1 Introduction¹

It is a curiosity of kaige research that despite years of hard labour the identification of kaige revision in the Septuagint still lacks a rigorous methodology. Scholars cannot confidently distinguish between those features that actually reflect kaige revision and those that are misleading traits. Whereas Barthélemy initially compiled a modest list of criteria that could be used to identify kaige revision, the number of suggested features has advanced out of control. This article provides objective data for identifying criteria of kaige revision and thus reduces dependence on subjective judgements. This is achieved through calculating with statistical precision the accuracy of each previously suggested criterion. Furthermore, this method is applied to discern several new criteria.

Kaige is a textual layer in which the Old Greek Septuagint was revised to accurately match the Proto-Masoretic Text (Proto-MT) at a lexeme level.² It appears to have originated in the first century BCE and its method evolved into the stronger isomorphic technique adopted by Aquila.³ The revision is witnessed among the Dead Sea Scrolls (8ḤevXIIgr) and parts of the received Septuagint such as 2 Kingdoms 10–3 Kingdoms 2:12 (βγ); 3 Kingdoms 22–4 Kingdoms (γδ).⁴ It moves away from the occasionally free translations in the Old Greek texts towards a stronger isomorphic approach that matches the Hebrew word order and standardises translation equivalents.⁵ The most basic isomorphic feature is the translation of καί γε to represent the Hebrew ‮וְגַם‬‎ or ‮גַּם‬‎ ‘(and) indeed.’ Throughout the relevant revised texts, this equivalent is so consistent that it supplies the name for the whole revision itself: kaige.⁶

Attempts to identify kaige revision began with several suggestions in Dominique Barthélemy’s study on the Greek Minor Prophets Scroll from Naḥal Ḥever (8ḤevXIIgr), though hints of this distinctive revision were already present in Thackeray.⁷ Beyond the occurrence of καί γε, Barthélemy identified eight further revision characteristics that he associated with the kaige phenomenon.⁸ For instance, the collective sense of ‮אִישׁ‬‎ is lost in kaige revision. The translation of ‮אִישׁ‬‎ varies in the Old Greek Septuagint; it is often translated with ἀνήρ, or ἄνθρωπος, but when it is used in its distributive sense, it is appropriately translated ἕκαστος. Kaige revision, however, consistently standardises this translation to ἀνήρ (e.g., Jonah 3:8, Mic 4:4). Advancing this research, Greenspoon gathered ninety-six potential kaige translation equivalents from all the relevant studies up to his day.⁹

The point of departure for this present study is Tim McLay’s article in Textus 19.¹⁰ McLay investigated these ninety-six suggested marks of kaige revision listed by Greenspoon. From this list, McLay enumerated the opinions of several scholars on the validity of each potential kaige feature. McLay’s tabular analysis underlined the mixed opinions on what constitutes a true kaige feature. Some are generally accepted, while others are rejected. McLay added a couple of markings to enhance the data and caution against certain opinions.

A quarter of a century since McLay’s study, research into the field has slowed down. Most focus has instead been on the style of kaige revision and has resulted in the recognition that the revisers were not mere automata, but showed some signs of sensitivity to the Greek text.¹¹ Nonetheless, the extraction of translation equivalent is still seen as an important task of kaige research.¹² Sixteen new suggested features have been catalogued by Tekoniemi and Tucker, thus bringing the number of potential features up to 112.¹³ Overall, however, the suggested criteria marking kaige revision have generally remained unchanged in decades.¹⁴ This is unexpected, given McLay’s observation that many of these criteria are not accurate, or should be excluded.

Therefore, the purpose of this article is to analyse statistically the accuracy and usefulness of all the features associated with kaige revision. I assess those suggested features linked to kaige revision in previous scholarship, and add several of my own. McLay concludes his article by writing: “Only a more refined methodology will provide the means to determine which texts, if any, may be so closely related as to justify the conclusion that they originate from the same author or the same school of translators.”¹⁵

This article presents just this refined methodology that is based on my own statistical analysis of the entire Septuagint and Greek Minor Prophets Scroll (8ḤevXIIgr). The results are important and provide a firm foundation for future research.

