Spoken Conversational Search

Abstract

This research investigates a new interface paradigm for interactive information retrieval (IIR) which forces us to shift away from the classic "ten blue links" search engine results page. Instead we investigate how to present search results through a conversation over a speech-only communication channel where no screen is available. Accessing information via speech is becoming increasingly pervasive and is already important for people with a visual impairment. However, presenting search results over a speech-only communication channel is challenging due to cognitive limitations and the transient nature of audio. Studies have indicated that the implementation of speech recognizers and screen readers must be carefully designed and cannot simply be added to an existing system. Therefore the aim of this research is to develop a new interaction framework for effective and efficient IIR over a speech-only channel: a Spoken Conversational Search System (SCSS) which provides a conversational approach to defining user information needs, presenting results and enabling search reformulations. In order to contribute to a more efficient and effective search experience when using a SCSS, we intend for a tighter integration between document search and conversational processes.

Towards a model for spoken conversational search

“…To this end we conducted a study to collect a set of utterances and search interactions from two actors communicating to fulfil an information need: SCSdata [53]. We developed an annotation scheme for SCS, the SCoSAS, analysed this and validated it with inter-rater reliability; further tested it with an independent data set, Microsoft Information-Seeking Conversation data (MISC) [45,46,50]. Our analysis provides insight into the interaction space and design recommendations for further research into SCS.…”

Conversational Search and Recommendation: Introduction to the Special Issue

“…Exchange of meta-information in conversational search is necessary to complete the seeker's state of knowledge. The authors refine the existing definitions of meta-information for conversational search and conduct quantitative and qualitative analysis on MISC [Thomas et al 2017] and SCS datasets [Trippas and Thomas 2019] for analyzing conversational interactions that contain meta-information.…”

“…They include but are not limited to conversational search conceptualization (e.g., Azzopardi et al [2018], Deldjoo et al [2021], and Radlinski and Craswell [2017]), effective conversational query re-writing (e.g., Yu et al [2020]), generating and selecting clarifying questions (e.g., Zamani et al [2020a, c]), conversational preference elicitation (e.g., and Zhang et al [2018]), and understanding user interactions with spoken conversational systems (e.g., Trippas et al [2018Trippas et al [ , 2020). The growing body of work in this area has been supplemented by an increasing number of recent seminars (e.g., Anand et al [2020]), workshops (e.g., , Burtsev et al [2017], Chuklin et al [2018], and Spina et al [2019]), tutorials (e.g., Fu et al [2020] and Gao et al [2018]), shared tasks (e.g., Dalton et al [2019Dalton et al [ , 2020, Ram et al [2018]), and datasets (e.g., Aliannejadi et al [2019], Choi et al [2018], , Reddy et al [2019], Thomas et al [2017], Trippas and Thomas [2019], and Zamani et al [2020b]). Despite this great success, there are still numerous open problems in this domain that require in-depth investigation.…”

VoiSeR: A New Benchmark for Voice-Based Search Refinement

“…There have been prior efforts in creating open multi-turn voice-based search datasets, but because of the lack of effective automated systems for these tasks, the datasets were collected in a lab using a Wizard-of-Oz approach, where a hidden human participant playing the part of the search engine, e.g., (Trippas et al, 2017(Trippas et al, , 2020Vakulenko et al, 2019), and (Trippas and Thomas, 2019). The dataset we contribute complements prior efforts by providing a large-scale collection of voice-based query reformulation, collected in a realistic environment with a real search engine, using automated ASR, thus providing a critical resource for training robust, high-performance models for voice refinement.…”