Titre : Towards Understanding Understanding: Dialogue, Robots, and Meaning
Résumé:
Hallmarks of intelligence include the ability to acquire, represent, understand, and produce natural language. Although recent efforts in data-driven, machine learning, and deep learning methods have advanced natural language processing applications, important challenges remain. In my talk, I'll give an overview of general trends in understanding language on machines, what we can learn from children who acquire language seemingly with minimal effort, and what that means for future research. I will then explain my own research on grounding language into different physical modalities, what role emotion could play, and the potential importance of embodiment.
Bio:
Casey Kennington is an assistant professor in the Department of Computer Science at Boise State University. He completed his PhD in Linguistics at Bielefeld University in Germany and his masters degrees in Computational Linguistics and Cognitive Science, respectively, from Saarland University in Saarbrücken, Germany, and Nancy 2 University in Nancy, France. His research at Boise State University brings together computer science, machine/deep learning, human-robot interaction, natural language processing, spoken dialogue systems, child development, and cognitive science.
Plusieurs auteurs ont noté que le fonctionnement cognitif humain est sensible à la complexité des structures (au sens de la taille de description minimale). La Théorie de la Simplicité est construite sur l’observation que de nombreux phénomènes cognitifs liés à la surprise, à l’intérêt narratif, à la pertinence, à l’intensité émotionnelle ou encore au jugement moral s’analysent sur la base d’une différence entre la complexité attendue et la complexité observée. Nous nous demanderons pourquoi cette capacité à repérer l’inattendu par la baisse de complexité est apparue au cours de l’évolution.
• Dessalles, J.-L. (2013). Algorithmic simplicity and relevance. In D. L. Dowe (Ed.), Algorithmic probability and friends - LNAI 7070, 119-130. Berlin, D: Springer Verlag.
• Saillenfest, A. & Dessalles, J.-L. (2015). Some probability judgements may rely on complexity assessments. COGSCI-2015, 2069-2074. Austin, TX: Cognitive Science Society.
Jean-Louis Dessalles est l’auteur de la Théorie de la Simplicité, qui offre un cadre théorique pour représenter la pertinence des paroles et des actions. Il travaille également sur la modélisation des signaux sociaux liés à l’origine du langage humain. Il est l’auteur de plusieurs livres : Why We Talk, La pertinence et ses origines cognitives, Le fil de la vie et récemment Des intelligences TRES artificielles.
Titre : Grounding word meanings in perceptual experience: A computer-vision approach
Résumé :
For about two decades, the fields of cognitive science and psychology have employed distributional semantic models such as LSA (Latent Semantic Analysis) as powerful computational models of semantic representation. These language-based models build meaning representations from the distributional patterns of words from large collections of natural text, and thus from approximations of the actual input experienced by humans. Until recently, this stood in contrast to computational models incorporating sensorimotor information, which was often approximated via participant ratings on sensorimotor features – the outcome of human experience. However, advancements in the field of computer vision allow us to model vision-based representations directly from visual input as approximated by large datasets of images. In this talk, I will present recent studies in which we employ these vision-based representations to investigate automatic (visual) sensorimotor activation (i) during word processing and (ii) in conceptual combination. In addition, I will present how a systematic mapping can be established between language-based and the vision-based representations, thus implementing a possible mechanism for the visual grounding of non-experienced concepts.
Références:
• Günther, F., Petilli, M. A., & Marelli, M. (2020). Semantic transparency is not invisibility: A computational model of perceptually-grounded conceptual combination during word processing. Journal of Memory and Language, 112, 104104. https://www.sciencedirect.com/science/article/pii/S0749596X20300188#f0005 preprint: https://psyarxiv.com/7dvpw/ • Petilli, M. A., Günther, F., Vergallito, A., Ciapparelli, M., & Marelli, M. (2019). Data-driven computational models reveal perceptual simulation in word comprehension. psyArXiV preprint. preprint: https://psyarxiv.com/98z72/
• Günther, F., Petilli, M. A., Vergallito, A., & Marelli, M. (under revision). Images of the unseen: Extrapolating visual representations for abstract and concrete words in a data-driven computational model. preprint: https://osf.io/45hdz/?view_only=c792b6eb276f413fac5a533d7a098976 (the file manuscript.pdf)
Bio:
Fritz Günther (born 1989 in Erfurt, Germany), studied Psychology and Mathematics in Tübingen, Germany (2008-2013) and did his PhD in Cognitive Science also in Tübingen (2013-2017), under the supervision of Barbara Kaup. During his PhD, he received a grant for a research stay abroad in Trento, Italy in summer 2015, under the supervision of Marco Baroni. From 2018 to 2020, he received a research grant for a PostDoc position at Marco Marelli’s lab at the University Milano-Bicocca in Milan, Italy. In March 2020, he returned to Tübingen to work as a PostDoc.
Titre : Grounding word meanings in perceptual experience: A computer-vision approach
Résumé:
Titre : Grounding word meanings in perceptual experience: A computer-vision approach Résumé : For about two decades, the fields of cognitive science and psychology have employed distributional semantic models such as LSA (Latent Semantic Analysis) as powerful computational models of semantic representation. These language-based models build meaning representations from the distributional patterns of words from large collections of natural text, and thus from approximations of the actual input experienced by humans. Until recently, this stood in contrast to computational models incorporating sensorimotor information, which was often approximated via participant ratings on sensorimotor features – the outcome of human experience. However, advancements in the field of computer vision allow us to model vision-based representations directly from visual input as approximated by large datasets of images. In this talk, I will present recent studies in which we employ these vision-based representations to investigate automatic (visual) sensorimotor activation (i) during word processing and (ii) in conceptual combination. In addition, I will present how a systematic mapping can be established between language-based and the vision-based representations, thus implementing a possible mechanism for the visual grounding of non-experienced concepts.
Références:
• Günther, F., Petilli, M. A., & Marelli, M. (2020). Semantic transparency is not invisibility: A computational model of perceptually-grounded conceptual combination during word processing. Journal of Memory and Language, 112, 104104.
preprint: https://psyarxiv.com/98z72/ • Günther, F., Petilli, M. A., Vergallito, A., & Marelli, M. (under revision). Images of the unseen: Extrapolating visual representations for abstract and concrete words in a data-driven computational model.
Fritz Günther (born 1989 in Erfurt, Germany), studied Psychology and Mathematics in Tübingen, Germany (2008-2013) and did his PhD in Cognitive Science also in Tübingen (2013-2017), under the supervision of Barbara Kaup. During his PhD, he received a grant for a research stay abroad in Trento, Italy in summer 2015, under the supervision of Marco Baroni. From 2018 to 2020, he received a research grant for a PostDoc position at Marco Marelli’s lab at the University Milano-Bicocca in Milan, Italy. In March 2020, he returned to Tübingen to work as a PostDoc.
