Ulrich Reimer

Topic essentials

An overview of TOPIC is provided, a knowledge-based text information system for the analysis of German-language texts. TOPIC supplies text condensates (summaries) on variable degrees of generality and makes available facts acquired from the texts. The presentation focuses on the major methodological principles underlying the design of TOPIC: a frame representation model that incorporates various integrity constraints, text parsing with focus on text cohesion and text coherence properties of expository texts, a lexically distributed semantic text grammar in the format of word experts, a model of partial text parsing, and text graphs as appropriate representation structures for text condensates.

Automatic generation of hypertext knowledge bases

A model of knowledge-based text condensation is presented which has been implemented as part of the text analysis system TOPIC. The condensation process transforms the text representation structures resulting from the text parse into a more abstract thematic description of what the text is about, filtering out irrelevant knowledge structures and preserving only the most salient concepts. The topical structure of a text, finally, is represented in a hierarchical text graph which supports variable degrees of abstraction for text summarization as well as content-oriented retrieval of text knowledge. Due to their non-linear organization, text graphs share a lot of similarities with hypertexts. Their contribution to this field incorporates a methodology for the automatic generation of hypertexts from given full-text files, a close coupling of basic hypertext notions (links, nodes) to the formal specifications of a frame representation model, and conceptual navigation and filtering facilities which allow a user-defined level of information granularity when accessing hypertext knowledge bases.

A frame-based knowledge representation model and its mapping to nested relations

Abstract The first part of this paper introduces the basic constructs of a frame representation model and gives a formal definition for them. Subsequently the nested relational model (NF2) is described as an extension of the classical relational model to support relation-valued attributes. In the second part of this paper both models are compared with each other and a mapping of frame representation structures to NF2 structures is specified. The structural similarities between frames and NF2 relations are made clear and it is shown that their main difference is due to the type polymorphism introduced by the concept hierarchy of the frame model. This causes type collisions to occur in the strictly typed NF2 model when frames are directly mapped to NF2 structures. Two solutions to this problem are suggested and compared. The paper concludes with a reformulation of query operations of the frame model in terms of NF2 algebra operations.

A representation construct for roles

Abstract This paper introduces a construct for representing concepts together with the roles they can play. Its definition is based on facilities for representing generalization (is-a), classification (instance-of), property domains and actual properties. Although for illustation purposes a frame data model is used as a formal basis, the role construct can be adopted by any other data model (or knowledge representation model) which supports these facilities. The semantics of the role construct is specified by integrity constraints which cover, besides other aspects, a facility for sharing properties among roles. Operational characteristics are described with respect to the most relevant modification operations. A comparison with the various intuitive notions of roles which currently exist shows that they all can be reinterpreted in terms of the role construct and their meaning thus be made explicit.

Inhaltsorientierte Navigation in automatisch generierten Hypertext-Basen

Der automatische Aufbau von Hypertexten aus Kollektionen linearer Texte erfordert Verfahren zur Analyse und Segmentierung von Texten, sowie zur Generierung von Hypertext-Kanten. In diesem Beitrag werden Theorien zur Beschreibung der thematischen Struktur von Texten aufgegriffen und zur Entwicklung von Kriterien genutzt, die es erlauben, inhaltlich begrundete Kanten zwischen Textfragmenten zu erzeugen. Textgraphen resultieren als netzwerkartige Reprasentationen von Texten und dienen einem auf objektorientierter Interaktion basierenden Dialogmodell als Ausgangsbasis zur semantisch kontrollierten Exploration der Hypertext-Basis.

Entwurfsprinzipien und Architektur des Textkondensierungssystems TOPIC

Ausgehend von einer informationswissenschaftlichen Motivation fur die Volltextverarbeitung in Informationssystemen wird zunachst ein Uberblick uber die konzeptionellen Grundlagen des Textkondensierungssystems TOPIC gegeben: partielles Textverstehen als fur das indikative Abstracting funktional adaquate Parsing-Strategie, Integritat der auf einem formal spezifizierten Frame-Reprasentationsmodell basierenden Wissensbasis, Erkennung von Textkohasionsund Textkoharenzmustern in Fachtexten durch einen entsprechend ausgerichteten Text-Parser sowie unterschiedliche thematische Aggregationsgrade der erzeugten Textkondensate (Textzusammenfassungen). Anschliesend wird eine strukturierte Beschreibung des gesamten TOPIC-Systems gegeben, das sich aus drei Hauptprozessen zusammensetzt: Text-Parsing auf der Basis extensiver Interaktionen zwischen Frame-Wissensbasis und Wortexperten-Parser, Textkondensierung auf der Basis dominanter Konzeptmuster der Textreprasentationsstrukturen und schlieslich der Aufbau eines Textgraphen, der die Wissensstrukturen des Kondensats auf unterschiedlichen Ebenen inhaltlicher Komprimierung enthalt, auf die beim graphisch-interaktiven Retrievaldialog flexibel zugegriffen werden kann.