How to Classify Codex Bezae’s Greek Text

For two centuries, Bezae’s classification has been thought of in terms of the so-called “Western” text type. Of course, the label “Western” tells us little if anything about Bezae’s relationships with other witnesses.

But Bezae does have relationships. Its parallels with 03 were noted by F. J. A. Hort.1 In Mark, Bezae’s parallels with 038 and 565 were pointed out by H. von Soden.2 Also in Mark, Bezae’s parallels with 032 were observed by H. A. Sanders to be a byproduct of the latter’s parallels with the Latin version.3 The question is, how does Bezae’s peculiar mix of relationships affect the classification of its text?

The most systematic attempt to classify Bezae’s text form to date is Text und Textwert (TuT), a project that seeks to classify every manuscript on the basis of agreement profiles at non-mainstream readings, which are found in the main list for each manuscript and book. When we inspect Bezae’s main list in Mark, we discover that its closest non-fragmentary relative is 038, which attests 40.4% agreement with Bezae’s profile — certainly not an impressive level of agreement.4 Overall, the impression we gain from Bezae’s main list in Mark is that of a text form isolated from the larger Greek tradition.

Of course, TuT considers only Greek manuscripts. For practical reasons, it relies on test passages, 196 in Mark. These test passages are not randomly selected and, hence, reflect a distinct bias in favor of passages of greater perceived text-critical interest. The result is a high proportion of test passages in which Bezae attests a unique reading in the Greek tradition, a situation that tends to exaggerate Bezae’s apparent isolation.

We can resolve the problem of bias in the selection of test passages with either a random sampling or a complete data set of variation units. According to my collation of transcripts prepared for the Editio Critica Maior (ECM) in Mark, in work for my dissertation, Bezae attests 1383 non-mainstream readings in Mark, setting aside readings that are singular or subsingular in the Greek tradition.5

Bezae’s closest Greek neighbors at these 1383 non-mainstream readings in Mark are 565 at 51.2% agreement, 038 at 47.3%, 032 at 39.3%, 01 at 35.3%, and 03 at 35.3%.6 So Bezae’s closest neighbor, 565, still only registers at 51% agreement.

Now according to TuT, 565 and 038 are closely related in Mark.7 So are 01 and 03.8 So we find at least three different traditions represented among Bezae’s closest witnesses in Mark: 565/038, 032, and 01/03. These results are summarized in the table:

Manuscript Non-Mainstream Agreements with Bezae in Mark
565 51.2% (706/1380)
038 47.3% (654/1383)
032 39.3% (522/1327)
01 35.3% (479/1356)
03 35.0% (482/1376)

Still, none of these witnesses exhibits an overwhelmingly high level of agreement.

Assessing the Unique Contribution of Witnesses

One question that remains unanswered by these figures is how much overlap exists between these agreements. To what extent do the 51% of agreements with 565 overlap the 47% of agreements with 038, the 39% of agreements with 032, or the 35% of agreements with 03? If we take 565 as the closest witness at 51%, which next witness accounts for the highest percentage of non-mainstream readings not attested by 565?

It turns out that the witness that contributes the most new agreements which are not already attested by 565 is 03 — even though this witness registers at fifth place in overall agreement. While 038, 032, and 01 all agree at higher individual percentages than 03, it turns out that many of their agreements are shared with 565.

On the other hand, 03 contributes the most new agreements. Sixteen percent of 03’s agreements at Bezae’s non-mainstream readings are not attested by 565. So between the agreements of 565 and 03 we can account for nearly two thirds or 67.5% of Bezae’s non-mainstream readings in Mark.

The witness that contributes the most new readings not attested by either 565 or 03 is not 038, but 032, which adds 9% to our cumulative percentage of agreement. This brings the total cumulative agreement between the three witnesses — 565, 03, and 032 — to over three quarters or 76.6% of Bezae’s non-mainstream readings in Mark.

With the fourth witness we start to encounter diminishing returns. The witness that contributes the most new readings after 565, 03, and 032 — 038 — contributes just 3.5% to the total cumulative agreement, bringing the total agreement to four out of five or 80.1% of non-mainstream readings. As we noted above, 038 is closely-related to 565 in Mark. So it is not surprising that it contributes so few new agreements.

