DS Research Projects
Print versionThe Dynamics of Conversational Dialogue (DynDial)
Conversation is the primary form of language use. The new Dynamics of Conversational Dialogue project aims to advance our understanding of human conversation by developing, implementing and testing a model of dialogue which reflects as closely as possible the incremental nature of human language processing. Its primary focus will be on ellipsis, in particular on the way in which, in conversation, turns are routinely fragmented and distributed across utterances and speakers, a phenomenon showing the tight coordination between the participants in the dialogue.
The project combines formal and empirical investigation of how people build utterances incrementally in dialogue. It will use corpus data to identify when people make use of part of what someone else said as an unspoken constituent of their own utterances: ellipses and split utterances. It will use new chat-tool techniques experimental techniques to test hypotheses about which utterance constituents are interchangeable across speakers in this way (Healey et al. 2003). This will lead to an implemented, incremental inter-speaker model of conversational turn construction. The primary theoretical framework to be adopted is Dynamic Syntax (DS: Kempson et al 2001, Cann et al 2005) which purports to directly reflect parsing process dynamics, with mappings onto Type-Theory-with-Records (TTR) (Cooper 2005) to provide cross-theoretical understanding of the processes involved.
The project combines formal and empirical investigation of how people build utterances incrementally in dialogue. It will use corpus data to identify when people make use of part of what someone else said as an unspoken constituent of their own utterances: ellipses and split utterances. It will use new chat-tool techniques experimental techniques to test hypotheses about which utterance constituents are interchangeable across speakers in this way (Healey et al. 2003). This will lead to an implemented, incremental inter-speaker model of conversational turn construction. The primary theoretical framework to be adopted is Dynamic Syntax (DS: Kempson et al 2001, Cann et al 2005) which purports to directly reflect parsing process dynamics, with mappings onto Type-Theory-with-Records (TTR) (Cooper 2005) to provide cross-theoretical understanding of the processes involved.
DynDial Aims and Objectives
The aims of the project are to achieve a radical step forward in our understanding of human dialogue by:
• Defining, implementing and testing a DS dialogue model for a broad range of ellipsis data.
• Explicitly formulating the correspondence between DS and TTR, thus providing a semantics for DS transitions and the DS model of context.
• Testing the model by a novel application of Healey's chat-tool methodology.
Our measurable objectives will be to:
• re-annotate two corpora of elliptical dialogue fragments (Purver 2004, Fernandez 2006) so as to incorporate the distinctions required by DS analyses for the various fragment types;
• develop DS-based analyses of the range of ellipsis data and phenomena revealed by these corpora not so far treated in DS, articulating the interaction of each with quantification, tense variation, and speaker/hearer switches;
• define the formal underpinnings to the DS concept of context as incorporating a record of sequences of parsing actions used, as used to explain various ellipsis phenomena (Purver et al 2006, Cann et al 2007)
• define a mapping from DS partial structures onto TTR record-type transitions. The goal is an incremental semantics for DS transitions, and will constitute a formal analysis of word-by-word incrementality
• develop earlier DS parsing and generation implementations to incorporatethe new analyses, resulting in a model handling dialogue phenomena including short answers to questions, clarifications, acknowledgements, and split utterances. Implementation will follow the formal DS/TTR mapping, to allow compatibility with both frameworks; compatibility will also be explored with the graph-theoretic notation being used in SDRT research so that mappings between all three systems can be developed in a later project, currently being planned (University of Bielefeld)
• extend Healey's experimental chat-tool methodology to test the predictions associated with the resulting DS model, in particular asserted symmetry of speaker/hearer processing in dialogue, hence split utterances. Tests will also be run with Branigan-style methodology, an established method of testing alignment, as a check on experiment methodology.
• Defining, implementing and testing a DS dialogue model for a broad range of ellipsis data.
• Explicitly formulating the correspondence between DS and TTR, thus providing a semantics for DS transitions and the DS model of context.
• Testing the model by a novel application of Healey's chat-tool methodology.
Our measurable objectives will be to:
• re-annotate two corpora of elliptical dialogue fragments (Purver 2004, Fernandez 2006) so as to incorporate the distinctions required by DS analyses for the various fragment types;
• develop DS-based analyses of the range of ellipsis data and phenomena revealed by these corpora not so far treated in DS, articulating the interaction of each with quantification, tense variation, and speaker/hearer switches;
• define the formal underpinnings to the DS concept of context as incorporating a record of sequences of parsing actions used, as used to explain various ellipsis phenomena (Purver et al 2006, Cann et al 2007)
• define a mapping from DS partial structures onto TTR record-type transitions. The goal is an incremental semantics for DS transitions, and will constitute a formal analysis of word-by-word incrementality
• develop earlier DS parsing and generation implementations to incorporatethe new analyses, resulting in a model handling dialogue phenomena including short answers to questions, clarifications, acknowledgements, and split utterances. Implementation will follow the formal DS/TTR mapping, to allow compatibility with both frameworks; compatibility will also be explored with the graph-theoretic notation being used in SDRT research so that mappings between all three systems can be developed in a later project, currently being planned (University of Bielefeld)
• extend Healey's experimental chat-tool methodology to test the predictions associated with the resulting DS model, in particular asserted symmetry of speaker/hearer processing in dialogue, hence split utterances. Tests will also be run with Branigan-style methodology, an established method of testing alignment, as a check on experiment methodology.