LLMs and the Future of Translation Technology: Think Global Forum AI Roundtable Key Learnings
Artificial intelligence (AI) and localization are intersecting to redefine the boundaries of communication and technology. The Think Global Forum AI Roundtable series brings together experts in the field to share insights and innovation in this area.
The most recent Roundtable event, titled "LLMs and the Future of Translation Technology & How to Leverage Language Data to Capitalize on Content Assets," not only shed light on the evolving AI and localization landscape but also served as a testament to the wealth of knowledge and experience brought forth by industry veterans Andrzej Zydroń and Bruno Herrmann. Our speakers discussed the implications of Large Language Models (LLMs) on translation and localization and the broader spectrum of human-AI interaction. For those keen to stay ahead of the curve in AI and localization, the learnings from this event are valuable.
The Think Global Forum Roundtable Guests:
Andrzej Zydroń, CIO at XTM
Bruno Herrmann, Executive Advisor and Strategic Consultant
Tom Doerr, Founder and CEO, Machina Marketing
María Roa, Communication Director, Think Global Forum
Simon Hodgkins, Founder, Think Global Forum
LLMs and the Future of Translation Technology
Andrzej Zydroń presented on "LLMs and the Future of Translation Technology," offering a glimpse into the industry's advancements. With over three decades of experience, Zydroń's work with XTM Cloud demonstrates a combination of robustness, scalability, and open standards, which are pivotal in today's digital ecosystem.
Andrezej’s presentation began with a cameo from XTM’s AI avatar, Lisa. AI Lisa presented how XTM plans to implement AI technology and explained that, as we are all aware, there is a significant change coming to computing in the AI space. It has the potential to have a dramatic effect on the direction of technology, such that we have not seen in over three decades. Large language models based on the latest AI algorithms are producing incredible results in text generation, covering not only basic question-and-answer queries but also translation and natural language processing. AI Lisa conveyed how large language models will have an increasing influence on natural language processing and translation. "You can think of them as the Swiss Army knife of AI, tackling translation and natural language processing functions such as grammar and spell checking, named entity recognition, et cetera." Zydroń emphasized the importance of collaborative platforms that are not only cost-effective but also user-friendly, removing technical barriers and democratizing access to advanced tools.
A standout revelation from the discussion was ChatGPT's performance on standardized empathy tests, which scored an unprecedented 100%, outperforming the average human score of 70%. This benchmark suggests that ChatGPT can deliver more empathetic responses than humans when provided with the appropriate prompts. For example, if instructed to consider gender bias in its responses, the tool can adjust with a sensitivity that might exceed typical human capability. ChatGPT's empathy score is a testament to its sophisticated design, which is capable of nuanced understanding and response generation that takes into account complex human emotions and social considerations.
María Roa progressed the discussion by sharing some ethical considerations to consider.
"The human factor is the one it's helping with; for example, cultural sensitivity. Linguists must possess these in-depth cultural sensitivities to ensure that the context of a text's meaning is accurately conveyed. This is especially important in fields such as law and medicine, where mistranslated documents can have serious consequences."
Ethical Considerations and AI Sentience
One of the most thought-provoking topics discussed was the question of AI sentience, particularly concerning empathy and consciousness. While some panelists expressed skepticism about AI's capacity for true sentience, Andrzej referenced conversations with models like Google's LaMDA as evidence of AI's evolving cognitive abilities. The debate over AI sentience continues, with each new development prompting further examination into the nature of consciousness and the ethical ramifications of AI evolution.
The Role of Human Prompts in AI Performance
The importance of human prompts in guiding AI behavior was another focal point of the discussion. The AI's performance, especially in tasks like translation and bias mitigation, heavily depends on the instructions it receives from users. This interplay between human input and AI output is crucial for achieving the desired results, underscoring the collaborative nature of human-AI interactions. XTM's experience when building an LLM was that despite sharing company data and information with their LLM, exact and strict prompts were required, along with a lot of trial and error to reduce AI hallucinations.
Building Custom LLMs
The conversation also covered the practicalities of creating custom LLMs, mentioning resources such as OpenAI's API and Hugging Face as starting points for those interested in developing their own models. However, the panelists cautioned that constructing an LLM from the ground up is a resource-intensive task, demanding significant computational power and a deep understanding of machine learning. For those looking to embark on LLM development, the journey involves a steep learning curve and substantial resource allocation, but the innovation potential is significant.
