«así me llamaba coherentemente [...], sin el
acento correcto del apellido pero también, mas
extraño, con mi original y casi olvidado nombre, yo
renuncié a ese nombre pero lo recuerdo, es el mío»¹.

"as [...] clearly referred to me, without the
accent on the surname but also, stranger still, with
my original and almost forgotten name, I renounced
that name but remember it, it's mine"².

When examining text generation, we realized that it consists in a set of tiny (and mostly unperceivable and above all not perceived by the translator while she is working) translation processes that continue to shuttle from the mental map to the verbal map.
  At one level there is reasoning, mental language, producing, rather than speech acts, single links to words. At a higher level there is syntactic planning that deals with finding a coherent structure for the words temporarily generated, both to express correct relations between what is symbolized by these words, and to create utterances answering both the demands of syntactic rules of the receiving culture, and of the language in which the (re)coding occurs.
  One of the recurring motives in translation studies is the definition of the notion of "translation unit", the magic text segment that every translator instinctively (or so it would seem) chooses as the right length for such a complex mental and verbal elaboration. Kirsten Malmkjær, in the Routledge Encyclopedia of Translation Studies, summarizes the situation:
  First of all, the length of the text (or sense) segment used as a working unit varies according to the degree of linguistic (and cultural, we add) competence of the actualizer, spanning from the single word to whole sentences. And the readability of a translated text depends on the length of the translation unit used: the shorter it is, the less readable the resulting text.
  But if text generation occurs in the way described in Incremental Production Grammar³, the input coming from lexicon and the incremental strategy come into play at this point. In this perspective, I don't now how much sense it makes to seek the translation unit. There are probably as many translation units as stages in the process, as far as both decoding and recoding are concerned. For example, microplanning has probably much fewer translation units than macroplanning, and the two different lengths are simultaneously employed (on the mental map level).
  In the case of intersemiotic translation, often only the whole of a given text may be considered its translation unit. Let us think, for example, of a masterpiece like Guernica by Pablo Picasso. To read it as a translation means to think of the Guernica battle as a (human? historical? visual?) text, and of Picasso's picture as a pictorial translation. It would be an equally diminishing didactic approach as one that would induce us to dismantle the masterwork into single compositional elements depicting one or another aspect of the battle: the work of art is a masterpiece because of what it is in its entirety, because as a whole it expresses pain, rage, impotence, confusion, mourning, desperation.
  On the other hand, there are researchers that, like the Russian translation scientist Barhudàrov, points out that even a single phoneme can be a translation unit, if transliteration is used as an example. Here is an example:

In this transliteration of the name of the Russian compositor in three different languages one can notice, for example, that the initial phoneme /t∫/, according to the different actualizations, is rendered as Č, Ch, Tch respectively, as a function of transliteration norms (or pronunciation rules) in force in the different receiving cultures. Barhudàrov is therefore undoubtedly right as far as transliteration is concerned.
  Without insisting on the extreme cases, and trying to use more common cases, let us look at a concrete example, from the classic psycholinguistic book by Levelt, Speaking. From Intention to Articulation. Let us suppose we translate, in this moment we do not care from which language, and that the final result is the English sentence:

the child gave the mother the cat⁴

In one stage the three key elements of the sentence, the three referents, as Levelt calls them, are identified: child, cat, mother. There is no reason to think that the mind works on the three referents in the order in which they are to be expressed, but probably they are among the first elements to emerge in the actualizer's consciousness.
  When the individual, having chosen among the more or less 30,000 lemma at her disposal, reaches the word child, that word discloses its properties to her as far as syntactic combination is concerned (rules, usage). In our specific case, the word child is a noun, it has a plural form, but it is singular here, and governs a verb in third singular person.
  Meanwhile, once the referent is chosen, parallel subroutines are activated having the goal of investigating the possible presence of articles, prepositions and parameter values. In our example, we realize that the referent is already known to the model reader, and that, therefore, an article should be used with it. In the instance of English, the process ends here, but that would not be so with other languages like French or Italian or Spanish, where the determinative article has a paradigm differentiating gender and number. In the case of languages that, like Russian, don't have articles, the procedure would be different still.
  Having settled that the child has a given function within the future sentence, and that its action is expressed by the verb to give, the location of this verb makes its syntactic properties emerge: it governs a direct object (what is given) and an indirect object (the person to which the thing is given). The location of that word, therefore, with its properties, sends feedback to the actualizer regarding the way in which the utterance can proceed.
  From this quick example we can get an idea of the quantity of translation processes mental-verbal-mental implied by even the formulation of the simplest sentence: if we had to say what translation units were used in the process, there is certainly a stage in which the translation unit consists of only the word child, and others where the unit increments until it covers the measure of the whole phrase. Maybe the emphasis of the debate should be shifted from the individuation of the translation unit par excellence to the study of the stages of the translation process and of the various units involved.

Bibliographical references

BARHUDAROV L. Urovni jazykovoj ierarhii i perevod, in Tetradi perevodčika, n. 6, 1969, p. 3-12.

BARHUDAROV, L. The problem of the unit of translation, in Zlateva, P., editor, Translation As Social Action: Russian And Bulgarian Perspectives, London, Routledge, 1993, p. 39-46.

KEMPEN G. Language generation systems, in I. Batori, W. Lenders, W. Putschke, editor, Computational Linguistics: An International Handbook On Computer Oriented Language Research And Applications, Berlin, De Gruyter, 1988.

LEVELT W. J. M. Speaking. From Intention to articulation, Cambridge (Massachusetts), The MIT Press, 1993 (original edition 1989). ISBN 0-262-62089-8.

MALMKJÆR K. Unit of translation, in Routledge Encyclopedia of Translation Studies, edited by M. Baker, London, Routledge, 1998, ISBN 0-415-09380-5, p. 286-288.

MARÍAS J. Negra espalda del tiempo, Punto de lectura, 2000 (original edition 1998), ISBN 84-663-0007-7.

MARÍAS J. Dark Back of Time, New York, New Directions, 2001 (translated by Esther Allen), ISBN 0-8112-1466-4.