Kolloquium/Ringvorlesung des Graduiertenkolleg Wissensrepräsentation

Das Model Checking soll entscheiden, ob alle Abläufe eines verteilten Systems (bzw. seines Modell) eine mittels einer logischen Formel ausgedrückte Eigenschaft haben. Typische hierbei verwandte Logiken sind temporale Logiken wie LTL oder PTL. Erfolgreiche Algorithmen für das Model Checking betrachten dabei die Abläufe als Woerter. Dies ist bei verteilten Systemen nicht angemessen, da ja Aktionen auf verschiedenen Komponenten parallel ausgeführt werden können. Zur Modellierung solcher Abläufe eignen sich Mazurkiewicz-Spuren. Verschiedene temporale Logiken für solche Mazurkiewicz-Spuren wurden in den vergangenen 15 Jahren eingeführt. Um für die Verifikation nutzbar zu sein, muss hierbei sichergestellt werden, dass das Erfüllbarkeits- und das Model Checking Problem in vertretbarer Komplexität gelöst werden können. Bisher wurden für alle neu eingeführten lokalen temporalen Logiken neue Beweise dafür angegeben, dass diese Probleme in PSPACE lösbar sind (häufig PSPACE-vollständig). Im Vortrag werde ich ein allgemeines Resultat herleiten, das auf alle bekannten (und viele weitere) lokale temporale Logiken anwendbar ist. Hierbei wird ein Zusammenhang zwischen der Komplexität der genannten Probleme und der Form der verwandten Modalitäten hergestellt werden. Ein Ausblick befasst sich dann mit der Frage, inwiefern sich diese Ergebnisse auf sog. Message Sequence Charts erweitern lassen.

Diese Untersuchungen wurden zusammen mit Paul Gastin (Paris) durchgeführt und zum Teil auf CONCUR 2003 vorgestellt.

Unlike the case for sequential and conditional planning, what existing work there is on iterative planning (planning where loops may be needed) still leans heavily on theorem-proving. I will propose a different methodology, describe an implementation, present a few examples, and sketch the beginnings of a theory.

Data Mining Methoden unterstützen die Suche nach verständlichen Mustern, Strukturen und Modellen in Daten und bilden somit den methodischen Kern einer automatisierten Wissensextraktion. Da interessante Muster in Daten häufig "unscharf" im Sinne der Theorie unscharfer Mengen sind, wird im Rahmen des Vortrags zunächst die Anwendung von Fuzzy-Methoden im Data Mining motiviert. Im Weiteren wird dann speziell die Frage nach der Bewertung unscharfer Muster näher untersucht. Mit der Idee, einen Datensatz (unscharf) in positive, negative und irrelevante Beispiele eines Musters zu partitionieren, wird ein generelles Konstruktionsprinzip für Bewertungsmaße vorgestellt.

In my talk, I will present LRI-Core, a core ontology for covering domains of law. After a decade of developing many ontologies for legal domains and applications, the need for a unifying core ontology that covers the main concepts that are common to all legal domains became very apparent. It can be argued that not only these legal domains have a predominant common-sense character -- legal cases give an account of social events -- but also that typical legal concepts such as norm, role responsibility, contract, etc. have still a grounding in abstract common-sense conceptualizations. This common sense grounding is lacking in other upper- or foundational ontologies developed thus far, so the most abstract layer of LRI-Core is in fact an upper ontology.

I will discuss a number of design principles that follow from this common-sense stance: the most important being cognitive plausibility, and an evolutionary account of the major categories In this perspective, knowledge about the physical world, with the central notions of object and process is taken as a basis for metaphorizing mental and abstract worlds. The intentional stance that differentiates the physical world from the mental/agent world is also the basis for understanding social worlds as consisting of roles. LRI-Core starts with four main categories: physical classes, mental classes, roles, and abstract classes. A fifth category consists of terms for occurrences, which are used to talk about instances (situations).

