textstat-py

Text analysis for Python. Zero dependencies.

NLTK is 40MB and requires a corpus download just to tokenize a sentence. textblob pulls in NLTK. spaCy needs a 50MB model file before it'll tell you anything. For most text analysis tasks — readability scores, vocabulary stats, sentiment, writing quality signals — none of that weight is necessary.

pip install textstat-py

What it does

$ textstat essay.txt
=== Text Statistics: essay.txt ===
  Words            : 1243
  Sentences        : 67
  Reading time     : 6.2 min
  Flesch ease      : 58.4  (0=hard, 100=easy)
  FK grade level   : 11.2
  Grade consensus  : 10.8  (avg of 4 formulas)
  Lexical diversity: 0.71  (unique/total words)
  Sentiment        : neutral  (polarity=0.02)
  Passive voice    : 0.18  (fraction of sentences)
  Adverb density   : 0.031  (>0.05 may signal weak verbs)
  Top words        : data(18), model(14), training(11), loss(9), layer(7)

Compare two versions of the same document:

$ textstat --compare draft.txt final.txt
Metric                      A: draft.txt          B: final.txt          Delta
------------------------------------------------------------------------
Words                       1891                  1243                  -648
Reading time (min)          9.46                  6.21                  -3.25
Flesch ease                 44.1                  58.4                  +14.3
Grade level                 13.2                  10.8                  -2.4
Passive voice ratio         0.31                  0.18                  -0.13
Adverb density              0.071                 0.031                 -0.04

Install

pip install textstat-py

Python 3.8+. No dependencies. Single file.

CLI

textstat document.txt          # full report
textstat --json document.txt   # JSON output
textstat --wpm 250 document.txt  # custom reading speed
textstat --compare before.txt after.txt  # side-by-side diff
cat text.txt | textstat        # stdin

Python API

from textstat import analyze, flesch_reading_ease, grade_level_consensus

text = open("essay.txt").read()

# Quick scores
print(flesch_reading_ease(text))    # 58.4
print(grade_level_consensus(text))  # 10.8

# Full analysis dict
stats = analyze(text)
print(stats["passive_voice_ratio"])  # 0.18
print(stats["adverb_density"])       # 0.031
print(stats["top_words"])            # [("data", 18), ("model", 14), ...]

Functions

Readability

• flesch_reading_ease(text) — 0–100, higher = easier
• flesch_kincaid_grade(text) — US grade level
• gunning_fog(text) — years of education needed
• coleman_liau_index(text)
• automated_readability_index(text)
• smog_index(text)
• grade_level_consensus(text) — mean across all grade formulas

Writing quality

• passive_voice_ratio(text) — fraction of sentences with passive constructions
• adverb_density(text) — fraction of words that are -ly adverbs (>0.05 is a signal)

Vocabulary

• lexical_diversity(text) — type-token ratio
• mattr(text, window=100) — moving-average TTR, stable for long texts
• herdan_c(text), yule_k(text) — length-robust vocabulary richness
• hapax_legomena_ratio(text) — fraction of words appearing exactly once
• vocabulary_richness(text) — all of the above as a dict

Counts & structure

• count_words(text), count_sentences(text), count_paragraphs(text)
• reading_time(text, wpm=200)
• sentence_stats(text) — min/max/mean/median sentence length
• paragraph_stats(text) — word counts per paragraph

Sentiment

• sentiment_polarity(text) — −1.0 to +1.0, lexicon-based, no model needed
• sentiment_label(text) — "positive" / "neutral" / "negative"

N-grams

• top_ngrams(text, n=2, k=10) — most frequent n-grams
• ngram_diversity(text, n=2) — unique n-grams / total positions
• ngram_stats(text) — bigrams + trigrams bundled

Misc

• top_words(text, n=10) — most frequent non-stopword words
• word_frequency_distribution(text) — total tokens, unique types, Zipf fit
• text_density(text) — content words / total words

License

MIT