How do you integrate post editing into your language service workflow and tools?

In managing localization at scale, choosing the right MT engine is crucial, and tools like TAUS DQF and MT-Hub can assist in this. **TAUS DQF** - **Pros**: Industry-accepted and focused on detailed quality metrics. - **Cons**: Steeper learning curve and potentially costly. **MT-Hub** - **Pros**: User-friendly with broader functionalities, including cost and speed. - **Cons**: Less comprehensive on the quality metrics front. At Meta, we used a hybrid approach, blending features from both tools, to select MT engines that allowed us to scale across multiple languages while maintaining quality. Your choice should align with your project's specific needs.

It may be useful to give a local MT engine a try if there's any. For example, Yandex (together with Google and DeepL) is one of the top performers for Russian MT projects. This is because Yandex has a huge local native Russian MT team.

In my language service workflow, post-editing is akin to fine-tuning a masterpiece. Clear guidelines act as my compass, ensuring each edit aligns with the project's tone and objectives. Collaborative tools become my virtual editing room, where real-time adjustments bring words to life. Feedback loops are not just corrections but conversations, fostering continuous improvement. Integrating machine translation post-editing is like having a trusty assistant—efficient and enhancing the overall quality. Through personal experiences, this approach has become my artistic process, refining language services to create impactful linguistic masterpieces.

Define the post editing scope and guidelines Post editing can be done at different levels, depending on the purpose and audience of the translated text. For example, light post editing aims to make the text comprehensible and error-free, while full post editing aims to make the text stylistically and culturally appropriate. You need to define the post editing scope and guidelines for your project, and communicate them clearly to your post editors. You can use existing standards like ISO 18587 or TAUS PE Guidelines, or create your own based on your client's requirements and expectations.

In my experience, light post-editing is slowly dying. More and more clients require full post-editing and a result that is similar to human translation.

Post-editing is not for just any translator; it requires a positive attitude towards the service, a specific skill set and expertise. A post-editor needs to undergo specialized training to understand, among other aspects, the intricacies of the specific MT engine or system used. Familiarity with the errors and limitations of these systems is crucial. Moreover, post-editors must grasp the techniques of MTPE to ensure that the final translation aligns with industry standards, such as ISO 18587.

One of the biggest lessons I've learned managing complex globalization and localization processes at a scale is that while translators excel at crafting content from scratch, post-editing is a distinct skill set, requiring adeptness in refining machine-generated translations. It's crucial to understand that not every skilled translator will naturally excel as a post editor.

Some skilled translators were reluctant to do post-editing because it required lots of retranslation and could kill their good writing style. Now, with the rise in the quality of MT and a shift of many LSPs to MT many of them have to do it to stay in the industry. It's important to give proper guidance and support to keep the best talent doing MTPE.

All of my post-editing jobs are done in the regular CAT tools, like Trados Studio, memoQ or client's proprietary online tools. For this matter, such jobs are no different from normal editing ones.

Integrating post-editing into our language service workflow involves a strategic approach to leverage valuable data and feedback. We meticulously collect and analyze post-editing data, examining edits, error categories, time, and cost savings, along with customer satisfaction metrics. Soliciting feedback from post-editors and clients is key, incorporating suggestions, addressing complaints, and acknowledging compliments. This wealth of information informs continuous improvement—fine-tuning MT engines, adjusting post-editing scopes, and optimizing workflows and tools. By actively involving post-editors and clients in the feedback loop, we ensure our language services evolve with precision, maintaining a commitment to quality & efficiency.