For most of us, being asked ``What do you give a hurt lemon?'' and being told, ``Lemon aid'' sounds like the occasion for deep depression. But the fact that a computer program was able to ask that question and supply that answer has implications for structural linguistics - and for artificial intelligence.

And, as Dr Kim Binsted told the Humour, Art and the Brain festival at Winchester in southern England recently, its applications may go far beyond the automated production of joke-bearing lolly sticks.

Binsted created the first Joke Analysis and Production Engine (Jape) in 1996, to combine her academic interest in artificial intelligence (AI) with her personal interest in improvisational comedy. Now, a fifth-generation upgrade of the program is being used in Standup - a three-year language-teaching experiment in Edinburgh and Dundee funded with 364,000 (HK$5.18 million) from the Engineering and Physical Sciences Research Council.

The ``System To Augment Non-speaker's Dialogue Using Puns'', to give it its full name, helps speech-impaired children incorporate humour into their exchanges. Other versions of the technology can be used in automated ``chatbots'' for second-language teaching.

``The program will chat back but also integrate jokes into the language about the topic of the week,'' says Binsted. ``Because the program is working with the same kids week in week out, it can integrate things it learns about them into the humour. So if the kid says their brother is tall, the assistant can make a somewhat lame joke about `I hear giraffes look up to him'.''

The principle behind the original Jape was, by concentrating on puns - which AI scientists seem to agree are the lowest form of wit - to generate algorithms that could produce simple jokes.

``We started with simple puns because they don't require much knowledge,'' says Binsted.

An example: Jape would look at its dictionary and perform a three-step operation. 1) Make a new ``word'' by, say, substitution: ``spook-tacles'' for ``spectacles''. 2) Find a plausible description for the item: ``glasses for ghosts''. 3) Arrange the relationship in Q & A form: ``What do near-sighted ghosts wear? Spooktacles.'' Jape's efforts were then tested on 120 eight- to 11-year-olds, who were asked to assess the results for ``jokiness'' and funniness.

``Because we were testing these jokes on eight- to 11-year-olds we can't put any sexual humour into the test,'' says Binsted. ``But, bless it, when it comes across a phrase, it will riff on that phrase until it's done. The first time I hooked it up to the big dictionary, one of the first phrases it found promising was `male orgasm' - and it went off on that for ages. The variations!''

A more complex program, Wiscraic - Witty Idiomatic Sentence Creation Revealing Ambiguity In Context - which works with idioms rather than simple puns, and was geared to helping language acquisition, was developed by Binsted's associate, Justin McKay. ``The performing lumberjack took a bough,'' it announced, for example.

There aren't many laughs in the title of a recent paper co-authored by Binsted, The Cognitive Linguistics of Scalar Humour. In fact, you would hardly be able to guess that it explains what you're really doing when you tell your friend: ``Yo mama's so fat, her ass has its own postcode.''

But ``yo mama'' jokes, according to Binsted, could be the next step up in complexity for computer-generated mirth. When you leave the realm of joke generation, and enter that of amusement, AI starts to shade into philosophy: Getting a computer to crack a joke is one thing; getting a computer to laugh at a joke is quite another. But if you ever wanted a computer that could surf the Internet, manage your finances, and help you learn French by insulting your mother, it may be around the corner.