Translation and what gets lost

Anyone who has tried to explain a joke across languages has already discovered the central problem, which is that meaning is not carried by words alone.

What translation actually involves

Beyond substitution.

Deciding what the text is doing, since a legal contract, a poem and a safety notice require different kinds of fidelity.

Choosing between accuracy to the words and accuracy to the effect, which frequently conflict.

Handling things that do not exist in the target culture.

Preserving register, which is invisible until it is wrong.

And making choices constantly, since exact equivalents are rare.

The things without equivalents

Categories of difficulty.

Words for concepts a culture has and another does not, which require either a borrowing or a paraphrase.

Kinship terms, which divide relationships differently across languages.

Politeness systems, where a language with several levels of formality must be rendered into one with fewer.

Grammatical categories that are obligatory in one language and absent in another — evidentiality, aspect, gender — which force the translator to add or discard information.

And idiom, which almost never survives literally.

Register and tone

Where translations most often fail invisibly.

A text can be lexically accurate and completely wrong in feel: too formal, too casual, too archaic.

Which is what makes machine output recognisable even when it is correct.

Judging it requires knowing how the target language actually sounds in that context, which is why translators generally work into their strongest language rather than out of it.

And why a fluent second-language speaker can produce accurate translations that read oddly.

Machine translation

What it does well and badly.

Modern systems handle common language pairs, standard registers and factual content well enough for comprehension, which is a genuine achievement.

They are weaker on: idiom, humour, ambiguity, register, culturally specific material, low-resource language pairs, and anything where the correct output depends on context beyond the sentence.

They produce fluent output regardless of accuracy, which is the specific hazard — errors do not look like errors.

Which means it is suitable for understanding a text and unsuitable, unreviewed, for anything published, legal, medical or consequential.

Using it sensibly

Practical guidance.

Fine for getting the gist of something.

Fine for a first draft that a competent speaker will revise.

Not for outbound communication in a language you cannot check.

Not for anything where an error carries a cost.

Back-translation — translating the output back and comparing — catches some errors and misses others, and is a rough check rather than a verification.

And short simple source sentences produce substantially better output, which is worth knowing.

Working with an interpreter

A related skill.

Speak to the person, not to the interpreter.

Short segments, pausing for consecutive interpreting.

Avoid idiom, humour and cultural references.

Avoid double negatives and complex conditionals.

Brief the interpreter beforehand on terminology and objectives.

Allow roughly double the time.

And do not use family members for medical or legal interpreting, which is a well-documented source of serious error.

Untranslatable words

A popular topic that is generally overstated.

Words presented as untranslatable are usually translatable in a phrase rather than a word, which is a difference of convenience rather than of possibility.

What is genuinely difficult is not the concept but the connotation, the frequency of use, and the cultural weight attached.

Which means a translation can convey the meaning and still lose what the word does in its own language.

Translating your own writing

A specific case.

Writers translating themselves generally take more liberties than translators would, and produce better results.

Which suggests the useful approach for anyone rendering their own material: rewrite for the target audience rather than translating.

Ask what this text is trying to achieve, and write a text in the other language that achieves it.

Which is what good marketing translation does, and it is called localisation for a reason.

What this means for learners

The practical implication.

Learning through translation builds a habit of routing everything through your first language, which is slow and produces unnatural output.

Which is an argument for learning words in context rather than as pairs, and for input in the target language rather than bilingual material.

And for accepting early that some things simply work differently, rather than looking for the equivalent that does not exist.