Rank-frequency distributions in several domains are often approximately power-law-like, with word frequencies the classic case.
Zipf's finding: the most common word appears about twice as often as the second and three times as often as the third. Exact exponents are not universal, and the city-size and firm-size applications show substantial variation and remain empirically debated.
In practice
- A support team finds a handful of question types cover most of its volume, and a very long list covers the remainder one case at a time.
- Concentration recurs across languages, city sizes and firm sizes; the exact exponent does not, and no agreed explanation covers the family.
- Possible explanations
- One explanation is preferential attachment, in which frequent items become likelier to be used again. Another is a balance between speaker and hearer effort. A third is that simple random processes produce the same shape, which is why the pattern alone does not identify a mechanism.
- In its favour
- The regularity in word frequencies is strong, holds across languages and corpora, and is a standard benchmark in language modelling.
- Against it
- Exponents vary, deviations at both tails are common, the city and firm applications are disputed, and because random typing reproduces the shape, the law describes without explaining.
- Standing
- Exact exponents are not universal; city-size and firm-size applications vary substantially and remain debated.
- Further reading
- Wikipedia
- Subject
- Science & Epistemology
- Essay
- The ways evidence misleads