Overview and details of the sessions of this conference. Please select a date or location to show only sessions at that day or location. Please select a single session for detailed view (with abstracts and downloads if available).
Qualitative approaches to comparative tasks in a forensic context have proved to produce good re-sults with very short texts due to their depth of description. Amounts of textual data that come up in investigations of social network communication require a different tackling. In a series of research studies, several automatic text comparison systems have been tested for their suitability in a forensic context. This means that an automatic system includes both the processing of linguistic data as well as the evaluation of this data as linguistic evidence (e.g. cf. Ishihara 2014 and 2017). Methodologically, N-gram analysis, Burrows’ delta (Burrows 2002) and cosine similarity have been used in these systems. For an evaluation framework, a likelihood ratio-based approach was implemented. In order to observe the effects of different parameters, various corpus-based and method-based variations have been applied resulting in a total of 2352 system variations each eval-uated as independent comparison system. For system validation, the metrics equal error rate (EER) and log likelihood ration cost (Cllr) have been used. The results (best-performing systems with EER of 5%, Cllr of 0.4) suggest that there is the possibil-ity to complement traditional approaches to authorship attribution – provided the data base is suited for an application of automatic systems.
4:30pm - 5:00pm
The Use of the Likelihood Ratio Based Approach in Forensic Authorship Analysis
The Australian National University, Australia
The use of likelihood ratio (LR) for conveying expert evaluative opinions to the decision makers, such as the court or juries, has been supported and recommended by relevant communities. In Aitken, et al. (2011), for example, which is a position statement signed by 31 individuals and supported by the Board of the European Network of Forensic Science Institutes (ENFSI), LR is stated as “the most appropriate foundation for assisting the court”. The ENFST also released new guidelines in 2015 for the practice of forensic voice comparison (FVC) - probably the closest field to forensic authorship analysis (FAA) - and they specifically recommend the LR approach for FVC (Drygajlo et al. 2015). Despite the tangible presence of the LR framework as the logically and legally correct framework for assessing and presenting forensic evidence in more and more different fields of forensic science, FAA studies based on the LR framework remain sparse; thus the LR framework has not yet made inroads into FAA. After the advantages as well as the theoretical/practical issues of the LR framework in FAA are described, this paper 1) argues, while attempting to refute the arguments against the application of the LR framework in FAA, that we need to start investigating how FAA can best profit from the LR framework, and 2) also discusses possibly how and where we can start the investigation. This requires a wide-open discussion at the theoretical and practical levels and an extensive amount of fundamental-empirical research.