Daniel sent us this one, and it's the kind of thing that sounds almost too obvious once you hear it, but almost nobody actually thinks about. He's asking about the hidden architecture of written language — not vocabulary, not alphabets, but the stuff underneath. Punctuation, capitalization, vowel systems, spacing. The conventions we treat as universal because we absorbed them so young we forgot they were conventions at all. And he's right that we tend to collapse all of this into "different language, different characters" and stop there. The actual question is: what else varies, why does it vary, and what does it cost us when we assume everyone's playing by the same rules?
It's a genuinely underexplored corner. Like, I've read a fair amount on comparative linguistics and even I keep bumping into things that surprise me. The inverted question mark in Spanish is a good entry point, actually.
That's where I'd start too. Because it's not just a quirk. It's a structural decision.
Right, so in Spanish, if you want to write a question, you open with ¿ and close with the standard question mark. Inverted at the front, standard at the back. And the reason that exists is interesting from a communication design standpoint. Spanish sentence structure doesn't always telegraph that a question is coming. Word order can be ambiguous. So the inverted mark is doing real work — it's signaling to the reader, before they've processed a single word, that what follows requires a different interpretive frame. You're primed going in.
English doesn't do that. You get the question mark at the end, and if the sentence is long, you've already built up an interpretation that you might have to revise.
Which is a real cognitive cost, actually. There's work on this in psycholinguistics — the idea that readers construct meaning incrementally, so a signal that comes late forces revision of an earlier parse. Spanish front-loads that signal. It's more efficient in that specific, narrow sense.
Yet English speakers encounter the inverted mark and the first instinct is usually to treat it as a typo or an autocorrect failure.
And that reaction is actually the thing Daniel's getting at, I think. We have this deep assumption that our own conventions are the default, and everything else is deviation. But there's no universal standard. The question mark itself has a contested history — one account traces it to a medieval Latin abbreviation for "quaestio," the word for question, which may have been contracted and eventually stylized into the mark we use now. That's not settled history, but it's a plausible lineage. The point is, someone decided. It wasn't handed down.
By the way, today's script is powered by Claude Sonnet four point six. Just so you know who to blame for any particularly elegant transitions.
Let's stay positive.
We'll see. So the inverted mark is one example, but what I find more interesting is why we don't notice these things until we're forced to. You don't think about punctuation as a system until the system breaks down for you. You're reading something in a second language, or a translation, and something feels slightly off, and you can't name it.
That's the globalization pressure Daniel's pointing to, I think. When most of your reading was in one language, your conventions were invisible. Now you've got multilingual content, international teams writing documents together, translation pipelines, AI-generated text that's been trained across a dozen languages — suddenly the invisible becomes visible, and sometimes it becomes a problem.
The places where it becomes a problem are not always where you'd expect. It's not just "the punctuation looks funny." It's deeper than that.
And that's what I want to get into, because the punctuation stuff is almost the surface layer. Once you go below that, into things like how vowels are represented, or whether a language even has a concept of capitalization, you're into territory that shapes how the writing system itself encodes meaning. And that's where it gets strange and fascinating.
I'm already leaning forward, which for a sloth is a significant commitment.
So let's dig in — there's a lot of ground to cover here, and I don't want to shortchange any of it. Corn, how should we think about the scope of what we're discussing?
The scope question matters here, because there's a temptation to just catalog every language's weirdest feature and call it a day. What Daniel's actually asking about is narrower and more interesting than that. Not phonology, not vocabulary, not even the alphabet itself. The stuff that sits on top of the characters — or in some cases, conspicuously absent from them.
Right, and I'd group it into roughly four areas. Punctuation conventions, which we've started on. Capitalization, which is far less universal than most people assume. Vowel representation, which gets into deep structural territory. And spacing — where words begin and end on a page, which sounds trivial until you realize some major writing systems handle it completely differently.
The reason these get overlooked is probably that they're invisible when they're working correctly. You learn to read in your first language and these conventions get baked in so early they feel like physics rather than policy.
