AI & Translation: Comparing cognitive processes in the human brain to AI models – using the learning conditions of interpreters as a testing ground.
Priority Area: Human and Machine The project compares processes in Large Language Models to those of the human brain.Focusing on the abilities of information uptake, consolidation and knowledge retrieval, we investigate the learning conditions of interpreters, who need to efficiently encode and consolidate new knowledge in new, special domains, to successfully produce target language while analyzing incoming source language. However, while LLMs show strong linguistic skills, they lack a memory component for efficient encoding, consolidation and retrieval of knowledge. To overcome both issues we design new learning tools for interpreters and a dedicated modularized LLM memory component, combining the expertise of natural language processing, interpreting science and cognitive neuroscience. With this we aim to simulate and compare processes of knoweldge processing in AI models and the human brain, for better understanding of both: cognition and AI models. Project leads:
Prof. Kerstin Kunz, Dept. for Translation and Interpreting, Heidelberg University,
Prof. Anette Frank, Dept. of Computational Linguisticsy, Heidelberg University,
PD Dr. Gordon Feld, Zentralinstitut für Seelische Gesundheit, Mannheim Project funded by: Exzellenzstrategie (BMBF and MWK Baden-Württemberg)