Poster #34 - Making use of largescale naturalistic audio recording: An introductory tutorial on HomeBank and the DARCLE Annotation Scheme_

• In Event: 2-001 - Poster Session 1 during Breakfast

Abstract

The last decade has seen the emergence of technologies that allow an unprecedented window into young children’s real world auditory experiences, by unobtrusively recording those daily-life experiences on a large scale. HomeBank, a component of the TalkBank system, provides access to a number of corpora of already-collected recordings, as well as associated annotation and transcription, and useful scripts for analysis. HomeBank corpora include both longitudinal and cross-sectional recordings of both typically developing children and special populations such as children with hearing impairment and children born to young mothers. The DARCLE annotation scheme (DAS) was developed as a means to harmonize annotations across corpora collected in diverse languages and with diverse research goals. In this tutorial, I will a) describe the corpora available on HomeBank and how to access them, and b) provide an overview of the DAS and give an illustrative example of how it is currently being used (the ACLEW project). Time will be provided at the end to discuss specific project ideas members of the audience may have for which HomeBank and/or DAS may be appropriate resources.
Proposal Edit Proposal
Over the last few years, there has been a growing network of researchers interested in pursuing questions involving highly naturalistic audio (and sometimes video) recordings of infants’ and young children’s everyday experiences. This research community (DARCLE – Daylong AudioRecording of Child Language Environments) comes in part, but not exclusively, from users of the LENA™ technology for clinical and research purposes. Two resources emergent from the DARCLE community – the HomeBank databank of recordings and transcripts (and its companion resource of scripts for analysis), and the DARCLE annotation scheme – will be described in detail. (N.B. Technical details of LENA usage will not be discussed directly in the tutorial, although we will mention aspects of HomeBank related to LENA, e.g. data outputs originating from LENA algorithms.) Together, these provide a rich and largely untapped resource for examining a broad spectrum of questions about early child development, both specific to speech/language and more broadly. The DEVSEC conference provides a unique opportunity to make these resources more widely known.

HomeBank. The HomeBank system consists of 1. a TalkBank-associated databank of long-running (from several hours up to a full day) audio recordings of infants and young children going about their day, and associated data files, and 2. HomeBankCode, a collection of programs (primarily written in Matlab, Python, Perl, and R) stored in GitHub that are available for researchers to use in their analyses of the HomeBank datasets. In the first part of the workshop, I will describe the various resources that can be found in HomeBank itself, including details of the specific corpora available, and the 3 levels of access for different corpora (fully public, password-protected, and secure-sensitive). I will also go through the steps for gaining access to the password-protected and secure-sensitive corpora. Gaining access involves a brief face-to-face interview, and I will be available both immediately after the workshop and by appointment at other times during the conference to conduct these interviews. I will also describe some of the code resources that other researchers have found helpful in conducting analyses on these datasets.

DARCLE Annotation Scheme. One of the challenges of working with these largescale, highly naturalistic, recordings is that existing transcription and annotation systems are not well-suited to the characteristics of these recordings nor to the research questions being posed. For example, these recordings often involve many speakers and feature highly noisy, speech-intensive periods and long stretches of non-speech, silence and other input (e.g. TV). In addition to transcription, researchers are often interested in demarking broader characteristics of the auditory environment, speaker/addressee identity, gross measures of child vocal maturity, etc. Additionally, the goal of HomeBank (as with other TalkBank databanks) is to allow researchers to make use of multiple corpora to analyze a single research question. In service of this goal, therefore, a group of researchers within DARCLE have developed the DARCLE Annotation Scheme (DAS), with a backbone in ELAN, but using a minCHAT format and easily exportable to other transcription frameworks such as CLAN. This annotation system is designed for use with largescale audio files, and was developed with the goal of being broadly applicable to a variety of research questions and cultural contexts, yet structured to allow for easy comparability across datasets for secondary analysis. Importantly, this comparability also greatly increases the ease of development of automated tools to streamline new secondary analyses across larger datasets. This second component of the workshop will provide an overview of the DAS and describe in detail an example implementation of it for a cross-linguistic/cross-cultural research project. This will allow attendees to see how the DAS might be of benefit for their specific research goals, with the added benefit that further DAS-compatible annotation can be shared back to HomeBank for use by other researchers.

Melanie Soderstrom is Associate Professor in the Department of Psychology at the University of Manitoba. She is a founding member and on the governing board of the DARCLE group, is a consultant on the NSF grant supporting the HomeBank initiative, and contributed to the development of the DARCLE annotation scheme. She has published peer-reviewed articles using largescale audiorecordings and has contributed to multiple workshops on HomeBank.

Author

• Melanie Soderstrom, University of Manitoba
  Presenting Author