|dc.description.abstract||Context in dialogue is at once regarded as a set of resources enabling successful interpretation and
is altered by such interpretations. A key problem for models of dialogue, then, is to specify how
the shared context evolves. However, these models have been developed mainly to account for
the way context is built up through direct interaction between pairs of participants. In multi-party
dialogue, patterns of direct interaction between participants are often more unevenly distributed.
This thesis explores the effects of this characteristic on the development of shared contexts.
A corpus analysis of ellipsis shows that side-participants can reach the same level of grounding
as speaker and addressee. Such dialogues result in collective contexts that are not reducible
to their component dyadic interactions. It is proposed that this is characteristic of dialogues in
which a subgroup of the participants are organised into a party, who act as a unified aggregate to
carry the conversation forward. Accordingly, the contextual increments arising from a dialogue
move from one party member can affect the party as a whole. Grounding, like turn-taking, can
therefore operate between parties rather than individuals. An experimental test of this idea is
presented which provides evidence for the practical reality of parties.
Two further experiments explore the impact of party membership on the accessibility of context.
The results indicate that participants who, for a stretch of talk, fall inside and outside of the
interacting parties, effect divergent contextual increments. This is evidence for the emergence of
distinct dialogue contexts in the same conversation.
Finally, it is argued that these findings present significant challenges for how formal models
of dialogue deal with individual contributions. In particular, they point to the need for such
models to index the resulting contextual increments to specific subsets of the participants||en_US