Context

As a text deemed heretical and suppressed for some eighteen centuries, Marcion’s Gospel (GMarc) does not exist in any known manuscripts. Scholars however concur that GMarc is a version (whether earlier or later) of the canonical Gospel of Luke. Its text is attested over 700 times by more than fifteen ancient Christian writers. Thus it is not lost to history nor necessarily relegated to obscurity. Reconstructions of GMarc involve the painstaking and judicious use of a wide variety of evidence: patristic polemical quotations, paraphrases, and allusions; variants in the manuscripts of Luke; and close parallels in the canonical Gospels of Mark, Matthew, Luke, and John, as well as the non-canonical Gospels of Thomas and Peter.

As a contributor to Thilo’s Codex apocryphus Novi Testamenti (Hahn, 1832), August Hahn became the first scholar to produce a fully continuous Greek text of GMarc based on his earlier study (Hahn, 1823) and compilation of attestations. In 1892, in his Geschichte des neutestamentlichen Kanons (Zahn, 1888–1892, 2.2: 455–529), Theodor Zahn produced an often discontinuous Greek text, pared down to correct and rival that of Hahn while making use of Tischendorf’s Editio octava critica maior (1869, hereafter EOCM) of Luke as a base text. While other 19^th century scholars such as Ritschl (1846) and Hilgenfeld (1850) undertook thorough analyses of GMarc and included many quotations, they did not offset a clear, running main text.

Thus Harnack’s critical study and reconstruction of GMarc in 1921, followed by a second edition in 1924, became history’s third major reconstruction and the scholarly standard for the better part of a century, translated into English several decades later (Harnack, 1990). Adopting Zahn’s discontinuous approach to restoration and many of his indications, Harnack nevertheless pares down Zahn’s text and attempts to correct many of his editorial decisions. More recently, Tsutsui’s (1992) and Roth’s (2015) reconstructions of GMarc both follow this discontinuous approach. The former is unique in rendering the main text in Latin instead of Greek, while the latter valuably compiles attestations, comparative citations, and critical notes in well-organized chapters and subsections. Reconstructions by Klinghardt (2015/2020; 2021) and Nicolotti in Gianotto and Nicolotti (2019) return back to Hahn’s maximalist, continuous approach, while providing for the first time in history a robust, consistent critical apparatus immediately after and/or below the restored text. The English reconstruction by BeDuhn (2013) finds a middle ground between a continuous and discontinuous approach. Differing approaches notwithstanding, all recent reconstructions of GMarc make use of newer critical editions of patristic sources (e.g., Tertullian, Epiphanius, Adamantius Dialogue) yet still frequently cite Harnack’s text and follow his decisions. As Roth (2015: 2n2) stated, “Harnack is still invaluable for Marcion studies, and some truth remains in Helmut Koester’s [1982] statement, ‘All further research is based on Harnack’s work.’”

While recent scholarship on GMarc has shifted from theology-based to text-based approaches, stylometric and/or statistical studies on GMarc have been infrequent and lacking in scientific method and rigor. Sanday’s (1876) stylometric claims about GMarc (made apart from serious engagement with its actual text) have held sway for nearly 150 years. Knox (1942) and Tyson (2006) both critiqued Sanday’s work and undertook their own statistical and stylometric analyses, but neither proved scientifically rigorous nor generally compelling (Cadbury, 1943; Filson, 1944; Spencer, 2008; Johnson, 2008; Roth, 2008). A recent chapter by Daniel A. Smith (2019) compiles verse count statistics based on Roth’s edition. My iterative First Gospel LODLIB (Bilby, 2021a) challenges the unscientific status quo in GMarc studies and reconstructions by integrating Computational Linguistics (CL), Historical Corpus Linguistics (HCL), and Open Data Science methods more generally.

Now in the public domain, Harnack’s text of GMarc is an ideal starting point for CL and HCL. Recent works may require copyright permission that is not always granted. Most original language texts in the Tufts Perseus project¹ are public domain, including its New Testament (Westcott and Hort, 1885). The PROIEL Treebank (Haug and Jøhndal, 2008),² a standard resource for HCL, uses Tischendorf’s EOCM. Stripping out parenthetical comments, marginalia, apparatus, and footnotes from the main text, these datasets are not mere print analogues or substitutions. Their data are transformative supplements, normalized to be read, analyzed, and connected to other data by humans and machines.

