We often hear and speak of all that is lost in translation — the turns of phrase that can't be replicated, literal meanings in one language that have no relevance in another, words that migrate from one language to another and on the way mutate or change their meaning (rail and rels, pier and pirs, ostranenye and defamiliariazation). Metaphors and idioms that need cultural contexts in order to make sense. Allusions that are not only culturally or socially or religiously but also linguistically specific. The notion of translatability immediately suggests its counterpart: untranslatability.
There are those who say that a translation is no longer the original author's work but rather the work of the translator. This is not totally true — unlike a true artist, the translator is working along readymade linguistic path — but it's accurate in the sense that translators are forced to make decisions at every single step. The work goes through them, and the consistency of its transference into another language depends on their ability to remain consistent in their decision-making process. Decisions are made on several levels, from specific detailed issues having to do with punctuation and word order, to larger returning linguistic issues based on thematic or stylistic concerns. That is, the translator has to understand or at least channel that which the artist him- or herself may or may not know consciously: the turns of their writing personality, the ways in which they make their work linguistically original.
What for the author comes out of the deepest place of self, in the translator has to be consciously and also intuitively identified and activated. A translator has to be able to sense those things that make the author's work what it is, and then use that perception in order to adapt it into the new language. That is, a certain linguistic tick may not translate well literally, but somehow it must be "transferred" into the target language.
A translator also has to have a sense of the whole book and what unifies it in order to allow some flexibility in the translation process. A certain linguistic originality that may not be able to be translated in one specific place may be able to be transferred to another where in the original work that linguistic tick didn't exist. That is, the fact of that tick is translated, but displaced because it fits better in the target language. When a translator intuitively and consciously understands this, he or she can make these kinds of calls while remaining, as we say, faithful to the original.
The very use of such a phrase as faithful gives us a hint to the fact that the translation process is a little like a relationship. First you are attracted to the work, you get to know it by reading it, at first superficially and later through reflection and return. It is a process of acquaintance, of getting to know, of building an understanding between yourself and the work that is not always based on the obvious, on what's said. Slowly you begin translating and through the process begin to understand more of what's between the lines — the place, as Maurice Merleau-Ponty said, where language rejoins the silence of painting. That is, all that is unsaid in the poem begins to come to light through the evolving attempt to transfer not only the words but also the way in which the author has used the language. And the translator taps into the deeper part of the linguistic flux — which flows like an underground spring — and makes use of this knowledge in order to adapt it into the new language.
Wolfgang Iser wrote that literary criticism "helps to make conscious those aspects of the text which would otherwise remain concealed in the subconscious." Translators are similar insofar as they are first and foremost readers and must understand a work enough to implant into the translation those same signposts which can either "remain concealed" or be again unconcealed. Truth is revealed through art, and the translator's job is to retain the kernel of art so that the same truth can reveal itself through the reading of it in a different language. The truth is indicated, not said, and it is this indication which must be preserved. Then the original truth will come through.
Through the process of doing this we find ourselves constantly dealing with the register of a writer's language. The scale of that register can be vertical — that is, low or high — and make use of words or syntax that are more or less easy to apprehend through the reading process. In Kafka, for example, the register is generally high not so much because of vocabulary, which tends to be fairly understandable, but because of the syntax and structure. As Albert Camus wrote, the beauty of Kafka is that he makes you reread his sentences over and over. The same can be said of Bruno Schulz and many others.
But a language's register can also be horizontal, with one pole which we could call local or idiomatic — that is, specific to a certain culture or subculture, with its own words, references, metaphors, and meanings. An example of this is Gogol's Ukrainian tales are almost impossible to translate. You'd have to essentially recreate them in a dialect of English — perhaps if William Faulkner or Flannery O'Connor had known Russian he might have made wonderful Southernized translations of Gogol.
The other pole of the horizontal register would be continental (or perhaps even intercontinental). It is transcontinental not only in its word choice — meaning that it utilizes words of Latin and / or Greek origin, which are pan-European and beyond, rather than ones with a local origin — but also in its address. That is, it does not take it for granted that its co-linguists will understand a word or a reference, but rather treats even its "own" and audience as if it were foreign. This linguistic aspect goes beyond into thematic and metaphoric use of language: such an author will make use of a language at the same time that he or she will tweak it, in the same way that he or she will raise ideas and memes that come from that linguistic culture but also treat them critically.
One important note I want to make is that a linguistic register can be both high and local, and in The Elements of Style, Strunk and White made a case for precisely such an approach when they advocated not only correct grammar but also, whenever possible, the use of anglo-saxon words over Greco-Roman ones in order to preserve as much as possible of (for example, to use the word weird, which is an Old English word originally of German origin, rather than strange, which is Middle English and comes from the Latin extraneus via the Old French estrange).
The language of literary communication is literary language, which is a specific case of creative, expressive, or signifying language. The communication, rather than being a direct assertion or proposition, is a suggestion of significance put forth by the author. The communication is the spectrum of significance suggested through language's second value, which is as allusive and unstated as it is creative and unveiling.
What we find in translation is hopefully what we found in the original: the artistic truth that is revealed (or unconcealed) to us through the work. The degree of a text's dynamism — its potential for a spectrum of meaning — depends on its ability to activate a variety of interrelated meanings in numerous readers. And the task of the translator, since that's the conversation at hand, is to preserve and make possible the dynamic of the text in a way that's as identical as possible to its original. The translator gets to know the text from inside, where there exists all that is not actually written. And what he or she translates, in the end, is not only what is written in one or another language, but also that which is said silently. What we find in translation is that which we find in no language at all.