TextBlob

A gentler on-ramp to NLP than raw NLTK, TextBlob wraps noun phrase chunking, part-of-speech tags, tokenization, spelling correction, word inflection, and sentiment scoring behind one friendly object: construct it from a string and read linguistic annotations off properties. Sentiment comes from two interchangeable backends, a lexicon-and-rules analyzer derived from the pattern library that returns polarity and subjectivity floats, and a Naive Bayes model trained on movie reviews via NLTK; both are word-level methods that routinely miss negation and sarcasm, so scores are best treated as quick baselines rather than production signals. Install is pip plus a one-time corpora download, after which everything runs offline on CPU. Steven Loria has maintained it under the MIT license since 2013, with the 0.19 release landing in early 2025 and community extensions covering German and French. It sits opposite spaCy on the tradeoff curve: less accurate than neural pipelines and slower per token at scale, but learnable in an afternoon, which keeps it a fixture in classrooms and exploratory scripts.