You're reading the public-facing archive of the Category Theory Zulip server.
To join the server you need an invite. Anybody can get an invite by contacting Matteo Capucci at name dot surname at gmail dot com.
For all things related to this archive refer to the same person.
Paolo Perrone (Jul 01 2020 at 20:30):Hello all! This is the thread of discussion for the talk of Giovanni de Felice, Elena Di Lavore, Mario Román and Alexis Toumi, "Functorial Language Games for Question Answering".
Date and time: Friday July 10, 12:10 UTC.
Zoom meeting: https://mit.zoom.us/j/7488874897
YouTube live stream: https://www.youtube.com/watch?v=CNdgmTMMUuw&list=PLCOXjXDLt3pZDHGYOIqtg1m1lLOURjl1Q
Paolo Perrone (Jul 10 2020 at 11:50):This talk will start in 20 minutes. Mind the "new" Zoom address, https://mit.zoom.us/j/7488874897
Jules Hedges (Jul 10 2020 at 12:36):So I have lots to say about this
Julio Song (Jul 10 2020 at 12:36):Alright I will post my earlier question here since we ran out of time in the Zoom meeting: I find the idea of “categorical pragmatics” very appealing (to the extent that we can categorically distinguish semantics and pragmatics)! Are there perhaps any previous work/papers on the topic?
Jules Hedges (Jul 10 2020 at 12:41):I think the next thing that is needed to do anything useful is to talk about playing games, which the original definition of open games isn't able to do. @Toby Smithe discussed a general framework for this in his talk. If you can do this, it should be able to connect to existing work on game theory and pragmatics, language evolution etc. One thing you can do that I'm thinking about heavily, and which @Elena Di Lavore is starting to work on, is this: If you have a (classical) repeated game, you can pretend that it's a Markov game by assuming that your agents have bounded memory. Markov games can be solved efficiently by dynamic programming methods, and that gives a way to get "pragmatic" approximate solutions to the original classical game. And to my understanding, the "theory of play" corresponding to Markov games is precisely what is called reinforcement learning. This is something that can be described using the machinery that already exists in open games: it's optics all the way down
Jules Hedges (Jul 10 2020 at 12:53):I could put it like this: The original definition of open games is about Nash equilibria, and Nash equilibria are themselves not pragmatic
Giovanni de Felice (Jul 10 2020 at 13:09):language-games.pdf
Here are the slides for my talk. And the animation I didn't manage to show can be found at this link: https://github.com/oxford-quantum-group/discopy/blob/master/docs/imgs/unsnake-sentence.gif
Giovanni de Felice (Jul 10 2020 at 13:15):@Jules Hedges We used the old definition of open games for simplicity. But I agree that the more recent definitions in terms of optics would be useful, especially if you want to model some stochastic games (such as the one for word-sense disambiguation I mentioned in my last slide). Also it would be interesting to understand how you can go from an open game to an open learner, replacing the best-response function (which is not pragmatic) with a _better_ response. The concept of Nash equilibrium is still useful because it tells us where the game goes asymptotically.
Giovanni de Felice (Jul 10 2020 at 13:21):@Julio Song Categorical pragmatics doesn't exists yet. There is work on game-theoretic pragmatics, here is a survey with a lot of references: Benz-and-Stevens-2018-Game-Theoretic-Approaches-to-Pragmatics.pdf
Julio Song (Jul 10 2020 at 13:24):Very cool stuff indeed. Thanks for the paper (and for the fantastic talk)!:pray:
Paolo Perrone (Jul 10 2020 at 16:20):Video here!
https://www.youtube.com/watch?v=9_rAyzSoC1c&list=PLCOXjXDLt3pYot9VNdLlZqGajHyZUywdI
