TextBlob Cheat Sheet

1Install & corporapip + nltk data

pip install textblob★
Pulls NLTK. Python 3.10+.
python -m textblob.download_corpora★
Downloads the NLTK corpora TextBlob needs (punkt, averaged_perceptron_tagger, wordnet, brown, ...). Do this once. ...download_corpora lite for a smaller set.
# LookupError: Resource ... not found -> run download_corporagotcha
Missing-corpus errors at runtime almost always mean you skipped the download step above.
from textblob import TextBlob★
The one import you need for most tasks. Word, Sentence, WordList also live in textblob.

2The TextBlob objecteverything is an attr

b = TextBlob("TextBlob is amazing. Try it!")★
Wrap any string. All analysis is lazy attributes/methods on b.
b.words # WordList(['TextBlob', 'is', 'amazing', ...]) b.sentences # [Sentence("TextBlob is amazing."), ...]★
.words is a tokenized WordList; .sentences splits into Sentence objects (each is itself a blob).
str(b) · b[0:8] · b.upper() · b + " more"
A blob behaves like a string: slice, concatenate, and call string methods, getting blobs back.
for s in b.sentences: print(s, s.sentiment.polarity)
Iterate sentence-blobs to analyze each one independently.

3Tokenize & POS tags.words / .tags

b.words # WordList of tokens (punctuation stripped)★
Word tokenization. b.tokens keeps punctuation as separate tokens (NLTK-style).
b.tags # [('TextBlob','NNP'), ('is','VBZ'), ('amazing','JJ'), ...]★
Part-of-speech tags as (word, tag) tuples (Penn Treebank tagset: NN=noun, VB=verb, JJ=adjective...).
b.word_counts["amazing"] # case-insensitive count
A dict-like counter over the blob's words.
from textblob import Sentence Sentence("Just this one.").tags
Same attributes on a single Sentence.

4Noun phrases.noun_phrases

b.noun_phrases # WordList(['textblob', 'noun phrase', ...])★
Extracts noun-phrase chunks — a quick keyword/topic signal. Returns lowercased phrases.
from textblob.np_extractors import ConllExtractor b = TextBlob(text, np_extractor=ConllExtractor())
Swap the extractor: FastNPExtractor (default) vs ConllExtractor (chunk-parser based, often better phrases).
"amazing" in b.noun_phrases # membership test
WordList supports the usual list ops (in, indexing, .count()).

6Inflection & lemmatizeWord objects

from textblob import Word Word("cats").singularize() # 'cat' Word("cat").pluralize() # 'cats'★
Word exposes inflection helpers. b.words[i] items are already Word objects.
Word("running").lemmatize("v") # 'run'★
Lemmatize to the dictionary form. Pass the POS ("v", "n", "a") — default is noun, so verbs need "v".
Word("octopi").lemmatize() · w.stem()
.stem() is the cruder Porter stem; .lemmatize() gives real words.

7Spelling correction.correct()

TextBlob("I havv goood speling").correct() # TextBlob("I have good spelling")★
Whole-blob spell-correction (Peter Norvig's algorithm). Fast to write, not always accurate.
Word("speling").spellcheck() # [('spelling', 0.98), ('spewing', 0.02)]★
Per-word candidates with confidence scores — pick your own threshold.
# .correct() is heuristic — verify on your domain before trusting ittip
It over-corrects rare/technical words. Good for noisy user text, not code or names.

8WordNet & definitionslexical database

Word("blob").definitions # ['a soft mass...', 'an indistinct shapeless form']★
Dictionary definitions via WordNet.
Word("car").synsets · w.get_synsets(pos="n")
WordNet synsets for word senses. Import from textblob.wordnet import VERB, NOUN to filter by POS.
dog = Word("dog").synsets[0] cat = Word("cat").synsets[0] dog.path_similarity(cat)
Semantic similarity between senses via the underlying NLTK WordNet API.

9n-grams & countsfeatures

b.ngrams(n=2) # [WordList(['TextBlob','is']), WordList(['is','amazing']), ...]★
Sliding n-grams — a quick feature source for simple models.
b.word_counts · b.np_counts # noun-phrase frequency dict
Case-insensitive frequency dicts over words and noun phrases.
b.parse() # chunk/parse tags via the Pattern parser
Full shallow parse string (POS + chunk + relation tags).

10Text classificationtiny models

from textblob.classifiers import NaiveBayesClassifier train = [("I love it", "pos"), ("awful", "neg")] cl = NaiveBayesClassifier(train)★
Train a classifier from (text, label) tuples in two lines. Also DecisionTreeClassifier, MaxEntClassifier.
cl.classify("This is great") # 'pos' cl.accuracy(test_set)★
Predict labels and score on a held-out set. cl.show_informative_features() to inspect.
b = TextBlob(text, classifier=cl) b.classify() # classify a whole blob / its sentences
Attach a trained classifier to a blob for per-blob/per-sentence prediction.

11Gotchas & ecosystemwhat's gone, what next

# b.translate(to="es") / b.detect_language() -> REMOVEDremoved
Translation & language detection were dropped — they relied on an unofficial Google Translate endpoint that broke. Use deep-translator, googletrans, or a cloud API instead.
# TextBlob = teaching / prototypes / baselines★
It's rule/lexicon based and English-first. For production accuracy, multilingual support, or speed, move to spaCy, NLTK directly, or Hugging Face transformers.
pip install textblob-de · textblob-fr # language add-ons
Community language packs add German/French analyzers & taggers.
# built on NLTK + a maintained fork of Pattern (nltk, pattern taggers)
Under the hood it orchestrates NLTK corpora + Pattern; you can drop to NLTK anytime for more control.

TextBlob cheat sheet