While specialists may view public domain texts as outdated and largely irrelevant for current scholarship, from a statistical and stylometric point of view the differences between older and newer editions are typically minimal. GMarc is an oddity in this regard, with unusually wide variations between reconstructions, stemming from the different a priori assumptions and methodologies of modern editors. Even so, clear relationships exist between texts. As we show in our First Gospel LODLIB, in total word count and numerous other metrics, the reconstructions of GMarc by Harnack and Roth are closely correlated, while those of Hahn, Klinghardt, and Nicolotti have many notable similarities among them.

Challenges and Resolutions

While the purpose of open Humanities datasets is not confined to the replication of a print reading experience, several challenges still present themselves to anyone attempting to build normalized datasets based on Harnack’s reconstruction of GMarc. The Greek text in the main body—found within the appendix whose pages are labeled with asterisks—is discontinuous and frequently interrupted by brief observations, summaries, and other kinds of descriptive indications in German, both inside and outside of parentheses. The term “unattested” / unbezeugt occurs frequently, indicating Lukan content not corroborated by patristic citations to GMarc. Sometimes unattested content is noted implicitly, as in the comments on GMarc 4.33–35 (184–85*), where 4.33 is skipped and “only” / nur some specific words in 4.34–35 are noted as attested for the passage. At other times Harnack clearly indicates words or verses as “missing” / fehlte, “erased” / getilgt, or “stricken” / gestrichen. Elsewhere Harnack states that “nothing is known” / ist nichts bekannt, as in 4.36–39 (185*). Whether absent or unattested or unknown, all such material is simply omitted from the datasets without any corresponding indication.

The term “allusion” / Anspielung also occurs frequently in Harnack’s main text, indicating attested passages and verses deemed too unclear to restore some or all specific wording. Footnote references also appear regularly in the main text, usually but not always abbreviated “s. u.” (siehe unten), sometimes paired with references to allusions and usually in places where Harnack refrained from restoring clear wording. The corresponding footnotes are an inconsistent hodgepodge: smatterings of quotations and translations of attestations to GMarc, brief lists of some related manuscript variants in Luke, and sometimes additional secondary analysis. For example, in the running text at 7.22 (197*), Harnack notes “s. u.” and in the corresponding footnote on the previous page (196*) quotes a German translation of Eznik of Kolb’s Armenian attestation to this verse, given without accompanying analysis or evaluation of how it might be used in a reconstruction. We render unrestored allusions and footnote references as empty parentheses (). We use this same indication when individual verse numbers appear without any subsequent content, as at 5.29 (189*), and also when individual verse numbers are skipped within passages generally attested as present, particularly when intervening content is necessitated by context and Zahn indicated the verse as implicitly present, as at 7.25 (197*).

Ellipses in the main text are used ambiguously, usually to note content implied or necessitated by the context of surrounding attested content, but sometimes to convey alignments between GMarc and Luke. When κτλ or “usw.” / “etc.” appears on its own or in front of ellipses, as at 18.20 (226*), it clearly indicates the subsequent alignment of GMarc and Luke. But the three ellipses across 18.10 and 14 (225*) likely indicate intervening unclear content, in keeping with Harnack’s description of 18.9–14 as merely “allusions” / Anspielungen. For all ellipses we make educated judgments to resolve ambiguities, rendering empty parentheses where we think Harnack conveyed unrestorable allusions, but rendering EOCM wording [inside square brackets] where we think Harnack conveyed alignment with Luke. Occasionally Harnack uses — (em-dash) to indicate alignments with Luke, an indication Zahn had previously used. An em-dash appears, for example, between 8.8 and 8.16, likely indicating (as Zahn previously did) that GMarc 8.9–15 aligns with Luke.

Parentheses are Harnack’s most ambiguous indication, with several possible meanings. 1) Clearly restored wording, often adjoined to brief summaries, as at 4.16 (185–86*), “Das Auftreten in Nazareth (ἐλθὼν δὲ εἰς Ναζαρέθ ὅπου ἦν κατὰ τὸ εἰωθὸς ἐν τῇ ἡμέρᾳ τῶν σαββάτων εἰς τὴν συναγωγὴν).” 2) More likely redings than the first readings given, as at 9.34, “ἐκ τοῦ οὐρανοῦ (ἐκ τῆς νεϕέλης wahrscheinlicher).” 3) Less confident readings, as at 8.25 (199*): “τίς (ἄρα).” 4) Apparent readings that follow from clearly attested wording, as at 16.17 (220*), “εὐκοπώτερον (δέ ἐστιν).” 5) Variant readings in general, as at 16.16 (220*), “ἐξ (ἀπ᾽) οὗ ἡ βασιλεία,” or introduced by “or” / oder, as at 9.8, “εἷς τίς τῶν ἀρχαίων προϕητῶν (oder προϕήτης τῶν ἀρχαίων).” 6) Highly doubtful variant readings, concluded with an internal question mark, as at 4.31 (183*): “(ἀπὸ τοῦ οὐρανοῦ?).” 7) Somewhat doubtful readings, with a standalone question mark or “uncertain” / unsicher note following a word, as at 10.4 (205*): “μήτε ῥάβδον (?).” This typology does not include a variety of other brief descriptions and summations of content. We resolve these ambiguities as follows: 1) by rendering clearly restored wording without parentheses; 2) by following the more likely reading, putting it in parentheses and replacing the less likely reading with empty square brackets; 3) by keeping less confident readings in parentheses; 4) by keeping implicit readings in parentheses; 5) by replacing variants with empty square brackets; 6) by rendering seriously doubted readings as empty parentheses; and 7) by wrapping a word in parentheses when a standalone question mark or “uncertain” note follows it.

