SBL Presentation by Andrew Wilson, London 2011
In his 1965 study, E. C. Colwell used singular readings to investigate scribal habits in P45, P66 and P75. His method has subsequently been followed by other researchers: Royse, Head and Hernandez, among others. The logic of this approach is simple and sound: if external evidence counts for anything, a reading found in only one manuscript is extremely unlikely to be the genuine reading of the NT text. Instead, it is more likely to be an individualism of its scribe. While it is possible that any given singular reading might theoretically be the authentic reading, it would be extremely unlikely that large numbers of singular readings were original. Singular readings studied in sufficient numbers thus provide us with our best guide to typical scribal errors. To use James Royse’s words, in singular readings, we attempt “to discover the actual habits of scribes on empirical grounds, in a manner as free as possible from any presuppositions about scribal behaviour”. Juan Hernandez Jr. writes that singular readings are “the safest place to speak confidently about scribal tendencies”.
The present paper reports results from a study of 4200 singular readings in 33 chapters from the NT, thus covering more than 10% of the entire New Testament text. It isolates singular readings using a combination of the following critical apparatuses: NA27, Swanson, Tischendorf, Von Soden, IGNTP Luke, the ECM in the Catholic Epistles and Hoskier for the Apocalypse. In the study, all four of the main transcriptional canons were investigated (Prefer the Shorter, Harder, Harsher and Non-Harmonised Readings). Today I will report on the results for two of the canons.
Prefer the Shorter Reading?
Among the 2800 singular readings that added to or omitted from the text, scribes omitted 1712 times and added 1088 times. Scribes tended to omit about 60% of the time and add about 40%. They did so in all of the NT chapters studied and in all of the MSS, with a few notable exceptions. This, of course, is the exact opposite of what the traditional canon suggests.
Griesbach, in his statement of this canon, argued that brief omissions were an exception to the rule. However, when singular readings were analysed according to the length of the text omitted, one and two word omissions constituted 67% and 14% of all omissions, respectively, amounting to 81% of omissions. If 81% of omissions can be dismissed as an exception, we might also say that sheep are naturally carnivorous, but on exceptional occasions nibble grass. Short omissions are not an exception, but the rule itself. The attempt to defend the canon by appealing to the short omissions ‘exception clause’ does not appear to be viable.
When possible mechanical explanations for omissions (like homoeoteleuton and homoeoarcton) were considered, they were found to be the primary possible cause among longer omissions (i.e. three words or longer), although they were a minority among shorter omissions. However, there were 124 omissions that were not attributable to homoeoteleuton or homoeoarcton among the 323 longer singular omissions. Thus, for 38% of longer omissions, there was no mechanical explanation which presented itself. This result will not surprise those familiar with early MSS like P66 with its 9 lengthy inexplicable omissions in the latter part of John’s Gospel. It would appear that scribes often omit longer stretches of text for no detectable mechanical reason.
When one word additions and omissions were analysed according to the part of speech involved, scribes tended to omit pronouns, articles, nouns, verbs, adjectives and prepositions rather than add them. Interestingly, scribes seemed to add and omit conjunctions and adverbs at roughly similar rates.
When singular additions and omissions were analysed according to the type of manuscript in which they were found, the results showed that papyri, majuscules, minuscules and lectionaries all tend to omit rather than add. These results align with earlier studies such as Royse’s, Head’s and Hernandez’s which called Lectio Brevior Potior into question. Contrary to Royse’s confident expectation, later manuscripts tend to omit rather than add just like earlier manuscripts, although not as frequently.
When corrections of singular additions and omissions were examined, the results show that correctors heavily tended to remedy omissions. Thus, among the 431 singular additions or omissions later corrected, there were 133 additions and 298 omissions corrected (31:69). Among the corrections which were themselves singular readings, there were 40 additions and 27 omissions (a ratio of 60:40). For example, in Sinaiticus, 90% of the singular omissions were corrected (56/62) but only 78% of additions were (28/36). Similar results can be seen in the correction of P66 (this is from Royse’s data). Notice: despite the fact that correctors remedied both additions and omissions, the text was substantially longer after correction (Aleph: 28 readings longer, P66: 40).