The respective contributions of the four witnesses that together agree with 80% of Bezae’s non-mainstream readings in Mark are illustrated in the chart.

How are we to understand this result?

It appears that Bezae is related in Mark, not to a single witness or family, but primarily to three witnesses or families in the Greek tradition, namely, 565-038, 032, and 03-01 and their relatives. The implication is that Bezae attests a composite text form in Mark, a result that can be explained by mixture.

Classifying Bezae’s Mixed Text Form

So to answer our initial question, how to classify Bezae’s Greek text, the answer is that there are at least two ways.

One way is to classify the text as a whole. We find then that Bezae’s text form is closest in the Greek tradition to that of 565 across all of Mark, with which it agrees at 51% of its non-mainstream readings. The downside of this all-or-nothing approach is to create the misleading impression that Bezae’s Greek text is not particularly related to any other traditions.

A second way is to partition Bezae’s non-mainstream readings according to the witnesses and families with which they most closely agree. So we acknowledge that different sets of Bezae’s readings are closely related to different parts of the tradition.

As a starting point, we might assume that the largest set of readings — in which Bezae agrees with 565, 038, or both — represents the base tradition before the mixture of readings from other traditions.9

If so, we can assume that, in the case of readings where 565 and 038 overlap with other traditions, the reading entered Bezae’s tradition through the base text, that is, through the tradition represented by 565 and 038. But the readings of traditions that do not overlap with this base tradition presumably entered through mixture.

So we classify Bezae’s text form in parts rather than as a whole. The upside is that Bezae is now shown to be thoroughly related to the rest of the Greek tradition, though not to any single part of it. The classification of its text form consists of sets of readings that are closely related to different parts of the tradition.

We find then that, while Bezae’s text form is perhaps unique among surviving witnesses in its pattern of mixture, its text form is not unrelated to the larger Greek tradition. In fact, it is closely related to multiple witnesses and families.

Profiling Codex Bezae’s Greek Text Form as a Composite Text

On Tuesday, I presented a paper at the SBL International Conference in Helsinki on profile-based classification of composite text forms, in which I highlighted Codex Bezae’s Greek text of Mark as a case study in the profiling of highly-mixed text forms. This paper previewed some of the research for my dissertation on Bezae’s Greek text of the Gospel of Mark.

Highly-mixed text forms, such as Bezae’s, present unique challenges when we compare them to other known traditions, as their total profiles are likely to be substantially unique — unless of course we can identify other manuscripts with precisely same patterns of mixture, an unlikely prospect for text forms that lie far outside of the mainstream.

Are we to conclude from this that such text forms have no significant relationships with the wider tradition at all? Intuition tells us that this could not be so. After all, mixture by definition implies the existence of relationships with a variety of other traditions. Thus, in Bezae’s text of Mark, we find agreements with a variety of witnesses — for example, 03, 032, 565 — that are not otherwise closely related. In fact, Bezae does agree with other manuscripts and with some manuscripts more than others.

To understand how these diverse traditions may have interacted in the development of Bezae’s final text form, we require a more granular approach to manuscript profiling that relates these traditions only at the specific set of readings that they share in common, rather than across their total profiles. In this paper, I discuss such a granular approach to profiling that addresses complex mixed text forms by splitting the total profile into a composite of smaller sub-profiles, each with specific alignments within the tradition and hence modeling different relationships in the development of the final text form. In this way, we can begin to reconstruct Bezae’s textual history and potentially the textual histories of other highly-mixed witnesses.

Can Greek manuscripts account for Bezae’s variation?

Bezae’s distinctive parallels with the Latin tradition are well known in the gospels. The so-called “Western non-interpolations” are among the better known examples. In Luke 22:19-20, for example, Bezae sides with VL(3 8 10 17) against all other Greek witnesses. Of course, Bezae is the earliest Greek manuscript to attest two of the more spectacular gospel variations, both well-represented in the Old Latin tradition, the pericope adulterae with VL(2 6 8 15 14) and the Markan long ending with VL(6 8 10 13 15 16). Bezae also stands alone with the Latin in some notable shorter variants, such as the angry Jesus reading in Mark 1:41 with VL(3 8 14) and the variant cry of dereliction in Mark 15:34 with VL(1 6 17). In these and many other places, Bezae stands out as the only (or earliest) Greek manuscript to attest distinctive Latin readings.

