I was recently reading an English translation of a non-theory-related non-fiction book that was published just a few years ago. Next on my list was re-reading Engels’ “Socialism: Utopian and Scientific”.
I can only describe my experience as the most severe whiplash; comparable only to reading the comments on a news thread on .world after checking the Lemmygrad thread.
I found myself reading so slowly. In 30 minutes I would have barely read a few pages, even though I was not struggling with the content itself.
I found that it’s just that the language is so needlessly impenetrable. So many run-on sentences, so many odd turns of phrases, archaic terms (not relevant to Marxist theory), and bizarre sentence structures.
I have never read the original German text nor any other language translation, so I don’t have a point of reference. I don’t know how much of it is Engels’ writing style, versus the translator’s 19th century English.
A translation written in 1892 was not written for English speakers in 2026.
My question is: Would these works not become so much more accessible if we had a modern English translation from the original text?
If this would be considered useful, how could we realistically realise this for all of the fundamental works from that time period?
-
I imagine that finding and hiring a translator who is familiar with Marxism to re-do the translations would be extremely difficult and expensive.
-
Community translations are probably not feasible as translation requires a very specific skill set, not just fluent speakers.
-
Machine translations might be possible, but it’s crucial to not lose any specific terminology and not misconstrue the meanings of certain quotes and phrases. The translator (whether human or machine) needs to have a strong grasp on Marxist theory and history to not distort the meanings in the text.
I would really appreciate everyone’s thoughts and suggestions on this. Maybe I completely dropped the ball on this topic.
P.S.
This is not just in regards to “Socialism: Utopian and Scientific”, it’s just the experience that inspired me to write this post.
I’m not unfamiliar with reading theory from that time period nor Marx and Engels’ writings in general. I’m also not completely unfamiliar with 19th century English, I’ve read plenty of fiction written in that time period as well.
Agreed on all points. I also definitely think we should leverage the fact that these works are digitised and available on the internet; we aren’t bound by the limitations of publishing in the 19th and 20th centuries.
We also have the benefit of later works that have built upon the work of those 19th century books and essays. I wouldn’t doubt that well-read MLs today would have a better understanding of the relevance of certain phrases and terminologies than the 19th century translator, and that they would be better equipped to relate and connect it to other relevant theory.
I’d be interested to see what @yogthos@lemmygrad.ml is working on if he chooses to publicise it.
Basically, I ended up going down a rabbit hole of seeing whether you could train a LoRA to do believable writing in a style of a given author. I spent a few weeks on it, but looks like it’s mostly working now. I ended up training it on Lovecraft cause I found his entire corpus conveniently compiled here. The trickiest part was ensuring that the meaning was preserved during the style transfer. I found that with smaller models like Qwen 2.5 with 8 billion params you didn’t quite get the right results. It wasn’t very consistent about either the style or the content. I hit a sweet spot with a 32 billion parameter model though. Ended up renting out some GPU time on runpod, and it cost me around 20 bucks total to train it. The results are great though.
For example, I had it style my last blog post just now. You can see the styled version here, and the original draft here:
What makes Greenland important is that is has valuable resources, while also providng a foothold in the Arctic where the melting ice creates new trade routes that Russia and China are well positioned to control.
What makes Greenland attractive: it wouldn’t be a long engagement, it’s just about the size of an operation that the US military is capable of carrying out successfully. It’s a big political stunt that will dominate the news. It signals that the US shift in policy away from international law. It demonstrates that the US no longer has enough geopolitical pull to dominate the UN Security Council, and is now reduced to take unilateral action as a rogue state.
The whole idea of Western unity was underpinned by an unspoken understanding that the vassals of the US would get to partake in the plunder of the world. Any atrocity could be excused under the guise of bringing freedom and democracy to the world, but it was never supposed to happen to the supposed allies of America. But the times have changed. As the world shakes off western parasitism, the share of the pie is shrinking, and the ring leader decides it now needs a payment in flesh from those who chose to subordinate themselves.
At its heart, Greenland is a disciplining action for Europe. The US already has a base there, it is a NATO protectorate, and NATO itself is an institution created by the US to subordinate Europe. There is no functional reason why the US would need to annex Greenland. The goal is to humiliate the Europeans, force them to publicly bow down and accept that the US can take whatever it wants from them with impunity, and all they can do is say thank you daddy. Everybody knows that when Trump goes to officially annex Greenland, it’s purely a symbolic move.
And of course, you might be saying to yourself, surely the Europeans won’t stand for it. Surely this will be the time that they stand up for their interests! But the reality is that they have no choice but to accept the humiliation because Europeans now have a powerful and angry Russia on their border, and they’ve burned all the bridges that could be burnt. Yet, they know full well that they’re not capable of fighting Russia on their own. They lack energy, industrial capacity, and even the ability to produce basic things like explosives. The armies of Europe are small and inexperienced, while Russian army is now well seasoned in its fourth year of fighting. Russia was forced to figure out its logistics out of sheer necessity, and its economy is now reorganized to support the war effort. Russia also enjoys the benefit of having abundant resources, large industries it has inherited from USSR, and vast energy reserves. Anybody looking at the situation objectively can see that Europe would not be able to fight Russia in its current state.