They're subgrammatical. Below the level where most language instruction even operates. When you study a foreign language formally, you learn vocabulary, grammar, verb conjugations. The punctuation chapter is usually half a page, if it exists at all. And yet these conventions are doing significant work in how the text communicates.
We're basically looking at the operating system rather than the application.
I'd go with that. And like any operating system, it's only when you try to run software designed for a different one that you notice the incompatibility. A Hebrew document dropped into an English-language workflow, a Spanish email read by someone who's never encountered the inverted marks — the content might be fine but something in the rendering feels wrong, and most people can't articulate why.
Which is the problem worth solving. So let's start pulling these apart.
Punctuation first, then, and not just Spanish. Because there are some surprising cases that I think most listeners haven't encountered. Greek is one. Modern Greek uses the semicolon — that mark, the one English uses for joining independent clauses — as a question mark. So the sentence "you are coming?" in Greek ends with what looks to an English reader like a semicolon. Completely different semantic load on the same glyph.
That's a good example of how arbitrary the mapping is. Same character, completely different job.
It goes further. Japanese written text traditionally doesn't use spaces between words at all. The segmentation is handled by the interplay of three different scripts running simultaneously — kanji, hiragana, katakana — and a fluent reader parses word boundaries from that mixture rather than from whitespace. Which sounds like it should be unreadable to an outsider, and it basically is, but for someone who grew up in that system, the spacing convention we rely on in English is actually redundant information.
The space is doing work in English that the script itself is doing in Japanese.
The script carries the disambiguation. And what's interesting about that from a design standpoint is that it's not inferior — it's just a different distribution of the cognitive load. Japanese readers aren't working harder. The load is just allocated differently across the system.
Though it does mean that when you're trying to process Japanese text computationally — search engines, translation tools — you have a word segmentation problem that English doesn't have. You have to infer the boundaries.
Which is a non-trivial engineering problem. There's a whole subfield of natural language processing devoted specifically to Japanese and Chinese word segmentation. English tokenization looks almost embarrassingly simple by comparison.
Because I think this is where a lot of people's assumptions really fall apart.
The first thing to say is that capitalization as a concept — upper and lower case as distinct letterforms — is not universal. It's a feature of certain script families, specifically Latin, Greek, Cyrillic, and Armenian. Arabic doesn't have it. Hebrew doesn't have it. You cannot capitalize a Hebrew word. There is no uppercase aleph. The concept doesn't exist in the writing system.
Which means when you're translating something into Hebrew and the original text uses capitalization for emphasis, for a proper noun, for the start of a sentence — all of that information has to be encoded some other way or it disappears.
It either disappears or it gets compensated for through word choice, through context, through explicit markers. And within the scripts that do have capitalization, the conventions vary enormously. German is the obvious case — German capitalizes all nouns, not just proper nouns. So "the dog ran across the street" in German has capitalized dog and street. Which looks strange to an English reader but is actually carrying useful information. You can scan a German sentence and identify the nouns visually before you've parsed the grammar.
There's an argument that it's more readable, not less.
For someone trained in it, almost certainly. English used to capitalize more freely as well — you see it in eighteenth-century texts, capitalizing nouns for emphasis or importance. We shed that convention over time, arguably losing some expressive precision in the process.
Now we use italics for the same job.
Or all-caps, which carries a completely different emotional register — it reads as shouting in digital contexts. And that's a convention that's also not universal. In some writing communities, all-caps is neutral emphasis. In English-language internet communication, it's aggressive. Same characters, completely different social meaning.
Which gets to something I find interesting about vowels, because the Hebrew situation is not just a quirk — it's a fundamental architectural choice about what the writing system is for.
This is where it gets deep. So Hebrew is classified as an abjad — a writing system that records consonants and leaves vowels to be inferred by the reader. Modern everyday Hebrew is written without vowels. No vowel letters, no diacritical marks in ordinary text. You see a string of consonants and your brain, if you're a fluent reader, fills in the vowels from context. The word for "writing" in Hebrew — ktivah — is written with four consonants. The vowels are not there.
For a native reader, this is apparently not a problem.
