What's in a Word Graph --- Evaluation and Enhancement of Word Lattices

During the last few years, word graphs have been gain-ing increasing interest within the speech community as the primary interface between speech recognizers and lan-guage processing modules. Both development and evalu-ation of graph-producing speech decoders require gener-ally accepted measures of word graph quality. While the notion of recognition accuracy can easily be extended to word graphs, a meaningful measure of word graph size has not yet surfaced. We argue, that the number of derivation steps a theo-retical parser would need to process all unique sub-paths in a graph could provide a measure that is both application oriented enough to be meaningful and general enough to allow a useful comparison of word recognizers across dif-ferent applications. 1