摘要：Legislative proceedings present a rich source of multidimensional information that is crucialto citizens and journalists in a democratic system.At present, no fully automated solutionexists that is capable of capturing all the necessary information during such proceedings.Evenif professional-quality automated transcriptions existed, other tasks such as speaker or vrhetorical position identifications are not fully automatable.This work focuses on improvingand evaluating the transcription software used by the Digital Democracy initiative, namedTranscription Tool.Human transcribers work to up-level state legislative proceedings usingthis tool.Five phases of tool improvements are introduced and for each phase, the resultingchange in efficiency is measured. We investigate over 12,00o individual transcription sessions(2,30o hours of video), where each session is the record of one bill discussion. A set of about3,200 sessions belonging to a single cohort of 2o transcribers is further evaluated.Throughintroduction of new tool features, human-assisted transcription efficiency can be improved by19.4% over five phases.Furthermore, investigation into transcriber usage patterns revealsthat transcription time is composed of passive time, speaker identification, text correction,tool startup, as well as splitting and merging utterances. We analyze and rank these as acontribution.