Barely conscious effort. The context constrains the possibilities so heavily that ambiguity is rare in practice. There are words that look identical in consonants but mean completely different things depending on vowels, and those are resolved by what surrounds them. It's a high-context system. The writing assumes a reader who already knows the language.
Which is a meaningful design choice. The script is not optimized for learners or outsiders. It's optimized for fluent readers who want efficiency.
And you can see the tradeoff clearly when you look at where vowels are added back in. Children's books use niqqud — the diacritical vowel marks — because children are still learning. The Torah uses a fully vocalized text in liturgical contexts for the same reason, precision and preservation. But the everyday newspaper, the text message, the street sign — no vowels.
Arabic does something similar but with a different default.
Right, Arabic is also an abjad, and also omits vowels in most everyday writing. But Arabic has a richer tradition of adding the diacritical marks — called harakat — in specific high-stakes contexts. The Quran is fully vocalized, always. Classical Arabic instruction uses full vowel marking. Modern Standard Arabic in journalism typically drops them. Colloquial written Arabic, in text messages or social media, drops them almost entirely and also mixes in a lot of conventions that are technically not part of the formal script.
Both systems have this tiered relationship with vowels, where the level of vocalization signals something about the register or the audience.
The cognitive demands shift accordingly. A reader of unvocalized Arabic or Hebrew is doing something qualitatively different from a reader of fully vocalized text. They're running more inference, drawing more on stored lexical knowledge. It's faster if you know the language well. It's nearly impenetrable if you don't.
Which creates real accessibility questions for people with dyslexia or other reading difficulties, because a lot of the support strategies for those readers in English rely on phonological cues — the sound-spelling correspondence. In an unvocalized script, that correspondence isn't on the page.
That's a underexplored area. There's research on dyslexia in Hebrew-speaking populations suggesting that the vowel-less script interacts with reading difficulties in ways that are distinct from what you'd see in English. The error patterns are different. The interventions that work are different. But the field is smaller and the findings haven't traveled as widely as the English-language dyslexia literature.
The writing system shapes not just how you communicate but how you struggle to communicate, and what help looks like when you need it.
Which is exactly the kind of knock-on effect that gets missed when you treat writing systems as interchangeable. They're not just different skins on the same underlying structure. The structure itself is different, and that difference propagates outward in ways most people never think to trace.
Translation tools are where that propagation becomes almost comically visible. Because a system like Google Translate is not just mapping words — it's making decisions about every single one of these layers simultaneously. Capitalization, punctuation, vowel inference, word segmentation. And it gets some of them wrong in ways that are very hard to diagnose if you don't know what to look for.
The Hebrew case is instructive here. When you feed unvocalized Hebrew into a translation system, the system has to resolve ambiguity before it can even begin translating. And it's doing that resolution statistically, based on patterns in training data. Which works surprisingly often. But when it fails, the failure is upstream of the translation itself. The system has silently committed to the wrong vowel reading, and everything downstream is wrong in a way that looks like a translation error but is actually a parsing error.
The user sees a bad translation and blames the vocabulary mapping, when the actual problem happened before a single word was translated.
Before the translation process even started, in a meaningful sense. And this compounds with directionality. Hebrew is right-to-left. Arabic is right-to-left. When you're building an interface that has to handle both Hebrew and English in the same text field — which is common in Israel, where you'll have a sentence that switches mid-thought — the rendering engine has to make decisions about text direction that the user never sees and has no control over. The Unicode bidirectional algorithm handles a lot of this, but it has edge cases that produce bizarre output. Punctuation marks that jump to the wrong end of a sentence. Quotation marks that flip orientation unexpectedly.
I've seen this. You type something perfectly reasonable and the cursor is suddenly on the wrong side of everything.
That's not a minor annoyance if you're a content creator working multilingually at scale. If you're producing marketing copy in five languages, some of which are right-to-left and some left-to-right, and your content management system wasn't built with that in mind, you are fighting your tools on every single piece. The conventions aren't just intellectual curiosities — they're engineering constraints that either someone accounted for or they didn't.
