Skip to main content

TS

A system to make machine analysis of communication available and accessible to communications researchers and non-technologists.

Authors

William Billingsley, Cindy Gallois, Andrew Smith and Marcus Watson

NICTA

Abstract

Analysing a conversation or a communication episode is a central activity in many fields of research, and it is also a key research area in many technology-based fields. Artificial Intelligence, Natural Language Processing, and Human-Computer Interaction research groups have invented many different techniques for analysing computable aspects of human expression, such as co-occurrence patterns of words, facial gestures, and measures of voice quality. At present, these tools are used piecemeal by communications researchers in their work. The problem is that each tool adds information about a single aspect of the conversation (e.g., lexical analysis works only on the transcript and does not take into account behaviours like pause durations or the vocal stress and other voice qualities of the speakers). In addition, the tool a researcher is using risks being the main way that researcher’s questions are framed, and thus the kinds of analysis and results the research will find. We have developed computer software that can bring together many different machine analysis tools, allowing their outputs to be visualised together. In this package, analysis tools are integrated as plug-ins. When a researcher wishes to study a communication episode (e.g., video with transcript), the software automatically processes it through the analysis plug-ins. The user can then open the viewer to interact with visualisations of the different analyses, together with the video and transcript, on a common timeline. The system supports combined analysis, or plug-ins and visualisations can use the output of other plug-ins in their own analysis. For example, a visualisation might identify the main themes in a transcript, based on co-occurrences of words, and then allow visualisation of their association with the level of vocal stress at each time period. The software is designed to be extensible, and new analysis plug-ins, visualisations, workflows, and configuration templates can be added. In this paper, we demonstrate the software and discuss the development process, including challenges like the need to find common units of time that are meaningful to researchers. Our hope for the software is that it will make machine analysis of communication more readily accessible, and that it will enable a closer link between technology-based and behaviourally-based researchers into language and communication. We also hope to make communication analysis more accessible to non-research users in ways that will be meaningful to them. For example, doctors (who have only limited awareness of communication analysis techniques) want to understand more about their interactions with their patients, and this tool may allow more access to research findings and interventions for them.

BibTeX Entry

  @inproceedings{Billingsley_GSW_10_3,
    booktitle        = {International Conference on Language and Social Psychology},
    author           = {Billingsley, William and Gallois, Cindy and Smith, Andrew and Watson, Marcus},
    month            = jun,
    year             = {2010},
    title            = {A system to make machine analysis of communication available and accessible to communications
                        researchers and non-technologists.},
    address          = {Brisbane, Australia}
  }

Download

Served by Apache on Linux on seL4.