Now it is generally assumed that Bezae represents the source tradition of these peculiar Latin readings. But every parallel has two possible directions. The Greek-source hypothesis is not the only possibility.

Before J. S. Semler (1725–1791), Bezae’s Latin parallels were routinely ascribed to Latin influence. For example, Semler’s mentor J. J. Wettstein (1693–1754) argued that Bezae’s Greek text had been systematically adapted to readings from at least two Latin traditions. Commenting on the scribe’s introduction of divergent text forms from the Latin, Wettstein writes:

“very often he made the Greek according to either the Latin or Italian version.” 1

Noteworthy is Wettstein’s explicit attribution of Bezae’s peculiarities to deliberate conformity of the Greek text to multiple Latin traditions. Significantly, Wettstein makes no claim that Bezae’s distinctive variations derive from its own Latin column, but rather from the broader Latin tradition, a point that has often been missed by subsequent critics. 2

Semler eventually reversed his opinion from that of Wettstein, arguing that sufficient diversity existed within the Greek tradition to account for Bezae’s variation:

“It is a dubious assertion that the Greek codices have been influenced and altered to the Latin. … We discover this by just looking at the codices written in Greek throughout the various Greek provinces. Noteworthy variations were already in place in the third, fourth, and fifth centuries.” 3

According to Semler, it is easier to suppose that Bezae’s Greek text came from a divergent tradition that lies behind the various Latin translations. Noting the number and variety of Greek witnesses discovered in his time, Semler argues that the diversity (magna varietas) of the Greek tradition is sufficient to account for the readings found in Bezae:

“Erasmus, R. Simon, Mill, Bengel, Wettstein, and nearly all other critics of this mass of Greek codices, which already more recently we have become acquainted with, designated the great multitude Latinized, that is, conformed to the Latin text because they agree with the Vulgate or a certain Latin translation. But we reject this with a single decisive observation, namely, that the Latin copies were originally translated from the same Greek codices from which many Greek copies of a different kind [i.e. Bezae’s] were transcribed. Among the Greek codices there was already at that time great variety, for a great many Latin translations already existed at the time of Jerome and Augustine derived from these Greek codices. Therefore, it certainly did not happen that these Greek books could ever have been altered to those Latin copies with which they already agreed.” 4

So in Semler’s view, there is no need to appeal to Latin codices to account for readings found in Bezae.

But can the diversity of the Greek tradition by itself account for Bezae’s distinctive Latin parallels? It is now clear that even the full Greek tradition leaves us short in accounting for Bezae’s unusual text form. As evident from results published in Text und Textwert, Bezae remains with an abundance of singular Greek readings even when we consider the entire surviving Greek tradition.

Here are the percentages of singular readings in Bezae by book according to Text und Textwert:

Book Witnesses Teststellen Singular Readings % Singular Readings
Matthew 1757 53 6 11.32%
Mark 1754 192 36 18.75%
Luke 1756 53 8 15.09%
John 1763 126 13 10.32%
Acts 607 72 31 43.06%

Clearly the diversity of the Greek tradition cannot by itself account for the full scope of Bezae’s distinctive variation, ranging from ten percent in John to over forty percent in Acts against over 1500 witnesses to each gospel and over six hundred witnesses in Acts. Is it plausible that so many readings generated in the first few centuries were subsequently lost to the Greek tradition from which they presumably derived, only to resurface in translation? Until a satisfactory answer is given, it seems premature to rule out any plausible alternative.

On the other hand, is it not more plausible to ascribe this surplus of variation to readings that we know already existed, even in another version? 5 Perhaps these variations arose in translation or entirely within the Latin tradition. Surely some Latin translators active at the time of Jerome and Augustine were capable of translating select passages into Greek. It seems all we require is a motive.

