Considerations for EA Content Localisation

Acknowledgements

I am grateful to Konstantin Pilz, Manuel Allgaier, Luis Costigan, Lorenzo Buonanno and Mike Pool for their insightful notes and additions (though they do not necessarily agree with/​endorse anything in this post). Any errors, of fact or judgment, remain my entirely own.

Introduction

With Open Phil funding people to translate web content into non-English languages, there has been an increase in work in this area. Instead of selecting a text and translating it, I propose taking a top-down approach and using content localisation as part of a communication strategy that starts with a theory of change.

Epistemic Status: I have worked on localisation projects in the last ten years in finance and textiles. There I set up a multi-role process for localising knitting patterns from UK/​US designers to German-speaking markets in the company I co-founded. However, I’m no professional in this area and can only go by my experience and thinking about the subject. Feedback on the first version of this post has updated my thinking that my approach might not be worth the effort in locations with a small EA community or with few English speakers.

Disclaimer: I wrote the first draft of this post before taking on the role of co-director EA Germany. I’m writing this in a personal capacity, although discussions around this text have influenced my thinking in the professional context.

Translation is not Localisation

While translation tries to transpose a text into another language in a way that loses as little meaning as possible, localisation cares about how the content is received. Take this example from “What is effective altruism?”:

Researchers in effective altruism argued as early as 2014 that, given the history of near-misses, there was a good chance that a large pandemic would happen in our lifetimes.

But preparing for the next pandemic was, and remains, hugely underfunded compared to other global issues. For instance, the US invests around $8bn per year preventing pandemics, compared to around $280bn per year spent on counterterrorism over the last decade.3

When localising this text, several questions have to be answered:

  • The link in the first sentence points to an English website. Should it be included in the translated text? If so, should there be a translated summary of the content? If not, should there be a footnote? Or should the text that was linked also be translated?

  • Should the dollar amounts be converted?

  • Is an example using the US appropriate for the target audience, or should a local one replace it?

[Reviewers of the first draft of this post pointed out that the points above might not be crucial but gave other examples, like a text for US/​UK audiences describing the readers as wealthy in global terms, which might not be the case for the target readers.]

The text also includes graphics which have to be translated and updated.

There are questions specific to the target language. One headline is “How can you take action?” In German, you can use either the formal “Sie” form or the informal “du” form. I can recount that either can lead to complaints about the text being too stand-offish or unprofessional.

Another question arises around the familiarity of the reader with certain words. One example is “longtermism”. Konstantin Pilz surveyed the translation of this word into German. However, I would differentiate based on the context. Suppose readers will likely use the translated text as a stepping stone to engage further with the topic in English. In that case, it might be easier to introduce the English term (especially in languages that regularly integrate foreign words). This might be different for readers who will probably not engage in other languages. One grantee commented that they leave the English word in parenthesis, and another that they have translators and proofreaders discuss the word and link to the discussion from a glossary.

Risks and Opportunities

Translations Can Be Harmful

In “Why not to rush to translate effective altruism into other languages”, Benjamin Todd argues that translations can create lock-in effects, that effective altruism is hard to translate, that existing EA material is often out of date, and that mass outreach is not the best way to promote EA in general. Instead, he is in favour of a very localised approach:

Rather than “translating” effective altruism into new languages, it seems better to think in terms of creating a local movement from scratch that’s inspired by the ideas of effective altruism, but is highly adapted to the local context. Imagine that Will MacAskill was Chinese: then what would he have written?
To do this well, we’ll need people who are both experts in the local culture and effective altruism in the West. We’ll also need people who are excellent writer and communicators in the new language.

In the comment section Jonas Vollmer, who led the EA foundation in Switzerland and Germany that translated EA content early on, agrees:

In the early days we made several mistakes that could have been prevented fairly easily. In particular, it seems hard to correct the perception that EA is not just about donating (to GiveWell top charities). It also remains very difficult to counter the impression that EA is mainly the practical implementation of Singer’s views; e.g. Singer’s views on infanticide get quoted in many media articles about EA.

He also comments on content updates:

Quickly translating English content is easy. However, it takes much more time to ensure high quality both in terms of language and framings/​nuance, and it’s even more challenging to keep these translations up to date. See the “fidelity model” blog post referenced above for more discussion of this.

The fidelity model post referenced here argues that mechanisms like mass media outreach might lose the nuance of ideas, leading to a “telephone game” effect.

Keeping these potential pitfalls in mind seems useful when considering who to reach through which medium and what content to use.

However, the commenters also pointed to errors of omission where the harm being done by being too careful might be more significant than not translating the content.

Better Outreach

Eli Rose argues that translations of existing web content will expose more people to longtermist ideas counterfactually. Konstantin Pilz thinks that localised content makes community building easier, will lead to more people discovering EA by chance and will also help with misconceptions:

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.

For German, he argues that there is no good description of longtermism when searching for the word on the web.

