Paris seminar on
Computation and Linguistic Theory
Computation and Linguistic Theory (CaLT)
Computation and Linguistic Theory (CaLT) is a recent but fast growing field of research at the intersection of theoretical linguistics, machine learning, and the theory of computation. CaLT explores the core conjecture that natural language has a rich and specific structure which allows for provably exact and efficient algorithms for language production, interpretation, and learning. On the one hand, CaLT tries to distill the algorithmic implications of powerful typological linguistic generalizations; on the other hand, it uses computational limitations, negative learnability results, and algorithmic bottlenecks as heuristic principles for the discovery of typological generalizations.
Goal of the seminar
This seminar offers mini-courses on CaLT targeted mainly to the linguistic community. The scope is construed broadly to include any areas of theoretical linguistics: phonology, semantics, syntax, historical linguistics, etcetera. Each mini-course consists of a few sessions with little time constraints, in a completely informal setting where attendees can freely ask questions and the presenters can freely digress at the blackboard. Readings are made available in advance.
Events for the academic year 2018/2019:
[1]
[2]
[3]
Workshop Analyzing typological structure: from categorical to probabilistic phonology
Tools for typological analysis usually only apply to categorical models. Yet, over the past two decades many phonologists have turned to quantitative data and worked extensively on patterns of stochastic variation and gradient acceptability. Such analyses often invoke probabilistic grammars, such as Stochastic OT (Boersma and Hayes 2001), Noisy Harmonic Grammar (Boersma and Pater 2016), and MaxEnt (Goldwater and Johnson 2003, Hayes and Wilson 2008). This work typically has the goal of showing that the models are rich enough to avoid undergeneration, but less attention has been paid to the question of overgeneration. The key question addressed by this workshop is how to analyze the typological structure induced by probabilistic models. This question is not trivial: while the typologies predicted by categorical phonology are usually finite, probabilistic frameworks generate an infinite family of different probability distributions.
Reading group on the learnability of syntax
This reading group discusses a broad range of studies on the learnability of syntax, including formal learnability results, computer simulations, and experimental findings from language acquisition. The seminar meets six times from September to November, on Mondays from 17h00 to 19h00 (possibly, followed by drinks). Meetings will take place alternatively at SFL (59 Rue Pouchet, 75017 Paris) and LLF (5, rue Thomas Mann, 75013 Paris). All sessions will be accessible also via the internet at https://rendez-vous.renater.fr/learnabilityofsyntax (preferably using Chrome). The program used to connect over Internet allows live interaction. The seminar will be conducted in English. The reading group is co-organized by Carlo Cecchetto (SFL), Caterina Donati (LLF), and Giorgio Magri (SFL).
M2 course Computational Phonology
This is an M2 course on phonology from a computational perspective. In the fall 2018, the course meets on Tuesdays from 13:30 to 15:30, in room 309, at the University of Paris 7. The topics covered this year are probabilistic constraint-based phonology and deep neural networks.
Archive
Information concerning events and mini-courses organized in previous academic years is available here.
Practical information
It is free to attend the seminar, and no registration is necessary -- you can simply show up on the day of the meeting. Information concerning meeting time and place is made available on the individual page of each meeting (as it may change with each meeting).
For any questions
If you would like to be added to the mailing list of the seminar or if you have any questions, please contact Giorgio Magri at magrigrg@gmail.com.
Acknowledgments
Financial support for this seminar has been provided by the MIT France Seed Fund, the Agence Nationale de la Recherche, and the France-Stanford Center for Interdisciplinary Studies.