LanGeLin (Language and Gene Lineages) is the acronym for the ERC-funded research project ‘Meeting Darwin’s last challenge: toward a global tree of human languages and genes’ coordinated by Professor Giuseppe Longobardi, PI, running from December 2012 to November 2017. The project addresses one question, formulated by Charles Darwin in The Origin of Species, namely whether the cultural transmission and differentiation of languages over the period of human history matches the biological transmission and differentiation of the genetic characters which define the populations of the world. The project is by definition interdisciplinary, and the work of the linguistic researchers in York is complemented by the participation of population geneticists and molecular anthropologists based at the University of Ferrara and the University of Bologna-Alma Mater Studiorum.
The FP7 Marie Curie Initial Training Network BBfor2 (Bayesian Biometrics for Forensics) consists of nine European research institutes and three associated partners. The Network provides regular workshops and summer schools, so that the PhD students and senior researchers can exchange research experience, insights and ideas. The main areas of research are speaker recognition (comparison), face recognition, and fingerprint recognition. These areas are studied both individually and in combination. The challenge of applying biometric techniques in a forensic context is to be able to deal with the uncontrolled quality of the evidence, and to provide calibrated likelihood scores.
Two projects are based at York, and York staff also co-supervise projects based elsewhere.
Multimodal speech and speaker recognition
- Research Fellow: Natalie Fecher
- Supervisors: Dominic Watt and David van Leeuwen (TNO & Radboud University, Nijmegen)
With various forms of biometric technologies becoming available, there is a growing need for scientists who are able to assess the merits of these technologies when applied to forensics. This project investigates multimodal speech and speaker recognition from a forensic perspective through assessments of the performance of human subjects and automated speech and speaker recognition systems where the quality and quantity of the information available in the audiovisual signal is manipulated experimentally. For example, the talker's face may be partially or completely obscured by clothing such as face-concealing garments or safety equipment worn for occupational, recreational or religious reasons, or for the commission of robberies, assaults or terrorist activities.
Calculation of likelihood ratios using phonetic and linguistic features
- Research Fellow: Erica Gold
- Supervisors: Peter French and Didier Meuwly (Netherlands Forensic Institute)
The most prevalent strain of forensic speaker comparison work across European experts and institutions involves auditory-phonetic and acoustic examinations of speech samples on a wide range of parameters including voice quality, intonation, rhythm and consonant and vowel realisations. Those working within this tradition express their conclusions within a variety of different frameworks. Most do not currently utilise a likelihood ratio (logical inference) framework, which is acknowledged to be the goal for forensic science generally.
There are practical difficulties in interpreting the results of a phonetic acoustic comparison of speech samples in a likelihood ratio framework. These include the use of speech data from the appropriate sections of the background population to estimate empirically the evidential value of the results and to combine them in an overall statistical assessment. The aim of the project is to find solutions to these problems, thus paving the way for forensic speaker comparison work to become conceptually aligned with more developed areas of forensic science such as DNA analysis.