There's a quote that gets repeated in linguistics: "A language is a dialect with an army and a navy." Max Weinreich, 1945. Usually cited as a witty observation that the line between dialect and language is political, not linguistic.
It's a good line. But I've been sitting with it, and I think it's stranger than it sounds.
Start with a specific case. Mandarin and Cantonese are both called Chinese dialects in official Chinese government policy. But a monolingual Mandarin speaker and a monolingual Cantonese speaker cannot have a conversation — zero intelligibility, according to actual tests. The linguist Yuen Ren Chao compared the structural difference to the difference between English and Dutch. English and Dutch are unambiguously two languages; nobody calls them dialects of the same thing. But Mandarin and Cantonese are "dialects of Chinese."
Now flip it. Serbian and Croatian — written with different scripts, associated with rival national identities, official languages of two separate countries. When Yugoslavia existed, they were one language. When Yugoslavia broke up, the language broke up too. Linguists who study the actual structure rather than the politics generally conclude they're varieties of the same thing: the same grammar, the same core vocabulary, the same sound system. But they are now different languages, complete with separate dictionaries and official commissions working to make them more distinct over time.
So: Mandarin and Cantonese are "dialects" despite being mutually unintelligible. Serbian and Croatian are "different languages" despite being mutually intelligible. The linguistic facts and the political decisions are pointing in opposite directions in both cases.
The usual response is: well, "language" is a social construct. The label doesn't track the thing. Fine. But then what is the thing? If a linguist studies "English," what are they studying? Not the political category — there has to be some actual object of inquiry. So there must be something they're pointing at. Something real enough to describe, measure, teach.
Maybe the object is a dialect continuum. Imagine a line from the west coast of Norway to the east coast of Sweden. At every point along it, people are speaking. Adjacent towns understand each other easily. But somewhere in the middle is a border, and a speaker at one end would struggle with a speaker at the other. Norwegian, Swedish, and Danish are three languages by any official account — but Norwegian occupies a middle position, understanding Danish better than Swedes do and understanding Swedish better than Danes do, for reasons that are partly historical (Norway used Danish as a written language for centuries) and partly geographic (the shared border with Sweden). The intelligibility isn't symmetric, isn't binary, and shifts with exposure. You can't point at a clean edge.
Mountains work the same way. "Where does the Rockies end?" has no correct answer, only conventions. The question sounds like it's asking for a geographic fact but it's actually asking for a decision.
Here's where I get stuck, though: mountains don't care about the line. Languages seem to.
When a community decides that their variety of speech is a language — not a dialect, a language — they often start standardizing it, writing it down, teaching it in schools, and using it to mark themselves off from their neighbors. Once that happens, the variety diverges from its relatives faster. The political decision doesn't just name the boundary; it starts to create it.
This is happening right now with Serbian and Croatian. Official commissions are deliberately reintroducing archaic Croatian vocabulary to replace shared South Slavic words. A Montenegrin dictionary introduces distinctions that weren't there before. The naming preceded the fact, and now the fact is following the naming. Two decades ago, linguists called it one language. At some point in the future, they may have no choice but to call it two. Not because of a discovery — because of a decision that was made and then enforced, word by word.
So maybe the army-and-navy joke understates the case. It isn't just that politics decides what gets called a language. Politics decides what becomes one.
I don't know where that leaves the original question — where does one language end and another begin. The honest answer seems to be: it ends where we decide it ends, and then it starts ending there for real. The line is drawn on a gradient and the gradient slowly bends toward the line.
That's not a comfortable answer. It suggests that the most linguistically accurate map of human speech would be a smooth color gradient with no edges — but that such a map would be wrong about what speech actually does in the world, which is mark belonging and exclusion and history and war. The edges matter even when they aren't there yet. Especially when they aren't there yet.