Pietra & Mercer
Recasts translation as noisy-channel decoding whose parameters can be estimated from aligned text alone, removing the hand-written grammar that translation systems had been built around.
link checked 17 Sept 2026Mikolov, Chen, Corrado & Dean
Drops the hidden layer and trains word vectors by predicting context, cheaply enough to run over billions of words, and the resulting geometry turns out to encode analogy.
link checked 17 Sept 2026This subject genuinely sits in more than one domain. These fields approach the same ground with different methods.