Dokument: Prototype Frames - A probabilistic account of typicality
| Titel: | Prototype Frames - A probabilistic account of typicality | |||||||
| URL für Lesezeichen: | https://docserv.uni-duesseldorf.de/servlets/DocumentServlet?id=69613 | |||||||
| URN (NBN): | urn:nbn:de:hbz:061-20250513-081531-2 | |||||||
| Kollektion: | Dissertationen | |||||||
| Sprache: | Englisch | |||||||
| Dokumententyp: | Wissenschaftliche Abschlussarbeiten » Dissertation | |||||||
| Medientyp: | Text | |||||||
| Autor: | Schuster, Annika Noel [Autor] | |||||||
| Dateien: |
| |||||||
| Beitragende: | Schurz, Gerhard [Gutachter] Prof. Dr. Hampton, James [Gutachter] | |||||||
| Dewey Dezimal-Klassifikation: | 100 Philosophie und Psychologie | |||||||
| Beschreibung: | In the 1970s, the psychologist Eleanor Rosch and her team found in a series of experiments that people overwhelmingly agree which category members are typical for their category (C) and which are not – apples are typical fruit, hammers are typical tools, robins are typical birds and chairs are typical furniture, while olives, scissors, emus and refrigerators are atypical examples of these categories. The typicality ordering of a C that is obtained from mean typicality ratings for many subcategories (SCs) is assumed to indicate an important component of the structure of mental representations of Cs, category concepts: in line with the thoughts presented in late Wittgenstein’s natural language philosophy, it is commonly assumed that there are additional, typicality-contributing properties stored in category concepts. Classical definitions, which consist of singly necessary and jointly sufficient conditions, cannot explain the internal structure of category concepts, because the fulfilment of these conditions makes all members equally representative of C. Typicality-contributing properties are shared between subsets of SCs, albeit not present in all of them, and constitute a family resemblance relationship between SCs, much like the physical features shared between family members. In addition to being the ripened reproductive bodies of a seed plant, apples, like many other fruits, taste sweet and have bright colours and olives do not, and in addition to being warm-blooded egg-laying vertebrates, robins are small, sing and fly while emus are big and do not sing or fly. Furthermore, not all properties are equally important for typicality. A food item that is sweet will be rated as more typical than one that is brightly coloured and a bird that flies will be rated as more typical than one that is small. A formal model of category representation that contains typical properties and reflects the extent of their typicality-contribution provides an enriched picture of mental representations and is a suitable basis for the investigation of phenomena like conceptual combination and reasoning patterns involving category concepts.
In this thesis, I develop such a model: probabilistic prototype frames. I investigate theoretically and empirically how typicality-contributing properties of Cs can be identified and how the extent of their typicality contribution can be quantified in a way that predicts and explains the typicality ordering of Cs. Schurz’ (2001, 2005, 2012) evolution-theoretic account of normality offers an intriguing way to identify typicality-contributing properties and base reasoning with prototype concepts on an ontologically justified probabilistic framework. He argues that not only the formation of biological categories, but also that of many other common-sense categories, is guided by the evolutionary principles of selection, variation and reproduction which lead to a prototypical norm state in which most category members are most of the time. This state is characterised by prototypical properties in the wide sense (iws), for which the conditional probability of the property P given the category C, pr(P|C), is high, and prototypical properties in the narrow sense (ins), for which additionally the reverse conditional probability, pr(C|P), is high. I term the former frequent and the latter diagnostic properties and I will argue that a property’s typicality contribution is best quantified with subjective probability estimations of its diagnosticity for and probability in the category. Frames enable a fine-grained representation of conceptual structure and are therefore a suitable representation format for this purpose. While it is assumed by many that category prototypes are identifiable, specific accounts of the identification of typical properties and parametrical representation of prototypes that predict typicality are rare. Two remarkable exceptions are the family resemblance score (Rosch & Mervis, 1975), which represents prototypes in property lists with application scores that reflect each property’s frequency in the category, and a frame-adapted version of Tversky’s contrast model (Smith et al., 1988), which represents prototypes in frames that quantify each value with its frequency in the category, measured as number of votes, and diagnosticity weights derived from the frequencies of the values in the category and in the contrast category. I will show that the probabilistic prototype model makes typicality predictions that are correlated with mean rated typicality of the same magnitude as the other models. Furthermore, they have the advantage of only relying on property generation data and probability estimations, without employing the unclear and possibly biased notions of property applicability and number of votes. Conversely, probability ratings can be easily used to represent the variables in the other models and then predict typicality with correlations in the same order of magnitude as the other models. Many promising accounts for various phenomena related to prototype theory have been brought forward. This thesis aims at contributing a way to find probabilistic representations of category prototypes that provide a solid foundation for further work. It is a contribution to the extension of frame theory developed in the CRC991 “The structure of representations in language, cognition and science” (e.g., Petersen (2007), Votsis, Schurz (2012), Löbner (2014), Kornmesser and Schurz (2020)) as well as to research questions on the structure and content of mental representations. Concepts are a topic of cognitive science and work on them is part of an interdisciplinary investigation involving disciplines such as philosophy, linguistics, psychology and information technology. While probabilistic prototype frames are a normative account of concepts and as such a philosophical contribution to the field, the use of psychological and statistical techniques for their empirical verification is indispensable and critically reflected upon throughout this thesis. | |||||||
| Lizenz: |  Dieses Werk ist lizenziert unter einer Creative Commons Namensnennung 4.0 International Lizenz | |||||||
| Fachbereich / Einrichtung: | Philosophische Fakultät » Philosophisches Institut | |||||||
| Dokument erstellt am: | 13.05.2025 | |||||||
| Dateien geändert am: | 13.05.2025 | |||||||
| Promotionsantrag am: | 04.05.2022 | |||||||
| Datum der Promotion: | 22.06.2022 |