What about typing speed? Because I've seen claims that the structure of Arabic or Hebrew input affects how fast you can actually produce text.
This is interesting and I want to be careful here because some of the comparative data is not clean. But there are studies on Arabic typing that point to a real structural issue, which is that Arabic letters change shape depending on where they appear in a word — initial position, medial, final, and isolated all have different forms. So a typist isn't just hitting a key, they're producing a glyph whose appearance depends on context. The keyboard input is the same but the rendered output varies. Early Arabic keyboard designs and early digital fonts handled this badly, and there's a legacy of workarounds that slowed things down considerably.
Which is a problem that doesn't exist in English because the letterforms are static. An A is an A wherever it appears.
There are ligatures in some fonts, but they're decorative rather than mandatory. In Arabic, the contextual shaping is mandatory — you cannot write correct Arabic without it. So the software layer has to handle something that English software never had to think about, and historically it didn't handle it well. Unicode and OpenType have improved this enormously, but the transition took decades and there are still environments where Arabic renders incorrectly or inefficiently.
This is the thing that strikes me about all of these cases — the conventions themselves are often elegant, they're internally coherent, they solve real problems for the communities that developed them. The friction appears at the interface between systems, not within any single system.
That's a really clean way to put it. Vowel-less Hebrew is not a deficient writing system. It's a system optimized for fluent readers of Hebrew, and it works. The problem surfaces when you try to run Hebrew through infrastructure built for English assumptions. The mismatch is in the interface layer, not in either system individually.
Which has an obvious implication for how we think about multilingual content creation. The advice is usually learn the vocabulary, learn the grammar — but the conventions are where the hidden costs live.
For a content creator, absolutely. If you're translating into Arabic and your translation vendor delivers perfectly accurate text but your publishing platform collapses the contextual shaping, what goes out the door is wrong and you might not notice until a native speaker flags it. The quality control problem is invisible to someone who can't read the script. And I'd extend this to localization more broadly — things like whether to use the Oxford comma style or not, whether to follow French spacing conventions with that narrow space before colons and exclamation marks, whether to use the British date format with full stops or the American format with slashes. These are not translation questions. They're convention questions. And they often don't appear in a standard translation brief.
Because the brief is written by someone who didn't know to ask.
And the people doing the translation may not flag it either, because they're working to the brief they were given. So you get technically accurate text that reads as slightly off to a native audience, and nobody can identify why. The cultural signal is wrong even though the words are right.
There's something almost philosophical in that. The words carry the denotative content, but the conventions carry something about register and belonging. Whether this text was written by someone who knows how we do things.
Readers pick that up unconsciously. A native German reader will notice if a text doesn't capitalize nouns correctly — not as a grammar lesson, but as a felt wrongness. Something is slightly off about this. Similarly, a reader of formal Arabic will notice if the harakat are absent in a context where they'd normally appear. It signals something about the care that went into the text, or the competence of the writer.
Or the competence of the translation pipeline.
Which circles back to AI, because these are exactly the kinds of signals that current large language models are still calibrating on. They're much better at getting the words right than at getting the conventions right, because the conventions are underrepresented in training feedback. A human evaluator rating a translation says "this is accurate" or "this sounds natural" — but they may not flag that the punctuation style is slightly off for the target audience, because they're not consciously tracking it.
The models absorb the same blind spot that human reviewers have.
It may be that the way to fix it is not more data but more structured evaluation — specifically asking reviewers to assess convention fidelity as a distinct quality dimension. Separate from accuracy, separate from fluency. Does this text follow the typographic and punctuation norms of the target language and register? That's a question that's rarely asked explicitly.
Which means someone has to know to ask it. You're back to the same problem.
The bottleneck is awareness. And I think that's actually the most useful thing a listener can take from this whole territory — not a specific list of rules, but the habit of asking: what are the conventions I'm not seeing? What are the things this text is doing or not doing that I've never thought to notice?
Because once you start looking, you can't stop seeing them.
It's a bit disorienting. You read a sentence in a language you're learning and suddenly you're noticing the punctuation style, the capitalization choices, the spacing. All the things that were invisible before.
