Counts syllables of your (English) triplet to determine whether is haiku. I mean haiku in the "5-7-5 syllables" sense that English-speaking computer programmers mean, not in the proper sense that literate Japanese mean.
If you run an internet haiku contest please consider vetting the entries against this webapp or something like it(1).
The first way it counts syllables is by looking up the answer in a big file from the CMUSphinx project. (Haikufinder is a project that does something very similar in Python. There may be others out there.)
The second-and-a-half way is by counting vowel runs and using a Porter stemmer. The Porter stemmer is a thirty-year-old algorithm for finding the common stem of similar English words; it can tell you that walk, walked, and walking all have walk as their stem. It's useful for text searching and indexing and stuff. It turns out also to work pretty well for improving naive syllable counts. For example, a naive count of vowel runs in the word hoeing erroneously predicts that there's one syllable; identifying the -ing suffix, noting that it's a syllable, and then counting the one syllable in the stem ho gives the correct count of two syllables.
The Porter stemmer is in raw Java. The rest is Clojure.
Not dissimilar approaches might be found elsewhere on github.
So English contains a bunch of words that can, for the same spelling, have a different number of syllables depending on the context, the speaker, and sometimes the speaker's whim:
- moped: "I moped about the theft of my moped."
- blessed: Can be one or two syllables depending, perhaps, on who did the blessing.
- doing: "If he keeps doing that to the rubber band it will snap with a loud doing!"
- Porsche: Words brought to English from other languages are often pronounced differently by different folks.
- In Texas, elm might have two syllables, and chimney might have three. In Boston, settler might have two syllables; elsewhere three.
- and so forth
I have chosen so far to ignore this problem.
Assuming you have Maven (2.x) all set up, you can download the dependencies and run the webapp by typing:
mvn clojure:run
and visiting http://localhost:8080. The first one you submit will be slow while the app parses the cmudict.txt file.
This code is running on http://www.daghlian.com/haiku so you can try it for yourself.
I used Dr. Porter's stemmer code, which is in the public domain, so all of this is, too. Drop me a line if you do anything interesting with it.
(1) I actually wrote this because there was this one week during which Daring Fireball and boingboing both featured clever little haikus that weren't even 5-7-5 (and definitely weren't Japanese-style poetry).