GoKawiilSide Project · 2026 Greek Alphabet
Cards A set of cards I made to help my kids learn the Greek alphabet through visual associations — each object is drawn so that it looks like the letter its name begins with.
The Idea We live abroad in China, and Greek is one of three languages my kids are learning. They were three and a half when I started these cards about five months ago, so I wanted something playful to nudge them along. My first attempt was an “A for airplane”-style deck — pictures of objects whose names start with each letter. After printing version one, I had the epiphany:
What if the object didn’t just start with the letter — what if it looked like it too?
The shape of the letter pulls up the object, and the object’s name pulls up the letter. Research seems to back this up — kids learn the alphabet far faster this way than by rote. Finding the Objects In the beginning I relied on my memory to come up with these associations, but you run out of ideas very quickly. So I got a bit more organised about it: Dictionary I downloaded an entire Greek dictionary that contained not just the words but also a list of words per letter, along with the frequency with which each word appears in Greek text. I used GreekLex, which contains 35,304 Modern Greek words ranging in length between 1 and 22 letters.
greeklex.csv · excerpt IDnr Word Length LemmaFreq WordFreq 223 φως 3 125 90.2 224 χαν 3 7 7.4 225 χαφ 3 14 13.5 226 χολ 3 6 5.6 227 χοπ 3 3 2.8 228 ψες 3 0 0.1 229 ψιτ 3 0 0.2 230 ωδή 3 3 1.4 231 ωθώ 3 17 0 232 ώρα 3 549 335.8 233 ώση 3 7 3.6 234 αβάς 4 2 1 235 αγάς 4 1 0.1 236 άγια 4 2 1.9 237 αγνή 4 4 3.2 238 άγος 4 3 1.1 239 άγρα 4 6 3 240 αγώι 4 0 0.1 241 αδάμ 4 4 4.4 242 άδης 4 2 0.4 243 αέρι 4 0 0 244 αθώα 4 6 5.7 245 αίγα 4 0 0 A small slice of the GreekLex corpus. Each row is a word with its length in characters, its lemma frequency (how often the base form occurs in the corpus) and its word frequency (how often this exact surface form occurs). The frequency columns are what let me filter for words a child would plausibly recognise.
Filtering I ran a filter to keep only words that were: between 3 and 10 characters long, and
had a frequency of at least 100 in the corpus, so they wouldn’t be too rare. The aim was a vocabulary that my kids would plausibly know. Visual candidate generation That still left 50 to 2,500 words per letter — too many to eyeball. So I fed them to ChatGPT in batches of 50, asking for each one whether its referent could be drawn to echo the letter’s shape, and how. The list usually came back down to 10–200 candidates. Ω Omega was the extreme case — essentially no match. Most suggestions were weak, but every batch had a few good ones.
triage · letter ε · sample output › ελιά — an olive tree can be stylised with a vertical trunk on the left and three rounded clusters/branches extending to the right, echoing the three arms of ε › ελαία — same idea as ελιά: a small olive tree with a slim trunk and three leafy bulges to the right can read clearly as ε › ελάφι — a deer’s head in profile could be stylised so the neck forms the spine of ε and the snout, chest, and lower jaw create the three outward curves
Sample of what ChatGPT returned for one of the letter ε batches. Each line is a candidate word with a suggestion for how its referent could be drawn to echo the shape of the letter. Most suggestions weren’t usable, but a handful in every batch were genuinely promising.
... continue reading