“Forensic linguistics: the contribution of humanities computing”

László Hunyadi University of Debrecen hunyadi@llab2.arts.klte.hu Enikő Tóth University of Debrecen teniko@pmail.arts.klte.hu Kálmán Abari University of Debrecen abarik@pmail.arts.klte.hu

Abstract:

The aim of the talk is to demonstrate how useful humanities computing can prove to be in solving issues of a seemingly as distant field as forensic science. We will present a case study of an actual forensic linguistic assignment whose aim it was to determine if a digitalized recording of a conversation had been tempered. The task, highly challenging due to its novelty both in applied linguistics and forensic practice, was carried out by investigating three independent aspects of the issue: those of experimental phonetics, situational semantics and computation. The results of the three approaches were synthesized to give a comprehensive basis for an answer to the initial question.

Introduction:

To find a proof for or against the assumption that a given document has been tempered with is one of the important tasks of forensic science. With the introduction of various voice-recording techniques it became especially important to decide whether or not such a recording can authentically represent a certain event of reference. The case of magnetic tape recordings is relatively simple, since any (either electronic or mechanical) modification of the tape leaves a trace behind, characteristic even of the kind of manipulation (cf. Gruber et al. 1995, Gruber et al. 1993, Poza 1979). However, with digital recordings gaining more and more popularity in our days, on the one hand, and digital manipulation becoming technically possible and easy, on the other, one might believe that the discovery of such a manipulation is highly unlikely. The novelty of the issue in scientific literature just adds to this challenge. In order to carry out the task, we had two assumptions: a. due to human voice having a highly complex structure, even digital tempering by a person might leave a significant trace, and b. due to conversations also having their strict internal structure, the removal of a segment of the given conversation might also be noticeable.

Discussion:

The findings of the three approaches were as follows:

1. A detailed spectrographic analysis of digitalized voice recordings found a characteristic pattern with a duration of 8-10 milliseconds at the place where a segment had been digitally removed. Although this pattern varied according to the actual immediate environment, its characteristic features could be established (a symmetrical increase of intensity as well as a 500 Hz increase of frequency). This spectrographic pattern was only manifested in the exact location of recordings with a digital cut. Applying the methodology to the given task, we found that no spectrographic trace of this kind of manipulation could be identified.
2. The aim of the situational-semantic study was to find out if an eventual cutting of a portion of the text could have possibly left a trace that can be identified as semantically significant. Our attention was directed to the analysis of the appropriateness of pieces of linguistic material with a referential function, including pronouns, determiners and names (cf. Brown et al. 1983, Kamp 1981). The analysis pointed at some places where the reference was not unambiguously computable from the immediate linguistic environment, but since in running conversation such immediate turns often happen, they could not be decided on the basis of linguistic content alone. These locations became thus subject to spectrographic analysis, and the latter concluded that no tempering could be identified there.
3. In our work we included a separate computational task to find out if segments of an eventual cut in the voice recording were still recoverable from the hard disk originally used for digitizing. Since it turned out that the hard disk had been completely reformatted, we could not complete this task. However, we elaborated a methodology for similar tasks and applied it in model situations. The applied statistical method of zero crossing proved to yield significant interpretable results in the differentiation of headerless segments of certain types of files. This method showed a significant difference between .bmp and .txt files, and .wav files also had a characteristic value for zero-crossing. Thus, we suggest that this methodology may be useful for the possible differentiation of at least a few types of files in future work.

Summary:

This case study showed us that humanities computing can have a significant contribution to forensic science, especially in the form of a combination of theoretical and applied linguistics as well as statistics and computing. This assignment was a real challenge for us, and it resulted in the elaboration of a new methodology both in experimental phonetics and computation. Thus proving the inspiring force of humanities computing across various fields of science.

References:

G. Brown G. Yule. Discourse Analysis. : CUP, 1983.

H. Kamp. “A theory of truth and semantic representation.” Formal Methods in the Study of Language. Ed. J. Groenendijk T. Janssen M. Stokhof. Amsterdam: Mathematical Centre, 1981.

J. S. Gruber F. Poza. “Voicegram Identification Evidence.” American Jurisprudence Trials. Lawyers Cooperative Publishing, 1995. 54: .

J. S. Gruber F. Poza A. J. Pellicano. “Audio Recordings: Evidence, Experts and Technology.” American Jurisprudence Trials. Lawyers Cooperative Publishing, 1993. 48: .

unknown. On the Theory and Practice of Voice Identification. : National Academy of Sciences, 1979.