Translating Justinian’s Digest with DeepL, a multiple challenge

The clivus Capitolinus, the slope of the Capitol Hill in Rome - image Wikimedia Commons
The clivus Capitolinus, the slope of the Capitol Hill in Rome – image Wikimedia Commons

This post is a tale of the unexpected. Last week I received a message about a new translation of the Digesta Justiniani, the famous sixth-century compilation of texts by the classical Roman lawyers. Soon two things became clear: It was not just a single new translation into one particular language, but into a number of languages, and these translations were not the fruit of just one scholar, but mainly the product of the online translation tool DeepL. Béla Pokol, a Hungarian lawyer at the Eötvös Lorand University in Budapest, used the powerful DeepL tool for translating the Digest into fourteen (!) European languages. For this project he had to process some 45,000 pages. You can download the translations as PDF’s from the section Law Working Papers of the online journal Jogelméleti Szemle. Journal of Legal Theory. A cordial exchange of questions and answers with Béla Pokol followed quickly. Here I will look first of all at the Dutch translation, but whenever necessary other languages will figure here, too. I will try to distinguish carefully between the input of DeepL and Pokol’s own efforts.

DeepL takes the plunge

I suppose normally you would use the translation function of the browser created by one of the Big Tech Firms for occasionally translating some information in languages which are clearly out of your range of linguistic capacities. In these cases you get a rough idea of the core of a text, with for example a number of words even left untranslated. The seduction of DeepL’s offer to bring really good translations from and to a great number of sounds for me simply too good to be true. So far I received just once a translation created by DeepL which I ignored completely by going straight for the English original of an article.

Enter Béla Pokol who at the very least has designed a kind of ultimate test for DeepL, translating from a dead language like Latin into modern languages, and not just a general text from classical literature, but the very core text collection of Roman law. Classical Latin has a complex syntax and a rich idiom. The classical lawyers seem to prefer a less rich vocabulary, but their concise and trenchant arguments set a great challenge for translators. To mention just a few modern examples, the Dutch team led by Job Spruit worked between 1993 and 2011 on their translation in twelve volumes of the Digest, Code, the Justinian Institutes and the Novellae [Corpus Iuris Civilis. Tekst en Vertaling]. The German team led by Okko Behrends started in 1995 and reached with the fifth volume published in 2012 only D. 34. The recent project for a version of the Digest with Latin text and an Italian translation was first published in five volumes between 2005 and 2014; the web version was launched in 2017. In 2011 I wrote here a post on recent and older translations for Roman law. At the page for Roman law of my legal history website Rechtshistorie I mention more translations, a number of them available online.

The idea for using DeepL to tackle the intricacies of the Digest came after Pokol had used DeepL to translate his book Juristocracy from English into seven languages, with surprisingly few errors. He enlisted the help of his daughters to deal with the new challenge of the Digest, because he had noticed only a small number of translations of the Digest into modern European languages. Hence the decision to aim at fourteen languages, starting with Hungarian, followed by French, German, Portuguese, Spanish. Italian, Dutch, Finnish, Rumanian, Czech, Slovakian, Slovenian, Polish and Russian. It took Pokol and his daughters a year to produce the 45,000 pages of these translations.

First impressions

For the aim of this post I decided to look first at the translated results, and only afterwards at DeepL. How should one quickly assess the quality of these translations? As a matter of fact many years ago I selected three passages in the Digest for translation into four languages which figure at my legal history website as Exempla iuris Romanorum. D. is a text on fraud, D. 19.2.59 a text about a building agreement, and D. 32.52.3 a text about hereditary law. I add to them the first case from the Digest I ever encountered, D., a case of deadly damage caused by a collision of two carts on the slope of the Capitol. A fifth check was quickly found, too, using the words plumbum, lead, and balineum, bath, both terms frequently used in connection with water, as can be seen in the Topoi database in Berlin on Roman water law discussed here in 2019. As a quick reference I used the Amanuensis tool of Peter Riedlberger and Günther Rosenbaum.

The Dutch translation created by Pokol using DeepL has 3921 pages. The first translation seems really good (p. 261), especially when DeepL succeeds in keeping everything in a single sentence. Spruit and Wubbe, the Dutch team for D, 4, used two sentences in their translation of D. The second example, D. 19.2.59 figures at pp. 1363-1364. Here I hesitate about the building being destroyed (verwoest), the verb concutio does mean to shake heavily. DeepL puts in the word ongeluk as a partial translation of acciderit. The case about a will speaking about books in D. 32.52.3 fares less well (p. 2164). In the first part the word bibliothecas has been translated as bibliotheken (libraries) but it is clear bookshelves are meant. In the second part the word scrinia does not mean writing tables but chests. Some words have more than one meaning, and it is vital to use the one clearly meant in the context of a case.

The case with the mules and two carts on the Capitol hill (D. killing a young slave is somewhat longer than the fragments here above. At first DeepL impressed me with a clear disposition of this complicated case (p. 681). The mule-drivers (muliones) become only once koetsiers, coachmen. However, translating the term lege Aquilia by “Lex van Aquilia” is decidedly odd. This law and another Roman law, with few exceptions, remains in Dutch untranslated. I cannot plod here through every occurrence of the word lead. In D. 32.35.3 the bath of Iulianus becomes the Julianabad (p. 3241), a very early homage to a former Dutch queen, instead of het badhuis van Julianus or het Juliaanse badhuis. Tibur has been left untranslated, but it is clearly Tivoli, and the word scitis has been promoted to Scitis. In the leges preceding this case DeepL has more luck with some difficult names of locations.

