MET workshops

Harnessing LLMs for translation: from theory to practice

Large language models (LLMs) have the potential to transform translation workflows, yet many translators remain uncertain about how to leverage them effectively. This workshop demystifies LLM technology and demonstrates practical prompt engineering techniques specifically tailored for translation tasks. Participants will gain insight into how these tools work, understand their capabilities and limitations, and learn strategies to integrate them productively into their workflows. Through theory and hands-on practice, participants will develop skills to use LLMs as powerful assistants while maintaining professional quality standards.

Facilitator: Dominic Currie

Purpose: To equip translators with foundational knowledge of LLM mechanics and practical skills for translation applications.

Description: The workshop will begin with an accessible explanation of how LLMs function, focusing on the mechanisms enabling them to handle translation. Then, we will examine prompt engineering techniques tailored to translation tasks, with examples from sustainability reporting covering both technical and non-technical content.

Common challenges, such as maintaining terminology accuracy and consistency, handling complex specialized content and avoiding overly literal translation, will be addressed. Participants will see demonstrations of effective prompting strategies for terminology research and extraction, machine translation customization and editing.

In the hands-on second half, participants will apply these techniques through guided exercises, including crafting effective prompts for terminology research, generating and refining translations, and avoiding errors and hallucinations. These exercises can be done using content provided or participants’ own examples. The workshop will conclude with a discussion of ethical considerations, including data privacy and environmental impact, and how smaller language models that can be used offline or on secure servers may address these concerns.

Participant profile: Translators who want to explore AI-assisted workflows. Examples will primarily use Italian-to-English translation, though the principles apply across language pairs. The workshop will be most relevant for those working into English. Participants should be familiar with basic translation technology, though no programming knowledge is required.

Outcome: Participants will understand how LLMs process and generate language, recognize the strengths and limitations of LLMs for translation tasks, and be able to craft effective prompts for common translation challenges. They will leave with a personal collection of English prompt templates and strategies ready to implement in their workflows, along with frameworks for quality assurance and risk management.

Preparation: Participants will need a laptop and a free or paid subscription to a generative AI provider such as ChatGPT, Claude or Gemini. They should familiarize themselves with the basic interface of their chosen platform. Participants will be sent additional information and resources before the workshop.

About the facilitator: Dominic Currie is an Italian-to-English financial translator based in Turin, Italy, with over 20 years’ experience specializing in financial reporting, investor relations, corporate governance and sustainability reporting. Throughout his career, he has developed extensive experience in translation technology, including neural machine translation. Since the emergence of generative AI, Dominic has conducted in-depth research and experimentation with LLMs for translation, developing systematic approaches to prompt engineering for professional translation contexts.