Crowdsourced translation has been with us for almost a decade, and has become a viable option for many companies including AirBnb, Twitter, WhatsApp, and Hootsuite. Many fast-growing companies look to crowdsourcing as a fast, low-cost way to launch in many locales simultaneously.
But is crowdsourcing the best way to accomplish translation? The answer depends on the nature of the crowd, and the nature of the project.
Crowds and Communities
Crowdsourcing generally boils down to two models: anonymous crowds and known communities.
Known communities are formed in a variety of ways. They may be recruited and qualified, just like a collection of contractors, or they may be culled from among a client’s large, enthusiastic user base. Communities may be paid by the task, they may receive non-cash compensation such as site credits, “soft” benefits such as an uptick in community reputation, or simply a personal sense of contributing to a larger effort.
For sites with a large, enthusiastic fan base, community projects are a great way to tap into the expertise of in-country users to launch localized products quickly across many target markets. Meanwhile, community projects often boost user interest and engagement.
Companies that don’t have large, enthusiastic user base to draw from may be tempted to crowdsource translation anonymously. Anonymous crowds are profoundly scalable and instantly available: hundreds of thousands of workers are on the bench through crowd engines like Amazon Mechanical Turk and eLance. The outsourcer divides work into discrete, repeatable tasks for the crowd to perform, and then the outsourcer reassembles their output as appropriate to the overall program. Crowd workers are typically paid by the task.
The Perils of Anonymous Crowds
As a solution architect, I have had lots of conversations with clients about the value of using anonymous crowds for translation, and generally I steer them toward known communities due to the following risks of anonymous crowds.
- Unverified credentials. Members of the crowd may not have the qualifications they say they do. There is no time to vet and qualify each and every crowd member to ensure they are representing themselves correctly. You may not get people with deep specialization or product knowledge, and if your job requires that — good luck. Also, buyers typically can’t review resumes and hand pick a resource; the crowd self-selects the jobs they want to do.
- Little or no quality assurance. You can’t do any quality checking on their work. You get what you get. When some task goes wrong you don’t fix it, you start over. Rework by that same resource is generally not possible.
- No opportunity for performance management. You can’t train the resources. You don’t own them, they are not paid for any time in training, and they simply come to you with the skills they already have. Also, there is no feedback loop and resources can’t improve over time.
- Inconsistency. For linguistic work, there is inconsistent use of terms, style. You cannot force compliance with style guides and terminology databases.
- Gaming the system. People can actually cheat, take shortcuts, or even outsource their own work to someone else. I’ve heard of work automations — a crowd worker takes on a task and then has a bot complete it for them. Unfortunately, these abuses may occur.
Types of Projects
Crowdsourcing can be great if you want an extremely simple task of relatively low importance to be done cheaply and quickly. Translating social media comments to generate business insight is a great example of such a task. But any public-facing translation is probably better done by a community of known users whom you can count on to abide by set processes.
For high quality linguistic work, you need process and quality controls, project management, resource qualification and specialization, and a way to manage resources and improve their work over time. If you can manage that via crowdsourcing, that’s wonderful. Get it done right the first time, and you’ll save yourself the potential embarrassment, delays, and costly rework of quality issues.
Have you ever tried crowdsourcing linguistic work? How did the experience stack up against your expectations?