Hi there, I have a few questions about the post-editing workflow and how its word count is calculated in Crowdin Enterprise.
When I use AI pre-translation and create post-editing tasks for our linguists using New Task screen, I need to select ‘Proofreading (by own proofreaders)’ in the Type. The post-editing tasks are sent to the linguists as Proofreading tasks. May I ask why there is no dedicated type for Post-editing when using AI or MT pre-translation?
In the Translation cost report, the Post-editing is listed in the Translation section. This is confusing. Currently, I’m taking the Post-editing word count from the Proofreading section because it aligns with the reality of how the post-editing tasks are sent to the linguists. However, the way you list the post-editing in the Translation section in the Translation cost report gives me the impression that I should not be doing it.
Similarly, one of our vendors extracts the word count of post-editing from the Proofreading section in the cost report and applies the human proofreading rate by mistake. (They should be applying the post-editing rate.) Because the name of step/type doesn’t match with the actual post-editing tasks they worked on.
I did Machine Translation Post-editing using other TMS before. In many situations, the word count of MTPE was differentiated from the word count processed in Human-Proofreading (a.k.a. Review) and Human-Translation. I think it was called Edit or Post-Edit step. That way;
it gave the teams more flexibility to adjust the workflow by language, by content type, according to their business needs and the quality of machine-output that time. Currently, with Crowdin, MTPE can only replace the Proofreading step, and its word count can be a mixture of proofreading of human translation (if the other tasks in the same project were sent to human translators) and post-editing of MT/AI output, where a different rate could be applied.
it was possible for the teams to use Proofreading/Review step for their own purpose, after MTPE, if needed. MTPE step wasn’t mixed up with the true Proofreading.
Q1. Is the creation of the dedicated Post-editing task type something your team planning? or not planned at all?
Q2. Maybe, there is actually a way to differentiate the word count between proofread and post-edited in Crowdin Enterprise but I’m not just aware of it. If so, could you please let me know how I can differentiate them?
Thank you for your detailed feedback on AI pre-translation, post-editing tasks, and reporting.
Crowdin’s task types focus on the action performed (like “Proofreading” for reviewing content), rather than the source of the pre-translation (AI or MT). This is why post-editing tasks using AI pre-translation fall under “Proofreading” in Crowdin.
You’re correct that AI/MT pre-translated content is initially counted under “Translation,” while the human post-editing effort is under “Proofreading.” This aligns with our statuses that are used for translation exporting, API, web-hooks, etc (translated/approved):
While there isn’t a separate “Post-editing” task type currently, here are the best workarounds:
Use very clear task names or descriptions, e.g., “Post-editing: [Content Type],” to instruct linguists and clarify the task type for vendors.
Ensure your vendor agreements explicitly define MTPE tasks and their rates, even if labeled as “Proofreading” in Crowdin.
In Enterprise, reports or API data might help differentiate these tasks.
While don’t plan to change this logic, at least for now, I will certainly pass this important feedback to our product development team for future consideration.
Thanks for your reply, Dima. My name is not Alex, though
Unfortunately, this doesn’t align with the professional localization workflow for post-editing. I didn’t feel that Crowdin takes the user experience seriously, as they prioritize the tooling logic over the real use cases. It’s a pity that Crowdin can’t produce the edited word count that is critical for the post-editing workflow, thus available in so many other TMS in the market. I’m submitting a feature request here.
We’ve already done the first and second points. However, since they can only see ‘Proofreading’ in the task type in the tool/report, they keep making the same mistakes. I wouldn’t blame them since it is clear to me that the tool is not helping them perform their work correctly.
That said, could you give me more information about the third point? How can we try to get the word count post-edited by using API or the existing reports? Thanks once again!
We have good support for post editing. But from what I understand, the UX is not ideal for you at the moment.
There is a categorization checkbox; we will rename it soon and enable it by default, which should improve the user experience.
To address your questions:
Q1. The creation of a dedicated Post-editing task type is not currently planned; for the time being, we will remain with Proofread/Translation tasks. However, I have passed your feedback to our product development team for future consideration.
Q2. For instance, the “Translation Costs” report can be used with specific filters to understand the volume of work done by post-editors.
These two options must be enabled, and you will receive a post editing report.
The first - so that if a person corrected the translation and approved that translation (either the cost of approval or an alternative translation is counted).
The second - so that the similarity with the pre-translated translation does not fall into the category of other translations.
I hope this information is helpful. If you require further assistance or have additional questions, please feel free to reach out.
I’ll test the options in the reports later. Just to understand, the post-edited word count will be shown in the Translation section? Not in the Proofreading section?
Once again, I find it to the broken logic because the post-editing task is assigned to a linguist under the type: Proofreading.
It depends on which options are selected during the report generation. If, for example, no options are selected, then everything will be counted: translation and proofreading (will be shown in both sections).
If Exclude Approvals for Edited Translations is selected, then only translation is counted and proofreading not (so it will be shown in the Translation section).
Additional information regarding those options you may find here:
I tried the approach with the following two options enabled. However, I don’t think the reports returned the correct figures.
Exclude approvals for edited translations
Use Specific Categories for Pre-translations
Just to confirm: what I’d like to get is the post-edited word count. We pre-translate strings by AI. The post-editing task is sent to the post-editor as Proofreading task type. The post-editor either approves the AI output without edit (if AI output is good as is) or makes edits (if AI output is not good enough), then saves the translation and approves it.
The thing is, when I make edits on an AI pre-translated string, saves it, and approves it, the edited word count is not added anywhere in the report, neither to ‘Translation & post-editing’ nor ‘Proofreading’, when the two options are enabled.
If you make edits to the translations at any step, this should be added to the report and will be shown as an action made by your account. If that does not happen, perhaps you could share some examples/screenshots?
If that does not happen, perhaps you could share some examples/screenshots?
I’d love to! However, it includes some business information such as rates, my business account name etc., I prefer to handle it privately, if possible. I can share the link to the actual projects and the post-editing tasks with the Crowdin team. That way, perhaps it’s easier for the Crowdin team to look into the exact case.
Actually, I posted the original question in the private Slack channel my localization team shares with the Crowdin team. The question wasn’t picked for days, I decided to post the question here. (I was also wondering if other localization professionals are looking for the same information and/or if they could share the tips.)
I suggest moving to the private Slack channel to continue our check I can’t find your name, but I see @Dima is there. I’ll ask him to forward the relevant links and the screenshots to you.
