Geneva, 25 November 2011.ppt

Geneva,uracyinrespeakingPabloRomero-Fresco(RoehamptonUniversity,CAIACresearchcentre)JuanMartínez(Respeakingconsultant)ITU-TWorkshopon“municationsrelayservicesforpersonswithdisabilities”(Geneva,25November2011)uracyinRespeaking97-98%accuracyBasicrequirementsforamodel1)Functionalandeasytoapply 2)IncludethebasicprinciplesofWERcalculationsinSR3)Differentprogrammes,differentediting 4)uraterespeaking paresubtitleswithoriginalspokentext 6)Includeotherrelevantinfo(delay,position,speed) 7)ProvidebothpercentageandfoodforthoughtintrainingTraditionalWERmethodsUSNationalInstituteofStandardsandTechnology N-uracyRate------------------------×100=% NBut...Well,youknow,youhavetotryandputoutagoodperformance,Imean,yeah,it’skindofasteppingstone,isn’tit,really?’ N-uracyRate------------------------×100=16% NBut...Well,youknow,youhavetotryandputoutagoodperformance,Imean,yeah,it’skindofasteppingstone,isn’tit,really?’=essibilityFocusGroupN–E–uracy------------------------×100=% NCorrecteditions:Seriouserrors:Assessment:NERModel205–3–2Accuracy------------------------×100=% 205NERModel226–13–1 Accuracy------------------------×100=% 226 Assessment:poorediting(notquantity,butquality)NERModel257–1–13 Accuracy------------------------×100=% 257 Assessment:poorrecognition(includingseriousmistakes)