The New Testament is full of difficult readings. There are probably hundreds of places where one scholar or another has argued that the text simply cannot be construed. Westcott and Hort, for instance, marked some five dozen passages with an asterisk as perhaps containing a primitive error. (A list of these passages is found in note 2 on page 184 of the second/third edition of Bruce M. Metzger's The Text of the New Textament.) Not all of these are nonsense, but all are difficult in some way.
In classical textual criticism, the response to such "nonsense" readings is usually conjectural emendation -- the attempt to imagine what the author actually wrote. Such an emendation, to be successful, must of course fit the author's style and the context. It should also, ideally, explain how the "impossible" reading arose.
The use of conjectural emendation in the classics -- especially those which survive only in single manuscripts -- can hardly be questioned. Even if we assume that there is no editorial activity, scribal error is always present. Thus, for instance, in Howell D. Chickering, Jr.'s edition of Beowulf, we find over two hundred conjectures in the text, and a roughly equal number of places where other sorts of restoration has been called for or where Chickering has rejected common emendations. All this in the space of 3180 lines, usually of four to six words!
In the New Testament the situation is different. There is one (badly burned) manuscript of Beowulf. The major works of Tacitus survives in several manuscripts, but they do not overlap, and while there are four manuscripts of the Agricola, it appears that three of them are descended from the fourth. Polybius and Livy, too, survive only in part. Asser's Life of Alfred exists only in a printed transcript. But for the New Testament, every passage survives in at least two hundred witnesses (excluding the versions), and outside the Apocalypse the number of witnesses rises into the thousands.
So how does this affect the tradition? In one sense it is an immense boon; it means that we can see our way around the peculiarities of any particular copy. Does this mean that there is no need for conjectural emendation?
Various scholars have answered this differently. Most contend that there should be no need for conjectural emendation. Others, such as Zuntz and Holmes, allow for the possibility; Holmes writes, ""That there is considerably less need for emendation of the NT text than that of comparable documents is indeed true, but we must not confuse less need with no need." (Michael W. Holmes, "Reasoned Eclecticism in New Testament Textual Criticism," printed in Bart D. Ehrman & Michael W. Holmes, The Text of the New Testament in Contemporary Research, 1995, page 348. This section, pp. 346-349, is probably the best brief summary of the need for a more "classical" style of criticism.) And Kenneth Sisam comments of the difference between printing an attested and an unattested reading, "To support a bad manuscript reading is in no way more meritorious than to support a bad conjecture, and so far from being safer, it is more insidious as a source of error. For, in good practice, a conjecture is printed with some distinguishing mark which attracts doubt; but a bad manuscript reading, if it is defended, looks like solid ground for the defence of other readings." (Kenneth Sisam, "The Authority of Old English Poetical Manuscripts," now available in Studies in the History of Old English Literature, p. 39. This volume, despite its title, is largely devoted to textual questions, and much of the advice, including the above, is capable of application outside the context of Anglo-Saxon.)
Of the theoretical possibilities for conjectural emendation there can be no question. It is likely that there are several New Testament books where all extant copies are derived from an ancestor more recent than the autograph. In the case of Paul, most copies are probably derived from the original compilation of the letters rather than the originals themselves. In each of these cases, errors in the remote archetype will be preserved in all copies. As a result, we see editors sometimes mark certain readings as corrupt (such as the aforementioned "primitive errors" obelized by Westcott and Hort).
But how does one detect these errors? Simply by looking for "nonsense" readings? But one scholar's nonsense is another's subtlety. In any case, can it be shown that all nonsense readings derive from copyists? I hardly think so. Much of the New Testament was taken from dictation. Can we be certain that even the original scribe had it right? And what proof is there that the original author was always grammatical and accurate? I have yet to see an author who never made an error in writing. And even if you think you've found an error, as Westcott and Hort did, how do you reliably correct it?
Take a concrete example, in 1 Corinthians 6:5. The Greek text reads diakrinai ana meson tou adelfou autou, "to judge between his brother." Zuntz, would emend to diakrinai ana meson tou adelfou kai tou adelfou autou, "to judge between the brother and his brother." (The Text of the Epistles, p. 15). This is technically not pure conjecture, since it has some slight versional support, but Zuntz thinks, probably rightly, that these are conjectures by the translators; he is just adopting their conjecture.
Now it's likely enough that Zuntz has the sense of this passage correct. But does that mean it is actually the autograph wording? People do leave words out sometimes. And there is at least one other possibility for emendation: instead of adding kai tou adelfou, we might emend ana meson -- i.e. to read something like "to pass judgment upon his brother" instead of "to pass judgment between his brother." Observe that, even if we are sure we need to emend (and we aren't), we are not certain how to emend. That's the heart of the problem.
