TOPIC CONTINUITY FOR DISCOURSE AND DIALOGUE
17 Oct 2018 Wednesday, 03:00 PM to 04:30 PM
AS6 Level 2
AS6-02-08 Discussion Room 5
Topic continuity, one of the major linguistic devices for text cohesion, plays important roles in multi-sentential scenario. It links multiple sentences to be a unit, e.g., a cohesive paragraph or a fluent dialogue, distinguishing them from a set of randomly collected sentences. In this thesis proposal, we study topic continuity in discourse and dialogues, two typical multi-sentential texts.
Firstly, we studied topic continuity in discourse through the task of implicit discourse relation recognition in Penn Discourse Treebank. We find it is the prerequisite for discourse relations.Specifically, it licenses individual semantic characteristics for each type of discourse relation to form corresponding discourse relation.
Secondly, we study topic continuity in a specific type of dialogue system, called task-oriented dialogue system. The topics continued across utterances are referred to as slots in dialogue research community. We propose a sequence-to-sequence model, called Sequicity, to capture such slots for both knowledge base search and conditioning response generation.
To complete a thesis, we propose to extend Sequicity framework into semi-supervised as well as unsupervised settings where the topics are not (completely) annotated. We hope the proposed method is able to capture the continued topics in dialogues even if there is no annotation at all. The extended framework can also be applied on non-task-oriented dialogue systems.