Social Behaviour in Police Interviews: Relating Data to Theories - Conflict and Multimodal Communication: Social Research and Machine Intelligence - page 311

indicated how the police officers or suspects felt about the interaction. Next, a
subgroup of the observers noted as many different terms as possible to describe
what was going on in the fragments (3)—these could, for example, be adjectives
describing the mood or nouns indicating behavioural traits. Based on these terms, we
created questions with variations of the format “To what extent is [term] the case?”
(4), see Sect. 16.2.2 . The original six observers then rated a random subset of the
same fragments from the corpus on a five-point Likert scale for every question, once
for the police interviewer and once for the suspect. For example, a question from the
point of view of the suspect was “To what extent is the suspect dominant?” (5).
We performed a factor analysis to find a clustering of correlated questions (6)
which we discuss in Sect. 16.2.3 . Next, a subgroup of the original six observers
reached consensus on the interpretation of the factors (7) and we selected well-
known psychological and sociological theories that addressed these interpretations
(8), see Sect. 16.3 . In Sect. 16.4 , we discuss how the concepts that these theories
employ were matched to the factors by selecting the concepts that fit each factor
(9 and 10). This also revealed the relations between the different concepts that were
selected (11), as for some factors, concepts from different theories were applicable.
This relation between theories allowed us to integrate the different theories into
one “meta-theory” that provides the terms and concepts to describe the interactions
in a police interview. We checked whether this “meta-theory” can describe what is
happening in a police interview in Sect. 16.5 (12). Our final step (13) is to create
a computational model from our “meta-theory” (Bruijnes 2013 ), but this remains a
future endeavour (see Sect. 16.6 ).

16.2.1 Corpus Description

The DPIT Corpus is a corpus that consists of police interviews conducted by trainees
of the Dutch National Police, recorded in 2012 and 2013. The police officers in
the corpus are novice to moderately proficient police interviewers. The suspects
they interview in this corpus are professionally trained actors. Due to privacy
requirements, the video and audio data of the corpus is not publicly available.
The corpus consists of 32 interviews from 6 scenarios (cases) with a total length
of approximately 13 h. The interviews vary in length from about 9 min to almost
an hour. Some scenarios were enacted several times (with different students and
sometimes also with different actors), while other scenarios were cut into separate
interviews. In the latter, the suspect was interviewed multiple times, for example
to give the police officers time to check facts. In these scenarios the same actor is
interviewed by different police officers. The scenarios are specified as follows:

Bruintjes Ms Bruintjes is suspected of having bought a stolen smartphone from
her cousin. She comes across as being not very bright but knew the phone was
stolen.