We are excited to announce that the NLP Research Team at the Centre for Artificial Intelligence (CAI) has successfully published two papers at this year's ACL main conference. They were selected out of more than 5'000 submissions from research teams all over the world.

Paper 1: Error-preserving Automatic Speech Recognition of Young English Learners’ Language

This paper introduces a novel dataset and methodology for training a Speech-To-Text (STT) module that accurately captures and preserves language errors made by young English learners in Switzerland. The research addresses the need for educational tools that provide realistic feedback, enhancing language learning for children in Swiss schools. This STT module is a component of a chatbot designed to assist in English practice. The study is part of the ChaLL Project and was partially developed by Janick Michot within his MSE Masters studies.

Paper 2: Favi-Score: A Measure for Favoritism in Automated Preference Ratings for Generative AI Evaluation

This paper presents the Favi score, a novel metric developed to detect favoritism in automated performance evaluations of generative AI systems. The Favi score quantifies biases in existing evaluation metrics, revealing how they can disproportionately favor or disfavor specific machine translation systems. This work, part of the UniVal Project, contributes to the fair assessment of generative AI systems and was partically developed by Pius von Däniken within his Ph.D. research.

Our team will present these findings at the ACL conference in Bangkok from August 11-16, 2024. We look forward to engaging with the global research community and discussing our latest advancements.