LANGUAGE:
LANGUAGE:
Content Services
- Technical Writing
- Training & eLearning
- Financial Reports
- Digital Marketing
- SEO & Content Optimization
Translation Services
- Video Localization
- Software Localization
- Website Localization
- Translation for Regulated Companies
- Interpretation
- Instant Interpreter
- Live Events
- Language Quality Services
Testing Services
- Functional QA & Testing
- Compatibility Testing
- Interoperability Testing
- Performance Testing
- Accessibility Testing
- UX/CX Testing
Solutions
- Translation Service Models
- Machine Translation
- Smart Onboarding™
- Aurora AI Studio™
Our Knowledge Hubs
- Positive Patient Outcomes
- Modern Clinical Trial Solutions
- Future of Localization
- Innovation to Immunity
- COVID-19 Resource Center
- Disruption Series
- Patient Engagement
- Lionbridge Insights
Life Sciences
- Pharmaceutical
- Clinical
- Regulatory
- Post-Approval
- Corporate
- Medical Devices
- Validation and Clinical
- Regulatory
- Post-Authorization
- Corporate
Banking & Finance
Retail
Luxury
E-Commerce
Games
Automotive
Consumer Packaged Goods
Technology
Industrial Manufacturing
Legal Services
Travel & Hospitality
SELECT LANGUAGE:
Generative AI (GenAI) and Large Language Models (LLMs) have taken off, and there’s no turning back, according to Vincent Henderson, leader of Lionbridge’s product and development teams. But are we there yet?
In other words, can we fully capitalize on the technology now? Spoiler alert: Not yet. At least, not entirely. But we can already reap noteworthy benefits that generate business value, including significant cost savings — if we change our habits and expectations of this era-defining paradigm.
Vincent examined this topic and more in the second webinar in our series on generative AI and Large Language Models.
If you missed the session, watch it on demand.
Want an overview of the session? Read on.
The history of Artificial Intelligence (AI) can be traced to the early 1800s, but the accelerating pace of change makes the technology so exciting in the present day. Consider this: We’ve already had more landmark developments in AI during the first half of the 2020s than in all the previous decades.
We are at a turning point today as GenAI and LLMs offer game-changing capabilities.
“LLMs and generative AI represent a true phase change in the history of AI. It’s an inflection point that we don’t take seriously at our peril.”
— Vincent Henderson
To appreciate the evolution of the AI paradigm and where we are now, we must reflect on past inflection points.
Every time a machine achieved a milestone — such as beating a chess player — technologists would conclude the machine is not really thinking. They continually raised the bar with an ever more challenging litmus test for the machine to demonstrate its intelligence.
What makes the present inflection point different is the ability of machines to understand language, problem solve, code, and produce meaningful content as a general capability, as opposed to being able to complete a specific task it was trained to do, such as detecting imperfections on sheets of metal.
The nature of the human-computer interface is changing because machines can, for the first time, understand the world and do things in a way that involves reasoning and problem-solving they’ve not specifically been trained to. Instead of clicking a button or uploading a picture, the new interface uses language, reasoning, and language-based expressions.
The inflection point that changes the interface between humans and computers is natural language. The computer’s ability to interpret and reason opens a whole category of new use cases and capabilities that computers will be able to do just by virtue of being able to read and understand language, reason, and solve problems.
Many linguistic considerations consume localization professionals. They question whether content properly represents the brand or whether the linguistic quality of the content meets predetermined quality thresholds. They apply a lot of energy to solve these problems, yet these efforts fail to impact business significantly.
Enter LLMs and the opportunity to get to the core of value generation. Here’s how the technology makes a difference. LLMs will increasingly address rote, core linguistic activities as they grow in their capabilities. Subsequently, this contribution will create more space for higher-value human activities, which is the key value of AI in global content.
