How To Devise The Perfect Tech Company Name
"Google" and "ChatGPT" violate the sound system rules of many languages
ChatGPT has made headlines again; this time, as the subject of an interesting neuroimaging study.
Today, though, I’d like to discuss how the LLM relates not to our brains — but to our vocal tracts.
The very phrase “ChatGPT” is phonetically illegal in many pockets of the world. And this detail is not trivial.
I recently wrote about the significance of brand names; they can reflect information about familiarity, relationships, and perceived agency. Naming is not cosmetic, but entwined with how our species makes sense of existence.
So what is it about the waveform “ChatGPT” that’s so impermissible around the globe?
Each language has what are called phonotactic constraints: combinations of sounds deemed unacceptable. Part of the beauty of fluency is having an intuitive grasp of your language’s phonotactics, even if you are not consciously aware of them (as you’ll see in just a second).
Now, of course language is a marvelously elastic faculty. Names can be bent, stretched, and restructured to fit just about any sound system.
But from a linguistic standpoint, some names carry more friction across languages than others. And for reasons unbeknownst to me (but that probably made sense at 3 A.M. in a Palo Alto coworking space), some tech companies seem to leave universal pronounceability by the wayside.
May We Test Your Phonotactic Instincts
Your job right now is to decide whether the following could conceivably be English words: (1) blick, (2) phin, (3) ngon.
The first two are fine, right? But that third one… not as much.
It may be a tempting defense that “ngon” is physically impossible to say. But that is far from the case. Even in English, you can utter “ring on” or refer to that famous Star Trek language, and “ngon” will glide with perfect ease across the tongue.
So, what’s going on? English has a rule that a syllable cannot begin with a “ng” (which actually, despite orthographic appearances, is a singular sound, the velar nasal “ŋ”). However, a syllable may end with it.
Therefore, while “ring” and “going” sound perfectly acceptable to the Anglophone ear, “ngoing” doesn’t. And this constraint is not universal. Tagalog’s word for “smile” is the delightfully fitting “ngiti” (whose high front vowels practically lift the corners of the mouth into a grin)!
These phonotactic rules don’t pertain exclusively to thought experiments, but to everyday names, too. Case in point: my own childhood nickname, “Zan” (which I cherish dearly).
Fortunately for me, those three letters also happen to be phonotactically well-received across many (though not all!) languages. English, Persian, Swahili, Italian, Slavic languages, Dutch and German all give “Zan” the green light.
However, the “z” sound does not exist in the phonemic inventory of Korean or Mandarin, among other languages. Also, in languages including Japanese, ending a syllable with a consonant is not okay (in linguistic terms, (C)V). So my diminutive is a phonetic no-go in Tokyo and Nagoya.
Let’s “Chat” ChatGPT and Google
It may not be shocking that “ChatGPT,” pronounced English-style, runs afoul of many global sound systems.
Even so, it is of course true that words can be modified and translated. For instance, “ChatGPT” in Mandarin is 聊天GPT (liáotiānGPT). While the literal word “chat” would violate Mandarin’s phonotactics (syllables can’t end in plosives like /t/), “liáotiān” actually means “chat”, and as it happens, the next three letters sound valid in the language:
In fact, on direct translation: the Apple corporation in Mandarin is 苹果公司
(Píngguǒ Gōngsī) — meaning, literally, "apple” (the fruit) “company!”
And don’t get me started on “Google.” That “gl” consonant cluster is not permissible in more languages than it is.
Thankfully, this obstacle is also resolvable. In languages that disallow “gl-”, phonotactic workarounds are straightforward. The Japanese Katakana transliteration is グーグル (Gūguru); Korean Hangeul is 구글 (Gugeul). Add an extra vowel, perhaps adjust a consonant or two, and pronounceability is restored.
What’s the “Perfect” Name?
For the sake of argument, imagine you’re naming a tech company — because “tech is the future,” if the volume of Google search results is any indication.
What if you wanted this name to be so phonotactically universal that it could meld well with most of the world’s languages?
In a terrific example from beyond the AI sector, I was impressed to learn that the European insurance company AXA is not an acronym, but rather a decided-upon name that speakers of many languages could say.
For your tech company’s name, I submit Kata, for two primary reasons:
It has a consonant-vowel, consonant-vowel (CV.CV) structure — the most universally accepted, and
It doesn’t contain sounds that don’t exist in the world’s major phonemic inventories (unlike, say, English’s infamous “θ” (as in think), or Norwegian’s “ø” vowel).
Kata’s semantic weight is also a dignified one, and across cultures, no less. In Greek, kata prefix means “down” — perhaps a timely reminder about humility by way of connection to the ground beneath us. In Sanskrit, katha is a “story;” in Bahasa Indonesia, kata is “word” itself. Those meanings gesture towards the relation between narration and synthesis — the latter of which, I’d argue, is the structure underlying all reality.
Something To Share With Every Person, Place, and Thing
To be honest, and despite what the past few dozen sentences may suggest, phonetics are not the end game. Sound is a powerful interface, but it is, in my view, an ephemeral one. On a cosmic level, not all beings — human or otherwise — will share our vocal tracts or auditory capabilities. In fact, if a universal medium for communication exists, Marv Minsky’s experiment decades ago suggested that it may just be arithmetic.
Minsky and a student ran the simplest Turing machines, and found that the ones that didn’t crash returned mathematical structures. Intriguingly, the fundamental principles of natural language are also “based on operations that yield something like arithmetic.”
The manipulation of numbers, then, may just constitute a substrate for all information transmission, including in a future age, long past the Holocene, when human fossils like OpenAI and Google become lost to tectonics and time.
So, I’m not trying to craft an AI company name that can be understood by aliens…yet. But during the most interconnected epoch of human history to date, tech companies (and really, any transcontinental organizations) may be well served in taking a cue from Minsky: extracting what is universal and consistent across systems, speakers, and time.
Because at the end of the day, I’ll venture that the most enduring names aren’t simply palatable — but scalable, too.