Bruno Herrmann, recently wrote a very insightful piece called Language Intelligence Needs Language Expertise and Recognition – The Power of T-shaped Language Experts on LinkedIn. We’d like to share some of the highlights that stood out for us.

Bruno explains that by combining holistic language effectiveness with in-depth knowledge you will receive better results. It’s something he’s advocated for and implemented throughout his career in the translation industry.

“…apply the T-shaped model to language management roles, first of all to extend and expand language expertise while elevating the people mastering it, whether they were project managers, linguists, vendor managers or content design specialists. In addition, it was a way to make these people more versatile through a combination of language, practice and domain expertise…”

With the proliferation of AI, it’s essential to have humans in the loop. As he says, “Human supervision does not start with AI-generated content editing and enrichment.” It starts well before that to avoid the dreaded garbage in – garbage out misery.

Language in LLM should be more than just a letter in an acronym. It should be seen as “a key driver on the generative AI highway.” Bruno might say he came up with that terminology, but we whole-heartedly agree with his statement. We see language as a driving force behind the projects we work on for our clients.

“If we consider the end-to-end generative AI supply chain, we clearly see that human supervision is necessary well before humans are asked to review and enrich AI-generated output. Much attention has been paid to downstream AI, when such output is evaluated and improved as needed and according to agreed effectiveness criteria … a complete process that starts with upstream AI, including data management, model training and prompt engineering.”

This aligns with DATAmundi’s way of working. Our projects start with a detailed intake to establish the guidelines for QA testing all along the way. We manage,

  • “The conversion of assets into training data
  • Training data from acquisition, curation, annotation, upcycling and maintenance to successfully train AI models
  • Prompt engineering
  • Output optimization, here too generated content can be enhanced by language experts …”

 

At DATAMundi we agree that language experts are the heart of what creates effective AI training models. Our pool of experts includes linguists who live in countries of the selected languages and have gained a deep understanding of the nuances of the language, the culture and the country. That helps lay the foundation of any good LLM.

“When language experts are enabled to demonstrate their value both in their role and in their collaboration with other roles, they can streamline and maximize complex workflows, from creation to delivery…”

If you want to learn more about how we at DATAmundi work with our pool of linguistic experts and how they can help get your multilingual data ready!