At GALA San Diego attendees had the opportunity to hear from Pavel Efimkov, an MT expert from Intento, on the technical checks and fine-tunings necessary so you can deploy effective translation solutions. Keep reading for insights on MT adoption, maintenance, planning, and more.
• • •
Adopting MT is a no-brainer for those looking to efficiently expand their business and operations. There’s a lot offered on the market today, from large, popular vendors to more niche vendors — offering everything from more mainstream language pairs to niche, under-resourced languages. It’s worth exploring our latest State of Machine Translation report to see a detailed overview of the current MT landscape.
To make sure that your MT is being managed correctly, it’s useful to start small and then see where you can expand from there. There’s no use in trying to bite off more than you can chew right off the bat. It’s not enough to simply integrate machine translation, even with all of the fine-tuning options available on the market. You need to know how to maintain your MT program, continually tracking your success metrics. It’s the details of this integral part of the process that we’ll be going over in today’s post. With a solid maintenance plan, you’ll be able to scale your MT program to new languages, content types, data formats, and platforms.
You wouldn’t drink expired milk — so why continue to use an out-of-date MT plan? Several elements of an over-the-hill program can be seriously harmful:
- Keep the state of your current model in mind, and be proactive. You don’t want to wait to see the ‘check engine’ alert — because at that point something’s already amiss.
- It’s essential to keep your product names, terms, and other nuances up to date and aligned with your company goals.
- Over time, you’ll have more translations that your MT models could benefit from. The more current data you have, the better your model will perform.
- There’s always something that can be improved, and we need a solid understanding of where our work will make the biggest impact.
- The machine translation landscape is constantly evolving — the number of Cloud MT Vendors has doubled in the last few years. So we need to be aware of new features and vendors and keep track of new providers emerging on the market. This way we can make the best decision on when to reroute our volumes.
• • •
It’s important to remember that MT maintenance is a recurring activity. You wouldn’t want to be driving a car that hadn’t been serviced in a few years, and you can think of your machine translation in the same way. To make sure that you stay on top of when you have to think about which aspect of the maintenance process, a timeline can be useful:
• • •
Terminology is the quickest way to adjust your model. This is also a good place to start introducing fine-tunings to your project. Every company has a vast array of content that needs nurturing, such as product names, DNT’s abbreviations, specialist and ad-hoc terminology, etc. Glossary updates are relatively easy to manage, making them a good place to start.
• • •
MT Quality Monitoring
You’ll want to monitor changes in the baseline models of necessary vendors. Keep an eye out for new vendors on the scene, and improvements in the baseline quality of each model. Changes are used as markers for evaluation, re-training, or replacing a model.
• • •
All models require fresh updates. Keeping your model updated will ensure that you’re getting the best quality possible. Vendors are constantly updating their models, training algorithm improvements, optimizing the data cleaning pipeline, and implementing new data injections.
• • •
A localization checkup is a series of analytical checks that allow you to narrow down your focus when needed. This way you can get to know your effort consumption, control your expansion, and identify bottlenecks — getting a comprehensive view of your entire process to see what needs help.
• • •
Properly maintained MT will have quick and obvious value gains for your team:
- Your content and terms will be regularly updated, making them easy to understand.
- You’ll gain an awareness of new vendors, allowing you to consider new areas of interest.
- Knowing about changes in quality will give you more flexibility in your program.
- Your models will be getting the necessary updates.
- New data injections will improve your efficiency of output.
- By knowing where your efforts are going, you can make adjustments before they become an issue.
- Seeing the big picture allows you to make better decisions moving forward.
- You’ll experience inevitable expansion.
• • •
For more information, please reach out to firstname.lastname@example.org, or visit inten.to for a live demo.