A multiple challenge

On closer inspection there are very real problems. The translation of the references to the works of Roman lawyers is a matter of some wonder. The book title membranae is translated as Perkamenten, parchments. Spruit cum suis opt for Notities (Notes). At some point an author Callistratus is mentioned, a name not mentioned at all within the Digest. The title page of the translation provides a clue to the origin of some of these problems. Pokol has not created a translation from the Latin original, but from the English translation by Samuel Scott, The Civil Law, including the Twelve Tables… (17 vol., Cincinnati 1932). This fact alone severely diminishes the value and possible importance of the translation under review here. It does matter much less which faults can be attributed to DeepL or to Pokol since the very starting point is awkward, and not what one would expect someone to do.

In his article ‘The enigma of Samuel Parsons Scott’, Roman Legal Tradition 10 (2014) 1-37 Timothy Kearley devotes pages 29 to 32 to an assessment of the value of Scott’s translation of the Justinian corpus. Reviewers accorded it mostly only value as a introduction for students and as a quick reference tool. Apart from mistakes in his translations they faulted Scott for ignoring the edition by Mommsen and Krüger and generally being less aware of the latest (German) scholarship. Kearley expands his views in his study Lost in Translations. Roman Law Scholarship and Translation in Early Twentieth-Century America (Durham, NC, 2018).

What is the value of Pokol’s efforts? He wrote to me his explicit aim was helping modern lawyers to have “a speedy online help” for Roman law and to make it a living heritage. Alas as a reference tool the current translation is marred by a lack of running titles indicating on each page the title of the Digest. In some titles, in particular D. 50.17, De regulis iuris antiqui, the numbers of the leges are not given correctly. Book 50 ends with leges numbered above 1000. This has simply escaped his attention. Pokol did not include the introductory constitutions and the Index Florentinus. It shows definitely he aims indeed at lawyers in general, and not at students and scholars who want to study Roman law for its own sake. Let it be clear Pokol did not at all attempt to translate the Latin original of the Digest. In his view the Hungarian translation of the Digest by DeepL is quite good.

Looking deeper into DeepL

It brings us to the final question of this post, the value of DeepL for translating classical Latin into any modern European language. DeepL offers currently 26 languages. Apart from European languages only Chinese and Japanese are now included. Arabic, Hindi and Swahili are absent. The inclusion of Finnish and Hungarian, two Finoegrian languages, is remarkable. Polish and Russian, too, are languages with a number of very real difficulties. Swedish and Danish are present as Scandinavian languages. There is simply no Latin to test here with DeepL at the moment of writing. DeepL does succeed in faithfully translating English into Dutch at a notable level for fairly difficult texts as the ones used here as tests. It might make you certainly curious about the way it would work as an assisting tool, for example when translating a textbook for Roman law into another language.

To give DeepL quickly a second chance to prove it can produce something convincingly adequate when faced with a text offering some difficulties I entered the text of my recent contribution ‘A dictionary for the Spanish colonial empire and canon law’ for a translation from English to Dutch. Using the free version of DeepL without a trial period this meant each time only 5,000 signs could be entered, roughly half of my post. The translation of the first half contained only a few problems, and in the second part with many Spanish words these were correctly left untranslated, and only some easily detectable glitches in the syntax occurred.

For your interest – and perhaps consolation! – I looked around briefly for other online translation tools which do include Latin. I found a few websites featuring both Dutch and Latin. Translatiz depends on Google Translate. In its Latin-Dutch translation of D. 32.52.3 the word legatis becomes luitenants, lieutenants. ImTranslator seemed at first to offer besides Google also Microsoft and PROMT for Latin-Dutch, but the two last do not offer this functionality. A search in Dutch for tools made in the Low Countries brought me to Webtran where you get only a word for word translation for Latin-Dutch filled with silly mistakes, and not even clear sentences at all. Opentran and the Dutch version of I Love Translation join the ranks of tools with insufficient qualities for translating legal Latin. Remembering just in time Cicero’s vehement sigh Quousque tandem abutere nostra patientia, Catilina? I will leave it at that for now, even though these tools did solve this particular question correctly.

It is a feat to climb the Capitol Hill of faithful translation from any language and for any subject! No doubt sooner or later an online tool will appear which will be able using artificial knowledge to produce translations from Latin, not only for regular classical texts with their own peculiarities, but also for legal texts. Classicists are keen in using digital humanities to sensible ends. The challenge remains to learn yourself sufficient Latin and to find reliable translations made in years of toil and endless care for details, and secondly to have all necessary capacities for understanding the way Roman lawyers thought, argued and acted. A good translation is an act of interpretation in itself, a necessary foundation for further research. Meanwhile we can benefit from a substantial number of older and newer translations for Roman law. As for having an impact on modern lawyers and legislators the quality and vitality of thinking and writing about law and legal history by legal historians should seduce people to enter the realm of Roman law in all its manifestations through the centuries until now.

2 thoughts on “Translating Justinian’s Digest with DeepL, a multiple challenge

  1. Andreas Moser

    DeepL has ruined my business as a legal translator. 😒

    It’s not good enough for briefs to be submitted to court, but for internal uses or to let clients know the gist of a document, most people are satisfied with DeepL. And because it’s free and instant, most people don’t have the patience anymore to wait for several days until a human translator will have completed his/her work.

    1. rechtsgeschiedenis Post author

      Dear Andreas, thanks for your comment! DeepL has clearly limits, some subjects are simply too difficult for this tool, but at a more general level such tools can certainly help you.


Leave a Reply

Fill in your details below or click an icon to log in: Logo

You are commenting using your account. Log Out /  Change )

Twitter picture

You are commenting using your Twitter account. Log Out /  Change )

Facebook photo

You are commenting using your Facebook account. Log Out /  Change )

Connecting to %s