Community Translation In The Real World
One thing our customers have recently been curious about, yet hesitant about trying, is to tap into their user community to translate their content.
Having been involved with such efforts, I can sympathize. It’s a daunting undertaking. Will it work? How do I make sure my community stays active? How do I deal with quality issues, or even detect them?
I also wanted to expand on my presentation at Localization World. Some interesting points were raised by the audience, and I wanted to share some of the issues we discussed and give those who were not present a chance to participate.
Why Use Community Translation?
While community translation programs are quite difficult to start, they are also incredibly gratifying: the ability to connect with your early users in a country is key if you want to win a market. As this TED video explains, you have to take care of your first followers if you want to start a movement :-).
How to Translate with Your Community
The three most common misperceptions about community translation that I hear are the following:
Most people have heard the horror stories: some site having to retranslate their whole user interface after a poor attempt at community translation. While some of these stories are true, what they do not mention is that most sites that had some early setbacks still use community translation to a certain extent, and some quite successfully.
An often forgotten point is that while there are certain pitfalls to avoid when starting down the community translation path, some of the same ones apply to working with vendors as well. There are stories of poor translations in the language service provider world too. I actually found that most of these stories share a common problem: throwing content over the fence to be translated with little collaboration doesn’t work. Translating anything requires a dialog with the initial creators of the content to ensure its meaning is properly interpreted.
Finally, an essential element of quality in a community translation program is the quality of the moderation program itself. We will get to that in a bit.
Another common misperception of the community translation solution is that it prevents any kind of deadline to be set. While some of it is true, it is certainly possible to work with a community of users and have them keep a good enough pace to keep a relatively large website translated. Some early Twitter languages took a while to be translated, but others like Traditional Chinese took just a week.
The critical aspect to speed in community translation is that it is a byproduct of a good moderation program. Steering users to the content with the most priority is important when dealing with their limited time. Keeping the most important content to translate at the top of their list ensures that it gets translated fast.
Most people believe that community translation is a cheaper alternative to hiring professional translators. Careful observation of the costs involved, in my experience, shows that not to be the case. Between the moderation program, research, and software development, most successful community localization programs incur costs that are even or larger than contracting with a language service provider. Salaries, servers, etc. can result in big monthly fees you have to pay even when you do not have a new language or new features to launch. Do not choose this model if you’re trying to cut costs!
An important thing to keep in mind is that monetary incentives are pretty much off the table. Orit Yehezkel, Head of Localization at Waze, says:
“If you give the community money, it’s not a community anymore. They become mercenaries.”
A better way to spend that money is to put it to work in community building efforts: organizing conferences, meetups for your community, creating programs to recognize them and providing other types of incentives like access to beta features. While these can represent a large investment, the money doesn’t go directly in the communities’ pocket. It is used to create goodwill, and comes with an investment in some of your time, which will be much more noticed and appreciated.
What to Translate?
There are so many different types of content that are produced and translated that attempting to list all of them would be fruitless. I’ve categorized them into seven different types below, but keep in mind that because your content is not listed here doesn’t mean it won’t work.
The most important aspects are to be aware of what kind of category your content fits in and to think about what might drive someone to translate that content or to become disinterested in it.
1) User Interface
This is typically the first thing attempted and historically the most successful. This is what drives most of your community to translate: they want to see the product in their own language.
There are variations on how you would approach this translation process depending on what type of software you’re translating (mobile app vs. website, responsive HTML5 vs. static content…) but those variations apply to how you would handle this content with a traditional language service provider so I will leave them to another blog post.
Short fun stories will be translated very quickly and well, even by volunteers. Some other content may require enlisting the help of a local marketing expert.
I’ve attempted to crowdsource the translation of a slogan at Transifex into French with some good success. The discussion that came out of it was very useful and thought-provoking.
3) Static Help, FAQs
Short, simple text can work, but it will require a large community and dedication that is closer to the type of efforts we see for some Open Source projects. Twitter, with up to a million translators, had to decide to translate this with a language service provider. However, forums and social media can also help build a knowledge base. Some companies have experimented with users supporting each other. There are great tools out there to do that already!
4) Support notes
Support notes are time sensitive and need to be right the first time. Translations need to be fast and accurate right away, and their content is often very dry. This makes it a tough sell for a community localization program.
The idea with this content is to have the community help create the terminology, but leave the interpretation/execution to professional translators.
One challenge with this idea is how fluid community translations can be sometimes. Having the right tools to lock down terminology and making sure that it is finalized before sending it to the professional translators can be a challenge.
Today, some tools can do an automatic glossary extraction. With a community translation process applied to the output of these tools, coupled with a modern TMS to perform the actual translation of the content it might be possible to try it out and get a good feedback from a limited community of users.