Of course the other option would be to accept that Russia won, and to try to find a path towards peaceful coexistence. Yet, the entire legitimacy of the current European leadership rests on having Russia as the enemy which requires the US to stay in Europe since Europeans know they can’t fight Russia on their own. And hence we see a scenario where Europeans are forced to accept any and every debasement to keep Americans in play.
Meanwhile, the US doesn’t really see Europe as strategic as during the Cold War when USSR posed an ideological opposition to the US model on the world stage. Today, China has risen to become the new ideological opposition, and Europe is just so much dead weight to be stripped it for parts and discarded. Now that the unipolar moment is over, the US see domination of its hemisphere as the primary focus for securing resources in order to contest China going forward.
And so, Europe is destined to suffer humiliation in its desperate attempt to keep its role as the junior partner in the western alliance. This stems primarily from two illusions. First, it hinges on the hope that “the US will be better if a different party takes power”; second, it holds a wishful thinking that “Europe will not become the next target of bullying.” This mentality reveals a harsh reality: Europe feels powerless to cope with various changes alone, and therefore regards its relationship with the US as a higher-priority “strategic asset” that must be carefully maintained. Consequently, some attempt to exchange compromises of principles for the preservation of their so-called core interests in the face of hegemonic behavior. This is a typical appeasement mentality, fantasizing about pacifying powerful forces through concessions.
Respect for and protection of the sovereignty, security, and territorial integrity of all states is a widely accepted norm of international relations forged through the blood and fire of World War II. It is also the foundation and soul of the UN-centered international system and of an international order based on international law. Historical lessons have long written a painful verdict on appeasement.
Eventually, the current liberal order in Europe will lose its legitimacy as a result of living standard will collapsing due to austerity to drive military spending on one side, and American predation on the other. The governments of all the major European countries are already hanging by a thread with their approvals hovering close to single digits. It is only a matter of time until nationalist parties such as RN, AfD, and Reform start sweeping elections. At that point NATO and EU will fall apart, with nationalists in better off countries pulling up their ladders. Europe will become a fractured continent again with different countries going their own ways, and historical grievances resurfacing.
Now I can really just jot my thoughts down and structure the narrative I want, then throw it at the model and get well written text out of it. I find this is really liberating for me, because previously I’d always get stuck on language and phrasing to get it to read the way I want. Now, I can just focus on the content and what I’m trying to say.
No offense, i’m sure you put a lot of effort into training your model, but your draft here is far more readable and logical than the article itself.
I struggled to make sense of parts of the actual article, even after reading the draft here, some of the sentences ended up being written in a very confusing way, some of the choice of vocabulary is unnecessarily pretentious, and a couple of parts were just grammatically wrong.
Content wise excellent but stylistically i really did not vibe with this one.
That’s fair, it is trained on somewhat archaic way of writing. Now that I have the process down, I can easily experiment with some other authors. It does illustrate the point though, you can take the content, and rewrite it in a new style that’s completely different from the original writing. And it will be consistent across, so for something like a wiki, this approach is great because you can have everything look consistent even when you have many authors editing the content.
Lovecraft was impenetrable even in his time, but the principle works. For wiki work or places where it’s important to have a cohesive style across entries this could be great, I’d love to try it on prolewiki pages because style varies a lot depending on who wrote the page.
I think I’m mostly happy with the way it works now. I just need to clean, and package it up here, and I’ll ping you once it’s ready to play around with.
This is quite interesting, but do you think it would be applicable to translation work? So if we use your example, the original version of the text would be the “draft” and it would be able to modernise the language? (not sure what it would be trained on though, maybe modern marxist texts?)
I would assume it’s possible to reign it in so it doesn’t attempt to modify the content itself too much.
Definitely, in fact there’s already translation happening internally. What I found out was that if you just gave original writing to the model, then it would often retain a lot of stylistic elements. What you have to do is neutralize the input first to strip out the style and simply keep the propositions. The first thing I tried was to get an LLM to try and rewrite the text in a neutral way, but that was spotty and sometimes it would drop some propositions. But then I had this idea to translate the text to Mandarin at HSK 5 level. It has completely different grammar, and the HSK5 forces it to use simple words and sentence. So that becomes the basis for the styler where it can now take what’s being said and write it using its own words.
And it doesn’t really matter what subject the writer whose style you’re adopting wrote about. What the model learns is the speech patterns, the adjectives, and so on, and then it’s smart enough to apply them in the context. So, you definitely could train it on some modern writing, and then feed old text to it to produce modern prose.
My only concern is that it could over correct and overwrite theoretically-relevant terms (proletariat, bourgeoisie, contradiction, labour power, work, etc.), which is why I believed it needed to be trained specifically on marxist texts to understand the relevance of these terms.
I’m not sure how much of this would be addressed through training or simply just prompting. In any case there would need to be multiple runs of manual review.
As with any LLM tooling, you do need a human to check over the work after, because it can’t guarantee correctness on its own. But it can save you a lot of time up front because now you just have to read through, and make sure things make sense instead of actually figuring out how to rewrite the text. So, manual review is key.
Training the model on Marxist texts will help too, especially more modern ones that use the right terminology without the archaic language. Basically the way the training works is that you create pairs of paragraphs, one with input text and one with the target output. You create thousands of these pairs, and then the model gets scored on how closely it managed to produce the desired output from the input during training, and learns the patterns through reinforcement. So, if you mix some Marxist text in that will help it see the terms and know they need to be preserved.