7

I need to have a large selection of free, publicly available texts in Latin on the internet, hoping to train an neural network (computer algorithm) on it. It would eventually be able to generate random Latin sentences, if I can find enough data.

So, my question is, where, online, can you find lots of Latin texts for free?

1
  • 1
    Welcome to the site! Do these questions (one, two) give what you need? // As a side remark, you may have trouble defining what a sentence should be. // Given how badly Google Translate works with Latin with its large database, I hope you are planning to do more than just throw a corpus at the network. An annotated corpus with some morphological tools would be more work but far more likely to produce sensible Latin.
    – Joonas Ilmavirta
    Commented Jun 9, 2021 at 18:42

2 Answers 2

13

Fortunately, almost all Classical Latin texts are in the public domain, since almost all Classical Latin authors have been dead for about two thousand years. Translations and commentaries are a different matter, but to the best of my knowledge, the exact words of Cicero and Vergil as transcribed from a manuscript are not under copyright anywhere in the world.

The Packard Humanities Institute corpus can be searched online. It's not currently available for download, but previous versions of the corpus were sold to libraries and institutions on CD; if you're affiliated with a university, it's worth checking if they have a copy. It consists of practically all Latin literature from before 200 CE—and "practically all" here means "including everything down to two-word fragments and quotations".

The Latin Library has a wide variety of texts available in HTML format. It's not quite as convenient for corpus studies as PHI's plaintext, so you'll have to do some preprocessing to gather all the pages you want and then strip out line numbers and the like, but on the other hand it's much easier to find a particular text than in PHI's arcane categorization.

The Perseus project similarly makes a wide variety of materials available, which can be read online or downloaded as XML (with convenient annotation of chapters, sections, quotations, etc). They claim everything there is licensed CC-NC-SA, but I question the enforcability of that—my understanding is that changing the format of a work (e.g. from a physical book to HTML) doesn't create a new copyright, and even for the translations, most of their sources are old enough to be in the public domain.

2
  • The Perseus project likely has a decent claim to work beyond merely changing the format—the aforementioned annotations, for instance. That makes a portion of their HTML copyrightable (I believe, I am not a lawyer), and they can license that copyrighted work however they wish and those using it would have to abide by it. I don’t know if stripping the HTML of annotations would help, either, though it might. But frankly the bar for copyright is pretty low—unless it was literally a trivial automated conversion (which it wasn’t), it probably can get copyright.
    – KRyan
    Commented Jun 10, 2021 at 12:55
  • 2
    Transcriptions which aim to be a faithful copy do not qualify for a new copyright. In the US, even a photograph of a painting does not qualify, if it aims to faithfully reproduce the original. Where Perseus and others gain the possibility of a new copyright is in the editorialising that takes place when choosing between competing manuscripts, and interpreting potential errors within them. However, transcribing (say) a Loeb edition from 1920, as a faithful copy, does not qualify for a new copyright. Removing new material / annotations should be sufficient to avoid any copyright claim.
    – Iacobus
    Commented Jun 12, 2021 at 17:43
4

Public domain texts here means “out of copyright and free to use”. While ancients texts are themselves public domain, a modern transcription may not be, as it may amalgamate different sources, make textual choices, correct errors and otherwise apply a degree of editorialisation to the Latin. Thus to be safely public domain, a Latin text must be a literal copy of a printed (and / or scanned) source document that is itself out of copyright, that is published before 1920-odd in the USA. Elsewhere, the editor must have died typically 70 years ago.

Perseus is the main contributer to the Open Greek and Latin project. there are other repositories that it amalgamates though.

This makes pretty clear what their view of the copyright vs public domain status of any given text is. Nevertheless there are other contributors to that project, and it seems Perseus is available as CC-BY-SA (sometimes stated as cc-by-nc-sa, but the repository seems to say cc-by-sa).

As has been said this is almost certainly an overstatement of the real copyright position, where these are copies of public domain editions, as they often are - it is the initial editing that entitled these ancient books to a new copyright, not digitisation.

There are also texts at Latin Wikisource which are clearly understandable as Public Domain, as they are literal transcriptions of public domain books. However, many texts there are not always linked to their original transcriptions.

The Latin Library is a good source of material as mentioned and usually but not always have the source of its texts, and says that they believe the texts presented are public domain. It seems reasonable to assume that if the source is stated and is public domain, then the literal copy is also public domain.

As a third easy source, there are some Latin texts transcribed at Project Gutenburg.

There are also many Bible transcriptions some of which are clear about their copyright status. The Clementine Vulgate Project in particular marks their transcription as public domain; it has also been imported into Latin Wikisource. You can also find copies on GitHub and Bitbucket.

All this said, there is another question as to whether a computer digesting texts, training an algorithm and creating a result is a potentially infringing act. In my personal view, it (morally) should not be, so long as you have the rights to access the text in the first place, as it is analogous to reading and making notes from a book, neither of which are infringing acts, rather than anything like publication; but the fact that copyright exceptions exist in the UK and Europe to “allow” machine reading and processing in certain circumstances suggest that this is not settled outside of the US, where it does seem more likely to fall into fair use.

4
  • Re TheLatinLibrary: thelatinlibrary.com/cred.html
    – cmw
    Commented Jun 13, 2021 at 16:09
  • Thank you CMW, I have corrected that!
    – Iacobus
    Commented Jun 14, 2021 at 7:37
  • @IácóbusdeBremesgrave a computer training an algorithm is not infringing, as otherwise Google would not be able to train machine learning models on the stuff it finds on the web.
    – osa
    Commented Aug 28, 2022 at 20:25
  • @osa In the USA it certainly is not; but depending on the output, the results can be. In the UK and EU it is still a matter of debate and uncertainty; hence exceptions being introduced for Text and Data Minining in the UK for example, which nevertheless only applies to non commercial acts gov.uk/guidance/…
    – Iacobus
    Commented Mar 18, 2023 at 13:37

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.