@phdthesis{LopezGambino2023,
  author    = {L{\´o}pez Gambino, Maria Soledad},
  title     = {Time Buying in Task-Oriented Spoken Dialogue Systems},
  doi       = {10.25932/publishup-59280},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-592806},
  school      = {Universit{\"a}t Potsdam},
  pages     = {148},
  year      = {2023},
  abstract  = {This dissertation focuses on the handling of time in dialogue. Specifically, it investigates how humans bridge time, or "buy time", when they are expected to convey information that is not yet available to them (e.g. a travel agent searching for a flight in a long list while the customer is on the line, waiting). It also explores the feasibility of modeling such time-bridging behavior in spoken dialogue systems, and it examines how endowing such systems with more human-like time-bridging capabilities may affect humans' perception of them. The relevance of time-bridging in human-human dialogue seems to stem largely from a need to avoid lengthy pauses, as these may cause both confusion and discomfort among the participants of a conversation (Levinson, 1983; Lundholm Fors, 2015). However, this avoidance of prolonged silence is at odds with the incremental nature of speech production in dialogue (Schlangen and Skantze, 2011): Speakers often start to verbalize their contribution before it is fully formulated, and sometimes even before they possess the information they need to provide, which may result in them running out of content mid-turn. In this work, we elicit conversational data from humans, to learn how they avoid being silent while they search for information to convey to their interlocutor. We identify commonalities in the types of resources employed by different speakers, and we propose a classification scheme. We explore ways of modeling human time-buying behavior computationally, and we evaluate the effect on human listeners of embedding this behavior in a spoken dialogue system. Our results suggest that a system using conversational speech to bridge time while searching for information to convey (as humans do) can provide a better experience in several respects than one which remains silent for a long period of time. However, not all speech serves this purpose equally: Our experiments also show that a system whose time-buying behavior is more varied (i.e. which exploits several categories from the classification scheme we developed and samples them based on information from human data) can prevent overestimation of waiting time when compared, for example, with a system that repeatedly asks the interlocutor to wait (even if these requests for waiting are phrased differently each time). Finally, this research shows that it is possible to model human time-buying behavior on a relatively small corpus, and that a system using such a model can be preferred by participants over one employing a simpler strategy, such as randomly choosing utterances to produce during the wait —even when the utterances used by both strategies are the same.},
  language  = {en}
}
@phdthesis{Galetzka2022,
  author    = {Galetzka, Fabian},
  title     = {Investigating and improving background context consistency in neural conversation models},
  doi       = {10.25932/publishup-58463},
  url       = {http://nbn-resolving.de/urn:nbn:de:kobv:517-opus4-584637},
  school      = {Universit{\"a}t Potsdam},
  pages     = {viii, 173},
  year      = {2022},
  abstract  = {Neural conversation models aim to predict appropriate contributions to a (given) conversation by using neural networks trained on dialogue data. A specific strand focuses on non-goal driven dialogues, first proposed by Ritter et al. (2011): They investigated the task of transforming an utterance into an appropriate reply. Then, this strand evolved into dialogue system approaches using long dialogue histories and additional background context. Contributing meaningful and appropriate to a conversation is a complex task, and therefore research in this area has been very diverse: Serban et al. (2016), for example, looked into utilizing variable length dialogue histories, Zhang et al. (2018) added additional context to the dialogue history, Wolf et al. (2019) proposed a model based on pre-trained Self-Attention neural networks (Vasvani et al., 2017), and Dinan et al. (2021) investigated safety issues of these approaches. This trend can be seen as a transformation from trying to somehow carry on a conversation to generating appropriate replies in a controlled and reliable way. In this thesis, we first elaborate the meaning of appropriateness in the context of neural conversation models by drawing inspiration from the Cooperative Principle (Grice, 1975). We first define what an appropriate contribution has to be by operationalizing these maxims as demands on conversation models: being fluent, informative, consistent towards given context, coherent and following a social norm. Then, we identify different targets (or intervention points) to achieve the conversational appropriateness by investigating recent research in that field. In this thesis, we investigate the aspect of consistency towards context in greater detail, being one aspect of our interpretation of appropriateness. During the research, we developed a new context-based dialogue dataset (KOMODIS) that combines factual and opinionated context to dialogues. The KOMODIS dataset is publicly available and we use the data in this thesis to gather new insights in context-augmented dialogue generation. We further introduced a new way of encoding context within Self-Attention based neural networks. For that, we elaborate the issue of space complexity from knowledge graphs, and propose a concise encoding strategy for structured context inspired from graph neural networks (Gilmer et al., 2017) to reduce the space complexity of the additional context. We discuss limitations of context-augmentation for neural conversation models, explore the characteristics of knowledge graphs, and explain how we create and augment knowledge graphs for our experiments. Lastly, we analyzed the potential of reinforcement and transfer learning to improve context-consistency for neural conversation models. We find that current reward functions need to be more precise to enable the potential of reinforcement learning, and that sequential transfer learning can improve the subjective quality of generated dialogues.},
  language  = {en}
}