2 A New Statistical Methodology

I turn now to supplant McLay’s study with a new methodology that can be used to confidently identify kaige revision. My method calculates the statistical accuracy of each suggested feature and indicates whether or not it predicts a kaige text. These results are displayed in Table 2. Features unique, or most common, to kaige texts will score the highest. If, however, a feature is inconsistently found in kaige texts, or is common in Old Greek texts then it will score lower. This accuracy is recorded in a new column alongside the number of instances where this criterion can be tested in the Bible. The accuracy figure is a weighted average of positive and negative matches: positive matches being where the criterion is found in a text, while negatives are mismatches where the criterion is not met. For a kaige feature, one would expect a high number of matches in kaige texts, and low number of mismatches in nonkaige texts, and vice versa. Two further columns show the likelihood that a feature is found in kaige (KR) and OG texts.

In order to obtain the accuracy of each criterion, I turned to the digital humanities and adopted a corpus statistical approach in a computer-assisted study. It required a complete analysis of every word in the entire Greek and Hebrew Bibles. To produce this analysis, I used the CATSS parallel database¹⁶ and my own alignment of the extant Greek Minor Prophets Scroll with BHS.¹⁷ Then I enhanced the CATSS data in places where its matching was suboptimal. This sometimes happened where parallel matches were at the phrase level, not individual words; that is, many-to-many mappings. Therefore, for each parallel, I identified the key word for consideration, which involved ignoring articles, pronouns, and occasionally other parts of speech. Where a suggested criterion involves sequential words, I wrote several programs to identify matches.¹⁸ I have not investigated criteria 7 and 95 since the manual identification required for these is too vast for now, although I hope to include these results in the future. Given that CATSS depends upon the text of Rahlfs’s edition, this will raise concerns with some readers. However, the large corpus means the differences between Rahlfs and the Göttingen Septuagint are negligible. On the rare occasions that an accurate criterion retains a negative result, I sometimes found an explanation in the Göttingen text or its apparatus.¹⁹

To derive an accuracy figure, I first marked known kaige or kaige-related text (Ruth; Kingdoms βγ, γδ; Ecclesiastes; Canticles; 8ḤevXIIgr; Theodotion-Job) against other nonkaige texts. I have not marked Theodotion-Daniel as kaige when making statistical calculations. This is in light of differences between this text and kaige texts,²⁰ but future research should include Theodotion readings in a separate column. The text from Theodotion-Job is far smaller, but included due to several shared features, some of which will be commented on below. Ecclesiastes displays a later development of kaige revision, but is included given the shared features. The Judges text from Codex Vaticanus (Judges B) could also be included as a kaige text, for it generally reflects a kaige revision compared to the text found in Codex Alexandrinus (Judges A). However, due to contamination in the textual tradition, I have left it out as a certain kaige text.²¹ Then for each criterion I count the number of positive matches (i.e., where a Hebrew word is translated by a Greek word or words) in kaige texts and negative matches (i.e., where the Hebrew word is not translated as expected for kaige) in nonkaige texts. This count is compared against the mismatches to produce a figure. Table 1 shows the results for the first criterion: ‮וְגַם/גַּם‬‎ translated by καί γε.

This means that in kaige texts, ‮וְגַם/גַּם‬‎ (criterion 1) was translated as καί γε 136 times, but only 15 times in Old Greek (OG) texts. Then ‮וְגַם/גַּם‬‎ was not translated by καί γε 510 times in Old Greek (OG) texts, but also 23 times in kaige texts it was translated surprisingly in another manner. Thus, of the total 685 occurrences of the Hebrew ‮וְגַם/גַּם‬‎, 94 percent followed the expected results. That is, it correctly predicts a kaige text 646 times (136 + 510), and fails only 38 times (15 + 23).

