Why I constructed BabyLang

Goo-goo ga-ga, mother freaker. Introducing the language for interpreting a baby's babbling.

The idea has an unsettling precedent. There's a story — attributed to Herodotus, and told about the Egyptian pharaoh Psamtik — of a ruler who raised a pair of infants in total isolation from language, certain that whatever word they spoke first would reveal humanity's original, God-given tongue. It's a cruel experiment and a bad one; what it actually shows is that language is instinctive as much as it's learned, not that isolation reveals some primordial Ursprache. I'm not repeating the experiment. I'm doing the opposite: trying to give meaning to the gurgling, cooing and slobbering sounds babies already make, unprompted, to express themselves.

A baby's mouth is nothing like an adult's. Neither is its brain. When children first start talking it sounds different to an adult — pitch is higher, and certain consonants are simply hard to produce while your baby teeth are coming in or falling out. I still call my sister "sister" instead of her name, Genevieve, because I couldn't pronounce "Genevieve" as a baby and it stuck. Babies also haven't learned to control their own saliva yet. A lot of what sounds like babbling is really just coincidental noise — vocalising while trying to manage slobber and move around an environment at the same time.

Building BabyLang meant making some unusual design calls. First: how much of the signal is in loudness versus pitch. A louder cry and a quieter coo can mean very different things, but Glossikon's phoneme system doesn't distinguish loud from quiet — sound intensity isn't a phonemic feature it models, and honestly the standard phoneme inventory doesn't fully cover everything a baby's mouth can produce either. Second: how much grammar to give it at all. I decided against building a language a baby's mouth could practically produce in full — that's not the brief. Instead BabyLang stays deliberately minimal: repetition and pitch standing in for a grammar the speaker doesn't have yet. Glossikon's typology tools describe it accordingly — isolating, head-final, and, tellingly, free word order. There isn't a real syntax to constrain; a baby babbling isn't building sentences, so the language shouldn't pretend it does.

For the script I went with Japanese hiragana, which turns out to be unusually well suited to this. Babies rarely make just one sound — outside of a long "aah," their vocalisations mostly come in pairs: ba-ah, g'ooh, g'ah. Hiragana already segments speech that way, with one glyph per consonant-vowel pair — あ for ah, が for ga, ば for ba — so BabyLang words fell naturally into reduplicated pairs of kana: mama, dada, tonton. Where Japanese itself didn't have a sound I needed, I bent the script rather than the babbling: ぶ゚ぶ゚ (brubru) borrows the handakuten that normally turns は into ぱ and puts it on ぶ instead, to notate a bilabial trill — a raspberry, IPA /ʙuʙu/. Not standard Japanese orthography, but neither is a nine-month-old.

A baby's world is small, so stage five — the lexicon — didn't need to be big. I split it across three domains. Family gives the core cast: baba is "me," mama and dada are self-explanatory, and hoho is deliberately vague — "whoever is holding me," rather than a fixed relation. Environment covers the limited geography of a crib: nunu for milk, bubu for bottle, bobo for the crib itself, mokomoko for a teddy bear. Activity covers what a baby actually does in there: gepugepu for burping, tonton for being patted, enen for crying, brubru for blowing a raspberry.

I've shared BabyLang (ばばば, Bababa) on Glossikon — go poke around the full lexicon, or run something through the pronunciation tool and hear what a raspberry looks like in IPA.

Shared language BabyLang (ばばば Bababa) →