(On possible motives for Latin assimilation, see here and here.)

What is Bezae’s place in Acts?

The textual tradition of Acts is clearly of great relevance for research on Codex Bezae. Not only is Bezae’s text longer and more divergent in Acts, but we also catch glimpses of it in other Greek witnesses, such as P38, P127, and MS 614 among others.

For a sense of Bezae’s close relationships in the Greek tradition, I will turn to the data in Text und Textwert (= TUT) from which I have constructed of a table of closely-related witnesses for Acts, described in a previous post. [1] Using the open-source graphing application, Gephi, I have made a graph depicting the structure of the Greek tradition of Acts based on close relationships between witnesses. The ten witnesses most closely related to Bezae in Acts (1162, 623, 619, 2718, 08, 945, 1704, 1751, 1884, and 2412) are indicated with yellow arrows and arranged in four groups, each with a distinct profile.

Acts Gruppierung (excluding 1-2, bezae)

Key

In the graph, nodes represent witnesses, while the edges denote close relationships and node size reflects the number of close relationships for a witness. Witnesses are color-coded according to “Lesarten 1 1/2 value” with purple farthest from the Majority Text and red closest. Note that some witness groups have no close relationships to witnesses in the main graph and hence they are not connected to the main graph. Some witnesses do not appear at all in the graph, either because they are not extant in enough test passages (such as P38 or P127) or because they have no closely-related witnesses (such as Bezae). A PDF version of the graph is available.

Interpretation

To suggest a tentative interpretation, we might note that witnesses of the so-called “Alexandrian” tradition (represented by purple nodes) appear in the center around a core of P74, Sinaiticus, and Vaticanus. Just above this “Alexandrian” group, a second group (with blue nodes) appears to the left of MS 1739. A third, smaller group (of green nodes) centered around MS 206 lies above the blue nodes. If we assume that the “Alexandrian” group represents the initial text (with a direction of flow towards the Byzantine tradition — represented here by the Majority Text), these “blue” and “green” groups appear to have developed in sequence through a natural but fairly controlled process of transmission. Below the “Alexandrian” group, the tradition quickly branches in a variety of directions into groups that are not as tightly connected, perhaps indicating a less-controlled pattern of transmission.

The large, tightly-coupled cluster (of red nodes) apparently representing the early Byzantine tradition seems not to connect to the main group based on our criteria. My guess is this group would connect near the bottom of the graph (near the orange nodes) if we lowered our thresholds for the Gruppierung criteria or counted 1/2 readings as non-majority. While there are many factors to consider, the radiating pattern seems consistent with models based on an “Alexandrian” initial text. (However, I would welcome alternative interpretations in the comments.)

The precise place of Bezae’s text is somewhat difficult to determine because it is no more than 36% related to any other witnesses that appear in the Gruppierung data. [2] In fact, given Bezae’s isolation and lack of close relationships, it is difficult to envision a natural transmissional pathway to (or from) Bezae’s profile from any other place in the tradition (as I suggested in my earlier reply to James Snapp Jr.). This disconnect with the rest of the tradition suggests that the variations we see in Bezae may have been introduced artificially by processes other than simple copying.


[1] This table is modeled on the Gruppierung nach Übereinstimmungsquoten tables found in the four TUT volumes on the gospels.

[2] This problem is discussed at the end of volume 1 in the TUT of Acts.

(revised) Top 10 Closest Greek MSS to Bezae in Acts

Important Note: This post supersedes a previous post that used an erroneous calculation to compute the Gruppierung nach Übereinstimmungsquoten table for Acts. Note that the figures here are still preliminary.

