Saggitarius: A DSL for Specifying Grammatical Domains
summary
Saggitarius defines metagrammars (DSLs whose denotation is a set of CFGs with existential parameters and constraints) and uses a generated giant grammar plus semiring parsing to extract rule formulas that feed a MaxSAT solver to pick the best grammar. The system returns a satisfying grammar in seconds on typical inputs, succeeds on a 110-problem benchmark, and infers CSV dialects for 84% of benchmarks within 60 seconds, trading some speed for generality while matching specialized-tool accuracy.
AI-written summary · about
chapters (11)
- 0:00Introduction to Sagittarius
- 1:26Grammar induction difficulty
- 3:13Constrained grammar induction
- 6:26Formalizing metagrammars
- 7:10Metagrammar for CSV induction
- 9:58n-column CSV grammars
- 11:22Inducing from metagrammars
- 13:07Evaluating on benchmarks
- 14:51Performance vs ProIn
- 15:33Sample efficiency & future work
- 15:54PhD candidates at SFU
related talks
46mGrammar a BNF like Ruby DSL Parsing
15mSynthesizing DSLs for Few-Shot Learning
21mSynthesizing Formal Semantics from Executable Interpreters
20mThe Lands Platform: Lan.guages and D.omain S.yntax
44mJava Code Analysis with Database and Domain Specific Language
24mDesigning a DSL That Perfectly Solves Your Problem
29mGrammar Inference for Ad Hoc Parsers
37mThe Art of Parsing
