Asked how she felt about being crowned the Nobel Laureate in Literature this October, the Austrian writer Elfriede Jelinek made some intriguing remarks to journalists. She explained her concerns that her works were "untranslatable" and that English audiences would never understand them. While her British publisher, Serpent's Tail, described her as a writer of "European sensibility", Jelinek emphasised that she is "a provincial writer, doomed to stay that way".

Asked to explain her remarks further, she broadened her definition of untranslatable works to encompass all "linguistically experimental texts". "I play with the sound of language," she said. "That can hardly be translated into another language. Each language has its own face and its own fingerprints, which are not identical with any other language."

As a response to winning the Nobel Prize in Literature, her statement was compelling in its perversity. The prize is rather like a United Nations of letters, a literary version of the Nobel Peace Prize. It is founded, like all the Nobel prizes, on belief in a universally intelligible project of global civilisation. As descriptions of its literary laureates suggest, the prize is often awarded to writers who are seen as the equivalents of UN goodwill ambassadors, exposing injustices, reaching beyond national boundaries. So, Günter Grass won in 1999 for portraying "the forgotten face of history", VS Naipaul in 2001 for "compelling us to see the presence of suppressed histories", Derek Walcott in 1992 for his "multicultural commitment", Dario Fo in 1996 for "upholding the dignity of the downtrodden", and Nadine Gordimer in 1991 for being "of a very great benefit to humanity".

Jelinek was already an unusual Nobel Laureate: a writer of spiky and rebarbative prose, her works are defined by their regional politics and her relentless loathing of Austria. With her remarks about untranslatable literature, Jelinek refused to play the Nobel game, insisting instead on the stark limits set by language, the impossibility of flawless literary communication between nations.

"Content can be translated more easily than linguistic play," she explained to me, and this may be why the Nobel Prize in Literature often emphasises content over form. "But the best works of literature always bind content and form together," said Jelinek. "It doesn't interest me, to present content in conventional language. It would be to pour new wine in old bottles."

The debate about translation is as old as the hills - or, rather, as old as language itself. It can be argued that it is impossible to translate any word into any other language without slightly changing its meaning. Still, some forms of communication are more translatable than others. If I ask for three loaves of bread, I can be confident that the quantity can be translated. But if I call my crime novel Room Thirteen, then translating the number immediately becomes more a question of cultural connotation than numerical precision.

Sounds and wordplay torment translators: translating a tongue twister like "Round and round the rugged rock the ragged rascal ran" might defy the most venerable polyglot. Puns are inevitably untranslatable, from a platitude such as "the quality of life depends on the liver", to a corny James Bond witticism such as "I think he's attempting re-entry".

With literary language, everything becomes still more complicated. If someone sighs "ou sont les neiges d'antan?", we can translate it as "where are the snows of yesteryear?", but something of the original is lost. Echoes inherent to language are the lifeblood of literature and are difficult to recreate in translation. So, as Virginia Woolf pointed out, the English word "multitudinous", for those who like to use it, always has a touch of Shakespeare's "multitudinous seas incarnadine" about it. A translation of the word would lose the Shakespearean resonance.

Innovation and idiosyncrasy, the clang of linguistic strangeness, also pose problems. How could one begin to translate Gerald Manley Hopkins's, "I caught this morning morning's minion, kingdom of daylight's dauphin, dapple-dawn-drawn Falcon"?

Each language has works that are often held up as untranslatable, like so many literary Everests, scaled only by the most intrepid translators. In English they include Shakespeare and Joyce, with his polyglot coinages and sentences such as "Sir Tristram, violer d'amores, fr'over the short sea, had passencore rearrived from North Amorica on this side the scraggy isthmus of Europe Minor to wielderfight his penisolate war".

In French they include Céline, who enjoyed a neologism or two; in Russian they include Pushkin, who littered his poetry with stylistic parodies. Samuel Johnson - quotable on most things - said that all poetry was untranslatable and that this was a good thing, as it forced people to learn languages. The only solution, Johnson decided, was for every reader to learn Greek, Latin, French, German, Spanish and Italian.

Yet all of this doesn't stop the translators. After a few peaceable centuries in which everyone in Europe wrote in Latin, literary translation went hand in hand with the development of national literatures in national languages. Geoffrey Chaucer translated The Romance of the Rose and The Consolation of Philosophy, paraphrased Virgil and Ovid, and collated stories he read in Latin, French and Italian.

From Chaucer, the influential translations stack up through the centuries: Golding's Metamorphoses, which Shakespeare used, Marlowe's Amores, Chapman's Whole Works of Homer, Edward Fitzgerald's Rubáiyát of Omar Khayyam, Constance Garnett's translations of Tolstoy and Dostoevsky, Scott Moncrieff's Proust. Ezra Pound claimed that every great age of literature was preceded by a great age of translation.

Pound, who translated works from Anglo-Saxon, Italian, Japanese, Chinese and Provençal, proposed that translation was one of the highest arts. He thought that it was possible for a writer to produce his or her finest works in translation: Swinburne, he said, never wrote so well as when he was translating; likewise Browning.

