PC2 Deep Language Processing Using Construction Grammars
Construction Grammar is an innovative approach to language that focuses strongly on semantics, cognitive processes and learning. As such, it has the potential to lead to more powerful language technologies and more comprehensive accounts of language processing than psychology and computational linguistics are able to offer today. This course will teach its participants how the key concepts of construction grammar can be implemented in computational models for deep language processing that are capable of both comprehension and formulation. Moreover, students will also learn how to implement learning operators for more robust processing, and how to evaluate their grammars. Each session will take the form of a hands-on atelier using Fluid Construction Grammar (FCG), an advanced computational platform for developing constructional processing and learning models.
Session 1: Introduction to FCG + Writing Lexical Constructions
Students are introduced to the core ideas of construction grammar and how they are formalized in Fluid Construction Grammar. They will learn how to write lexical constructions.
Session 2: Writing Grammatical Constructions for Argument Structure
Students will learn how to write more complex grammatical constructions, particularly for handling word order and argument structure.
Session 3: Meta-layer Problem Solving and Learning using Anti- and Pro-unification
Language is abundant with innovations, incomplete phrases or errors. This lecture teaches the participants how to use FCG's meta-layer for achieving more robust language processing and learning.
Session 4: Evaluating Construction Grammars + Wrap-up
Students will learn how to use and expand facilities for evaluating computational construction grammars.
This course will teach its participants how the key concepts of construction grammar can be implemented in computational models for deep language processing that are capable of both comprehension and formulation. Moreover, students will also learn how to implement learning operators for more robust processing, and how to evaluate their grammars.
Adele E. Goldberg (2003). Constructions: A New Theoretical Approach to Language. TRENDS in Cognitive Sciences 7(5):219--224.
Luc Steels (in press). Basics of Fluid Construction Grammar.
Remi van Trijp (in press). A Computational Construction Grammar for English(available soon)
Paul Van Eecke & Katrien Beuls (in press). Meta-layer Problem Solving for Computational Construction Grammar.(available soon)
Tânia Marques & Katrien Beuls (2016). Evaluation Strategies for Computational Construction Grammars. Proceedings of COLING 2016.
Lecture Room 4Course requirements
Students must bring their own laptop with Babel2 installed (which includes the FCG software). Students may receive a notification before the School with a new download link if there are critical updates to the software.
Students must have basic programming skills, preferrably in Common Lisp or related programming languages such as Python.
Students who are unfamiliar with linguistics and construction grammar are advised to also follow the course "SC9 Construction Grammar and Usage-Based Linguistics: Studying Language with Large Corpora by Martin Hilpert."
Remi van Trijp and Paul Van EeckeEmail
Dr Remi van Trijp is the head of the language research unit at Sony Computer Science Laboratory Paris and one of the chief architects of Fluid Construction Grammar. He has pioneered the first constructional processing model for argument structure that works for both language comprehension and formulation, and has developed state-of-the-art agent-based language models for the emergence of grammar. In his latest work, he has focused on how robust models of constructional language processing can be applied for broad-coverage grammars.
Paul Van Eecke is a member of the language research unit at Sony Computer Science Laboratory Paris. He has been involved in the development of Fluid Construction Grammar since 2014. His main research topics include (1) algorithms that make constructional language processing more robust and flexible and (2) algorithms that learn more abstract constructions from concrete observations.