Detecting Speech Interruptions for Automatic Conflict Detection - Conflict and Multimodal Communication: Social Research and Machine Intelligence - page 366

18.1.1 Model of Conversational Interaction

Conversation is a social interaction between two or more people, where taking turns
to talk is naturally observed. In the pioneering work of Sacks et al. ( 1974 ), an
organizational model of turn-taking for conversation that is context-free, capable
of context sensitivity, and having a cross-cultural validity was investigated. The
constraints of their model were set in reference to the high cross-cultural flexibility
of conversation accommodation, with a wide range of interaction in which there
is a variety of persons and numbers of persons who are taking part. The authors
proposed a model that relies on two components that are related to the turn-
constructional units (TCUs, the basic units of talk) and the turn allocation at the
end of each TCU for the next unit (the next speaker's TCU). TCUs end with points
of possible completion (e.g., gap, query) called transition-relevant places (TRPs),
in which the turn transition could be relevant but is not necessary. Observed in
any conversation, 14 facts were listed. An excerpt of this list is the following: (a)
mostly one party talks at a time; (b) the vast majority of turn-taking transitions
is composed of transitions that have no/slight gap and no/slight overlap; (c) the
turn size varies; (d) overlapping speech is common, but brief; (e) two basic turn-
allocation techniques are used: the “current selects next” technique when a current
speaker can select a new speaker (e.g., addressing a question) and the “self-select”
technique when a speaker can self-select in starting to talk; (f) repair mechanisms
exist for addressing turn-taking violation; e.g., when overlapping speech occurs,
one (or more) of the speakers will stop prematurely. A set of rules was edited
for addressing turn transitions from TRP in such a way as to minimize the gap or
overlap in the transitions. The turn transfer is defined according to the construction
of the TCU, regardless of whether the “current speaker selects next” technique is
used as well as the eventual application of “self-selection.” The rules are based
on the purpose of no-gap-no-overlap transitions, for which ability is required in
anticipating the precise moment at which a TCU is going to come to a completion
point (i.e., a TRP). In related work (De Ruiter et al. 2006 ), the lexical and syntactic
content of TCU was shown to be necessary for this anticipation, while the intonation
contour was neither necessary nor sufficient for this projection. According to the
turn-taking rule-set applied to a multiparty conversation, overlap is expected in the
neighboring transition-relevant places: when a possible completion of the current
TCU is wrongly projected by a party or when parties are competing in a self-
selection mode for a next turn. In a work that is related to turn-taking organization
and that is beyond the ordinary conversation and is mostly unconstrained in terms
of a role, a wide range of publications have studied the turn-taking practices
and characteristics within various contexts of multiparty interactions. Distinctive
features of turn-taking were found in institutional interactions in which a turn-
taking organization is more constrained and specialized according to the roles that
are assigned to the group members (e.g., interviewer vs. interviewee, chair vs.
participant). Studies on turn-taking management were investigated in institutional
settings such as in a classroom (Mac Houl 1978 ; Mehan 1985 ; Lerner 1995 ),