Both inside and outside of parentheses and/or quotation marks in the main text, Harnack also sometimes uses German or Latin to indicate the presence of Greek words or expressions ambiguously. For example, 5.33 (189*) has “(Christi Jünger)” instead of clarifying whether this meant the Lukan phrase “but those who are yours” / οἱ δὲ σοὶ, the Markan (2.18) “but those who are your disciples” / οἱ δὲ σοὶ μαθηταὶ or the Matthean (9.14) “but your disciples” / οἱ δὲ μαθηταί σου. In this instance, following from Harnack’s view that GMarc was a later, abridged version of Luke, we default to the corresponding EOCM wording of Luke, placed in square brackets. As another example, the Latin “coetus” inside of quotation marks is the only word given for 5.17 (189*), corresponding to Tertullian’s paraphrase “amidst a throng” / in coetu (Marc. 4.10.1). Luke 5.17 describes “Pharisees and lawyers” but lacks a specific term corresponding to Tertullian’s “throng”, hence Harnack’s vagueness. In this case we treat the term as an allusion, rendered as empty parentheses.

To complicate matters further, Harnack also used angled brackets to indicate variants, as at 5.14 (189*): “<τοῦτο>.” He also used square brackets inside of parentheses to indicate variants within doubted readings, as at 4.31 (184*), “(πόλιν τῆς Γαλιλαίας [Ἰουδαίας]?).” We omit variant wording but indicate these omissions as empty brackets.

In conclusion, for all content that is not clearly restored in the main text, we use two simple indication symbols to cover four different batches of material: 1) (apparent and lower confidence restorations within parentheses); 2) empty parentheses () for footnotes, allusions, seriously doubted content, and most individual verse numbers skipped and all bare verse numbers lacking subsequent wording; 3) [alignments with EOCM Luke in square brackets]; and 4) empty square brackets [] for variants both inside and outside of parentheses. This allows for explicitly restored (batch 1) and implicitly restored [batch 3] words to be counted and analyzed in certain scenarios, while avoiding the problems inherent in any attempt to apply statistical analysis to the unclear content in batches 2 and 4. In this way we distill Harnack’s noisy text of GMarc into clear, scientifically useful data. These normalized indications and data types are also applied to our other forthcoming GMarc datasets in the interest of consistency and meaningful comparison.

Quality and Version Control

To enrich the Greek text and provide for deeper CL analysis, we also manually created a second dataset with lemmatization and morphological tagging for all words. The BibleWorks Greek Morphology (BGM) schema (Bilby, 2021b) was an ideal choice for this work, given its open license for non-commercial use, its familiarity to many scholars, its lightweight schema that is easy to create and to query in word processors and advanced CL environments, and the previous collaboration of scholars to apply BGM tagging to the canonical Gospels, which often have close or exact parallels to the text of GMarc. For ambiguous options (e.g., conjunctive vs. adverbial καί), BGM also allows for multiple tags separated by a forward slash.

Practicing rapid, agile, and iterative development, we offer these UTF-8 encoded txt file datasets as a starting point for CL research on GMarc. For quality control, we ran extensive cross-checks between our Harnack and Zahn datasets, performed segmented word counts and made them openly accessible in our First Gospel LODLIB, confirming a total of 4338 words in both Harnack datasets. We welcome feedback from scholars and will gladly release corrected versions in the future. We also welcome collaborations to convert these datasets to schemata such as MorphGNT (Tauber, 2017) and especially enriched TEI XML with variants and notes placed in the markup so as not to interrupt the visualized flow of the main text.

Dataset Creators

Mark G. Bilby (California State University, Fullerton) manually created both datasets.

Languages: Postclassical Greek. English

License: CC-BY-NC-ND

Repository name: Journal of Open Humanities Data Dataverse

Publication date: 2021-09-10