Heaps's law

Heaps's law describes how a text corpus’s vocabulary size grows approximately as a power of its total number of words. Its parameters are estimated from observed text and vary across corpora.

Connect

Vocabulary growth: The increase in the number of distinct words as more text is observed. Heaps's law models this growth as a power of corpus length.
Corpus linguistics: The study of language through systematically collected bodies of text or speech. Heaps's law is estimated from observed corpora.
Information retrieval: The field concerned with finding relevant information in documents or other data. Vocabulary growth informs estimates of the distinct terms an index must represent.
Good–Turing frequency estimation: A statistical method that estimates probabilities of unseen events from frequencies of observed events. It estimates unseen word mass, whereas Heaps's law predicts vocabulary growth with corpus size.
Power law: A relationship in which one quantity varies as a fixed power of another. The law’s defining curve is a power-law relationship.
Corpus: A curated collection of texts or speech assembled for linguistic analysis. Corpus composition determines which vocabulary-growth pattern is measured.
Full-text indexing: The construction of searchable indexes from the words in a document collection. Heaps's law helps estimate how an index vocabulary expands with added documents.
Mandelbrot's law: A rank-frequency distribution model that generalizes Zipf's law with an offset parameter. It models word frequencies by rank rather than vocabulary accumulation over corpus length.
Type-token ratio: The ratio of distinct words to total word occurrences in a text. Unlike this ratio, Heaps's law models how vocabulary changes with corpus size.
Tokenization: The process of dividing text into units such as words, subwords, or characters. Different token boundaries change both the total word count and vocabulary size.