Skip to content

Pliny’s Letter about Christians

March 4, 2016

Among ancient sources used by historians to chart the rise of early Christianity, Pliny the Younger’s letter to Emperor Trajan is commonly taken as crucial.[1]  In a newly-published article, however, Enrico Tuccinardi employs some recent method-developments in stylometric analysis of texts and presents the proposal that the letter has interpolations (putatively inserted by Christians, I presume).[2]

As I am among those (many, I suspect) scholars not engaged or expert in stylometric analysis, I have had to devote a good deal of time trying to absorb at least the basics of some recent developments in methods in order to engage (or even to understand) Tuccinardi’s article (and the bibliography to Tuccinardi’s article is a good guide to the relevant literature). That various methods are still being developed and refined suggests that the aim of producing an authorship test sufficiently reliable to be used in law courts, for example, has not yet been achieved.

The fundamental assumption behind all stylometric analysis is that each author has a characteristic stylistic habit, or pattern, or “profile,” and (what is still more essential) that this profile is unconscious to the author, meaning that the author does not (and therefore cannot) change it, even if he/she changes topic or genre. That assumption, it seems, is crucial; but it is really what all the various approaches also seek to verify.  Over the years, scholars have experimented with word-frequencies, sentence-lengths, punctuation, and other features.

As I understand it, Tuccinardi’s particular experiment basically involved the recently-developed approach of tabulating a set number of character-sequences (called “n-grams”) in a given text. These n-grams can be of varying sizes, typically from two to five characters.  Tuccinardi used n-grams of four to six characters.

A text is fed through a programme to produce all n-grams of a chosen character-length, and the most commonly recurring n-grams then form the basis of a hypothetical author-profile. So Tuccinardi produced a “profile” of the most recurring n-grams in book 10 of Pliny’s letters (his correspondence to Emperor Trajan).  Note that this is a composite profile produced from treating all of Pliny’s letters in book 10 as one textual body.

I can’t go into all the other details of his procedure here. But in Tuccinardi’s tabulation the letter about Pliny’s interrogation of Christians (letter 10.96) seems to be something of an outrider in comparison to this composite profile.  There is notable variation among all the samples he studied, but it appears that 10.96 is somewhat more of an outrider than the other samples.  How much of an outrider, however, depends partly on what calculation-model is chosen.  Curiously, Tuccinardi contends that the profile of letter 10.96 suggests the presence of interpolations in the letter, i.e., material inserted that throws off the n-gram data for this letter.  But he doesn’t (or can’t) indicate what the supposed interpolations are.  In email correspondence, nevertheless, he proposed that Pliny’s letter should not be used as a source in historical inquiry as it has been used.

It is, of course, always appropriate to reconsider widely-accepted positions and widely-trusted sources and data, and the historical importance of Pliny’s letter makes it fully right to consider any new light to be cast on it. I leave it to those more expert than I in the highly-specialized stylometric work reflected in his article to assess the specific technicalities of Tuccinardi’s experiment, and I’m sure that he, too, would welcome engagement from such experts.  I will simply lodge a few observations on matters within the limits of my own competence.

First, strictly speaking, contrary to Tuccinardi (pp. 6-7), his data don’t of themselves “suggest” interpolations. His data surely indicate that letter 10.96 has a distinctive n-gram pattern in comparison with his composite book 10 profile.  But the suggestion of interpolations (and that suggestion only) is from Tuccinardi, not the data.  For, assuming the validity of his data, they are compatible with at least four hypotheses:  (1) the letter is a forgery, (2) there are interpolations that corrupt its stylistic character, (3) Pliny’s stylistic profile is varied, and letter 10.96 simply exhibits that, and so/or (4) the method may be inadequate for the task and need some further tuning.

