Natural Language Generation (NLG) systems in interactive settings often face a multitude of choices, given that the communicative effect of each utterance they generate depends crucially on the interplay between its situational circumstances, addressee and interaction history. This is particularly true in interactive and situated settings. Traditionally, the generation process has been divided into distinct stages of decision making, e.g. content selection, utterance planning and surface realisation. However, this sequential model does not account for the interdependencies that exist among these stages, which in practice can become manifest in inefficient, ineffective communication and an increased cognitive load for the user.This book presents a joint optimisation framework for NLG in dialogue that is based on Hierarchical Reinforcement Learning and learns the best utterance for a context by optimisation through trial and error. The joint model considers decisions at different NLG stages interdependently and produces more context-sensitive utterances than a model that considers decisions in isolation. To enhance the human-likeness of the presented framework, we integrate graphical models, trained from human data, as generation space models for natural surface realisation. The proposed technique is evaluated in a study with human participants and confirms that the hierarchical learner is able to learn an adaptive policy that leads to smooth and successful interactions. Results also suggest that a joint optimisation leads to substantially higher user satisfaction and task success and is better perceived by human users than its isolated counterpart.

To solve complex real-world search and optimization problems it is a common strategy to divide the problem into different sub-problems. A consequence of this divide and conquer approach is the appearance of interdependencies between the sub-problems, that makes coordination necessary.The field of coordination is investigated by researchers from different areas. An important result of this research is that there exists no universal coordination method. Consequently, a suitable coordination mechanism has to be identified for each single coordination problem. This book presents the ECo process, which is the first approach that addresses this selection problem. Coordination mechanisms have to respect the context in which the coordination has to be performed. This is encoded in coordination requirements that guide the selection process. Additionally, a new classification scheme is presented to efficiently pre-select coordination mechanisms. The coordination mechanisms are classified according to these classification scheme that indicate the characteristics for re-use of a coordination mechanism. The implementation of coordination mechanisms is supported by the CoPS process and framework. The CoPS process structures the decision making during the implementation phase. The CoPS framework provides a set of basic functionality to implement coordination methods among software agents.Within the CoPS process, techniques are presented for the design and implementation of conversations between agents. These techniques are not limited to the design of coordination mechanisms, but for multiagent systems in general. The ECo-CoPS approach has been successfully validated in two case studies from the logistic domain.

Natural Language Generation (NLG) systems in interactive settings often face a multitude of choices, given that the communicative effect of each utterance they generate depends crucially on the interplay between its situational circumstances, addressee and interaction history. This is particularly true in interactive and situated settings. Traditionally, the generation process has been divided into distinct stages of decision making, e.g. content selection, utterance planning and surface realisation. However, this sequential model does not account for the interdependencies that exist among these stages, which in practice can become manifest in inefficient, ineffective communication and an increased cognitive load for the user.This book presents a joint optimisation framework for NLG in dialogue that is based on Hierarchical Reinforcement Learning and learns the best utterance for a context by optimisation through trial and error. The joint model considers decisions at different NLG stages interdependently and produces more context-sensitive utterances than a model that considers decisions in isolation. To enhance the human-likeness of the presented framework, we integrate graphical models, trained from human data, as generation space models for natural surface realisation. The proposed technique is evaluated in a study with human participants and confirms that the hierarchical learner is able to learn an adaptive policy that leads to smooth and successful interactions. Results also suggest that a joint optimisation leads to substantially higher user satisfaction and task success and is better perceived by human users than its isolated counterpart.

Diese Arbeit beschäftigt sich mit der Frage, wie eine Infrastruktur für digitale Objektgedächtnisse aussehen muss. Ziel der Arbeit ist die Erstellung eines Architekturkonzepts für die Repräsentation und die Verwaltung von, sowie den Zugriff auf digitale Objektgedächtnisse inklusive abgestimmter Werkzeuge, die die Erstellung neuer und die Migration bestehender Systeme erleichtern.Aufgrund der heterogenen Datenlandschaft in einem „open-loop“-Szenario, werden an ein Objektgedächtnis besondere Anforderungen gestellt. Zum einen muss es flexibel genug ausgelegt sein, um mit den unterschiedlichsten Daten umgehen zu können. Zum anderen muss ein einfacher und strukturierter Zugriff auf die Daten ermöglicht werden, der je nach Anwendungsfall auch mit Konzepten des rechte- und rollenbasierten Zugriffs und der Versionsverwaltung ergänzt werden muss.Diese Arbeit zeigt daher zuerst ein Datenmodell zur Strukturierung von Objektgedächtnisdaten. Dieses erlaubt mit Hilfe von Metadaten auch in heterogenen Szenarien einen domänenübergreifenden Datenaustausch. Im Anschluss wird eine Softwarearchitektur vorgestellt, die eine Speicher- und Zugriffslösung für Objektgedächtnisse anbietet und die Gedächtnisse zusätzlich mit einer Aktivitätskomponente ausstattet. Abgerundet wird das Konzept mit einem Satz an Werkzeugen, die es Entwicklern erlauben, zum Beispiel bestehende Datenbestände zu migrieren, neue Strukturen auf Basis von semantischen Daten aufzubauen und diese mit anwendungsspezifischen Visualisierungen dem Benutzer wieder zugänglich zu machen.

The computer-assisted authoring of mathematical documents using a scientific text-editor requires new mathematical knowledge management and transformation techniques. The challenge is that, throughout an assistance system like the ΩMEGA system, various kinds of given and derived knowledge units occur in different formats and with different dependencies. If changes occur in these pieces of knowledge, they need to be effectively propagated.We present a Change-Oriented Architecture for mathematical authoring assistance. This work includes an efficient method for the computation of weighted semantic changes and an invertible grammar formalism for the automated bidirectional transformation between interface documents. The presented architecture provides an adequate basis for the computer-assisted authoring of mathematical documents with semantic annotations and a controlled mathematical language.