Future Applications in Translation and Localization
The potential applications of LLMs in translation and localization are extensive and multifaceted. The ability to generate content directly in target languages or to refine translation processes presents a spectrum of challenges and opportunities. The panelists concurred that while AI will undoubtedly reshape the industry, human oversight will remain indispensable, particularly in regulated sectors such as finance and healthcare. The future of translation and localization is inextricably linked with AI advancements, promising enhanced efficiency and innovation while highlighting the enduring value of human expertise.
How to Leverage Language Data to Capitalize on Content Assets
Bruno Herrmann's presentation, "How to Leverage Language Data to Capitalize on Content Assets," underscored the strategic role of language data in content creation and global market engagement. Bruno shared information for a value-driven and inclusive strategy that included:
Enabling Change in Operations with Language Data
Calibrating Operations with Language Data to Create Value
Calibrating Operations for AI-Driven Roles to Include People
Leveraging Language Data for Multilingual Content
Leveraging Language Data for Customer Experiences and More
Bruno also covered key areas of importance, such as unleashing the power of language data for customer-centric operations and how to meet customer needs with data from a technology, operations, and customer experience perspective. The discussion widened to include leveraging language data services, including content and data services in an ecosystem at scale and working with strategic language partners.
A thought-provoking area shared was a detailed breakdown of re-shaping content operations as we move to an AI-powered content supply chain. Bruno detailed the importance of defining roles and responsibilities, as well as creating and transforming functions to maximize and value language skills. This discussion moved into greater detail around scoping and phasing work to cover the creation, management, and delivery of data through defining activities and tasks.
Finally, Bruno covered enhancing customer experience, augmenting and enriching customer reality with data to include micro experiences. Herrmann's approach to data and digital content is pragmatic and insightful, advocating for a customer-centric and data-driven methodology. His presentation included the essential aspects of moving upstream to data and not only focussing on the wow factors of AI.
Panel Discussion
The panel discussion that followed the presentations was an insightful melting pot of ideas and opinions, with each panelist, Andrzej Zydroń, Bruno Herrmann, Tom Doerr, María Roa, and Simon Hodgkins bringing their unique perspective to the table. The conversation ranged from the technical intricacies of AI in translation to the broader implications for the localization industry and global business strategies. The panelists, equipped with diverse backgrounds and expertise, provided a multifaceted view of the challenges and opportunities that lie ahead. Tom Doerr took on the challenging question, “Does empathy require some degree of consciousness?” he shared his experience of how he has been able to produce content that is strikingly effective in terms of its ability to empathize with a target audience, as much as a human writer or strategist would. He outlined human QA and also some of the risks in relying solely on AI, offering, however, that humans also make mistakes.
Key Takeaways
Key takeaways from the panel discussion included the debate and views on technological innovation and human expertise. The panelists discussed that while AI and machine learning are indispensable tools, the human element is a significant topic of debate. Andrzej and Bruno shared differing viewpoints and perspectives, although the panel agreed that LLMs are poised to revolutionize the language services industry.
While the technology introduces new challenges, it also unlocks many opportunities. It is finding a balance between harnessing AI's potential and maintaining the necessary human involvement to ensure quality, ethics, and contextual relevance. However, Andrzej suggested that the human requirement may be limited to the next five years. As we stand on the cusp of a new era in AI, does the language services industry need to navigate the delicate equilibrium between embracing cutting-edge technology and preserving the irreplaceable human element? The panel’s message was clear: the future of translation and content technology is bright and inherently global.
To watch the full event recording, click here.
Registrations Now Open: AI Roundtable Part Three
Registrations for Part Three of the Think Global Forum AI Roundtable series are now open. "How Will AI Transform the Work of Language Professionals?" will feature a presentation from Arle Lommel, Senior Analyst, CSA Research. Panelists include Alison Alexander, Co-Founder, COO, and director of Ethics and Social Policy at Metacampus; Patrick Nunes, Director of Global Communications and Design at Rotary International; Julio Leal, Head of Localization at Spendesk, and Simon Hodgkins of Think Global Forum. Join us as we discuss the future of language professionals in an AI-driven world!
How Will AI Transform the Work of Language Professionals?
Nov 22, 4 PM (GMT) | 5 PM (CET) | 8 AM (PDT) | 11 AM (EDT)