The Unseen Architects of AI: Translators in the Age of Machine Learning

In today's rapidly advancing artificial intelligence landscape, the backbone of new machine translation systems is largely built on the labor of translators. This dynamic has transformed their work into foundational data capital, important for the evolution of neural and statistical machine translation. The question is, why are these important contributors largely invisible?

Translators: The Unsung Heroes of AI

Translation memories and parallel corpora are the lifeblood for training multilingual large language models. These resources ensure an accurate one-to-one correspondence between source and target text, making them indispensable. Models like neural machine translation (NMT) and the Transformer architecture couldn't exist without this wealth of data.

Despite their critical role, translators often lose out on moral, creative, and economic recognition. Their work, bought under contract, is dissected into technical data and processed under copyright frameworks. This raises a important question: How can we address this oversight?

Legal Perspectives and Invisible Contribution

Two concepts can help unpack this issue. First, there's 'appropriation without consumption'. Here, works aren't appreciated as art but are mined for statistical features, a practice legitimated in Japan under Article 30-4 of the Copyright Act.

Second, translators endure 'invisible teacherisation'. They effectively serve as instructors for AI systems through post-editing and quality assessments, yet remain uncredited. The invisible contribution of translators, akin to teachers in a classroom of AI students, is a striking oversight.

A Call for Change: Recognition and Redistribution

This paper also dives into the data supply chain, tracing paths from translators to language service providers and AI model developers. It compares legal frameworks across Japan, Europe, and the US, highlighting discrepancies and potential for reform. As human-generated data becomes increasingly valuable, the time for addressing these inequities is now.

So, what are translators truly concerned about? It's not just about losing jobs to machines. It's about being sidelined in a space where their contributions are central. The market map tells the story: translators want recognition, and the industry must pivot toward redistributive design.

As AI models continue to dominate, prioritizing human-generated data, the industry must rethink how it values this essential labor. How can we ensure translators receive their due recognition and fair compensation in an AI-driven world?