Translating EA Online Content: Motivation and Learnings from a German Project

By constructive @ 2022-09-11T08:52 (+38)

TL;DR

I think translations are valuable because they improve public perception of EA in the local context, make community building easier, and potentially make people discover EA by chance more often. I've run a project for translating key longtermist content into German for three months about ten hours per week with both translation services and an EA translator. I recommend experimenting with different translators at the start of the project. Proofreading is the most time-consuming work that can only be done EA-internally. I think publishing content well is extremely important but have not done so yet.

If you are interested in running a similar project, please get in touch with me or apply for funding at Open Philanthropy.

Epistemic status

I don't have any formal background in translations. While I've spent the last four months thinking about this topic on the side, it's possible that there is some aspect I've so far overlooked and I encourage others to investigate the topic more thoroughly. I also expect this to differ considerably across languages.

The general motivation for translating EA content

I think there are three potential theories of change for translating content:

1 Shaping the public perception of EA in the local context

By public perception, I mean reports about EA in popular media which in turn influence the general attitude towards EA topics e.g. in academic circles or politics. Having high-quality content in the native language decreases the risk of negative or low-fidelity press coverage and provides a reference when EA topics are discussed.

Currently, upon searching for “Longtermism” on the German web, there is no EA-related resource explaining the concept. Instead, you come across two articles in major German newspapers that portray the idea as far from reality and potentially dangerous. I have heard similar reports from other languages. Having high-quality texts about longtermism available won’t be a guarantee for positive media coverage but I think clarifying misconceptions by translating a good introductory text and longtermism.com’s FAQ will make it much more likely.

2 Having more promising people discover EA content independently

Since many people usually browse the web in their native language, it is unlikely that they will come across EA content in English. Providing content about important EA topics such as biosecurity, nuclear risk or AI safety makes it possible to discover EA through an interest in these topics.

I remain uncertain about how impactful this is because:

1. The most impactful EA work requires strong English skills.^[1] Those individuals who possess these skills are more likely to browse the web in English. I don’t have any data here to see how much Germans (or others) browse the web in their native language vs English. 
2. A brief investigation into typical search volumes in German showed that “doing good with your career” or “most important problems” hardly ever happen. I did not check this with different prompts such as “global pandemic” or “AI risk” which could be important. I expect to get more data on this once the translated texts are published and I can look at website traffic.

3 Making community building easier

Having good EA content at hand makes it a lot easier to reach out to people, introduce them to EA and facilitate discussions.

I think it’s likely that people engage more with content in their native language because:

• People have higher reading speeds in their native language, which means it takes less time and effort to read texts. It also seems likely that comprehension is higher in your native language although I did not find any studies on this.
• The content is perceived as less foreign/ more approachable, especially if examples are adapted to the local context
• People with lesser English skills can participate and engage. I remain uncertain how important this is in the German context as most educated individuals speak English fairly well. However, this looks very different in other countries such as France or Japan.

Overall, I’m most enthusiastic about 1. while I believe that 2. and 3. need more empirical evidence to determine how promising this is.

The specific motivation for my project

Despite currently being the biggest non-English EA community, very few resources exist in German. While effective giving is covered to some extent, no longtermism-related content exists. I believe that longtermism is one of the most important ideas developed by EA and that it deserves much more attention than it is currently receiving. Additionally, current media coverage of longtermism is mostly negative and it seems important to establish a pool of reliable, high-fidelity texts for the broader German public.

I initially planned this project as an experiment to discover how well translations work because I had a bigger project in mind. See the section “Would an international organization to translate content into many languages be high-impact?”

The texts I translated and my reasoning behind it

To provide a good introduction to longtermism I translated the following pieces: 

• An introduction to longtermism from longtermism.com
• The Our World in Data article on longtermism which describes how vast our future could be
• The FAQ from longtermism.org
  • Hints at common misconceptions, quite important for the media
• There already exists an article on x-risk by FLI that is also translated to German (though I prefer the term “existentielle Gefahr”)

I then moved on to introduce the most important longtermist cause areas. For most of them, I found the 80k article to be the most thorough introduction.