Pound thought there were two sorts of translation. There was functional translation, showing the reader "where the treasure lies", as he put it, materially assisting "the hurried student who has a smattering of a language and the energy to read the original text alongside the metrical gloss". The "other sort" consisted of "cases where the `translator' is definitively making a new poem".

The question seems to be whether accuracy or atmosphere is more important, and whether one must always preclude the other. Nabokov - who wrote in Russian, French and English, and translated his own novels from Russian to English and English to Russian - was a great stalwart of literal translation, spending several years labouring on a remorselessly accurate translation of Pushkin's Eugene Onegin. Nabokov aimed to banish from his Pushkin "everything that honesty might deem verbal velvet". He produced a literary curio, an awkward and fascinating work, written in a contorted language which is not quite English. He called it a "pony" - a pony to carry the student, nothing more.

Nabokov was unusual among writer-translators in his faithful adherence to the code of literalism. Often the writer-translators are on the other side, crafting their non-literal creations, their metaphorical representations of the original spirit of the work, and then finding themselves roundly rebuked for their inaccuracies. The American poet Robert Bly produced a remarkable translation of the Norwegian writer Knut Hamsun's wild, proto-modernist novel Hunger. Bly conveyed the nervy and energetic atmosphere of the original beautifully, but was later attacked by the translator Sverre Lyngstad. Lyngstad produced his own translation with an angry appendix pointing out discrepancies between Hamsun's original and Bly's version. But Lyngstad's translation lacks the atmosphere and pace of Bly's, and his prose often sounds stilted.

There's a long tradition of writers translating from languages in which they are hardly card-carrying experts, relying rather on their proficiency in their own language: WH Auden translated Icelandic poetry; Tom Stoppard translated Chekhov; Seamus Heaney translated Beowulf. These celebrity translators are like famous singers performing cover versions of well-known songs: they are expected to improvise a little, to add their own subtle touches to the tune.

For many career translators, their work is a form of self-effacement. Tony Briggs, who is translating War and Peace for Penguin Classics, modestly explains that "professional translators are generally mediocre people like me, not great poetic geniuses. Within these limitations, you are trying to give to the English reader the same experience as the Russian reader has reading the original."

For David Constantine, translator of Goethe and a poet in his own right, the key is to "convey the completely irreducible individuality and idiosyncrasy of the original work". Accuracy is essential, says Constantine: "It would be irresponsible to put into circulation something that is lexically incorrect." A piece of poetic licence can condition the understanding of generations. Yet there's also a need for realism, says Constantine: "Translation is a pragmatic business, and all translators are pragmatists."

Stanley Mitchell, a translator of Pushkin, explains the process of negotiation: "When translating poetry, one wants to be faithful to the original as much as possible, but given the exigencies of meter, one simply can't. You have to give things or take them away. But if you add something to an image, then what you add has to keep within the logic of the whole."

The founder of Harvill Press, Christopher MacLehose, says: "It's a question of the expectations of the author and the reader. I would say that many so-called untranslatable works are also very widely read in their native country, such as Doctor Zhivago. So you have a strong argument that they deserve to be available elsewhere. You try always to buy Beethoven and ask Schnabel to perform him. You want the best possible writer and the best possible translator."

As an example of the intricacies of the process, MacLehose mentions the translation of WG Sebald, who was first published by Harvill. By the time Sebald began to write fiction in German, he had lived and worked in Britain since the 1960s and had a near-native command of English. Yet he decided not to translate his own works, though he collaborated closely with his English translators Michael Hulse, Michael Hamburger and Anthea Bell. For The Emigrants, Sebald received three anonymous sample translations, and from these he selected Michael Hulse to translate the book. There followed an intricate series of negotiations between writer and translator, with versions going backwards and forwards. Finally, Sebald had the whole book translated again by a colleague at UEA, and checked one version against the other before permitting Hulse's version to be published.

There are writers who opt boldly for DIY translation - after initially collaborating with translators, Beckett translated himself from French into English and vice versa. Even then, he was unhappy with the English version of Fin de Partie (Endgame). Joseph Brodsky eventually discarded his translators; his self-translations were full of linguistic curiosities, but they were his own.

The Nobel prize is a testament to this dystopian state of affairs after Babel. It is judged by a committee of Swedish academics and authors who, though assisted by linguistic referees, are inevitably reading most of the works they assess in translation. The Nobel committee's judgments over-emphasise more translatable elements - political content, the biography of the writer - and under-emphasise less translatable elements such as style.

This separation of form from content is a fundamentally unliterary way of assessing literature, as Jelinek pointed out. Yet the pragmatism of the Nobel prize with its partial parameters is necessary for any community of letters hoping to transcend linguistic boundaries. The translation of literature is impossible to avoid, and in Britain it looks set to become ever more important, if the teaching of modern languages continues to decline. As we know more of the world, and less of the languages of the world, it seems we will turn increasingly to our translators.

In the abstract, there are the untranslatable texts, the literary insurmountables. And then there are the translators busy trying to translate them anyway. That is, unless we throw in our lot with Johnson, and brush up on seven languages.