Democratizing MT with Language Weaver’s new feedback editor feature

Ollie Stott 07 Jul 2023 6 mins
Language Weaver
Historically, businesses have faced difficulties in achieving natural-sounding translations for their industry if they did not have a team of localization experts at their disposal. This typically results in reliance on third party localization service providers who, while being linguistic experts, are not necessarily experts in the subject matter. Running a comprehensive localization program takes significant time and comes at a high cost, especially when multiple review iterations are required. Without this, though, translations may show a lack of understanding of the industry.
However, recent advancements in AI and Natural Language Processing (NLP) are challenging this dynamic. It is worth considering that today, with Adaptive Machine Translation (MT), your entire business can contribute to translation efforts – not just the localization team.
Taking advantage of this new opportunity comes with its own difficulties, of course. Businesses must address the challenge of enabling teams that lack knowledge of typical translation workflows, and of AI, to leverage both of these tools. The solution? To provide them with an easy-to-use product that empowers them to teach AI to speak their language.
Language Weaver’s feedback editor is democratizing the ability to teach AI and get the most out of Machine Translation. In doing so, employees from many roles and departments are now able to actively participate in, and benefit from, translation projects.
Businesses embracing this technology will be able to reduce reliance on third parties. They will also see vastly improved translations based on AI models that are uniquely theirs, trained by the real experts in their business and industry – their employees themselves.

Why has it been difficult to achieve natural-sounding translations for my business until now?

Naturally, businesses want translated content to look and feel fluent, and reflect their deep expertise in the industry and subject matter. In trying to achieve this, however, they typically find themselves coming up against barriers. The problem here is that, although MT tools are constantly improving, the use of generic machine translation may still require involving localization experts at the final stage to refine the output and ensure that it sounds natural and relevant.
In the localization industry, this workflow is referred to as MT Post Editing (MTPE), or Expert-in-the-Loop (EITL), and generally requires external Computer-Assisted Translation (CAT) tools in addition to the MT platform. This is largely why professional linguists’ expertise are required. CAT tools take time to learn, can be expensive, and most employees with multiple priorities simply don’t have the time to be post-editing MT output time and time again.
There are solutions to these challenges currently facing businesses, and they are coming from the increasing power of Neural Machine Translation (NMT). NMT is a type of AI which, using neural networks, learns to gradually improve its translation accuracy as it is exposed to more and more training data. Increasingly, modern MT tools can facilitate many of the requirements that have previously been met by MTPE, whilst simultaneously offering solutions to the challenges faced.

What is Adaptive MT and how is it democratizing MT’s user base?

As Neural Machine Translation has matured, developers have refined the models to allow for greater flexibility in how we can train them. One of the biggest things to come out of this has been Adaptive MT. Adaptive MT enables the users themselves to actively train MT models with their own data so that the output will become increasingly more specific to them and their industry, thereby reducing the need for post-editing over time.
Generally, the most thorough way to adapt MT is to engage the AI’s deep learning capabilities through training with large bilingual data sets. However, this benefitted those in the localization industry the most – they are typically the ones who have access to this sort of data.
Further developments around Adaptive MT led to the creation of feedback loops, which made MT model adaptation available to businesses that did not have access to large bilingual data sets. With feedback loops, Adaptive MT is now available to new user bases, regardless of their functional focus area.
Using feedback loops, users can provide feedback on MT outputs and suggest improvements by directly editing segments that require intervention. The model will then learn these improvements and suggest them when the same source text is input in future. An issue with this, however, is that adapting MT models in this way takes longer than using large data sets. After all, it is just one user gradually training the model, bit by bit.
So, what if an entire team could be utilizing feedback loops to train the model at the same time? This summer, Language Weaver are releasing an advanced feedback loop feature that will empower all types of users to translate content and adapt the MT models together, sharing the benefits.

The feedback editor

From July 2023, the feedback editor will be available in all Language Weaver portals, enabling users themselves to enter the loop and adapt MT at every level. Users will be able to input real-time, direct feedback to the MT output by suggesting preferred translations. They will also be able to edit the output for entire documents or individual segments directly, natively, and securely within the application – without having to use a dedicated CAT tool.
The extent to which businesses utilise EITL-style workflows will be up to them. Admin users will be able to automate approval of feedback based on user roles, thereby maintaining the desired quality control that EITL workflows offer for some users, while allowing the most trusted and skilled users to start training the MT output from the get-go. 
Importantly, all users will be able to benefit from approved feedback. Those providing feedback will not be training the models in isolation, but rather adapting the model for the entire business. Ultimately, this will lead to increased productivity as the model learns from more and more feedback.
By doing all of this within the Language Weaver portal, without the need to move data between systems, users are presented with a convenient solution that enables them to achieve end-to-end translation and editing within one easy-to-use UI. For the same reasons, businesses will also be able to always guarantee the security and privacy of their data.
Finally, by combining the feedback editor feature with auto-adaptive capabilities, businesses will be able to take the power of MT to the next level, continuously and automatically training their models with feedback submitted by users accumulated across the organization over time.
The feedback editor is a powerful tool that empowers the user to become the expert in a secure loop that never leaves the Language Weaver portal, enabling feedback and adaptability at every level. No more post-editing the same documents time and time again, no more third parties. Just natural-sounding MT outputs, trained by your employees, leveraging your own unique AI.
For more information visit the Language Weaver feedback editor webpage.
Ollie Stott

Ollie Stott

Business Product Manager
Ollie is a Business Product Manager at Language Weaver focused on delivering exceptional, end-to-end customer experiences from the very first interaction all the way through to final business outcomes.
All from Ollie Stott