View Expanded

The final new columns (KR and OG) refer to the likelihood that a feature is found in a kaige or Old Greek passage respectively. These figures complement the percentage supplied in the overall ‘accuracy’ column through supplying a different angle on the data. These columns are particularly useful for features that tend to occur in kaige revision, even if they are also found in OG texts. In this example, the translation of ‮וְגַם/גַּם‬‎ with καί γε occurs 86 percent of the time in kaige texts (136/(136 + 23) = 86 %). In the OG, however, it is only found 3 percent of the time (15/(15 + 510) = 3 %).²² The value of this data is demonstrable when looking at criterion 3. Here it indicates ἀνήρ translates ‮אִישׁ‬‎ 89 percent of the time in kaige texts and also in 50 percent of OG texts. Kaige texts generally standardise other translations of ‮אִישׁ‬‎ such as ἄνθρωπος and ἕκαστος, but sometimes retains ἄνθρωπος, especially in the set phrase ἄνθρωπος τοῦ θεοῦ.²³ The fact ἀνήρ is so common in the OG means its appearance cannot alone be considered a revision. It is, however, 80 percent more likely to be found in kaige revision than in OG texts. Therefore, this criterion could be used as evidence of revision with sensitivity to further evidence.

Table 2 retains McLay’s annotations on the opinions of other scholars regarding the validity of the criteria. Where McLay evaluated a study in a different way to what an author concluded, he marked that author’s name with #. Where studies were based on fewer than three examples, so less significant, McLay marked an asterisk indicating hesitation to include the opinion. The studies he includes, with their author’s abbreviated name in parentheses, are as follows:

• (S) James Donald Shenkel, Chronology and Recensional Development in the Greek Text of Kings (Cambridge, MA: Harvard University Press, 1968)

• (O) Kevin G. O’Connell, The Theodotionic Revision of the Book of Exodus: A Contribution to the Study of the Early History of the Transmission of the Old Testament in Greek (Cambridge, MA: Harvard University Press, 1972)

• (B) Walter Ray Bodine, The Greek Text of Judges: Recensional Developments (Chico, CA: Scholars Press, 1980)

• (Gr) Leonard J. Greenspoon, Textual Studies in the Book of Joshua (Chico, CA: Scholars Press, 1983)

• (Gn) Peter John Gentry, The Asterisked Materials in the Greek Job (Atlanta: Scholars Press, 1995)

• (M) Tim McLay, The OG and Th Versions of Daniel (Atlanta: Scholars Press, 1996)

For ease of reference, I have marked the best criteria in dark grey. These score higher than 70 percent accuracy (column 4) and are found in kaige texts with a significantly higher percentage than in Old Greek texts.²⁴ Other promising criteria are marked in light grey. I retain the numbering of Greenspoon’s criteria, but reorder their position chronologically in the new table. Therefore, the criteria are grouped by published date and not the order Greenspoon chose to introduce them into his study on Joshua.²⁵ I have also pointed the Hebrew words and enumerated the Greek or Hebrew words under specific roots when necessary.

This analysis confirms that the suggested criteria are generally helpful in identifying kaige texts. Forty-two criteria score higher than 70 percent overall accuracy, and occur with a significantly higher proportion in kaige texts, highlighted here in dark grey. They should be consulted when discussing kaige revision. A further twenty-eight criteria score higher than 70 percent accuracy, but are not established significantly in kaige texts to provide certainty, and are highlighted here in light grey. However, some criteria are unhelpful. Rare criteria could present a dangerous sense of overconfidence.

I am especially hesitant about O’Connell’s criteria, since many of these are rare words found only in accounts of the tabernacle construction from Theodotion readings in Exodus.³⁴ For example, criteria 39, 48, 51, 52, 57, 59, 60, 65, and 70 are only found in Theodotion (e.g., Exod 28:4, 30; Exod 38:28).³⁵ Many of these are coloured in light grey reflecting their omission outside of Theodotion-Exodus; see, for example, criterion 48. Likewise, criterion 52 appears to be the same standardising tendency detectable in the Minor Prophets scrolls, and not a specific mark of kaige revision.³⁶ These Theodotion suggestions are distinct from kaige and do not occur frequently enough to be useful kaige identifiers.

