Distinct vocabulary in a corpus grows as a sublinear power of total tokens.
Herdan and Heaps independently. The practical version: a corpus ten times larger yields maybe three times the vocabulary, which is why dictionary compilation hits diminishing returns.
In practice
- A support knowledge base of 500 articles covers most questions; at 5,000 articles it covers slightly more and is far harder to search.
- A style guide of forty rules catches most errors; four hundred rules catch a few more and are never read.
- Further reading
- Wikipedia
- Subject
- Science & Epistemology
- Essay
- The ways evidence misleads