The operating system becomes visible.
Which is when you actually start to understand how the thing works.
Given all of that — given that the bottleneck is awareness — what does someone actually do with this? If you're a listener who works multilingually, or you're learning a language, or you're building something that has to handle more than one script, where do you start?
The most concrete thing I can offer is: treat punctuation conventions as a checklist item, not an afterthought. When you're preparing content for a new language market, the question "what are the punctuation norms here?" should be sitting right next to "is the translation accurate?" Because as we've been saying, you can have a perfectly accurate translation that still signals wrongness to a native reader because the spacing before colons is off, or the question marks are formatted to English convention rather than the target language's.
That's not a heroic research task. Most of that information is findable. It's just that nobody thinks to look.
Right, and for anyone building multilingual tools or working in localization, Unicode's documentation is useful here. The Unicode Consortium has detailed character and script specifications that include information about how contextual shaping works, how directionality is handled, what the expected behavior of bidirectional text is. It's not light reading, but it's authoritative. And there are Unicode converter tools that let you inspect what's actually happening at the character level in a piece of text — which is useful when something is rendering strangely and you can't tell why.
That's the kind of tool that pays for itself the first time you use it to diagnose why your Arabic copy is displaying backwards in a content management system.
And for language learners specifically, I'd say: when you start a new language, ask your teacher or your textbook or your app what the punctuation conventions are. Ask whether capitalization works differently. Ask whether the script uses spaces the way you expect. Most courses skip this entirely and you end up learning it by accident, or by making embarrassing mistakes in writing that would never have happened in speech.
Because in speech, none of these conventions exist. They're purely a written-language problem.
Which is why they get overlooked. The oral version of the language doesn't carry them, so you can get fairly fluent conversationally without ever confronting them. Then you sit down to write something formal and suddenly you're not sure whether to capitalize a noun, or where to put the question mark, or whether the spacing around punctuation is right.
The gap between spoken fluency and written competence is wider than most people expect, and this is a big part of why.
I think for listeners who aren't language learners but are working in global communication — marketing, documentation, anything that crosses language boundaries — the practical takeaway is to involve native speakers in reviewing not just the content but the form. Ask them specifically: does this look right? Not does it read correctly, but does it look like something a native writer would have produced? Because those are different questions.
One tests the words. The other tests the conventions.
Both matter, but only one of them usually gets asked.
Which brings us to a question I keep circling back to. If globalization keeps accelerating, if more and more text is crossing language boundaries every day, does that pressure eventually flatten some of these conventions? Do we end up with a kind of punctuation esperanto?
I don't know, and I think anyone who claims certainty here is guessing. There are two forces pulling in opposite directions. Digital communication has already nudged some convergence — emoji function as a kind of universal paralanguage, and the informality of messaging apps tends to erode local punctuation conventions in favor of whatever is fastest to type. But at the same time, formal written registers seem to be holding their conventions fairly stubbornly. The inverted question mark in Spanish is not disappearing from newspapers.
Convergence at the casual end, preservation at the formal end.
That's my read. And the formal end is where it matters most for professional communication, for legal documents, for anything where getting it wrong costs something.
Where AI is increasingly being asked to operate.
Which is the open question I find most interesting going forward. Right now, large language models handle these conventions inconsistently. They're improving, but the evaluation frameworks to measure convention fidelity as a distinct quality don't really exist yet at scale. As AI takes on more multilingual drafting and translation work, whoever builds those evaluation frameworks is going to shape what "correct" looks like across dozens of writing systems. That's a significant amount of influence concentrated in a small number of decisions.
The people who write the rubrics define the standard.
They may not know what they don't know. Which is, in a sense, where we started.
The conventions that feel like physics are actually choices, and choices can be made differently.
Which is most of the battle.
Thanks to Hilbert Flumingtop for producing this one, and to Modal for keeping our infrastructure running. This has been My Weird Prompts. Find us at myweirdprompts.com, and if you're enjoying the show, a review goes a long way.
Until next time.
