Previous attempts at getting machines to understand humour have failed miserably, because what is funny to humans is subjective and complex -- and fiendishly difficult to programme.
But, says New Scientist, Julia Taylor and Lawrence Mazlack of the University of Cincinnati in Ohio have devised a prototype joke-detection software.
They began by loading a programme with a database of words, extracted from a children's dictionary to keep things simple, and then supplied it with examples of how the same word can have different meanings depending on the context.
When presented with a text, the programme uses that knowledge to work out how new words may relate to each other, and what they probably mean.
If it fails to find a word that matches its context, it rummages around in a digital pronunciation guide for similar-sounding words.
And if any of those words are a better fit for the rest of the sentence, the passage is flagged, ha ha, as a joke.
So far, the joke-bot only understands rather leaden puns and still delivers a blank look when facing more complex stuff or dead-pan humour.
Even so, the researchers hope it will add a kindlier touch to robots of the near-future which will act as human companions or helpers.
Here's an example of what tickles a circuit board:
- Mother to boy: "My, you've been working in the garden a lot this summer."
- Boy: "I have to, because teacher told me to work a lot" (thus a pun on working the soil and doing schoolwork).
The research, reported in next Saturday's New Scientist, was presented last week at a conference of the American Association for Artificial Intelligence in Vancouver, Canada.