This might be a good sweet-spot for an organization who wants to try out community translation with a small investment.
6) Style Guides
There are tons of standard ones. An option to involve again the community in this process could be to use a wiki to have the community participate in the definition and creation of how the product talks to them?
7) Terms of Service
This is probably best handled by a lawyer. Most of the time, the work required is not just to translate, but to adapt to fit the customs and rules of the local country. This makes this type of content a poor choice for community translation.
Beyond those different types of content, there are also surprising differences in which languages happen to be successful at doing community translation. While it’s difficult to point to any particular trend, it’s worth keeping in mind a few things:
Some smaller languages are dying to help you translate, so don’t overlook anyone. For instance, you may have a very loyal following in a lesser-known community: a small community of users in Catalonia who email your CEO for an official translation effort so much that he had to crack and ask their engineering team to do it :-).
One nice side effect of community translation, because it is a heavy investment, is that it needs to scale across many languages to be cost-effective. But being able to access a lot of these smaller pockets of users quickly can be an effective strategy for some!
In some languages, most early adopters are “perfectly fine with English” and thus will have a limited interest in helping translate your website in English. If this market is important for you, you might have to invest in a professional translation. The risk here is that the translation may not be necessary.
An important feature of a successful community localization program is the quality of its moderation team. While moderation is good in moderation, there is an art to doing it well.
You might think that a core competency for moderators would be to be able to tell the difference between a good translation and a bad translation. For instance, our initial tests showed that the best moderators were not those who provided the best translations, but those who voted up the best translations.
Another important job for the moderators of these communities is to actively find “broken windows” in the community to repair. These jobs can vary from putting out a flame war between translators on which form of the word “you” should be used in your product, to organizing a “translatathon” where translators get to meet on IRC, or a brief chat to discuss various glossary terms that have been misused.
Finally, as a moderator you have to act as a control and monitoring tool for the health of each community, making sure it has what it needs to be active and productive. This can mean being innovative in creating content, to keep people coming back:
When it comes to translations with the community, with enough volume you can think of the community as a forest fire. It spreads outward translating everything, leaving inactivity behind in the middle. You want the strings and translations that are active to be the most useful ones, or the ones with the highest priority. As the activity increases, there’s more and more trolls on the limited strings available for translations. This makes for a poor user experience. I learned a few choice words in Greek that way…
There are three approaches to moderation:
1) Automated Moderation
In this model, the system is self-policing. Each user is both responsible for contributing moderation and content. This typically uses a reputation system based on votes. In general, 3-5 votes are enough to promote a translation to reviewed.
- Pros: This model can scale to lots of volunteers.
- Cons: With this large of a community, they will potentially feel like you are out of touch. Implementing this takes time.
There are a few ways to automate the moderation, running the gamut through what I call a “laissez-faire” approach to a fully gamified experience.
Laissez-faire: This uses votes as a signal to decide moderation. The underlying assumption is that the community, in this case, relies on intrinsic rewards; Getting the site translated is an end in itself.
Gamified: This allows you to extend community efforts beyond translations, but actually editing content or personalizing the experience for users in that country.
2) Controlled Moderation
In a controlled moderation environment, users provide the translations, but a paid team of linguists/reviewers/community managers adapts and corrects the translations for consistency. We recently experimented with this approach at Transifex and translated our site into FIGS with the help of our community and e2f.
- Pros: Human-sized, easy to implement.
- Cons: Cost to scale, hard to find experts in some languages.
Quality control can be done using different processes:
a) On signup, using an invite-only system: Users can only be invited to translate, using some kind of a referral program, or after having been vetted by a human. This also provides a good way for you to ensure your translators have access to a contributor license agreement if your legal team requires one.
b) After the fact: Users provide the translations, but a paid team of linguists/reviewers/community managers adapts and corrects the translations for consistency.
3) Crowdsourced Moderation
In this model, the responsibility of choosing moderators falls on the community itself.
- Pros: Promotes a fully collaborative environment, scales to tons of languages easily.
- Cons: Delivery dates can be very fluid, can stir conflict and flame wars.
Used frequently in the Open Source world, this model requires you to steer your community towards a collaborative model rather than a competitive one. The gamification techniques mentioned earlier may thus need to be adjusted.
Another interesting area to explore once the translation effort has settled is to expand the localization effort to other types of work. Now you’re asking your community to not just add translations, but to actually act as editors or curators for some of your content. Waze comes as a very successful example in this area, with users contributing mapping data, gas prices, hazards, with very specific location area. Twitter has a similar effort on-going currently with their country-specific suggested user lists.