Post-Production Checking

At the time of testing, the only post-production checks in JAPE-1 were that the punchline of the riddle was not a real noun phrase (i.e. that it was not in the lexicon), and that none of the key lexemes used to build the riddle were accidentally identical (i.e. they were identical only if linked by an identity link). Other less obvious checks were not included in pre-evaluation JAPE-1in the hope that the evaluation would reveal their heuristic power so that they could be added later.

Some good heuristics were expected to be:

question length:

Questions that are too long seem unwieldy; questions that are too short don't provide enough information to suggest the punchline.

alliteration and rhyming:

Punchlines made up of alliterative or rhyming words are often more successful. Spoonerisms and related types of joke rely largely on this effect for their humour.

`funny letters':

It is a commonly-held belief among stand-up comedians that certain letters (particularly ``k'', ``q'', ``v'', ``w'', and ``z'') are inherently funny - of a pair of synonyms, the one containing the most such letters is funnier. This may be a myth; on the other hand, it is may explain why one joke is a little bit funnier that another similar one.

subject matter:

Jokes that have `funny' subjects (sex, religion, politics, etc.) are generally funnier than similar jokes about more mundane issues.

accidental associations:

It was expected that some jokes produced by JAPE-1 would be funny in unforeseen ways. For example, the question part of the riddle could be syntactically ambiguous, thus making the punchline a more humorous answer; or a semantic link not included in the lexicon could be suggested by an accidental juxtapostion of words in the riddle.

Obviously, some of these potential heuristics are easier to implement than others. For example, if accidental associations were easy to spot, they would not be accidental - they would be included in JAPE-1's lexicon, schemata, or templates. Also, none of these post-production heuristics could absolutely determine whether or not a given joke is funny; however, they could be used to partially order JAPE-1's output in terms of quality.