Upcoming event summary – ALS 2022 and SST 2022

In November and December, the Hub will be presenting at two upcoming conferences. These are:

ALS 2022 – The Australian Linguistics Society annual conference, to be held at The University of Melbourne (Nov 30- Dec 2).

SST 2022 – The 18th Australasian Speech Science and Technology conference, to be held at ANU in Canberra (Dec 13-16).

At ALS in Melbourne, we will be presenting a paper called  Developing an accountable evidence-based process for producing reliable transcripts of indistinct forensic audio. This will be on Day 2 of the conference, on Thursday December 1st at midday, and is a paper authored with our colleague Ute Knoch from the Language Testing Research Centre (LTRC). Our ALS presentation is a joint Hub/LTRC research paper which focuses on the results of a study in which we asked participants to transcribe (cold) forensic-like audio. Even though our transcribers were very experienced in transcription, this was an extremely difficult task due to the nature of the audio. We then worked on the development of a process for scoring transcriptions – this was challenging but we have made some good progress in this space. This project has kept us very busy in 2022 and we are looking forward to sharing preliminary results. The abstract with some more detail can be found further below, and the ALS program can be found here.

The SST conference in Canberra has a day devoted to tutorials – Tuesday December 13. Tutorials are described as “intensive instructional sessions that provide a comprehensive introduction to established or emerging research topics of interest to the SST community”. Our contribution will be in the afternoon(1.30-5pm), a tutorial called Specifying new scientific knowledge required by forensic applications.  In this tutorial we will be focusing firstly on legal procedures governing admission and use of speech evidence in criminal trials, as well as specifying scientific knowledge needed to provide expert evidence in ways that help juries use it effectively. We will be talking about some research topics which can contribute to answering problems in forensics. Our abstract is available further below, and you can read about the other tutorials on this page.

As well as the tutorial, the Hub will be participating in a special session at SST on Wednesday December 14 at 11am, called Multi-disciplinary approaches to forensic speech science: from different starting points to a shared goal. This is organised by Yuko Kinoshita, who has collaborated previously with Helen (read about some of their previous research here). There will be four talks in this special session (see details here), and there will be some time set aside for discussion. We expect this to have good participation from the SST audience, as well as the ASSTA Forensic Speech Science Committee (Helen is Chair of this committee, Yuko is Deputy Chair, Debbie is a member).

ALS ABSTRACT – FRASER, LOAKES & KNOCH

Developing an accountable evidence-based process for producing reliable transcripts of indistinct forensic audio

Covert recordings provide powerful evidence in criminal trials. The problem is that they are often of extremely poor quality, to the extent that they cannot be understood without the assistance of a transcript (Fraser and Loakes 2020). Current Australian law allows transcripts to be provided by police investigating the case (French and Fraser 2018). This has been shown to be unacceptable, even when safeguards embodied in legal procedure are followed properly (Fraser 2018). In 2017, ALS led a ‘call to action’ asking the judiciary to review and reform legal procedures for handling transcripts of indistinct forensic audio in English and other languages.

It is expected then, that at some point police transcripts will no longer be allowed, making it necessary for linguistics to develop evidence-based processes for providing demonstrably reliable transcripts. This requires more than just applying existing scientific knowledge. While transcripts are used frequently in many branches of linguistics, we are rarely called upon to transcribe extremely indistinct audio when we do not know the content and the context is potentially misleading (Fraser 2022a).

The present paper investigates how transcripts with demonstrable reliability can be achieved, drawing on insights from language testing research (Knoch and McQueen 2020). The paper offers results of a preliminary study using a 2-minute section of an indistinct forensic-like recording, for which we were able to prepare a reliable master transcript. Forty participants recruited via the ALS mailing list each transcribed the audio, with no prior knowledge of the content or context. Each participant’s transcript was divided into intonation phrases (IPs) and each IP was scored against the master transcript, with four ratings: a global rating and three analysis ratings (the latter showing how much of each IP was misinterpreted, missing or added) to give an impression of what factors contributed to the global score.

Results confirm that it is unrealistic to expect individual transcribers with no contextual information to produce demonstrably reliable transcripts (cf. Love and Wright 2021). Rather it is necessary to ensure that accredited transcribers follow an evidence-based process, designed and managed by experts (Fraser 2022b) – in a manner similar to those used by other responsible forensic sciences such as DNA analysis. The paper concludes with next steps for developing the evidence-based process, seeking discussion and input from ALS members.

ALS abstract references

Fraser, H. (2018). Forensic transcription: How confident false beliefs about language and speech threaten the right to a fair trial in Australia. Australian Journal of Linguistics, 38(4), 586–606.

Fraser, H. (2022a). A framework for deciding how to create and evaluate transcripts for forensic and other purposes. Frontiers in Communication. 7: 898410.

Fraser, H. (2022b). Forensic transcription: Legal and scientific perspectives. In C. Bernardasci, et al (Eds.), Speaker Individuality in Phonetics and Speech Sciences: Speech Technology and Forensic Applications (pp. 19–32). Milano: Officinaventuno.

Fraser, H., & Loakes, D. (2020). Acoustic injustice: The experience of listening to indistinct covertrecordings presented as evidence in court. Law Text Culture, 24, 405–429.

French, P., & Fraser, H. (2018). Why “ad hoc experts” should not provide transcripts of indistinct forensic audio, and a proposal for a better approach. Criminal Law Journal, 42, 298–302.

Knoch, U., and Macqueen, S. (2020). Assessing English for Professional Purposes. Routledge.

Love, R., & Wright, D. (2021). Specifying challenges in transcribing covert recordings: implications for forensic transcription. Frontiers in Communication: 6: 797448.

SST TUTORIAL ABSTRACT – FRASER and LOAKES

Specifying new scientific knowledge required by forensic applications

Speech scientists often assume that giving forensic evidence requires merely applying existing scientific knowledge. Actually it poses challenges that demand generation of new knowledge, with implications for speech science more broadly. This tutorial aims to open these challenges for discussion, and lay a foundation for new research projects connecting forensics more closely to other branches of speech science.

Part 1 aims to ensure clear understanding of the legal procedures governing admission and use of speech evidence in criminal trials. Since ASSTA joined the 2017 ‘call to action’, many know these procedures embody misconceptions about language and speech that affect the fairness of trials (e.g., police give opinions as ‘ad hoc experts’). However, scientists also hold misconceptions about the legal process that can affect efforts to provide reliable evidence in useful ways – indeed that may unwittingly entrench rather than overcome legal misconceptions.

Part 2 aims to specify the scientific knowledge that is needed to provide expert evidence in ways that help juries use it effectively – showing that some of this knowledge is not yet fully available, and considering the projects that can generate it. Topics include forensic voice comparison, enhancing poor quality audio, automatic speech and speaker recognition, transcription of indistinct forensic audio.

Target audience: The tutorial is open to researchers at any level, especially those looking to develop projects with real-world impact for themselves or their students. Participants do not need to have a background specifically in forensics – though of course those who do have forensic experience are very welcome. Participation by those who work in other areas interfacing with societal misconceptions about language and speech would also add value.

Requirements: No special equipment is needed.