It was probably in the 1980s (not sure) and probably in Scientific American (not sure) that I read an article reporting on research that had analyzed the frequency of letters, pairs of letters and triplets and quadruplets of letters in four or five languages. Two of the languages were English and Italian. I don't remember the other two or three.

The authors -- or rather, their computer -- then "wrote" a paragraphs in each language using the frequency rules for that language, first for letters only, then for letters plus pairs of letters, then adding triplets and finally adding quadruplets.

It was fascinating the way the paragraphs progressed from pure nonsense resembling nothing, to nonsense closely resembling each language. To a non-Italian speaker, the last Italian paragraph -- the one including the quadruplet rules -- looked like Italian. The English looked as though it ought to make sense....maybe with a few glasses of wine it would.

Does anyone recognize this article or this research?

Has this research been followed up? Are there applications of this research, or linguistic insights resulting from it?

  • I can't think of a single article for this because the idea is so basic. it is called n-grams (bigrams for 2 letter or two word combinations). en.wikipedia.org/wiki/N-gram. Have you tried a google search?
    – Mitch
    Nov 2, 2015 at 13:40
  • Multilingual Word Processing? by Joseph Becker? scientificamerican.com/search/…
    – Mitch
    Nov 2, 2015 at 13:52
  • @Mitch Thanks, I will check out the entire article later today. The article I remember may be basic now, but it looked pretty exciting when I read it more than 30 years ago -- perhaps because increased computing power was only just making large-scale playing with the concept feasible. Write it up as an answer, and I will upvote it.
    – user9732
    Nov 2, 2015 at 14:54
  • it's not really na answer because it was just half random noodling on the web. You need to do some serious noodling yourself on the web (that SciAm article is behind a paywall) and I'm not sure it is the one you are looking for, especially since it is a half-remembered guess.
    – Mitch
    Nov 2, 2015 at 15:00
  • @Mitch It isn't the one I am looking for -- I can tell from the description, but at the price of a sandwich, I am willing to look at it for whatever it has, including references.
    – user9732
    Nov 2, 2015 at 21:08

1 Answer 1


I don't recognize the article, but you're describing generative orthotactics (possibly as implemented in a Finite State Machine). (See also 'phonotactics', which does the same thing on sequences of phonemes/phones.) You may also have noted various 'Markov' Twitter accounts, which perform this generation using sequences of words, again instead of letters. The concepts of this approach are still used for different applications in fields like automatic speech recognition, phoneme-grapheme conversion, psycholinguistics (see 'wordlikeness'), morphology/language innovation, etc.

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.