EXPERT COURSE

Understanding and Using LLMs in Localization

8 hours • 4 live sessions

AI is now a localization skill. Localization is changing quickly and adopting AI at a large scale. However, the right approach is not simply to use more AI. It is to understand where AI is useful, where it fails, and how language professionals can design workflows that keep quality, culture, and accountability in the loop.

Why AI Literacy Matters Now

According to Nimdzi Insights, the language services industry reached USD 71.7 billion in 2024 and was projected to reach USD 75.7 billion in 2025. That scale makes AI literacy a practical career skill for translators, reviewers, language leads, and localization teams.

The Gap Between Demos and Real Workflows

Many AI initiatives look promising in a demo but fail when they meet real content, terminology, multilingual ambiguity, client instructions, and production constraints. MIT NANDA reports that only 5% of custom enterprise AI tools reach production, with poor integration, weak adaptability, and workflow misalignment among the main barriers.

This course is designed around that gap: it connects the inner workings of LLMs with the decisions localization professionals must make in real projects.

A linguist-centered approach to AI. Students will learn how tokens, vectors, embeddings, attention, sampling, fine-tuning, prompting, grounding, evaluation, and agentic workflows shape the behavior of AI systems. More importantly, they will learn how to translate that knowledge into better prompts, stronger quality checks, safer AI use cases, and more realistic integration strategies for localization teams.

Practical confidence, not hype. The course does not ask language professionals to become machine learning engineers. It gives them the conceptual map and operational vocabulary they need to work confidently with AI, challenge weak assumptions, design better human-in-the-loop processes, and make smarter decisions about when AI should assist, automate, or stay out of the way.


Conditions: Please read our course and subscription plans terms and conditions carefully. With your registration, you confirm that you have read, understood and accepted our conditions and agree with them. 

If you have any questions, please visit the FAQ section (for courses or subscription plans) or get in touch with us.

Add to Calendar

Apple Google Office 365 Outlook Outlook.com Yahoo

  • This webinar includes:
  •     Expert tutor: Marina Pantcheva, Director of Linguistic AI Services at RWS
  •  Lifetime access to the course and extra contents
  •  In English
  •  TranslaStars Certificate of Completion
  •  Course duration: 8 h (2 h/day approximately)
  •  When: 06-09 October 2026 (18.00 CET)

SEO AND GEO IN THE AI ERA

Understand how AI experiences are reshaping search. Learn SEO/AEO/GEO best practices,
tool-specific behavior, and how to power AI-assisted SEO/GEO strategies

Course Curriculum

8 hours across 4 live sessions • 2 hours per session

Session 1 - LLMs and the Representation of Language

2 hours
  • ● From n-gram models to transformers: the evolution of language modeling
  • ● Tokens as currency: chunking text, tokenization strategies, and why tokenization changes meaning
  • ● Embeddings and vector representations: how words become numbers, semantic proximity, and multilingual vector spaces
  • ● The transformer architecture: how attention moves beyond individual words to understand context
  • ● Why this matters for localization: how LLMs handle ambiguity, word order, and multilingual input

Session 2 - How LLMs Are Trained and What They Do

2 hours
  • ● Next-token prediction and output probabilities: how models generate text one token at a time
  • ● Softmax, sampling, and temperature: why LLM output is probabilistic and non-repeatable
  • ● Epochs, checkpoints, overtraining, and the curse of multilinguality
  • ● Post-training: classification fine-tuning, instruction fine-tuning, supervised fine-tuning, and RLHF
  • ● Hallucinations, sycophancy, and the limits of LLMs: numeracy, common sense, form vs meaning, and reasoning
  • ● RAG and autonomous function calling: grounding models in real data for reliable output

Session 3 - From Inner Workings to Outer Instructions

2 hours
  • ● Six basic concepts in prompting: prompts, tokens, completions, shots, context windows, and primary content
  • ● Prompt design errors and the curse of knowledge: common mistakes and how to avoid them
  • ● Parameters that control output: temperature, top-p, top-k, presence penalty, and frequency penalty
  • ● System vs user vs assistant prompts: building reliable instruction structures for production use
  • ● Grounding techniques: RAG, tool grounding, search grounding, critic LLMs, and rules with examples
  • ● Decomposing complex requests and designing step-by-step agentic workflows

