Koen Claessen
Chalmers University of Technology, Sweden
Ladda ner artikel
Ingår i: Proceedings of the NoDaLiDa 2017 Workshop on Constraint Grammar - Methods, Tools and Applications, 22 May 2017, Gothenburg, Sweden
Linköping Electronic Conference Proceedings 140:1, s. 1-5
NEALT Proceedings Series 33:1, p. 1-5
Publicerad: 2017-07-06
ISBN: 978-91-7685-465-5
ISSN: 1650-3686 (tryckt), 1650-3740 (online)
We present a method for automatic synthesis of Constraint Grammar rules from a corpus. The method is designed to aid grammar writers interactively in coming up with new rules or improving existing ones, but also to synthesize a whole set of rules autonomously from scratch. A SAT-solver is used to compute the “best” rule at each stage, according to some measure. Initial experimental results on two corpora look promising: suggesting one rule to improve an existing set of rules typically takes seconds; synthesizing a whole set of rules takes minutes to hours.
[1] Samuelsson, Tapanainen, Voutilainen. Inducing Constraint Grammars. International Colloquium on Grammatical Inference. 1996.
[2] Lindberg, Eineborg. Learning Constraint-grammar style disambiguation rules using Inductive Logic Programming. COLING. 1998.
[3] Sfrent. Machine Learning of Rules for Part of Speech Tagging. MSc. thesis. Imperial College London. 2014.
[4] Een, Sörensson. An Extensible SAT-solver. SAT conference. 2003.
