String Shuffling over a Gap between Parsing and Plan Recognition

We propose a new probabilistic plan recognition algorithm YR based on an extension of Tomita’s Generalized LR (GLR) parser for grammars enriched with the shuffle operator. YR significantly outperforms previous approaches based on topdown parsers, shows more consistent run times among similar libraries, and degrades more gracefully as plan library complexity increases. YR also lifts the restrictions on leftrecursion imposed by approaches based on top-down parsing algorithms. We further propose that context-free shuffle grammars, more than traditional context-free grammars, should be seen as the appropriate analogue of HTN plan libraries in the correspondence of plan recognition and parsing.