What does Unbabel define as suspicious behaviour?
We’re proud to count thousands of honest, hardworking editors in our community. The vast majority of these editors are unlikely to ever commit any kind of infraction or break our Terms of Service, but unfortunately there are people out there who are determined to game the Unbabel system for illicit gain. These people not only deliver translation quality that is far below what we expect from our editors, but also take work from our hardworking editors. That’s why we are unequivocally committed to finding these people, revoking their access and, where appropriate, seizing any funds gained through illicit means.
There are many different kinds of suspicious behaviour, but they can broadly be placed into three main categories. These are 1) using a third-party translation service, such as Google Translate, and claiming it as your own work, 2) working incredibly slowly in order to game the system in order to artificially inflate earnings, and 3) working incredibly quickly to the point that the text is not being read properly and/or is being completed with zero edits.
Using a third-party translation site and claiming the work as your own
In these cases, suspicious behaviour is defined as the partial or complete use of translated text that has been completed by an automatic translation service and the input of this content into the Unbabel platform. This is generally a particularly problematic form of fraud and perpetrators can expect swift disciplinary action.
Even in cases where the editor might still be planning to work on the text, this still counts as a severe contravention of our Terms of Service; once you copy text from our platform, it violates our data rules and jeopardises the security of customer data. This isn’t up to us - we have a duty to protect customer texts and the second it’s out of our platform, we violate this. In a nutshell, no matter how much you think it might help your process, just don’t do it. It’s not worth it.
Very slow editing
Our system automatically detects tasks and reviews that have a very high cost per word (CPW). It’s normal that CPW should vary from task to task, just as the difficulty of the translation will vary depending on the content. However, we know when someone is trying to game the system and it will not be tolerated.
This is a particularly sensitive topic for reviews, where reviewers have a flat rate and are trusted to complete work in an appropriate amount of time. Clicking repeatedly on the review to avoid idle mode, over-editing (e.g. deleting and re-inserting the same word) or otherwise artificially extending the review time are all considered fraudulent behaviour and will be subject to disciplinary procedures.
Very fast editing and/or completing with zero edits
Some editors, particularly those in language pairs with lots of tasks, sometimes think that the best thing to do is to submit every task as quickly as possible without editing it properly, if at all. This is particularly damaging because it means that a lot of customers end up with a lot of bad translations, jeopardising not only Unbabel’s relationships with their clients, but thousands of hardworking editors.
Often, we find that editors are not necessarily completing tasks with zero edits, but are completing their jobs with a lack of care that is not befitting to their role. We deal with these instances on a case-by-case basis, but bear in mind that this is still considered a serious violation of our expectations and will be dealt with under the same process.
Whilst these are the three main behaviours we encounter, there are others that we consider to be suspicious. These include, but are not limited to:
- Using any unauthorised third party programs, including translation apps, bots, crawlers and any automation programs that interact with Unbabel in any way, for any purpose;
- Avoiding, bypassing, removing, deactivating, impairing, descrambling or otherwise circumventing any technological measure implemented by Unbabel or any third party to protect or control access to Unbabel or any part thereof;
- Creating more than one account for yourself, creating another account if we have already disabled your account, attempting to access the Services through unauthorised third-party applications, soliciting login credentials from other users, or buying, selling, renting, or leasing access to an account or a username.
The discovery of any of these behaviours will be dealt with on a case-by-case basis.
How we detect suspicious behaviour and what we do about it
We have a number of tools to detect suspicious behaviour, including (but not limited to) automatic alerts, observation of editors’ actions on the platform, and data analysis. If you’ve received a message talking about suspicious behaviour, it’s because we have clear evidence of a contravention of the rules and denying these actions will not result in a different outcome.
If we find evidence of any of the behaviours listed above, we will (depending on the severity) either send a warning or demote the editor immediately and indefinitely. There are no second warnings; if you have received a warning, please alter your behaviour immediately. If the behaviour is found again, you will be demoted permanently.
Please note that demotions may not be limited to content type or language pair. If you’re found to be gaming the platform for reviews, in some cases we may just remove your review access, but depending on the case it’s also possible that we can revoke your access to tasks altogether. In particularly serious cases, we may decide that even though the infraction(s) occurred when the editor was working on one language pair, the severity of those actions means that the editor should be demoted from all their language pairs.