Please use this identifier to cite or link to this item: doi:10.22028/D291-25260
Title: What's in a word graph evaluation and enhancement of word lattices
Author(s): Amtrup, Jan W.
Heine, Henrik
Jost, Uwe
Language: English
Year of Publication: 1996
SWD key words: Künstliche Intelligenz
Free key words: artificial intelligence
DDC notations: 004 Computer science, internet
Publikation type: Report
Abstract: During the last few years, word graphs have been gaining increasing interest within the speech community as the primary interface between speech recognizers and language processing modules. Both development and evaluation of graph-producing speech decoders require generally 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 theoretical 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 different applications. This paper discusses various measures that are used, or could be used, to measure word graph quality. Using real-life data (word graphs evaluated in the 1996 Verbmobil acoustic evaluation), it is demonstrated how different measures can affect evaluation results. Finally, some algorithms are presented that can significantly improve word graph quality with respect to specific quality measures.
Link to this record: urn:nbn:de:bsz:291-scidok-53757
Series name: Vm-Report / Verbmobil, Verbundvorhaben, [Deutsches Forschungszentrum für Künstliche Intelligenz]
Series volume: 186
Date of registration: 7-Jun-2013
Faculty: SE - Sonstige Einrichtungen
Department: SE - DFKI Deutsches Forschungszentrum für Künstliche Intelligenz
Collections:SciDok - Der Wissenschaftsserver der Universität des Saarlandes

Files for this record:
File Description SizeFormat 
Vm_186.pdf16,87 MBAdobe PDFView/Open

Items in SciDok are protected by copyright, with all rights reserved, unless otherwise indicated.