Crowdsourcing User Preferences and Query Judgments for Speech-Only Search

Abstract

The analysis of spoken interactions between human participants, for the purpose of understanding tactics and strategies for performing a task such as collaborative search, involves a number of steps, completing with the identification and classification of operations designed to interpret search results or to progress the interaction to a successful outcome. This paper provides guidelines for transcription and analysis of conversational search recordings which are critical steps for the overall goal in identifying and classifying operations in conversational search. Many decisions need to be made when transcribing utterances, such as determining pauses, punctuation, or when a speaker stops or starts uttering. To our knowledge, no publicly available guidelines are available for conversational search transcription and analysis, thus we adapted several different guidelines from other fields. In this paper we provide fundamental principles for transcriptions and a protocol based on those principles. We also introduce annotation tools, transcription quality assurance guidance, analysis and coding instructions.

Publication
SIGIR 1st International Workshop on Conversational Approaches to Information Retrieval