Integrating Generic and Contextual Knowledge
15/05/2020: Note that you can only access our Slack space if you use the invitation link you received upon registration. If you did not receive/save the link, please get in touch with us at firstname.lastname@example.org
11/05/2020: Our first pre-recorded keynote talks are now available
11/05/2020: Our Slack space is now open! To register, click on the invitation link you received upon registering.
22/04/2020: Registration for the virtual symposium is open: Registration
30/03/2020: GeCKo will be turned into an online event in light of COVID-19. It will still take place on the same day, May 18th. Details to be announced later.
For both humans and computational models, it is essential to be able to abstract away from specific instances to broader categories, e.g. to build a generic concept for "bird" from instances of birds. But modelling specific situations is equally essential; for instance to be able to understand the sentence "that bird is about to peck you" and react accordingly. Current data-driven models excel at distilling generic knowledge acquired over time, such as knowledge reflected by which words or objects tend to co-occur. However, they can still struggle with specific situations, which require fast recognition of individual elements, such as entities, events, and relationships, and the ability to reason about them. In the literature, the relation between generic and contextual knowledge surfaces in dichotomies such as:
- lexical/conceptual/generic vs. contextual(ized)/grounded/situated/referential/utterance-specific information;
- category/type/kind vs. instance/token;
- semantics vs. pragmatics;
- semantic vs. episodic memory;
- slow vs. fast learning.
The challenge of how to combine generic knowledge with situation-specific information arises in many applications of Computational Linguistics and connected areas and applications, such as Machine Translation, Natural Language Inference, and Language and Vision. We think that it would be beneficial to address this challenge in an integrative fashion, drawing inspiration from across the field. The GeCKo symposium seeks to 1) understand the issues involved in the integration of generic and situation-specific information in Computational Linguistics, across applications and research areas; 2) identify ways forward; and 3) cross-fertilize Computational Linguistics with Machine Learning, Linguistics, and Cognitive Science researchers working at this junction.
The GeCKo symposium will feature talks by invited speakers, contributed talks, as well as a poster session preceded by lightning talks. We seek contributions of published work as well as new research, of the following, non-exclusive types:
- Analysis: what kind of situation-specific information can current models capture; what can’t they?
- Modelling: models and/or tasks aimed at integrating generic and situation-specific information in any area related to CL/NLP.
- Cross-fertilizing: how can computational models of language integrate current findings in Linguistics, Cognitive Science, and Machine Learning about generic vs. situation-specific information? How can our results inform those fields in turn?
- Position papers and surveys outlining the issues involved and ways forward.
- Submission closed as of February
7th14th 2020, 11:59PM UTC-12:00 ("anywhere on Earth") Notification of acceptance: March 13th 2020 Decision of whether to postpone the symposium in relation to COVID-19: March 30th 2020
- Registration opens: April 22nd 2020
- GeCKo symposium: May 18th 2020