Session 4 - AI Integration Strategies for Localization Teams

2 hours
  • ● The machine-human intelligence continuum: mechanical checks, linguistic AI, and human judgment
  • ● Localization use cases that work: TM cleanup, language ID, source quality scoring, terminology extraction, adaptation, and variant conversion
  • ● Where LLMs still struggle: non-repeatability, instruction following, tags, numbers, rare words, and sycophancy
  • ● AI LQA and evaluation methods: true/false scoring, score-based assessment, pre-labeled data, and human alignment
  • ● From pilot to production: chat-based vs API prompting, token costs, batching, and processing at scale
  • ● Human roles around the loop: HITL, HOTL, HOOTL, and choosing the right level of oversight for each workflow

Understanding and Using LLMs in Localization

8 hours • 4 live sessions • Live Online

This course gives localization professionals a grounded, practical understanding of Large Language Models — not hype, not theory divorced from practice, but the inner workings, real capabilities, and operational boundaries of the AI tools that are reshaping the industry.

Students will learn how tokens, embeddings, attention, fine-tuning, prompting, grounding, and evaluation shape AI behavior, and how to translate that knowledge into better prompts, smarter quality checks, safer AI use cases, and realistic integration strategies for localization teams.

Designed from a linguist's perspective. No coding or ML background required. Every concept connects directly to the decisions localization professionals make in real projects.

Meet Your Coach

👩‍🔬

Marina Pantcheva, PhD

Director of Linguistic AI Services at RWS

Marina holds a PhD in Theoretical Linguistics and leads a team of linguistic AI professionals at RWS developing cutting-edge linguistic AI solutions. Her career bridges academia and industry, from research in theoretical linguistics to leading teams across linguistic quality, technology, and service innovation in localization.

She is a co-founder of the AI Localization Think Tank, an international public speaker, and a science communicator known for making complex AI concepts accessible to language professionals. Her focus: helping localization teams understand what LLMs can and cannot do, and how to build AI workflows that are linguistically sound, culturally aware, and operationally realistic.

Who Is This For?

This course is designed for language and localization professionals who want a grounded, practical understanding of LLMs. It is ideal for:

  • → Freelance translators and reviewers who want to understand how LLMs behave so they can use AI tools critically, not blindly.
  • → Localization PMs and program managers who need to evaluate AI proposals, coordinate pilots, and explain AI risks and benefits to stakeholders.
  • → LSP operations and production teams who are exploring AI workflows for TM cleanup, terminology, source quality, adaptation, LQA, and post-editing.
  • → Linguistic quality specialists and language leads who need stronger methods for evaluating AI output, diagnosing recurring errors, and designing human review loops.
  • → Language technologists and solution designers who want a linguist-friendly framework for moving from chat-based experimentation to scalable, governed workflows.
  • → Localization educators and team leads who want to train their teams in AI literacy and responsible AI use.

No coding or machine learning background required. The course is designed from a linguist's perspective and is accessible to anyone with professional experience in translation, localization, or language technology.

Requirements & Resources

Required: Stable internet connection and a modern browser. Basic familiarity with translation, localization, linguistic review, project management, or language technology workflows.

Recommended: Access to at least one approved LLM environment (ChatGPT, Gemini, Claude, Copilot, or similar). No specific vendor is required. A small non-confidential localization sample (translation, terminology, or content) for reflection exercises.

Optional: API access or a sandbox environment for students who want to test integration ideas. The course can be followed without coding or API access. CAT/TMS experience is useful for Session 4, but not required.

Important note: Students should not upload confidential client data to public AI tools.

Marina Pantcheva

| Director of Linguistic AI Services at RWS
| Co-Founder of AI Localization Think Tank
About MARINA
Marina is Director of Linguistic AI Services at RWS Group  where she leads teams building cutting-edge linguistic AI solutions for real world localization challenges.
She holds a PhD in Theoretical Linguistics and is known for bridging rigorous research with practical innovation, speaking and publishing on topics like LLMs for terminology, LQE metrics, and GenAI readiness for language teams.