The Words We Have to Borrow

There is a genre of article, endlessly recycled, which presents a list of beautiful untranslatable words from around the world and invites the reader to marvel. The Portuguese saudade. The Japanese komorebi, light filtered through leaves. The German Fernweh, an ache for distant places, formed as the exact mirror of homesickness.

The genre is pleasant and its central claim is false, and the way in which it is false turns out to be more interesting than the claim itself.

None of these words is untranslatable. I have just translated three of them. What they are is unlexicalised in English: the concept is perfectly expressible, but it requires a phrase rather than a word, and so it is not available for casual use. This is a real and consequential difference — a concept with a name can be reached for instantly, deployed in conversation, used as a premise in an argument — but it is a difference of convenience, not of possibility. Every natural language can express anything, given enough space. What languages differ in is what they make cheap.

The observation that matters

The formulation I find most useful was made by Roman Jakobson in 1959, and it inverts the popular framing entirely. Languages differ essentially, he wrote, not in what they may convey but in what they must convey.

Consider what English forces you to specify. If I say I met a friend yesterday, I have obligatorily marked tense — the meeting is past — and left unspecified the friend's gender, our relative ages, whether you already know this person, and how I came to know about the event. A speaker of French or Spanish would have been compelled to mark the friend's gender. A speaker of Mandarin would have needed no tense marking at all, but if the friend were a sibling, could not have said merely brother: the language distinguishes elder from younger, and the choice is not optional.

More striking still is the category of evidentiality, grammaticalised in a number of languages of the Amazon and the Andes and in Turkish. In such systems, a verb must carry a marker indicating the source of the speaker's knowledge — whether the event was witnessed directly, inferred from evidence, or reported by another person. A speaker of Tuyuca cannot simply say he played football. The sentence must specify whether the speaker saw it, heard it, deduced it from the muddy boots, or was told.

English can express all of these distinctions. Apparently, evidently, I gather, I saw him. The difference is that English lets you omit them and Tuyuca does not. You may be vague in English. In some languages vagueness of that particular kind is ungrammatical, and a speaker who tried it would sound not evasive but incoherent.

Whether this shapes thought

The obvious next question has an unfortunate history, having been answered far too confidently in both directions.

The strong claim — that language determines thought, that concepts unavailable in your language are unthinkable — was popular in the mid-twentieth century and does not survive contact with evidence. People routinely think things their languages do not lexicalise; if they did not, no language could ever acquire a new word, and borrowing would be impossible rather than universal.

The weak claim is more modest and better supported: that habitual linguistic distinctions influence attention, memory and speed of processing in measurable but bounded ways.

The colour experiments are the cleanest example. Russian obligatorily distinguishes two categories where English has one, marking lighter blue and darker blue with separate basic terms rather than modifying a single one. Speakers of Russian discriminate between shades that straddle that boundary slightly faster than shades equally distant within a category. English speakers show no such advantage at the same point on the spectrum. The effect is real, replicated, and small — a matter of milliseconds — and it disappears when participants perform a verbal task simultaneously, which suggests language is being recruited during the judgement rather than having permanently reshaped perception.

The most substantial findings concern spatial reference. Some languages describe location using relative terms — left, right, in front — while others use absolute cardinal directions for everything, including small-scale space: the cup is to the north of the plate, you have an insect on your southern arm. Speakers of the latter maintain accurate orientation continuously, in unfamiliar buildings and in the dark, at a level that speakers of relative systems find frankly implausible. They must, because they cannot form a grammatical sentence about position without it. Here the linguistic requirement has produced a genuine and substantial cognitive difference, sustained by constant obligatory practice.

The translator's actual problem

Practising translators, it should be said, spend very little time on lexical gaps, which are the amateur's preoccupation. Their difficulties lie elsewhere.

They lie in register, which rarely maps cleanly. Many languages grammaticalise a distinction between familiar and formal address that English lost some centuries ago, and a novel in which the moment two characters shift from one to the other is the emotional pivot of a chapter presents a problem with no solution — only compensations, achieved by adjusting names, contractions and rhythm elsewhere.

They lie in what the source text presumes. A single word may carry an allusion that every domestic reader receives without effort and no foreign reader can receive at all. The translator must then choose between a footnote, which stops the reader dead, and an expansion, which flattens the text into explanation.

And they lie in a choice so fundamental that it is often invisible: whether the translation should read as though it were written in the target language, or should retain an audible foreignness that reminds you it came from elsewhere. The first is comfortable and quietly erases the original's strangeness. The second is honest and harder to read. There is no neutral option, and every published translation has silently taken a position.

A note on machines

Automatic translation has become extremely good at precisely the thing that is easiest to evaluate — producing fluent, idiomatic output in the target language. Its systematic bias is therefore towards the first of those two choices. It smooths. It resolves ambiguity that the author left open, because it must produce a single output and ambiguity has no single rendering. It selects the most probable phrasing, which is by definition the least surprising one.

For a contract or a manual, that is exactly right. For a text whose meaning includes its strangeness, the fluency is the loss, and it is a loss of a kind that a reader who does not know the original has no way of detecting.

Which returns us to the list of beautiful words. Its real lesson is not that other languages contain treasures English lacks. It is that every language, including your own, is quietly obliging you to specify some things and permitting you to leave others vague — and that you have never once noticed which, because you have never had to say it any other way.