Rubric’s Approach
At Rubric our tech suite is constantly evolving to serve the needs of our clients. We pride ourselves in making best use of technology whilst ensuring our services are “human at the core” — helping our clients to navigate the ever-changing tech landscape and orchestrating their global content efforts.
AI is a rapidly evolving technology with a plethora of potential use cases in localization. As such it is an important part of our offering, but using AI effectively in localization means more than just buying a subscription to a new tool. It requires a deep understanding of the suitability, ROI, capabilities, and limitations of such tools. We must then apply them skilfully to the steps in our localization workflows where they will have the most impact.
The key is to focus on your entire localization workflow and be ready to add new tools as they become available.
During our recent Rubric Roundtable on AI in localization experts from Rubric and localization market research company CSA Research discussed AI in localization. Rubric’s IT Director, Dominic Spurling, summarized Rubric’s approach:
"We continually invest in workflow optimization. As AI services mature, we are ready to plug them in at any stage of the localization pipeline, from authoring right through to release and publishing."
How we use AI will continue to evolve, but how are we using AI at Rubric now? During our recent Roundtable, Alicja Weikop, Program Manager at Rubric, introduced 5 of the most common AI applications that we incorporate into our clients’ global content workflows.
Application 1: Machine translation post-editing
Machine translation (MT) has been the most heavily used form of AI in localization for decades. It is a core technology that helps companies to translate higher volumes of content efficiently and consistently.
MT workflows often include a human element — post-editing. This is when a human translator checks the automatic translation to correct any errors that may have arisen.
As Dominic said:
"Machine translation is something that we have done for many, many years. Usually, because of the nature of our clients' content sets and requirement for quality, we almost never do MT in its pure form. We always get humans to check."
With newer AI tools, it is now becoming possible to use AI to perform some post-editing checks, such as performing quality estimation and providing post-editors with context-specific suggestions. This can reduce the work humans are required to do and further streamline the process of MT localization.
Application 2: More realistic voices in text-to-speech
The basic function of a Text-to-speech (TTS) algorithm is to convert text into speech. Professional-grade human audio recordings are expensive — requiring voice actors, directors, expensive equipment, and expertise. As a result, TTS can offer global companies exceptional ROI by slashing audio recording costs and reducing turn-around time.
TTS technology has been around for years, but from the early vocoders in the 1930s right up to more recent systems, the voices have often sounded too "robotic" to be acceptable for many content types. However, this is changing and TTS voices are becoming more realistic and more accepted by end users.
During the roundtable, Alicja said:
"We have been using TTS for years and it is an area that is fast developing. The voices have gotten so good that there is a much higher acceptance rate now in various target languages."
Application 3: Regional content adaptation
A common situation for global companies is to serve multiple markets with the same language. For example, translations into French might be used in both France and Canada or Spanish could be used in up to 20 different markets.
It is often advisable not to use exactly the same translations in different markets. Markets that use the same language will have local cultural and linguistic differences that make it necessary to adapt the content slightly. Depending on the scale and complexity of these changes, it is now possible to use AI tools to perform this type of regional content adaptation by following a set of rules.
Alicja explained:
"AI can easily tackle issues such as adapting German content for Germany into content for Switzerland. For example, by following a specific list of grammar or punctuation rules that we would tell it to adapt."
Application 4: Extra efficient transcription
AI transcription is also an application that we regularly use at Rubric. Transcription tools can make the localization process more efficient, allowing us to process larger volumes of spoken content. This makes it possible to scale localization projects that would have been prohibitively expensive or time-consuming in the past.
Generally, AI transcription has some limitations that you need to manage. These include difficulties with jargon or slang, inaccuracies when voice is recorded in non-studio environments, and potential loss of information when automated.
But the technology is improving all the time.
As AI tools overcome these limitations, automated transcription becomes more and more efficient. We continue to incorporate these improvements into client workflows, which makes their localization more efficient.
Application 5: Source content optimization
Global companies rarely realize how much of their localization budget they could save by optimizing their source content.
Alicja said:
"We look to standardize the content. We often find that clients have multiple copywriters writing copy that is similar but not exactly the same. Then they're translating that copy into multiple languages and paying multiple times for translation. If we can standardize the source, we bring savings further down the line in translation costs."
With AI tools, it is now possible to improve the consistency of the source content through automation. This creates even more ways to scale localization, without adding to the human workload.
Some ways that AI can help optimize source content include identifying similarities in meaning between different pieces of content, adapting content to meet style guide requirements, and automatically detecting redundancies between different source files.
This opens the door for many small but significant automated steps in a localization process.
Conclusion
The 5 applications listed above are just some of the many ways that AI can be applied in localization workflows. They are the ones that we are currently focusing on most at Rubric, however, we are always on the lookout for more cutting-edge tools that we can use.
If you are interested in learning how you can apply these applications to your company's workflow, just get in touch.