Here are the (revised) top 10 closest Greek MSS to Bezae in Acts based on the computed Gruppierung data:

GA Non-Majority % Total % Profile
1162 36.4% (8/22) 34.7% (25/72) 18.4 23 26 42.4 46 57 61 74
623 35.3% (6/17) 32.1% (17/53) 23 26 42.4 46 57 62
619 35.0% (7/20) 30.9% (17/55) 23 26 42.4 46 57 61 74
2718 33.3% (6/18) 30.6% (19/62) 23 42.4 46 57 62 68.3
08 31.8% (7/22) 34.7% (25/72) 2 18.4 25.3 26 46 61 62
945 31.8% (7/22) 31.9% (23/72) 18.4 21 23 26 46 57 68.3
1704 31.8% (7/22) 31.9% (23/72) 18.4 21 23 26 46 57 68.3
1751 31.8% (7/22) 31.9% (23/72) 18.4 21 23 46 57 61 68.3
1884 31.8% (7/22) 32.9% (23/70) 2 15.4 25.3 26 46 61 62
2412 31.8% (7/22) 37.5% (27/72) 23 42.4 46 49.4 57 62 72.4

The first thing to observe is that none of these “top ten” MSS is particularly close to Bezae. For example, none of these MSS would appear in the Gruppierung nach Übereinstimmungsquoten table in the printed edition because they do not satisfy the “second” Gruppierung criterion that Non-Majority % must be greater than 50%.

Still there are some noteworthy patterns:

  1. MSS 623 and 619 share a similar profile
  2. MSS 945, 1704, and 1751 share a similar profile that is different from that of 623 and 619 (945 and 1704 have the same profile)
  3. The Greco-Latin bilingual, Codex Laudianus, appears in the list
Notes on Method

I am following the same essential method used in the gospels (e.g. see John vol. 1, pp. 50-53; Luke vol 1., p. 25; Mark vol. 1, p. 44) with two modifications:

  1. Due to Bezae’s free character, basing agreements on exact matches tends to exaggerate Bezae’s distinctiveness with the result that it may appear more isolated. (see Epp, “Textual Clusters,” 2013) To mitigate this effect, I am including TUT‘s “variant” agreements (denoted as capital letters) as matches (unlike the TUT calculations which exclude these).
  2. I am not counting Bezae’s five 1/2 readings (10, 35, 52, 55, 76) or its 1/2 variant (1/2L) as initial-text readings.

Which MS is closest to Bezae in the gospels?

Where does Bezae’s text fall in the Greek gospel tradition? Is it totally isolated or does it have some close relationships?

To get a rough sense, we can turn to the Gruppierung nach Übereinstimmungsquoten tables in Text und Textwert. By excluding agreements with secondary Majority readings, these tables are helpful in taking account of distinctive agreements with the greatest potential genealogical significance.

Not surprisingly, the result depends on which gospel we are examining. Only in Mark of the four gospels does Bezae lack any close witnesses according to the Gruppierung tables. The data for the other three gospels are given below.

In Matthew, two family 1 MSS appear closest to Bezae, followed by the palimpsest Codex Dublinensis (Z) and the fragmentary majuscule 0281 from Saint Catherine’s monastery:

Witness Without MT Total
1582 62,5% (15/24) 55,1% (27/49)
1 61,5% (16/26) 52,8% (28/53)
035 (Z) 60,0% (6/10) 50,0% (9/18)
0281 60,0% (3/5) 50,0% (5/10)

While the fragmentary nature of the latter two witnesses diminishes the significance of the data, the close alignment of a segment of Bezae’s readings with core Family 1 MSS suggests the influence of a Greek tradition in Matthew similar to that of Origen (see Anderson 2004).

In Luke the significance of Bezae’s agreement with Codex Zacynthius (Ξ) is questionable (like that of 035 and 0281 in Matthew) given the fragmentary nature of this codex:

Witness Without MT Total
040 (Ξ) 85,7% (6/7) 43,8% (7/16)

In John it is well-known that Bezae attests a high number of distinctive parallels with Codex Sinaiticus (א) in the first eight chapters (Fee 1968), a result that is confirmed by the Gruppierung data:

Witness Without MT Total
01 (א) 57% (31/54) 46% (57/123)

Although only the first half of John (through 10:41) is represented in the test passages, this level of agreement in even part of a gospel over so many test passages must be considered highly significant.

So what might we gather from the Gruppierung data for Bezae?

