Microtalks
Track: Microtalk | M7 | Everyone | MICRO TALK |
Thursday, October 16, 2025, 1:30pm – 2:15pm
Held in: Steinbeck 2
Presenters:
Helena Batt - TED Conferences
Helen Corfield - Translated
Marina Pantcheva - RWS
Iti Sahai - Klaviyo
Mathijs Sonnemans - Blackbird International
Pavel Soukenik - Acolad
Host: Daniel Goldschmidt
M7: Localization Management Interfaces — From Plugins to the Perfect Project Management Environment
Presented by: Mathijs Sonnemans, Blackbird.i0
As localization workflows evolve, the role of interfaces in managing translation processes is shifting. Traditionally, plugins have served as the primary bridge between content creation and localization workflows. However, modern approaches increasingly integrate localization directly into existing project management and productivity tools, allowing for seamless synchronization and control without the need for separate plugins. This presentation explores the technical and operational advantages of this shift, highlighting how organizations can streamline localization by embedding workflows into the tools they already use. Attendees will gain insights into best practices, key challenges, and the future of localization management in a more connected and automated ecosystem.
Takeaways: Traditional plugins have major downsides: technical overhead, lack of customization, duplicate interfaces; integrating in existing PM tools creates more controls, insight, and flexibility to improve workflows.
M7: Closing the Loop: Leveraging User Feedback for Global Product Success
Presented by: Iti Sahai, Klaviyo
The presenter will explore strategies for systematically collecting and integrating user feedback from diverse markets to inform product development, ensuring global relevance and user satisfaction.
Takeaways: Implementing structured feedback loops ensures that product decisions reflect real-world user needs across markets; localization teams can play a crucial role in interpreting regional user feedback for product optimization; AI and automation tools can help scale feedback collection while maintaining cultural and linguistic accuracy.
M7: AI Dubbing at Scale: Balancing Automation, Quality, and Global Reach
Presented by: Helena Batt, TED Conferences
AI-driven dubbing is reshaping video localization, making content more accessible, engaging, and scalable across languages. This session explores real-world implementation, focusing on TED’s journey in AI dubbing—how automation and expert-in-the-loop workflows optimize quality while maintaining speaker authenticity. We’ll discuss lessons learned, key challenges, and the evolving role of AI in localization. Attendees will gain practical insights into integrating AI dubbing into their workflows, balancing automation with human expertise, and navigating quality, cultural adaptation, and scalability.
Takeaways: Proven strategies for blending AI with human expertise to deliver flawless dubbing quality at scale; tips for optimizing speaker opt-ins and building trust in AI-driven localization; key challenges and solutions in deploying AI dubbing for diverse global audiences.
M7: How to Make an LLM Reason Like a Linguist. Bridging Human Expertise and AI for Next-Level Localization
Presented by: Helen Corfield, Translated
In a world of multilingual AI and high-speed content demands, can a machine capture the nuance and quality of a human linguist? This presentation reveals how large language models (LLMs) and linguists working in tandem outperform either alone. The presenters will introduce a collaborative approach where AI is trained to “think” like a linguist, enabling it to catch contextual nuances, maintain style guidelines, and even flag cultural or legal concerns — all at scale. The result is a symbiosis that boosts productivity and quality, showing that human expertise and AI together can elevate global content beyond what traditional methods achieve on their own.
Takeaways: Best practices for adapting traditional linguistic assets for LLM based workflows; how to train an LLM to match specific client localization styles and requests; how linguists can use this approach to expand their client portfolio faster and be more productive.
M7: SPA-RAG: A Smarter Way to Validate and Fix Terminology in AI Workflows
Presented by: Pavel Soukenik, Acolad
Ensuring, evaluating, or automatically correcting glossary adherence in AI workflows is a challenge that often relies on computationally expensive and unreliable retrieval-augmented generation (RAG). This talk introduces SPA-RAG, a special-purpose algorithm RAG that uses one-time LLM pre-processing of the glossary to enable filtering out of compliant entries and precise retrieval for the generative AI step. By invoking LLM only where it adds value rather than noise, SPA-RAG ensures faster, more deterministic glossary compliance. Attendees will see a live demo and statistical comparisons demonstrating how this method improves quality, reduces waste, and improves efficiency in automated localization workflows.
Takeaways: A concrete workflow for automated glossary adherence evaluation and fixes that significantly improves accuracy and efficiency of standard AI with RAG; understand how pre-filtering and deterministic lookups improve speed and accuracy in AI workflows; practical insights on what tangible improvements they can expect from SPA-RAG for cost-effective terminology QA and implementation.
M7: Breathing Life Into AI-Generated Text
Presented by: Marina Pantcheva, RWS
Have you ever started reading a text, only to realize within minutes that it was AI-generated? Beneath its polished surface, the writing feels banal and dull — full of clichés, void of meaning. As our industry moves towards AI-generated content, the ability to create authentic and engaging text becomes crucial. This talk presents techniques for bringing life and depth to AI-generated writing. The presenters begin with a brief linguistic analysis of AI language and share practical strategies for cleaning AI-generated content from the typical AI traits. Additionally, they discuss how to create prompts that help generate meaningful and authentic texts.
Takeaways: Attendees will gain insight into the structural and stylistic properties of AI-generated language, including why it often lacks depth and originality; learn how to refine AI-generated text to make it more expressive and human-like; learn how to design prompts that lead AI models toward producing more meaningful, coherent, and authentic content.