• 80k’s article on global priorities research
• 80k’s new article on AI risk 
• 80k’s article on GCBR + additional information on the best GCBR careers in Germany and the EU from German biosecurity EAs that has also been posted here.
• 80k’s article on nuclear risk 
• Climate change, a report by GWWC 
  • Here, I decided against 80k as I think their climate article is more likely to scare people off than to attract them to EA (though I think the content is true). This is mainly because it argues that climate change is less important compared to other longtermist causes and that it’s likely not an x-risk.

As an introduction to why you should think more about your career:

• 80k: most important decision 

I thought about translating some EA fundamentals as well and ended up with:

• CEA’s introduction to EA
• Scope insensitivity 
• Cause X 
• Expected Value 
• EA is a Question (not an ideology) 

My conclusion here was that it’s probably better to write original German texts about EA fundamentals and I encourage anyone with a solid EA philosophy education to do so. I also think it is valuable to have more native-language content in general, as original texts can incorporate the nuances of a culture much better than translations.

I also got some articles from the Most Important Century series translated but have not managed to proofread all of them:

Proofread:

• Most important century: Summary 
• All possible futures are wild 

Not proofread: (Reach out if you have the capacity to do this)

• This Can't Go On
• Forecasting Transformative AI, Part 1: What Kind of AI?
• Why AI Alignment Could Be Hard With Modern Deep Learning
• Forecasting Transformative AI: What's The Burden Of Proof?
• Forecasting Transformative AI: Are We "Trending Toward" Transformative AI?
• Forecasting Transformative AI: The "Biological Anchors" Method In A Nutshell
• AI Timelines: Where The Arguments, And The "Experts," Stand
• How To Make The Best Of The Most Important Century? 

My approach

1. I wrote down my reasoning behind the project and why I thought I would be a good fit. I then reached out to several funding sources and was funded by Open Philanthropy. They are now openly looking for more people interested in translations. If you are unsure if your project is valuable, reach out and we can discuss it.
2. I identified the texts I thought to be most important (see my reasoning above) and got feedback from others in the German community. I then asked the authors of the texts for permission to translate them.
3. I held a small competition to find good German translations of the term “longtermism” and conducted a survey on MTurk to evaluate which term was preferred.
4. I had already looked for potential translators and met an EA who had a background in translations that was interested in working on the project. Since they had limited capacity I decided to also use professional translation services. I contracted several services and evaluated them. Quality differed considerably and I finally used “Ibidem” for most of the texts.
5. After the initial translation, I proofread all the texts (both from the EA translator as well as from the service). I had to spend a significant amount of time improving the texts, mostly to make the style appropriate.

The texts are not yet published and I thus don’t have any experience to share here. However, I think this is a crucial step and should ideally be done by someone with knowledge of SEO and web design as well as a good understanding of the topics. (If this could be you and you speak German, please get in touch with me.)

My learnings

Can you just use AI translations?

I ran a small test where I compared a machine translation with one that was done with the AI + the translator and the latter one was easily recognizable and much better. I thus think having a translator involved reduces the time needed for proofreading significantly and strongly advise using trained translators. However, this might vary from one language to another.

Use of EA’s with a background in translation vs professional translation services

I think texts about scientific topics should ideally be translated (or at least proofread) by someone with a background in the topic. I noticed this most strongly when translating 80k’s article on GCBR that contains many scientific phrases. Since my translator did not have a background in life sciences, I had to change most parts of the text significantly. It might thus be useful to rely on translation services that have specialized in scientific texts.
Texts that talk mainly about EA ideas likewise benefit from translators that are familiar with the ideas. However, I think that professional translators are able to understand and convey these ideas well in most cases.

I think my main takeaway is that it’s worth experimenting a lot at the beginning of the project, i.e. test different services and have translators do a small test translation. This somewhat supports Eli Rose’s view that the difference between EA-adjacent translators and translation services is small.

In either case, I strongly suggest either paying professional EA-adjacent translators or using translation services instead of having volunteers translate material, as this requires much less managing, is more reliable and the final output is of higher quality.

If you would like to pay for professional services, I recommend applying to Open Phil.

Would an international organization translating content into many languages be high-impact?

Earlier this year I proposed the following idea for the FTX Future Fund Ideas Competition:

EA-related texts are often using academic language needed to convey complex concepts. For non-native speakers reading and understanding those texts takes a lot more time than reading about the same topic in their native language would. Furthermore, today many educated people in important positions, especially in non-western countries, do not speak or only poorly speak English. (This is likely part of the reason that EA currently mainly exists in English-speaking countries and almost exclusively consists of people speaking English well.)

