extracao-de-metricas-estilisticas-cognitivo-literarias

Compute a style vector containing type-token ratio, Flesch-PT readability, and sentence-length standard deviation for both texts, then calculate the Euclidean distance between them. A distance below the configured threshold (default 0.25) indicates the text stays within the author's style zone.

Use the Flesch-PT adaptation: 248.835 minus 1.015 times average sentence length minus 84.6 times average syllables per word. The original English Flesch-Kincaid constants produce systematically inflated scores for Portuguese because its words average more syllables.

spaCy with the pt_core_news_lg model provides Portuguese tokenization and lemmatization, while NLTK or Pyphen can be adapted for syllable counting. These are required to filter content words (nouns, verbs, adjectives, adverbs) for the lexical richness metric.

A minimum of 3 chapters or 5,000 words of manually written text is required to calibrate the author's reference style vector. Smaller corpora produce unstable baseline metrics that make drift detection unreliable.

Avoid it for fixed-form poetry such as sonnets or haiku, where sentence length and syllable counts follow metrical rules rather than stylistic choice. It is also unsuitable for simple grammar or spelling checks, which standard proofreading tools handle.