This paper is a survey of qualitative decision theory focusing on available decision rules under uncertainty, and their properties. It is pointed out that two main approaches exist according to whether degrees of uncertainty and degrees of utility are commensurate (that is, belong to a unique scale) or not. Savage-like axiom systems for both approaches are surveyed. In such a framework, acts are functions from states to results, and decision rules are derived from first principles, bearing on a preference relation on acts. It is shown that the emerging uncertainty theory in qualitative settings is possibility theory rather than probability theory. However these approaches lead to criteria that are either little decisive due to incomparability, or too adventurous because focusing on the most plausible states, or yet lacking discrimination because or the coarseness of the value scale. Some suggestions to overcome these defects are pointed out.

Die in der Wissensrepräsentation verwendeten Begriffssysteme und philosophische Kategorientheorien haben hinreichend viele Gemeinsamkeiten, um aus der Feststellung von Unterschieden konkreten Nutzen zu ziehen. Der Vortrag wird das GOL-Projekt mit dem Programm der General Process Theory (GPT) in einigen Aspekten vergleichen: inbezug auf methdologische Aspekte, wie Erklärungsanspruch, Modellierungsskopus, und Adäquatheitsnachweis, sowie inbezug auf die Modellierungsleistung in konkreten Anwendungen, insbesondere die Behandlung von Aussagen über Persistenz und Veränderung. Das weitergehende Anliegen des Vergleichs ist die Untersuchung der Frage, ob und wie sich eine geeignete Schnittstelle zwischen einer GOL top-level Ontologie und GPT erstellen liesse, um die in GPT entwickelte Typologie der Prozesse für die Zwecke von GOL nutzbar zu machen.

We introduce the OWL Plugin, a SemanticWeb extension of the Protégé ontology development platform. The OWL Plugin can be used to edit ontologies in the Web Ontology Language (OWL), to access description logic reasoners, and to acquire instances for semantic markup. In many of these features, the OWL Plugin has created and facilitated new practices for building Semantic Web contents, often driven by the needs of and feedback from our users. Furthermore, Protégé?s flexible open-source platform means that it is easy to integrate customtailored components to build real-world applications. This document describes the architecture of the OWL Plugin, walks through its most important features, and discusses some of our design decisions.

The talk describes the FF planning system, and explains its success in many of the traditional planning benchmarks. I give a brief introduction to AI Planning, then I explain the main algorithmic techniques underlying FF. I outline the dramatic runtime breakthrough achieved with these techniques a few years ago. I then provide a formal explanation of this success, by giving the results of an investigation of the local search topology in the planning benchmarks, under FF-style heuristic functions. While the results contained in the talk are mostly relatively old (1-4 years), they are still highly relevant for current AI Planning research.

CONALD is a spoken language dialogue system to conduct rational dialogues addressing an application system, e.g. to collect information, to negotiate about resources, or to control a technical system. To enable flexible, robust and generic dialogue management, we use a plan-based approach to conduct task-oriented dialogues which aim at accomplishing concrete tasks. For linguistic analysis and generation, we claim that a ``pragmatics-first'' view on rational interaction provides an appropriate framework for flexible and scalable dialogue modeling. CONALD combines deep syntactic and semantic analysis, discourse processing, and language generation and features a complex semantics-pragmatics interface. Semantics is defined in terms of an extended version of DRT. Word lattices as produced by a speech recognizer are analyzed by a chunk parser which in turn transforms what can be recognized from the user's utterance into a semantic representation, from which the dialogue system derives a goal for the application. For the reasoning part, i.e. knowledge representation and inference for the interpretation of dialogue as well as for planning to satisfy user goals in the application domain, we argue for a computational logic framework. Discourse and application knowledge are considered independent; user utterances affect operations in an application domain when their speech acts are executed. Dialogue coherence is determined by analyzing the content of contributions to a dialogue with the help of a partial logic: for each new utterance, the dialogue manager evaluates how it contributes to the completion of one of the currently active plans. Due to the clear functional separation between the language model, the dialogue model, and the domain model, the feedback cycle between the dialogue manager and the application is of vital importance, as is the close interaction on the linguistic side between the speech parser, the dialogue manager and the text generation module. CONALD has been implemented in a multi-agent system framework which facilitates quick configurability for various applications.