It is clear that only in Matthew and John are the numbers significant enough (statistically speaking) to relate Bezae to another part of the Greek tradition. But here the data are historically significant because they seem to connect a sizeable body of Bezae’s readings in these gospels to recognized old texts, whether of Family 1 in Matthew or of Codex Sinaiticus in John.

The situation is especially intriguing in John, because Sinaiticus departs from its usual pattern to agree with Bezae and the Old Latin gospels. If we accept this as evidence of a relationship, it might imply, for example, that something like Bezae’s tradition was already well-established in Greek by the mid-fourth century (accepting a ca 350 date for Sinaiticus). On the other hand, if Bezae’s bilingual Greek tradition reflects any degree of influence from Old Latin traditions that were still developing in the mid-fourth century, this would seem to push Sinaiticus to the end of the fourth century at the earliest.

Top 10 Closest Greek MSS to Bezae in Acts

The Gruppierung nach Übereinstimmungsquoten table found in each of the four Text und Textwert (= TUT) volumes on the gospels offers a good high-level sense of the closest textual relationships in the Greek tradition. Unfortunately, this useful table is not present in the TUT volume of Acts. Recently, I have written a Python script to parse the Verzeichnende Beschreibung data and compute Gruppierung data for Acts (see below).

[Note: The list below has been revised in a more recent post.]

Here are the top 10 closest Greek MSS to Bezae in Acts based on the generated Gruppierung data (NA28 readings in plaintext, special readings in bold, work shown here):

GA Non-Majority % Total % Teststellen
1853 66.7% (6/9) 37.5% (27/72) 4 18.4 46 57 62 72.4
XII Athos acts paul
1646 66.7% (4/6) 36.1% (26/72) 21 26 61 72.4
1172 Athos gospels acts paul
1610 66.7% (4/6) 33.3% (20/60) 18.4 46 57 62
1463 Athens acts paul
1893 62.5% (5/8) 33.9% (20/59) 23 42.4 46 57 74
XII Jerusalem acts paul revelation
2201 60.0% (3/5) 32.8% (19/58) 23 46 57
XV Elassona gospels acts paul revelation
623 60.0% (6/10) 32.1% (17/53) 23 26 42.4 46 57 62
1037 Vatican acts paul
619 58.3% (7/12) 30.9% (17/55) 23 26 42.4 46 57 61 74
X Florence acts paul
913 57.1% (4/7) 35.4% (23/65) 18.4 46 57 62
XIV London acts paul
1162 57.1% (8/14) 34.7% (25/72) 18.4 23 26 42.4 46 57 61 74
XI Patmos acts paul
436 55.6% (5/9) 34.7% (25/72) 18.4 42.4 46 57 62
XI/XII Vatican acts paul

I am following the same method used in the gospels (e.g. see John vol. 1, pp. 50-53; Luke vol 1., p. 25; Mark vol. 1, p. 44 in order of helpfulness, note Matthew has same description as Luke) with a few modifications:

  1. Due to Bezae’s free character, basing agreements on exact matches tends to exaggerate Bezae’s distinctiveness with the result that it may appear more isolated. (see Epp, “Textual Clusters,” 2013) To mitigate this effect, I am including TUT‘s “variant” agreements (denoted as capital letters) as matches (unlike the TUT calculations which exclude these). For example, at Teststelle (= TS) 8 (Acts 2:31) I am counting Bezae’s “singular” variant 2B (ενκαταλειφθη) as an agreement with the initial-text reading 2 (εγκατελειφθη), though this TS is not included in the Haupliste. The result is that in my calculations Bezae has 13 agreements with the initial text (including two additional TSS 2 and 57), rather than the 11 listed in the Hauptliste (4, 8, 21, 23, 26, 46, 58, 61, 62, 74, 75).
  2. Also unlike the TUT Acts volume, I am not counting Bezae’s five 1/2 readings (10, 35, 52, 55, 76) or its 1/2 variant (1/2L) as initial-text readings.
  3. Note that as in the Hauptliste, special readings do not include singular readings. For Bezae there are nine non-singular, special readings (‘15.4’, ‘18.4’, ‘25.3’, ‘42.4’, ‘44.4’, ‘49.4’, ‘68.3’, ‘71.3’, ‘72.4’)