The need to translate more content at a much faster pace without compromising quality has paved the way for Linguistic AI solutions. Many of our corporate customers are embracing the latest technologies and are using them to improve their content strategy.
However, the picture is a little more complex for the Regulated Industries (RI). Sectors like Finance, Legal or Life Sciences need to take many additional factors into consideration when looking to localize their content: keeping track of national and international regulations; retention rules; different content types and formats; specific data privacy laws; and the challenges that go hand-in-hand with very specific quality expectations. With the added pressure due to the global pandemic, it has now become imperative for the RI sector to keep up with other industries already benefitting from AI both in terms of money and time to market.
Linguistic AI for Regulated Industries requires a nuanced and carefully orchestrated approach.
When we decided to introduce AI to our RI business in the form of Neural Machine Translation (NMT), we needed to carry out a thorough risk assessment. Our top priority was to avoid any disruption in the production process. As part of a first phase content strategy we focused on the financial domain. Highly regulated financial source content is often suitable for a Machine Translation (MT) approach. A good example of this content type are KIIDs; these are structured “fact-sheet” style documents containing critical information about a fund which are aimed at helping investors understand the nature and key risks before making an investment decision.
This is just one example of frequently requested, highly structured content with tight turnaround times within the financial sector, making Finance the ideal candidate for an MT solution.
When introducing MT to RI, content suitability is not the only challenge. Translators are usually highly specialized and may not have been previously asked to post-edit. Often it was also the first time for project managers to introduce a post-editing step in their workflow. This meant that education and training really were key. Training focused on NMT technology, tools and processes on how to successfully implement MT and PE. Translators were able to access our successful Post-Edit Certification Program which covers key NMT behavioral phenomena.
To establish valuable data points, diverse financial content was tested on a range of languages. We carried out formal MT productivity testing on representative samples to identify if post-editing was more productive than human translation. The financial pilot showed significant productivity gains across languages and content types and proved that MT and PE is more than up to the challenge of delivering a high quality outcome while improving turnaround times and cost-effectiveness.
An intelligent, repeatable and measurable workflow for regulated content
The success of the financial pilot was the basis on which we could extend the testing framework to other RI fields. We continually revise our processes to ensure that they deliver the best results and are sustainable and repeatable across different verticals.
Analyze: We analyzed the taxonomy of each RI vertical including data privacy requirements for whole sectors plus specific customers; standard timeline requirements; supply chain status and any additional information relevant to the specific content type
Define: We planned the MT testing project with a view to engage the main stakeholders based on content mapping; quality expectations; customized processes for each vertical and clearly communicated timelines
Execute: Execution was based on the following steps
- Collecting samples for all verticals
- Evaluating content in different ways to guarantee meaningful results: automated evaluations help us to understand the potential quality of the NMT models while human evaluations provide valuable feedback on the understandability of the content and the expected productivity gain over human translation
- Communicating the results to all stakeholders with the aim of increasing MTPE adoption
Document: Documenting the results in the format of a suitability matrix and sharing them across internal platforms
Business As Usual: Ensuring that all stakeholders are aware of the results and expected productivity and margin improvements
Our tests showed that the success of the financial pilot was repeated across the Legal and Life Science sectors. Significant productivity gains for a number of content types clearly showed the potential of our NMT technology and paved the way for a truly exciting collaboration. In addition, this process provides a template for other RI sectors and use cases and strengthens the case for an MT-driven content strategy even for challenging industry sectors.
Expanding capabilities to different content types - what does it take?
Expanding capabilities to sectors such as RI and increasing capacity at the same time would not be possible without the help of experienced resources. Throughout the process of evaluating the content, performing MT testing and then implementing AI in a new workflow, you need the right roles and personas to drive change.
Our interconnected teams of MT developers, researchers, computational linguists and Linguistic AI consultants have created a solid framework for introducing high-quality MT solutions on a large scale. The support of the production teams – post-editors and project managers – is essential when implementing an MT solution.
How did it work in practice for our Regulated Industries business?
We very much took a strategic approach when introducing MT to RI, identifying high impact content with tight turnaround times. The project management teams provided consultancy on which projects and content types would benefit from MT, worked on anticipating potential risks and strategized on how to communicate the benefits of MT to their customers. Translators also had to adapt to the new process, with Supply Chain making sure that enough trained resources were available for post-editing. The goal was to upskill existing translators with project and domain experience rather than having to train new resources.
Our technology is highly adaptable and we maintain a close feedback loop between MT users and R&D to design the best fit-for-purpose solutions. Customized approaches are taken where appropriate, ranging from specialist data preparation to training custom models.
Can AI improve your content strategy?
The close collaboration between the Linguistic AI and Regulated Industries teams with a focus on the best outcome for the customer have helped to open up new ways of converting strategic content workflows to MT and PE. Our tailored – but at the same time flexible – approach is making localization more sustainable in the long run and reduces the need to make difficult choices between quality, speed and cost.
We have been able to demonstrate that AI in the form of MT can be safely introduced to Regulated Industries, with no risk of compromising data or losing quality. Our technology can be customized where needed and offers reliable and sustainable quality even for the most challenging RI sectors.