The eruption of AI-powered solutions will empower human creativity and strengthen their involvement in the following three areas:
Higher-value services like transcreation will become more economically attainable for companies. The upshot? GenAI-powered solutions will ultimately enable brands to deliver content that better resonates with their buyers and is more convincing and trustworthy to buyers in different countries.
Because of the emergence of GenAI/LLMs, customers will increasingly turn to Localization Service Providers (LSPs) to provide services in two major categories: LLM Development Support and LLM Content Production.
Since there will always be localization workflows, we anticipate a high demand for LLM-related services that further automate and enhance workflows. There are many opportunities for generative AI to be impactful.
From source content preparation to content reviews, expect LLMs to enhance every step of the localization workflow as they evolve.
Here are a few ways the technology will affect localization workflows:
During source analysis — LLMs can conduct source analysis and determine whether the source content is fit for efficient localization. Automation of this step is increasingly important as non-native English speakers produce an increasing amount of English content associated with products. LLMs can simplify this source content and make it cleaner upstream so it is better suited for localization.
During translation — While LLMs are no match for existing Machine Translation (MT) engines due to higher costs and slower translation speed than MT engines, LLM technology will still provide new possibilities during this step in the workflow, as it can translate with variance or translate using special instructions.
During post-editing and quality assurance — Ask an LLM to review an existing MT translation, and you’ll really see it shine. The LLM can significantly reduce the load of post-editing work required by a human. Similarly, have the LLM look at a quality assurance report and ask it to determine what to do about existing issues. The LLM can decide if the item in question is a nonissue, an easy fix, or something that needs attention from a linguist.
The most promising use of LLMs right now is in the post-editing realm. But there’s a hitch. Companies must be open to a new interpretation of language quality, especially since no objective measure exists for it. Our testing bears out the subjectivity of evaluation.
Three professional reviewers reached no consensus on the quality of LLM post-editing output when we presented them with the same segments. One reviewer found the work acceptable, while another gave the same segment a significantly less favorable rating. In each of our test cases, at least one reviewer deemed the quality good, leading us to conclude that there was nothing egregious in the output and that LLMs are now a worthy tool for partial post-editing.
If you accept that quality is about fitness for purpose rather than a linguist’s opinion, LLM technology is ready for partial post-editing, giving you access to significant cost savings.
Capitalizing on LLMs throughout the localization workflow will greatly improve localization's outcome, effort, and cost.
We expect LLMs to have a material impact on the post-edit process, envisioning the technology to assess content after initial MT translation, with linguists completing the post-editing process. However, LLMs won’t initially be suitable for every language pair, industry, and subject. We are identifying best-fit scenarios that make sense based on outcome and economics.
According to our research, generative AI / LLMs can potentially reduce localization costs by up to 25 percent, depending on the language pair, when utilized for post-editing after initial MT translation. We are still evaluating the precise economic impact for various use cases and domains. Our initial research suggests it will be significant.
LLMs are here to stay, and they will change localization forever. Lionbridge is developing applications to leverage LLMs to their fullest capability to automate the localization workflow further.
As GenAI advances and proliferates, watch for an evolving regulatory environment.
While GenAI/LLM capabilities are bound to increase dramatically, regulatory agencies may disrupt progress to some degree as they set out to protect users and navigate the real dangers and ethical considerations that are associated with AI.
Generative AI is already producing concrete business benefits, particularly with its use for post-editing work. However, we face a long road ahead, exploring all the use cases generative AI enables before using the technology to its full potential.
One of the defining problems will be bridging the trust gap between us and the machine. We will ask ourselves to what extent we trust the machine to produce acceptable output and to what extent we trust our evaluation of the machine.
Every use case requires evaluation and testing. We will perform these evaluations for months and even years, one solution at a time.
For a more in-depth analysis of generative AI / LLMs on localization, watch the webinar on-demand now and visit the Lionbridge webinar page to access the other webinars in this series.
Ready to start using AI tools to save money and increase your bandwidth? Reach out to us today to find out how.