French Face-to-Face Interaction: Repetition as a Multimodal Resource - Coverbal Synchrony in Human-Machine Interaction - page 144

used in this study on repetition, it is important to consider that the
project aimed not only to provide and develop conventions and
tools for multimodal annotation but also to define the organization
of annotations in an abstract description from which a formal XML
scheme could be generated (Blache and Prévot, 2010).
2.1 Corpus
For a few years, numerous programs have been conducted in different
countries to provide large-scale spontaneous speech interactions
involving the creation and development of resources (in terms of
both corpora and annotations). Among others, one can mention the
Map-Task corpus (Anderson et al., 1991) that is one of the first semi-
elicited corpus and which has been reduplicated in many languages,
the Columbia Game Corpus (http://www.cs.columbia.edu/speech/
games-corpus/), the Buckeye Corpus (Pitt et al., 2005), the Corpus
of Spontaneous Japanese (Furui et al., 2005), DanPASS (the Danish
phonetically annotated spontaneous speech corpus, Grønnum, 2006),
as well as corpora annotated at the gestural level such as the Göteborg
Spoken Language Corpus (Allwood et al., 2000), the MIBL Corpus
(Multimodal Instruction Based Learning, Wolf and Bugmann, 2006)
or the D64 Corpus (Campbell, 2009), among others (for a more
exhaustive list, Knight, 2011). The Corpus of Interactional Data (CID)
(Bertrand et al., 2008) described here is an audiovideo recording of
conversational French (eight dialogs of 1 hour each, 110.000 words).
Participants were filmed by a single camera and recorded with a
head-set microphone (one track per speaker, in order to enable the
acoustic analysis of speech and segments produced in overlap by both
speakers). Participants were asked to tell about conflicts or unusual
events in their personal lives.
2.2 Corpus transcription
The first and most important step in the annotation process is
transcription because most of the annotations in the different domains
are based on this particular level.
In a preliminary stage, the speech signal was automatically
segmented in inter-pausal units (IPUs), speech blocks surrounded by
200-ms silent pauses. The transcription process takes this series of IPUs
as input. The transcription conventions adopted in the project derive
from the ones defined by the GARS (Blanche-Benveniste and Jeanjean,
1987). They take into account some remarkable and frequent phonetic
phenomena: non-standard elisions, phoneme substitution or additions,