Redundancy-free Island Parsing of Word Graphs

Island parsing is a bidirectional parsing strategy mostly used in speech analysis, as well as in applications where robustness is highly relevant and/or processing resources are limited. Although there exists an efficient redundancy-free island parsing algorithm for string input, it has not yet been applied to word graph input, an application which is central for speech analysis systems. This paper describes how the established algorithm can be generalized from string input to word graphs, increasing its flexibility by integrating the selection of island seeds into the search process inherent to parsing.