With all these factors in mind, it is worth noting that conjectural emendation is not entirely dead; the UBS text prints a conjecture in Acts 16:12 (the reading is supported by codices Colbertinus Theodulfianus of the Vulgate, as well as by the Old Church Slavonic, but these are clearly variants peculiar to the version rather than their underlying text). But it should be frowned upon; we note that, when selecting a reading from among variants, one generally choses the one which best explains the others. But when adopting a conjectural emendation, one should only accept a reading which completely explains the others. This happens so rarely that we can almost ignore it -- particularly since such corrections can still be wrong. An example comes from Langland's Vision of Piers Plowman. In the editio princeps, which for a long time was the only text available, the very first line read
In a somer seson whan set was the sonne
("In a summer season, when >>set<< was the sun")
"Set" is perhaps meaningful, but does not scan. Therefore attempts were made to correct it. The most popular emendation was "hotte," "hot."
The correct reading, as now known from many manuscripts, is "softe," "soft." Thus the proposed emendation, although perfectly sensible and meeting all the desired criteria, in fact gives a meaning exactly opposite the true reading.
Or we might illustrate an example from Beowulf, where we do not know the correct reading. Line 62, as found in the manuscript, reads (in Old English and translation):
Which doesn't make any more sense in Old than Modern English. There is a missing noun. The context is a list of the children of Healfdene; we are told there are four, and three have been listed (Heorogar, Hrothgar, and Halga); we expect the name of a fourth. Old English word order would allow the name to appear in the next line -- but it doesn't. And this line is defective, missing a stress and an alliteration.
What's more, there is no known King Ela for this unnamed girl to marry. This suggests an easy emendation: "ela" is short for "Onela." If we insert this likely emendation and the verb was, as well as expanding the abbreviation for that, we get
Now we need a name. It must be feminine, it must complete the alliteration, it must fill out the line.
The moment I saw this, without a moment's hesitation, without even knowing Old English, I suggested the emendation "Elan," which meets every requirement. And it would explain how the error came about: A haplography elan1...elan2. In other words, our line would become
This conjecture has been proposed before -- and rejected because there is no evidence that Onela had a wife Elan. (Of course, there is also no evidence that he didn't -- if we had good evidence about this period, we very well might have another copy of Beowulf, and the whole discussion would be moot.)
As a result, at least two other conjectures were offered for the name. One suggested the name Yrse (Grundtvig, Bugge, Clarke). This, too, faces the problem of being a poorly-attested name. So a third suggestion was "Signi" (or similar). This is on the basis that the "real" Signi was the sister and bedmate of Sigismund, and our unnamed wife of Onela is also accused of incest. The problem is that, if we wish to preserve the alliteration, this forces further emendations to the line, changing (On)ela to "Saevil" or some such.
Still others propose to leave the line as it is and emend in a half line below this. (Though it appears that no such emendation really works). A fifth proposal is to emend the line to omit any name of the woman and just read "a prince," or some equivalent non-name, for Onela.
I happen to have eight complete editions of Beowulf (mostly in translation, but some in Old English), plus an essay by Norman E. Eliason on this very subject (Norman E. Eliason, "Healfdene's Daughter," pp. 3-13 in Lewis E. Nicholson and Dolores Warwick Frese, editors, Anglo-Saxon Poetry: Essays in Appreciation. The various solutions they adopt are as follows (first the name of the girl, then the name of the man who married her):
So here is the situation: We have an obvious error, and an obvious emendation, and no one accepts the obvious emendation, and we see two different alternate conjectures, two other conjectures for the form of the line, two different primitive errors marked, and one editor who refuses to admit that nonsense is nonsense. It's not the most impressive performance.
For these reasons, with all due respect to Zuntz et al, who correctly point out that conjectural emendation may be needed to restore the original text, we must always be cautious of going too far. As Duplacy remarks (quoted in Vaganay & Amphoux, An Introduction to New Testament Textual Criticism, English translation, p. 84), "The supreme victory of internal criticism is... conjectural emendation, especially when it is the original text itself which is emended." Unless we are certain we are not making that mistake, conjectural emendation should be avoided.
To give a concrete New Testament example, consider the third part of Matthew's genealogy, Matt. 1:12-16 (the portion of the genealogy after the exile, where we have no other sources to compare against). Matthew 1:17 implies that there should be fourteen names here, but there are only thirteen. It may be that Matthew goofed (in fact, it's quite clear that this genealogy cannot be complete -- thirteen names spread across 570+ years is 45+ years per generation, which is simply not possible). But it is also reasonable to assume that one name was lost from the genealogy at a very early date -- in other words, there is a primitive error here. But can we correct it? The answer is simply no. We may think a name is missing, but we have no grounds whatsoever for determining what it might be or where it is lacking. Although we see the need for emendation, we have no tools for correctly performing it.