Consider the first verse of the gospel of John, and consider its usual English translation:
| En arch hn o logos kai o logos hn pros ton qeon kai qeos hn o logos | In the beginning was the word, and the word was with God, and the word was God. |
Now consider retroverting the latter back into Greek. Chances are that a translator, lacking any knowledge of the Greek, would produce something like
En h arch hn o logos
kai o logos hn sun tw qew
kai o logos hn o qeos.
Note that, while the English translation is more or less an adequate rendering of the Greek (except, perhaps, for the interesting flavour of the Greek preposition pros instead of sun or meta), it is simply impossible to move from the English to the Greek. It doesn't preserve the same attributes.
This is a constant difficulty, and one rarely addressed in either the manuals of textual criticism or those on translation; both leave it for the other.
Fundamentally, to the translator, variants can be classified into four groups based on two criteria:
1. Meaningful variants, and
2. Translatable variants.
The former list is almost the same from language to language; the latter differs from tongue to tongue.
Using English as our target language, let's give examples of each class:
It will presumably be evident that variants of the first class are the most important, and variants of the last class can be ignored. We will return to this subject later.
More complex are the cases where the distinction is blurred. Take the disciple whose name was either Lebbaeus or Thaddaeus. This is clearly a translatable distinction. But is it meaningful? Not necessarily, since neither name occurs elsewhere in the New Testament. If this disciple had been called "James Francis Edward Stuart the Old Pretender," it might set us wondering about anachronisms, but it wouldn't affect the plot, if I may so call it, of the gospels. It would affect synoptic studies, but those should be conducted based on the Greek text anyway.
Or, similarly, take the parable of the two sons (Matthew 21:28-31). We know that the son who went to the vinyard is the one who did the will of the father. But is that the first or the second son? This is a difficult question textually. The meaning, however, is the same either way. Is this a translationally significant variant?
There is also the question of textual support. Matthew 1:16 has a major variant concerning the paternity of Jesus -- but the real variant is found only in the Sinai Syriac. Is that enough reason to note a variant? Or 1 John 5:7-8 -- the work of a known heretic, with no significant textual support at all. Is that worth noting?
So which variants should go in the margin of a translation (if any)?
The answer to this depends very much on the intended audience of the translation. Obviously a translation intended for children should not have any marginalia at all if it can be avoided. But a translation for educated adults certainly should note places where the text is doubtful.
The number of variants still depends on the intended audience. As well as on the style of the translation. A severely literal translation, we should observe, ought to have more textual variants noted in the margin, because readers are trusting it to say what it says. By the same argument, a translation with a high number of marginal notes on the translation should have a high number of textual notes, because the text affects the translation.
The obvious temptation is to take the United Bible Societies' edition -- which, after all, has variants selected for translators -- and simply follow the variants there, or perhaps those marked as being of only the "{C}" level or higher, indicating significant uncertainty. This is presumably why they provided those variants, and for a translator with no text-critical background, it's certainly better than nothing. But there are several problems with this. First is the fact that it is generally conceded that the UBS editors are overconfident -- the fourth edition, in particular, marks many variants as more secure than they should be. Second, their selection of variants is somewhat questionable. And third, there is the problem of how this will be used. My experience is that the notes in a translation are most often used by groups such as small Bible Study classes. These groups will usually have several translations in use -- including, perhaps, someone with a King James Bible. The UBS apparatus omits many variants where it differs significantly from the Byzantine tradition behind the King James Bible. A good translation needs better notes than UBS provides.
A point I don't often see addressed is the different types of marginal notes. Typically, a translation, if it has notes at all, will feature both notes on the text and notes on the translation. This, of course, is perfectly reasonable -- but it's not obvious that they should be grouped together. (Note that most other critical editions with marginal glosses -- e.g. editions of Shakespeare or Chaucer -- have textual and linguistic glosses firmly separated.) In the case of translational differences, you put the best rendering you can in the text (either you think anwqen means "again" or it means "from above," but it means what it means). If something is in the margin, it's a less likely rendering. Textual variants are fundamentally different: Only one can be correct. There is no doubt of meaning; there is doubt of reading. It makes a different demand on the reader. A note on the translation often makes our understanding of the text richer. But a note on the text says that there are two different traditions about what is read here.
Then, too, most editions don't really indicate the nature of a variant. Is it highly uncertain? Is it included only because it's found in the King James Bible (e.g. 1 John 5:7-8). Admittedly a translation probably shouldn't be a textual commentary. But a strong case can be made that it should be more than it is: That it should include nearly every translatable and meaningful variant where there is significant doubt about the text, and that it should also include translatable and meaningful variants where the reading is not really in doubt but where some well-known edition has included the reading anyway -- and that these two classes of readings should be clearly distinguished.
There is no absolute and final rule for how to deal with textual variants in translations. There is no doubt in my mind, however, that more needs to be done.