To make EA widely known and easy to understand there needs to be a translation service enabling e.g. 80k, important Forum posts or the Precipice to be read in different languages. This would not only make EA easier to understand - and thus spread ideas further - but also likely increase epistemic diversity of the community by making EA more international.

The idea was one of the winning entries, however, on their website they published only parts of the idea, focusing mostly on translations in general and not claiming a centralized organization would be valuable.

I used the German project as an experiment to get information on whether such an organization to translate texts into several languages would be valuable. Here are some takeaways:

• Most of the work required to produce good translations has to be done by native speakers:
  • Finding good translators
  • (Translations themselves - though you can easily use professional services to outsource this)
  • Proof-reading the texts
  • Publishing the texts well
• Some work can be done centrally:
  • Asking the authors for permission to translate the texts
  • Applying for funding
• Work that can be done centrally, though would be preferable to be done from within the local context:
  • Choosing which texts to translate
  • Communicating with others in the local context to get feedback and develop a good strategy
  • Coordinating translators and proof-readers

Since most of the work is best done from within the local context, and a central organization seems to have more of the downsides I list in the next section, I am currently less enthusiastic about a more centralized organization. However, I think some centralized communication would be valuable:

1. Compiling a list of texts that are valuable to be translated (See Open Phil’s recommendations or mine above)
2. Compiling a list of already existing translations (There is one here)
3. Having an overview of ongoing projects so people can cooperate
4. Getting author permissions to translate texts

There’s a #non-english-speaking-groups channel on the EA Groups Slack, where discussions on translation often take place. Laura González has been organizing some coordination efforts there and is working on improving the current permissions system.

Downsides

I think there are some risks with translating content, though most of them can be avoided by acting carefully.

Lock-in & loss of the opportunity to experiment

Approaches

I think Ben Todd’s article on the topic is still worth reading, although being written back in 2018 and focusing mostly on China.

Translating content means you transfer ideas from one cultural and academic context to another. If counterfactually similar ideas would have been developed from inside the country this may lead to a loss of novel perspectives on EA topics.

It also means that you copy the framings and approaches to EA, which has an opportunity cost, as other framings and approaches can not be explored. This is especially important for places where EA is still in a fledgling state.

I thus strongly recommend thinking about experimenting with new presentations of EA before you commit to translating already existing content.

Words & Phrases

The first translation of a key idea may lead to very long-lasting downstream effects. I think this should be done carefully and with a lot of feedback from other EAs in the community. I think it’s worth reviewing relevant literature in the target language and using existing phrases. For the German project, I held a small competition for the best translations of the word “longtermism” and conducted a small survey to get an insight into the German public’s preference.

Laura González shared this glossary of important EA-related words and phrases. I personally think you don’t have to come up with a good translation of every one of these words (and for many, a translation likely already exists), thinking about the top 25 EA words such as “longtermism”, “cause neutrality” or “AI alignment” is probably sufficient.

Being perceived as foreign

If translated content is not adapted to the cultural context well, it can easily be perceived as foreign and is likely to be opposed. I think this can be mitigated by exchanging examples, using more familiar framings, and connecting to ideas from the local context.

However, I think that introductory material is best written from scratch in the target language, especially in non-western contexts, as the audience’s demographics are likely very different. In general, I think it makes sense to favor original content over translated content whenever possible, with the exception of more scientific texts such as 80k’s cause profiles.

Attracting attention

While this is to some extent desired, it can also be a problem. If popular media becomes interested in EA, it is important to have people the media can reach out to. It helps to be well connected to and communicate with other EAs in your language. Ideally, you would develop a plan for what to do once you attract more attention.

1. ^{^}
   1. It's much harder to update your worldview via discussions with other EAs if you're not fluent in English. This seems crucial to find a high-impact career.
   2. Most EA causes rely on academic and scientific texts that require at least decent English skills to comprehend. Almost all the intellectual output of EA is in English and so are the discussions of the topics.
   3. EA organizations operate almost exclusively in English, they are also mostly situated in English-speaking countries. High-impact jobs outside of EA orgs often also rely heavily on English, e.g. campaigning, government jobs, and consulting.
   4. Movement building seems to be one exception, though I'd like to argue that a good connection to the international community & keeping up to date with ongoing discussions seem very important and require at least decent English skills.