Seminar Series: Abstract

15.30
September 04 2009
BPB 4.44 Baldessin Precinct Building

Murder Fraud and Forensic Speech Science
Philip Rose (School of Language Studies, ANU)

Over about the last twenty years much attention has been given to the proper evaluation of forensic evidence, resulting in what has been heralded as a major paradigm shift in many areas of forensic identification science, including forensic voice comparison. The paradigm shift was ultimately due to the post-1968 "new evidence scholarship" debate, and the increased incidence of statistical evidence associated with forensic DNA profiling. The National Research Council of the USA recently reported to Congress on serious deficiencies in forensic science, specifically in the lack of research demonstrating the reliability of forensic methods; and in the way evidence is presented by analysts in the courtroom. The UK too has recently recognised the need for greater accountability in forensic science. Spurred on in part by some recent spectacular miscarriages of justice due to faulty statistical reasoning and junk science, the Law Commission of England and Wales recently issued a consultation paper on how to determine evidentiary reliability and admissibility of expert evidence.

In Australia, and in particular at the ANU, we paradigm-shifted long ago, and have been very much in the forefront of evidence evaluation, at least as far as forensic voice comparison is concerned. In this talk, I will explain what this paradigm shift entails, and how it applies to two areas of Forensic Speech Science: Forensic Voice Comparison, and Disputed Utterances. In particular I will explain how one properly (i.e. legally and logically) estimates the strength of evidence in support of a typical prosecution claim like “the incriminating speech came from the suspect”. I will illustrate the application of the ideas from two recent high profile cases: a $150 million telephone fraud in NSW, and the “confession” in the David Bain trial (New Zealand’s most celebrated murder case). Because knowing your error rate is one desideratum for admissibility, I’ll also give examples of research into testing how well we can identify someone by their voice under forensically realistic conditions.

One of the points I hope to make is that the ability to do this kind of work presupposes (as well as a lot of other things of course) a solid knowledge of Linguistics, especially phonetics and phonology.