Stergos Afantenos (Université Paul Sabatier, IRIT)
Analogies: a brief attempt at understanding what they are and an even briefer one at detecting analogies between pairs of sentences.
Abstract: Analogies have been preoccupying thinkers at least since the post-mycenaean period of the Greeks (e.g. Euclid, Theon, Aristotle, etc). More recently they have been characterized as being at the core of cognition (Hofstadter 2001, Hofstadter and Sanders 2013). In this talk we will attempt to briefly shed some light on the various points of view via which analogies have been approached, without being of course exhaustive. For the second part of this presentation we will retain the classical approach of analogies as proportions between four terms a:b::c:d (a is to b as c is to d) where a, b, c and d are sentences. We will present a series of experiments trying to understand whether the classical postulates hold for analogies between sentences, focusing mostly on the central permutation. We will also explore how well recurrent neural networks and transformer based ones are capable at detecting analogies between sentences.
Short Bio: Stergos Afantenos is an Associate Professor at the Université Paul Sabatier at Toulouse. His research focuses on Natural Language Processing. He has worked on the tasks of Summarization for events that evolve through time, Structured Prediction methods for learning Discourse Structures in text, Multi-Party Written Dialogues (chats) and Argumentation, as well as the detection of Fake News. Lately he has been focusing his research on Analogy Making.
Claire Gardent (CNRS, LORIA)
Neural Approaches to Analogies [slides]
Abstract: Over the last decade, neural methods have been proposed to process analogies i.e., quadruplets of the form A:B :: C:D such that the relation that holds between A and B also hold between C and D. In this talk, I will review some of this work focusing on the models used, on the representations proposed and on the motivations underlying the different proposals. The talk will focus mostly on neural approaches to analogy for text, structured data and images.
Short Bio: Claire Gardent is a senior research scientist at the French National Center for Scientific Research (CNRS) and is based at the LORIA Computer Science research unit in Nancy, France. She works in the field of Natural Language Processing with a particular interest for Natural Language Generation. In 2017, she launched the WebNLG challenge, a shared task where the goal is to generate text from Knowledge Base fragments. She has proposed neural models for simplification and summarisation; for the generation of long form documents such as multi-document summaries and Wikipedia articles; for multilingual generation from Abstract Meaning Representations and for response generation in dialog. She currently heads the AI XNLG Chair on multi-lingual, multi-source NLG and the CNRS LIFT Research Network on Computational, Formal and Field Linguistics. In 2022, she was awarded the CNRS Silver Medal.
Dave Raggett (W3C/ERCIM)
The application of qualitative metadata to analogical reasoning [paper] [slides]
Abstract: Analogical reasoning can be used for plausible inferences based upon direct similarities or structural mappings involving properties and relationships. This can be implemented on top of a combination of symbolic knowledge plus sub-symbolic qualitative metadata, with matching based upon structural or causal similarities, and noticing interesting differences, in essence, abstracting from similarities and dissimilarities, and will be applied to examples of the form “A is to B as C is to ?X”. A further challenge is to support the use of literal and figurative analogies in natural language, e.g. comparing life to the wheel of fortune, when you want to highlight the role of chance. An easy to use syntax will be presented for expressing knowledge, along with a web-based proof of concept demonstrator, and a unifying cognitive architecture for human-like AI. This builds upon work by Alan Colins on plausible reasoning and Dedre Gentner on analogies.
Short bio: Dr. Raggett is a senior researcher for W3C/ERCIM and has been involved in developing Web standards since the early nineties. He has participated in numerous EU projects and is especially interested in the study of human-like AI and developing practical techniques to mimic human perception, memory, reasoning, learning and action, building upon the hard-won insights by researchers in the fields of experimental psychology and the cognitive sciences. He co-chairs the W3C Cognitive AI Community Group.