Further, from the detailed stylistic analysis of Pliny’s correspondence by Gamberini, it’s clear that book 10 is markedly different from books 1-9 of Pliny’s letters, and also that within book 10 itself there are noticeable and multiple variations in Latin style, forms, and topics. Moreover, it appears that Pliny edited his letters for the collection in which they are preserved.[3]  In light of this, it appears that Pliny may have been able to vary his writing a good deal, and even adjust his style for the edited collection, and so whatever method we use to authenticate his writings we must take this into account.  Indeed, Pliny was trained in an educational culture that prized the ability to vary your writing style and imitate that of others.  So, Tuccinardi’s analysis may reflect Pliny’s particular gift of stylistic diversity rather than signal inauthenticity or interpolations in letter 10.96.

Second, in considering “external” evidence for the letter, Tuccinardi discusses Tertullian’s reference to it (Apologeticum 2.6-7, composed ca. AD 197), but dismisses Tertullian as insufficiently trustworthy, in light of other works cited by him that are likely legendary in nature.  But, of course, whatever one thinks of Tertullian, the far earlier external witness (not mentioned by Tuccinardi) is the reply to Pliny’s letter by Emperor Trajan (letter 10.97 in Pliny’s correspondence), in which Trajan comments on a report sent to him by Pliny specifically concerning his handling of Christians.  That is, Trajan’s letter confirms that Pliny wrote a letter to the Emperor about his interrogation of Christians in Bithynia, and from Trajan’s reply it seems that the contents were pretty much what we have in the familiar letter 10.96.

Third, from the published stylometric work that I’ve read, I can’t recall any that attempts to detect small interpolations of the sort that Tuccinardi suggests. Instead, the typical aim is simply to identify the author of a given text (and preferably texts of at least several hundred words length).  So, Tuccinardi’s use of the n-gram method to detect small interpolations, e.g., of sentence-size, is novel, and all the more, thus, has to be treated with some caution, at least until there are confirming experiments on other texts.

Further, if there are interpolations in a given text (not an impossibility by any means), it is crucial that they be specified (which Tuccinardi says that he can’t do). This is usually done by identifying anachronisms, or other irregularities that raise our suspicions about this or that bit of a text.[4]  So, what specifics in Pliny’s letter about the Christians raise such suspicions?  So far as I can judge, its contents seem to have secured the confidence of historians immersed in the period.[5]  Many decades ago, to be sure, the claim of interpolations was raised, but for many decades now has been judged fallacious, at least as we are able to judge on the basis of historical information and Latin stylistic data.

It seems to me dubious, or at least premature, therefore, to follow Tuccinardi’s advice and set aside Pliny’s letter, simply on the basis of his particular experiment in the use of a particular stylometric technique. As in all things historical, obviously, the matter remains open for further investigation.  But, for now, it seems to me that we are justified in continuing to cite the letter as a valuable piece of evidence about earliest treatment of Christians by this Roman legate in early second-century Pontus and Bithynia.  The letter exhibits a distinctive n-gram profile, but the reason for this is, so far as I can see, yet to be established.

[1] E.g., Mary Beard, John North and Simon Price, Religions of Rome (2 vols; Cambridge:  Cambridge University Press, 1998), 1: 225-26, 237-38; 2: 277-79.

[2] Enrico Tuccinardi, “An Application of a Profile-Based Method for Authorship Verification:  Investigating the Authenticity of Pliny the Younger’s Letter to Trajan Concerning the Christians,” Digital Scholarship in the Humanities (Advanced Access, 14 February 2016):  doi:  10.1093/llc/fqw00l.  I thank Tuccinardi for alerting me to his article, and I congratulate him of publishing his work in such a venue where it can be engaged by scholars.  I also thank him for patiently responding to my queries relating to his article in several emails over the last couple of weeks.

[3] Federico Gamberini, Stylistic Theory and Practice in the Younger Pliny (Hildesheim/Zurich/New York: Olms – Weidmann, 1983).

[4] The textual witnesses are very few, as noted by Tuccinardi (2), and so it is not possible to approach the matter from a text-critical standpoint in the way that the abundance of early witnesses to NT writings, for example, permits us to do.

[5] E.g., T. D. Barnes, “Legislation against the Christians,” Journal of Roman Studies 58 (1968):  32-50; A. N. Sherwin-White, The Letters of Pliny: A Historical and Social Commentary (Oxford: Clarendon Press, 1966); and the confident use of the letter in Beard, North, Price, Religions of Rome, cited in n. 1.