Some of these criteria fail since they flatten a subtlety behind kaige revision.³⁷ Criterion 61, for example, states the Hebrew root ‮עבד‬‎ is translated in kaige texts with forms from the Greek root δουλ-.³⁸ The data, however, suggests this is true only 64 percent of the time. The translation of ‮עֶבֶד‬‎ with δοῦλος or παῖς is complex and explained in various ways.³⁹ Kim argued that kaige revision consistently differentiates according to social status, with δοῦλος denoting a servant or slave, while παῖς is used for persons of higher rank.⁴⁰ The varied translation of ‮עבד‬‎ to fit the context shows kaige revision was more subtle than sometimes realised and undertaken with attention to small details.

Criterion 94 deserves special comment. It shows a low accuracy because transliteration is found throughout most Old Greek texts; that is, it is not unique to kaige revision. Transliterations, however, are more common in kaige texts. While kaige texts only constitute 10 percent of the texts I analysed, they contain 17 percent of all transliterations. This means transliteration is almost twice as common in kaige texts as in Old Greek texts (0.22 % to 0.12 % across all words). Therefore, this criterion may be used with care. For example, when variants differ between a Greek translation or Hebrew transliteration, this criterion should be used. In other cases, mere transliteration without variant readings does not necessarily suggest kaige revision.

3 The Method Applied to New Features

This new statistical methodology does not need to stop after reviewing past criteria. Hence, I adopted and extended it to identify potential new features that were previously unidentified (Table 3). Through searching all the translation equivalents found in the kaige portions of Kingdoms I identified five hundred potential criteria. Among these suggestions I present the twelve that were most promising and several further possibilities that merit comment.⁴¹ The first features I now highlight are those that they appear to be the most accurate and occur with a sufficiently high frequency that they cannot be overlooked.

4 This Method and the Distribution of Kaige across the Septuagint

Beside identifying kaige revision, this method also makes it possible to plot the occurrences of diverse features across the entire Septuagint.⁵⁶ Thus it provides a visual representation of the distribution of kaige revision in the Bible. Therefore, we now turn to four pertinent charts based on different criteria for marking kaige revision. The first chart shows by book the absolute number of instances where ‮וְגַם‬‎ or ‮גַּם‬‎ is translated by καὶ γε (Figure 1). The second chart includes only Barthélemy’s nine suggested features and plots the average number of their occurrences per chapter (Figure 2). The third chart displays the distribution of all 112 suggested features across the Septuagint (Figure 3). The fourth chart is the most important, for it retains only the best criteria for identifying kaige revision. That is, it filters out the weakest criteria as calculated by my statistical method. Further, unlike the third chart, it plots the ratio of positive to negative matches. This provides more accurate results than simply the average number of positive results per chapter. A negative result where a kaige reading is not found is often just as important as a positive result.

All 7,436 occurrences of the 124 possible kaige features across the Septuagint can now be conveniently plotted for analysis. Each of these 124 tagged criteria could be selected or hidden (toggled), in order to produce any combination of different charts. The data is supplied from a collation of every positive or negative occurrence of the kaige features across the Septuagint. A negative occurrence is where a Hebrew form is not translated with the expected kaige equivalent. For example, if ‮וְגַם‬‎ was translated by καὶ γε, then this would be a positive result, but if the translation was only καί or γε, then it would be negative. Red bars indicate books associated with kaige revision, while blue bars signify books where the Old Greek remains.⁵⁷ Despite not including texts such as Judges B or Theodotion-Daniel in my statistical calculations as a precaution, for these charts I have coloured them in red to highlight their similarities.

View Full Size

Figure 1

‮גַּם‬‎ translated by καὶ γε

Citation: Textus 32, 2 (2023) ; 10.1163/2589255X-bja10038

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Figure 2

Barthélemy’s kaige features per (extant) chapter

Citation: Textus 32, 2 (2023) ; 10.1163/2589255X-bja10038

• Download Figure
• Download figure as PowerPoint slide

The similarities between charts 1–3 demonstrate that, despite kaige criteria increasing tenfold, there has been little advance in our understanding or identification of kaige features. The reasons for this are: first, scholars have looked at obscure features in Theodotion translations, such as in Exodus; second, kaige is a more complex phenomenon than realised. The texts as we have them are not as homogeneous as might be expected. It is possible the methodology is looser than expected.⁵⁸ Thus, criteria to establish kaige texts have been insufficiently accurate.

View Full Size

Figure 3

112 kaige features per chapter

Citation: Textus 32, 2 (2023) ; 10.1163/2589255X-bja10038

• Download Figure
• Download figure as PowerPoint slide

View Full Size

Figure 4

Ratio of best refined kaige features to nonkaige features

Citation: Textus 32, 2 (2023) ; 10.1163/2589255X-bja10038

• Download Figure
• Download figure as PowerPoint slide

Figure 4 is the most accurate chart and deserves the most attention. It shows the improved results when only the best criteria are used. These are the criteria marked dark grey in Table 2. It displays the ratio of positive kaige matches to negative mismatches. Hence, Zechariah shows if there were ten potential kaige features in the text, the Minor Prophets scroll would likely contain nine of them, only missing one expected feature. That not all features are found consistently reflects the diversity in kaige features.

In this chart, the books containing kaige revision consistently score higher than the other texts. The unique outlier is the single chapter of Obadiah, which I have left uncoloured given the small sample size for this ratio. Semi-kaige revision is also detectable in Kingdoms. The improved identification of the Greek Minor Prophets Scroll, and lower ratio in nonkaige Kingdoms is especially clear. Excluding the Minor Prophets scroll, kaige texts are found in the Megilloth, and Former Prophets, but not in the Pentateuch or Later Prophets. Kaige features are detectable throughout the Septuagint, which suggests some contamination in transmission history or the borrowing of kaige features from these books.

I marked several books in purple since they appear to be contaminated with kaige readings. The text of Judges from Codex Vaticanus (B) shows more kaige features than in Codex Alexandrinus (A), as correctly discerned by Barthélemy. Judges A, however, still contains some marks of kaige revision. Ezra-Nehemiah⁵⁹ shows numerous kaige features, which could group the book in the kaige tradition. Otherwise, the usual suspects appear Kaige-Kingdoms, Ecclesiastes, Lamentations, Theodotion-Daniel, and the Greek Minor Prophets Scroll.

It is unclear why kaige revision occurs in some books but not others. However, trends appear when books are ordered according to the Masoretic not Christian tradition. Figure 4 highlights that Canticles, Ruth, Lamentations, and Ecclesiastes all show similar kaige influence.⁶⁰ It is uncertain why they are similar. Either they were transmitted together on a revised kaige scroll, or they were the last books to be translated, so translated in the style of kaige revision.⁶¹ Unfortunately, this puzzling question is beyond the scope of this article.

5 Conclusion

This article has filled the need for a rigorous methodology to identify kaige revision. This is seen in an enhanced list of the best features to identify kaige revision. It has demonstrated that certain criteria for identifying kaige revision are statistically more significant than others. Several criteria should be discarded, while others should only be used with care. The results are necessary for textual criticism and future kaige research. Furthermore, the new statistical method advanced in the paper was then applied to identify a list of new kaige features. This method should be applied to all future suggestions of kaige revision, in order to confidently establish true marks of kaige revision. Finally, the adoption of this methodology leads to a firmer grasp of kaige revision and the ability to confidently detect its influence in the textual traditions.