Language learning
What learning a non-Latin alphabet actually takes
Learning the letters of Armenian or Greek takes a few days. Reading them without consciously decoding takes months. Almost every tool optimises the first part, which was never the bottleneck, and abandons you in the second, which is where people quit. This post is about what that middle stage actually requires, using the two alphabets MemDecks supports most deeply.
The letters are the easy part
Greek has 24 letters. Armenian has 39. Both are alphabets in the strict sense — one letter, one sound, with far fewer exceptions than English manages. A motivated adult can name every Greek letter in an afternoon and every Armenian letter within a week.
Which is why "learn the alphabet in 20 minutes" content is technically honest and practically useless. Naming a letter in isolation is recognition. Reading is something else: converting a string of shapes into a word, fast enough that the meaning arrives before working memory drops the beginning of the word.
The real bottleneck is decoding speed
Between "I know the letters" and "I can read" sits a stage that lasts months. In it you can decode any word given enough time, and that time is the problem. Reading a five-letter word letter by letter takes long enough that by the time you reach the end you have lost the beginning, and the word never assembles into meaning.
The fix is not more letter practice. It is volume of real words in the real script, in contexts where the meaning is already available so that the shape and the meaning can attach to each other. This is the stage where a learner needs the most support and typically receives the least, because it is unglamorous: there is nothing to unlock and no visible milestone, just repetition until the shapes stop being shapes.
Transliteration is a crutch, and you need crutches
Transliteration — writing բարև as barev — is genuinely a problem. If it is always there, the eye takes the free path and never learns the script. Plenty of learners have spent a year on a language and still cannot read a shop sign, because every card they ever saw came with the Latin version underneath.
It is also indispensable at the beginning. Without it a beginner cannot pronounce anything, cannot check their guess, and cannot use audio to correct themselves. Removing the crutch before the leg is ready does not accelerate the leg.
MemDecks shows transliteration on cards in the nine study languages whose scripts are not Latin, and treats it as a support that a learner should notice themselves relying on. Armenian and Greek use deterministic transliteration written for those alphabets specifically — the same input always produces the same output, so the mapping you internalise is real and consistent rather than a probabilistic guess that varies between words.
What actually helps in the decoding stage
| Practice | Why it works |
|---|---|
| Real words rather than letter drills | You are training word-shape recognition, not letter naming. Letters in isolation do not transfer. |
| Audio attached to every word | Sound and shape have to be encoded together, or you learn to read silently and wrongly. |
| Producing the script, not only reading it | Spelling a word forces attention to every letter; recognition lets you guess from the first two. |
| Gradual removal of the Latin support | The crutch has to come out eventually, and gradually beats abruptly. |
| A picture instead of an English gloss | It cuts English out of the loop, so the script attaches to the meaning rather than to a translation. |
How MemDecks handles it
Four concrete things, all live as of 2 September 2026.
- Every card carries the script, a transliteration and audio together. The three arrive as one object rather than as three optional extras, so the association forms from the first exposure.
- Alphabet Quest. A game built for exactly this stage: text that begins in English and gradually shifts into the target script, so decoding happens inside sentences you can already understand. It supports Armenian, Greek and Russian.
- Word Type. Rebuilding a word letter by letter from a mixed pool is production, not recognition, and it is where sloppy letter knowledge gets exposed.
- An on-screen keyboard. MemDecks ships a virtual keyboard for the nine non-Latin study languages, so you can type in the target script without changing your operating-system input method — which is the friction that stops most people ever typing in the script at all.
Armenian specifically
Armenian's 39 letters include several near-identical pairs that separate only by a small stroke, and a set of aspirated-versus-unaspirated consonant distinctions that English does not mark and English speakers therefore do not hear. Both are visual-confusion problems, and both respond to production practice rather than reading practice: you find out you cannot tell two letters apart when you have to choose one.
The MemDecks word-difficulty scorer counts visual confusion explicitly, so words containing easily-confused letters carry higher card HP and therefore demand more repetitions before MemDecks treats them as done.
How long it takes
Honestly: a few days to name every letter, several weeks of daily contact before short words arrive as words, and months before reading stops being an activity you notice yourself doing. The scale is the same for Greek and Armenian, and there is no way to compress it that does not involve more contact with the script.
What can be changed is whether those months are pleasant. That is the entire argument for putting the script inside games rather than inside drills.