Meetings Sommer 96

18. Juni 1996

Traditional philosophical approaches to reference take neither time nor learning into account, let alone the referring individual. On the contrary, their models assume an ideal agent which can determine the referent of a term perfectly and instantaneously. However, this could only be possible in Plato's heaven. In the real world, even the simplest form of reference, empirical reference, is a complicated process that involves associating visual objects and auditory signals. We believe that learning empirical reference is the first step in language development, and other forms of reference such as indexicals are based on it. In this talk we will discuss the problems arising with empirical reference and the requirements necessary for modeling referential behavior. We will suggest a (connectionist) model consisting of an auditory, a visual, and an associative component which seems to be suitable for learning empirical reference.

4. Juni 1996

Abstract in preparation

21. Mai 1996

ACHTUNG: Der Vortrag muß aus familiären Gründen entfallen!

(Der zu Semesterbeginn angesetzte Vortrag von Astrid v. Stein entfällt)

7. Mai 1996

23. April 1996

26. März 1996

Das "Beobachterproblem" hat besonders fuer die Erforschung der Kognition Bedeutung, da der Untersuchungsgegenstand selbst Beobachter ist. Wenn wir nach der Relevanz unserer Modelle fragen, uns mit Semantik beschaeftigen oder das Verhaeltnis von Objektivitaet und Subjektivitaet untersuchen, kommen wir um das Beobachterproblem nicht herum.

Ausgangspunkt fuer meinen Exkurs zum Beobachterproblem ist die Physik. Ich werde ein einfaches Resultat der Einstein'schen speziellen Relativitaetstheorie vorstellen und daraus Aussagen ueber Beobachtungen ziehen, die sich auf Information und Informationsverarbeitung anwenden lassen. Eine Erweiterung der sehr einfachen Beobachtungen der speziellen Relativitaetstheorie um Nichtlinearitaet und Nichteindeutigkeit eroeffnet einen Zugang zu Chaostheorie und zur Heisenberg'schen Unschaerferelation. (Keine Angst, alle Schritte lassen sich einfach und in Beziehung zu Konzepten aus der KW darstellen.) Abschliessend werde ich die gewonnene Sichtweise auf die KW anwenden und eine Art Unschaerferelation der KW formulieren (zwischen natur- und geisteswissenschaftlichem Zugang).

19. März 1996

In this talk I will introduce a connectionist model for concept formation and word learning, relying very much on the idea of concepts being the result of a categorization process making state vectors of different concepts orthogonal to each other. Several hypotheses raised by the model and a concrete experimental setting to test these hypotheses will be described and discussed.

The approach, mainly based on elemnts of competitive learning and adaptive resonance theory, is in contrast to other connectionist models based solely on multilayer perceptrons and backpropagation. I will discuss the differences between the approaches, the difficulties with each of them with respect to important properties of word learning, and propose a possible way of combining the ideas into one model.

In addition to aspects arising in word learning I will also talk about several more general issues that arise in the context of building connectionist cognitive models. They concern both weaknesses and strengths of the approach. I will argue that many aspects of connectionism have not been fully explored yet, and that a more thorough analysis of the models will be necessary (i.e. we cannot rely on "black-box" models that somehow seem to fit the observered data.)