Coding classes in East Ayrshire

The learner loads a Project Gutenberg collection of Burns's poems and songs, splits it at blank lines into 4,042 stanzas, and keeps the 688 that have exactly six lines. Then come two tests. The shape test counts rough syllables in each line by counting groups of vowels, and asks whether lines one, two, three and five are long, seven syllables or more, while lines four and six are short, five or fewer. The rhyme test compares the last letters of the end words and asks whether lines one, two, three and five match each other, and four matches six.

The rhyme test throws away most of the shapes, and a look at the rejects shows why it is wrong to trust it. Among the first stanzas it rejected are the verses of Poor Mailie's Elegy, true Burns stanzas every one, with end words like "prose, nose, close, remead, woes, dead". The book’s own glossary gives "remead" as remedy, and to the ear it rhymes with "dead"; to a program comparing the last two letters, "ad" and "ed" do not. Another verse ends "shape, raip, gape, dread, crape, dead": "raip" is meant to rhyme with "shape" but shares none of its final letters. A stricter three-letter test halves the matches again.

So which number is right? Probably something closer to the shape count than the rhyme count, but the honest answer is that the program cannot tell without a better idea of sound. This is the everyday trade-off of pattern recognition: a strict test misses real examples, a loose one lets in false ones, and the only way to know which you have is to look at the mistakes. Computer scientists call these recall and precision, and every search engine and spam filter balances the same two.

So a young person in East Ayrshire should learn to code in 2026 to test language tools on the words that matter locally. The longer argument is in why teenagers should still learn to code in 2026.