From → Uncategorized

  1. As an analyst in context of the business environment of a large organization I would like to re-frame the question that underlies the core of the paper: “what is being measured?” I will stipulate that stylistic variations have been detected by the modeling involved.

    These are letters – directed communications to specific audiences. Somehow I suspect that a similar analysis of the totality of my email traffic would detect variations between my communications to upper management (Trajan), to personal friends (Tacitus), and to other various recipients. And probably variations within those communications depending on my emotional investment in the topic.

    Then there is the matter of the media mechanics. There is no way to account for variations introduced in composition via dictation to literate slaves. Pliny was a Roman official at the time of this letter and somehow I doubt that the extant material represents the totality of Pliny’s official written communications (which also calls into question the universe of what is being measured). Think about the modern business context before the advent of word-processing software. How many business executives of any consequence did their own writing? That’s what the secretarial pool was for.

    It seems to me the measurements are capturing artifacts of the cultural environment, and not evidence of interpolation.

  2. Sean permalink

    Wouldn’t the use of an amanuensis affect ones stylometric analysis?

    • Possibly. But if one used the same amanuensis, this wouldn’t account for significant variations among one’s letters.

  3. Tuccinardi further explained his thoughts on a recent post on our discussion forum. It goes like this: “From the findings of my analysis Ep.96.10 should be excluded from Book 10, but I’m not an advocate of this extreme solution. Because I’m fully aware that large insertions in the letter might justify the anomaly (the abnormal length of the letter if compared with all other letters of Book 10 pushes in the same direction).”
    So I am under the impression that his excercise wants be a seminal work open to further developments, based on a new approach.

  4. Rather than the ministrae/ancillae (so the comment above) or some anonymous interpolator, I suspect the “contamination” derives from Livy AUC 39.18 (regarding the suppression of the Bacchanalia). Pliny’s appropriation of Livy’s language here was recognized by R.M. Grant, “Pliny and the Christians” HTR 41 (1948), 273-274, and (apparently independently) by F. Fourrier, “La lettre de Pline à Trajan sur les Chrétiens (X, 97)” RThAM 31 (1964), 161-174.

  5. What happens to the stylometric analysis when an author cites another author? Isn’t Pliny writing /because/ he doesn’t know? Perhaps the two deaconesses, for example, answered questions from their torturers, had their answers written down, and Pliny is copying and/or heavily referring to that document? I don’t know, but maybe this could be a fifth hypothesis.

  6. I assume some sort of multivariate statistical analysis was used here. If that’s the case an interpolation would represent a different mechanism for generating the data than the rest of his sample. Unless the interpolation was minor (and hence probably undetectable) it should be possible to account for that in the model and tease that out and separate it from the “true” sample. If that’s not possible then forgery or intentional variation of style would seem more likely.

  7. While I haven’t read Tuccinardi’s article yet (library proxy is playing up), from the chatter it appears that he is using Potha and Stamatatos’ profile methodology for investigating n-gram stylometry from PAN2013 (

    What would be really interesting is to repeat the analysis utilising the whole of Pliny’s corpus as seperate n-gram matrices and utilise Latent Semantic Analysis for document comparison rather than the profile methodology. Satyam et. al. 2014 ( describe a methodology for using LSA to analyse n-grams, although the step of also using n-grams may not be required for LSA document comparison. This would allow for a much larger corpus comparison, and be able to see whether there are parallels to 10.96 in the rest of Pliny’s corpus.

    LSA and stylometry are techniques that should shed reasonable light on ancient manuscripts, but it is still early days for the techniques as a whole, let alone their application to historical documents. FWIW if any entrepreneurial research students are looking to learn more on LSA then it is all relatively freely available at, especially this chapter of the Handbook that is freely accessible: Also see Landauer et. al. 2007 for the Handbook (
    Overall the field of computational linguistics (one of my previous disciplines) tends to be rather good at being open access, including all the past years of PAN papers and plenty of advanced tools that are freely available.

Comments are closed.

%d bloggers like this: