The Neath projectSorting Welsh street names by the Welsh alphabet
240 street names, one line of Python that gets them wrong, and a sort key that gets them right.
One query to OpenStreetMap for named roads in and around Neath returned 1,197 road sections with 678 different names. On 413 of those sections mappers had also recorded a Welsh name, giving 240 distinct Welsh street names. Heol begins 70 of them, Stryd 47, Clos 25 and Rhodfa 10. Three contain an accented letter, among them Stryd y Dŵr. The learner's first move is the obvious one, sorted(names), and the second is to ask what a Welsh reader would expect instead.
The expected order comes from the Welsh alphabet as the Unicode Common Locale Data Repository sets it out for computers: ch comes after c, dd after d, ff after f, ng after g, ll after l, ph after p, rh after r and th after t, each counting as one letter. Capital letters and accents are ignored when deciding order. In Python this becomes a key function: fold the case, strip the accents, then walk through each word taking two characters at a time whenever they form one of those letters, and turn every letter into its position in the alphabet.
240 Welsh street names in Neath, compared with the Welsh alphabetical order, our Python run| Method | Names in the wrong position | Pairs in the wrong order |
|---|
| sorted() with no key | 67 | 79 |
| Lower case and accents removed | 39 | 42 |
| Welsh letters as single units | 0 | 0 |
Plain sorted() failed in three separate ways. Capitals: it puts every capital Y before every small y, so Cwrt Y Gollen lands ahead of Cwrt y Cadno. Accents: the ô in Clôs Llwyneryr has a larger character number than any plain letter, so that street drops below every Clos name. And the Welsh letters themselves, which cause most of what is left once case and accents are cleaned up (a few more come from hyphens and apostrophes, which our key treats like spaces): in Welsh, Lon Fedwen comes before Llys Andrew, because l is an earlier letter than ll, and Heol Forster comes before Heol Ffranc. Python's character-by-character order sees two letter ls and puts Llys first.
Five names contain the characters n and g together: Bryngwyn, Heol Llangatwg, Heol Longford, Stryd Whittington and Teras Rockingham. In none of them is it the Welsh letter ng. Bryngwyn joins bryn and gwyn, Llangatwg joins llan and a form of Catwg, and the other three are English names. A key that always reads ng as one letter is wrong five times here. Yet treating ng as n followed by g changes no position in the sorted list, because no two of these names differ only at that point. The learner writes that down as a lesson about testing: a bug that no current data exposes is still a bug.
Ages 8 to 11
Cards with Welsh words: sort them English style, then Welsh style, and count the swaps.
Ages 11 to 15
Use sorted() with key=str.lower, then explain the names it still puts in the wrong place.
Ages 15 and up
Write the full Welsh sort key, test it on all 240 names and design a test that catches the ng problem.
Sources and limits
Street names are from OpenStreetMap under the Open Database Licence, as mapped on 30 September 2026; some Welsh names may be missing or mistyped. The letter order is the CLDR Welsh collation. The word splits in Bryngwyn and Llangatwg are standard Welsh, and the counts are from our own code.