A more strategic approach

Looking at the discussion of the topic on the EA Forum, I feel that a more strategic approach might be helpful, especially in countries where the level of English in the target audience is relatively high and when approaching more than basic material like the EA handbook. While it is true that we might reach more people with translated texts about longtermism, we might also cause harm by selecting a specific subset of texts, making mistakes in localising and using the wrong channels to promote the texts. Focussing on the translations more than on the content distribution might also reduce the impact.

Instead, I favour starting with the impact we’re looking to have. Then we can go back to see who we’re looking to reach and through which channels to determine what contents to localise and how to do so.

Localisation Strategy

To choose the target groups, channels and contents for localisation, I propose setting goals and developing a theory of change. This might read like a very time-consuming approach. However, mapping out thoughts for a few hours might help formulate a rough strategy and processes.

Goals

Achieving impact through localisation can mean that people, for example,

  • change their career plans to more impactful options

  • start impactful organisations

  • use their careers to yield influence in impactful ways

  • donate more effectively

  • take a donation pledge

The goals can be used as the endpoint for a theory of change.

Theory of Change

Getting someone to donate effectively might be a relatively short process where someone searches for a keyword, reads an article, fills out a donation form and is finished.

On the other hand, career changes might need multiple contact points over a more extended period. They may involve different channels, including external media sources, websites, newsletters, local groups and career advice. If, for example, we’re doing student outreach at a university in the local language, handing out flyers that promote a website and then an introductory talk at a local group that leads to a fellowship and career research, then we already have several contact points. If it is impossible to localise the whole process, the audience must be proficient enough in English to switch languages. If this is the case, why localise the content? There might be good reasons, like getting the users’ attention more quickly, placing content in external localised media channels, briefing journalists or providing content for local groups that discuss content in their language.

Still, the impact of localising different contents might differ widely, depending on the underlying theory of change and the content’s role.

Channels and Target Groups

The intermediate outcomes in the theory of change should include channels, target groups and contents. Channels could be websites, newsletters or social media channels owned by EA organisations. Still, they could also be external blogs, podcasts, newspapers, local discussion groups, fellowships, or introductory talks. The target groups will differ based on the channel, and together they will drive the level of localisation needed. A text localised for an EA fellowship might be much closer to the original than one that briefs journalists or is on the main page of a general audience website.

Localisation Process

After having developed a strategy, the actual process of content localisation starts. It will differ based on the channels and contents. For example, using the content of “​​What is effective altruism?” on a localised EA website might include:

  • Contacting Effective Ventures Foundation, as the owner stated on the website, and getting permission

  • Getting editable graphic files from EVF

  • Getting information about the target group, style and tone from the editor of the EA website, perhaps using personas

  • Deciding what examples, links and other elements to change

  • Selecting a glossary if one already exists

  • Commissioning a first translation of the content

  • Deciding which unique expressions to translate and how

  • Finalising the content

  • Proof-reading the content for errors and readability

  • Proof-reading the content by a native speaker that is an expert in the content

  • Making changes to the content

  • Changing graphics

  • Final correction

  • Publication

  • Monitoring the original content for changes

  • Re-Starting the process for any changes, based on the size

As commentators on the first draft have noted that this might seem overwhelming, some of these points might not apply or be minor, depending on the project. The goal is not to show how complicated the process is but rather to help not forget any steps.

Roles

The localisation work includes several different roles. One or a few people might hold some or all. However, specialisation may help in scaling the process.

  • Project owner: Overall owner of the process, makes sure that goals are set and tracked

  • Strategic planner: Defines goals, theory of change, channels, target groups and measurement of goals

  • Localisation specialist: Defines per content and channels what elements should be localised and manages glossaries (unless glossaries exist on a per-channel basis already)

  • Project manager (might also be the owner): Manages the overall process and interfaces with content owners and editors of channels

  • Translator: Translates the contents

  • Content expert: An expert for the specific content that can proof-read the localised version and give feedback on the accuracy of the localisation

  • Proof-reader: Checks the contents for mistakes in spelling, grammar as well as readability

  • Graphic designer: Localises graphic elements

  • Target Channel Editor (per channel): A person responsible for the target channel that can define the target audience, style and tone. Checks the content before publication and sets up the update process if the source content is updated. An example would be convincing 80k to add local content to their existing website or a newsletter owned by an organisation in the target country.

Using the different roles can help start the process from thinking about the overall goal through fleshing out the strategy, contacting channels, and working on the contents through publication, goal measurement and revisions.

Conclusion

Taking a strategic approach to localising EA content can increase the chance of translation projects having a positive impact by including considerations of target groups, distribution channels, and the roles needed for the localisation process. This should consider the potential harms and mistakes in previous localisation projects. Depending on the region’s size, the national organisation or community builders might be in an excellent position to coordinate and help with the strategy, staffing and outsourcing.