3IA PhD/Postdoc Seminar #9

Natural Language Argumentation Quality Assessment maps

Abstract: With online fora increasingly serving as the primary media for argument and debate, the automatic processing of such data is rapidly growing in importance. Several techniques for argument mining already exist but very few assess the quality of said arguments. In this work we propose argument quality assessment guidelines for persuasive essays in order to understand and characterize the quality of the arguments. Following the guidelines we annotated a new quality dataset on persuasive essays and proposed different models in order to automatically assess the quality features.

Cortex-inspired multimodal AI approach based on self-organizing 

Abstract: Among artificial intelligence architectures, brain-inspired architectures are attracting more and more interest. Classical neural networks have long become the state of the art in the domain of artificial intelligence, but other models, closer to natural neural networks, are beginning to experience very promising results. In this work, carried as part of the 3IA Institute's strategic axis "AI for Biology and AI Inspired by Biology", we are developing a framework inspired by the work of the cerebral cortex and exploring the self-organization of information and error correction of one modality using information from another. The proposed neural architecture uses a rather powerful unsupervised learning methods. This original approach will conduct to the development of digital technologies breaking with the Von-Neumann paradigm taking advantage of embodied and brain-inspired computing, to go further in the exploration of an "electronic brain".

Open discussion on the two contributions