Three factors were at work to produce an expansion of the text over time due to correction. Firstly, because the dominant error was omission, correction predominantly reversed this. Secondly, on top of this, there was a marked bias in favour of the correction of omissions. Nearly 70% of corrections involved omissions, while only 30% responded to additions. This over-sensitivity to omissions and more-indulgent attitude to additions accelerated the expansion of the text. Thirdly, when correctors themselves created singular readings (thus introducing textual novelty), these readings tended to be additions, not omissions. Thus, the process of correction over-reacted to scribal omission, and ultimately produced a longer text relative to our early witnesses.
Where does all this leave the text? Our current critical text prefers a shorter reading about 60% of the time, in line with the traditional canon. Thus, in Matthew, NA27 has a longer reading 256, a ‘middle’ reading 20 times, and a shorter reading 379 times, the majority of which involve short add/omit variation units. If we were to re-adjust the text to reflect the evidence from scribal habits, we should have to promote over 100 longer readings to the mainline text. Extrapolating these figures to the rest of the Gospels, there would need to be nearly 400 changes to our text. This figure is, of course, a rough guess, and each case would have to be taken on its merits, all other things being equal. However, the fact that the traditional canon is wrong means that there must also, of necessity, be a change in the text, to a larger or smaller degree.
Prefer the Harder Reading?
For the investigation of the Harder Reading Canon, 11 chapters of the NT (out of the 33 chapters earlier studied) were examined and all of the singular readings (not just additions and omissions, but also substitutions and transpositions) were catalogued. From these chapters 2279 singular readings were analysed for their effects upon the text. The results were categorized as nonsense, harder sense, harder style, neutral, easier style or easier sense.
The slide shows the percentages of nonsense (24%), harder sense (11%), harder style (27%), neutral (30%), easier style (8%) and easier sense (0.4%). By far the most significant result here is that of easier sense readings, readings that removed difficulties or improved the text semantically. Only 8 readings out of over 2000 fell into this category (0.4%). By contrast, there were 30 times as many readings where the scribe made the text more difficult, yet without producing nonsense (244 cases). Set against the expectation that scribes tended to improve the sense of the text or remove difficulties, a result of less than 1% fails to inspire confidence in the canon.
This is an extraordinary and surprising result. On its own, this result is unlikely to convince many textual critics to abandon such a long-standing and venerable canon. However, interestingly, other studies corroborate the findings here.
In Colwell’s study, out of a total of 1014 singular readings (after he had excluded itacisms), he mentions only 5 readings that actually remove difficulties or improve the sense of the text: a figure of 0.5%. I won’t run through these readings now – they are listed on the handout. However, in the five pages of his study dealing with (and entitled) Editorial Changes (pp 118-123 of the Brill edition, 1969 (NTTS IX)), rather than discussing readings that improve the sense, Colwell instead spends most of his time discussing matters of style. He refers to terms like style, clarity, conciseness and smoothing 33 times in this section of the study. By spending virtually all of his time discussing trifling stylistic editorial changes, Colwell appears to unwittingly confirm that scribes rarely ever editorialise in ways which improve the sense or meaning of the text.
Similarly, in Royse’s study, out of 1125 ‘significant singulars’, Royse only comments on 9 cases of readings which actually remove a difficulty or improve the text semantically (0.8%) - eight per thousand. In fact, the Harder Reading canon is not even mentioned in Royse’s index. As an indication of how rare these improved readings really are, consider Tarelli’s opinion that the reading of P45 at Luke 9:50 is ‘the one singular reading of the papyrus which materially affects the meaning of the text’.
