Spike-based acquisition of grammatical structure: on the importance of low-level brain mechanisms for high-level cognitive skills
Please use this identifier to cite or link to this item:
https://doi.org/10.48693/405
https://doi.org/10.48693/405
Title: | Spike-based acquisition of grammatical structure: on the importance of low-level brain mechanisms for high-level cognitive skills |
Authors: | Lehfeldt, Sophie |
ORCID of the author: | https://orcid.org/0000-0002-7630-1024 |
Thesis advisor: | Prof. Dr. Gordon Pipa |
Thesis referee: | Prof. Dr. Jutta L. Mueller Prof. Dr. Peter König |
Abstract: | Cognition can be studied across a variety of interfaces. While the 'brain-cognition' interface highlights the aim to build explanatory models for the implementation of cognition in the biological brain, 'low-high' and 'subsymbolic-symbolic' interfaces demonstrate the scope of possible representation and computation types underlying cognition. This thesis aimed at contributing potential connecting links across these interfaces by taking a closer look at the underlying learning mechanisms of human grammar acquisition (i.e. a high-level and possibly symbolic cognitive skill). Motivated by the hypothesis that especially early childhood language acquisition phases are characterised by associative, statistical learning types, the acquisition of isolated and nested non-adjacent grammars was modelled by associative, statistical learning in a spiking recurrent neural network (i.e. a low-level and subsymbolic mechanism). By demonstrating that grammar learning outcomes of the model (i.e. distributed synapse assembly strengths) complied with a variety of grammar acquisition performance patterns of human learners, associative, statistical learning was identified as an essential contributor for successful grammar learning and potentially also language acquisition. Moreover, given that a low-level and subsymbolic mechanism accounted for aspects of high-level human grammar learning, this cognitive skill might potentially be grounded on a subsymbolic basis. Finally, by including core brain components and processing principles into the model (i.e. a generic recurrent neural network, distributed encoding and unsupervised learning), a potential minimal set of neurobiological requirements for non-adjacent grammar learning was identified. Taken together, this thesis provided a connecting link between the fields of psycho- and neurolinguistics and neuroinformatics that was identified by a salient commonality in both, namely the central role of associative, statistical learning. The model especially described how a naive learner might acquire first knowledge about the statistics of the surrounding environment by passive exposure. However, other cognitive mechanisms that potentially build up on the formed low-level knowledge representations might go beyond mere associative, statistical learning. This thesis therefore provided a potential starting point for future research with a focus on understanding in more detail the potential interplay of different cognitive strategies: starting from low-level and subsymbolic learning towards reaching high-level and symbolic cognition. |
URL: | https://doi.org/10.48693/405 https://osnadocs.ub.uni-osnabrueck.de/handle/ds-202310049807 |
Subject Keywords: | Cognitive Science; Neuroinformatics; Psycho- and Neurolinguistics |
Issue Date: | 4-Oct-2023 |
License name: | Attribution 3.0 Germany |
License url: | http://creativecommons.org/licenses/by/3.0/de/ |
Type of publication: | Dissertation oder Habilitation [doctoralThesis] |
Appears in Collections: | FB08 - E-Dissertationen |
Files in This Item:
File | Description | Size | Format | |
---|---|---|---|---|
thesis_lehfeldt.pdf | Präsentationsformat | 10,73 MB | Adobe PDF | thesis_lehfeldt.pdf View/Open |
This item is licensed under a Creative Commons License