我要吃瓜

Citation

Li S, ANDRADES MORENO A, MAROTO N, FERNANDEZ VIVANCO A, Chang Y, FERNANDEZ PRIETO MJ, NGUYEN NT, Huang Q, FURMSTON T, KEANE J & ARGUELLO CASTELEIRO M (2024) Low-Code/No-code AI for Machine Translation: democratising generative AI for Blue Transformation. In: Handbook on Smart Health. Studies in Health Technology and Informatics Book Series. Sage.

Abstract
Background: Malnutrition and disease affect human health. Planetary Health emerges as a cross-disciplinary field, emphasising the interconnection of human health and the health of planet Earth. The Blue Transformation is an ambitious strategy for the transformation of aquatic food systems worldwide, aiming to eradicate hunger and poverty and improving the living standards of all. Objective: This study investigates if neural Machine Translation (MT) with generative Artificial Intelligence (AI) may reduce the language barriers, promoting access to relevant information, such as the key terms underpinning the Blue Transformation. Methods: This study evaluates the quality of MT for a Blue Transformation dataset (a multilingual glossary of 188 term pairs and an English vocabulary of 217 unique terms and phraseological expressions) using automatic quality metrics and human judgment to assess the semantic quality (adequacy) of MT. We introduced a low-code/no-code AI approach to MT using open-source python libraries and open-source (free of charge) Large Language Models (LLMs), including multimodal LLMs that can produce content (text or speech) as output (generative AI) for text translation. Results: The study confirms the viability of the low-code/no-code AI approach presented to MT using three lines of python code, revealing a moderate agreement among the three experts about the relative highquality of fully automated MT with Facebook's SeamlessM4T. Conclusion: Our findings aligned with the 2024 AI index report about the lower performance of open-source LLMs in comparison with closed-source LLMs, e.g. OpenAI's GPT4.