Similarly, Hernandez’s result was 4 readings from 322 singulars: 1.2%. Now, maybe there are readings that I have missed in these studies, but there simply are not dozens, or hundreds, of them. Trawling through Royse’s 3000-odd footnotes will increase your respect for Royse’s magisterial and meticulous study, but it won’t increase by very much your count of singulars that improve the text. I submit that the reason these studies comment upon so few singular readings that improve the sense is that few of these readings really exist. If historians finding two independent accounts of the same event think they have hit the historical jackpot, the alignment of these results would appear to indicate that we are standing on text-critical solid ground. They show that scribes tended to improve the text about 1% of the time. Correctors removed difficulties, yes, (in fact, it was their job description), but scribes themselves tended to do the exact opposite – they created difficulties (hence the need for correctors).
There are two sorts of objections that could be raised against these results, logical and methodological. Firstly, it could be argued that lectio difficilior potior must be true on logical grounds alone, irrespective of any evidence. The argument is a familiar one: assuming that scribes did not intentionally make the text more difficult, we should prefer a harder reading unless some obvious accidental explanation suggests itself. This argument, however, appears to commit a logical fallacy: the fallacy of the excluded middle (a.k.a. the false-dilemma). It ignores the possibility of readings created sub-consciously, that is, as a result of a wandering mind, a lapse in concentration, haste or fatigue. The many singular harmonisations are evidence of scribes operating on ‘auto-pilot’, the mind drifting back to or anticipating events in the immediate context, or wandering off to parallel accounts in different books altogether. The argument also ignores readings created in more than one stage, as salvage readings of earlier errors, as botched corrections or due to the confused interpretation of corrections in a Vorlage. Colwell and Royse give many examples of imperfect or two-stage corrections or harmonisations. My own study found literally hundreds of harder readings among the singulars. The simplistic nature of the ‘common sense’ argument seems either to ignore these realities or to lack acquaintance with the complexities of scribal behaviour.
Science has two components: theory and observation (or research). Theory without research is mere speculation; observation without theory is mere stamp-collecting. The idea that we may dismiss the best line of research we possess (singular readings) in favour of ‘common sense’ is a rejection of evidence-based approaches altogether in favour of theoretical speculation. It is on a par with insisting that a particular folk remedy is able to cure ailments, despite clinical trials showing no evidence of this happening. A canon based on ‘common sense’ is little more than an article of speculative faith.
Secondly, methodologically, some might perhaps argue that singular readings are ill-suited for investigating the harder reading canon. For example, it could be argued that later copyists would tend to preferentially perpetuate easier readings or readings with less difficulties with the result that easier readings would no longer show up among singulars, but instead attract other documentary support. To test this objection, sub-singular readings (readings only found in two or three mss) were also catalogued in this study in the same chapters, to see whether there was any increase in the percentages of easier readings. The results (4/750, 0.5%), show no jump in easier readings (either as a result of genealogical relationship or scribal coincidence). There is still very little improvement going on in sub-singular readings. This shows that the results from singular readings cannot be dismissed as poor sampling. In any case, the attempt to discredit singular readings as sources of evidence is troubling: why should scribes always be “flying under the radar”, so that their attempts to remove difficulties or improve the sense rarely show up? If Prefer the Harder Reading were true, we should catch more than a few scribes in the very act of creating novel improvements – among singular readings. The problem is that the vast majority of scribes do nothing of the sort.
Let me conclude by listing eight summary points:
- Scribes of all eras tend to omit, rather than to add
- Brief omissions are not an exception to any rule, but the rule itself
- Among longer omissions, roughly one third have no detectable mechanical cause
- Ad hoc correction played a significant role in the expansion of the text over time
- There is very little evidence among singular readings for Lectio Difficilior Potior
- Sub-singular readings also provide little support for the Harder Reading canon.
- Colwell’s and Royse’s studies show little evidence of the canon in action.
- The common sense case for this canon lacks acqaintance with scribal realities.
The accumulating empirical evidence about the tendencies of scribes is important. It offers significant help in the tasks of shaping the form of our critical text and studying its history. NT textual criticism, after all, must be based upon evidence.