A holistic approach to artificial neural network models of language emergence and language acquisition - Case studies on pragmatic reasoning and language-perception interactions
Please use this identifier to cite or link to this item:
https://doi.org/10.48693/379
https://doi.org/10.48693/379
| Title: | A holistic approach to artificial neural network models of language emergence and language acquisition - Case studies on pragmatic reasoning and language-perception interactions |
| Authors: | Ohmer, Xenia |
| ORCID of the author: | https://orcid.org/0000-0002-1295-0973 |
| Thesis advisor: | Prof. Dr. Michael Franke Prof. Dr. Peter König |
| Thesis referee: | Prof. Dr. Thomas Brochhagen |
| Abstract: | Humans interact with each other and with their environment. On the one hand, language is shaped by these interactions as we communicate about our experiences. On the other hand, language shapes these interactions as we use it to act in the world (by informing, requesting, promising, and so on). As a consequence, language is intertwined with many cognitive functions, such as perception, action, and social reasoning. To understand how language emerged and evolved, as well as how language is learned and used, it is important to take these connections into account. This thesis presents three case studies that investigate interfaces between language and other areas of cognition. The case studies use computational, agent-based models to study language acquisition and language emergence phenomena. We follow a long tradition of modeling interactive language use in the form of communication games. In all case studies, the agents play games that involve generating and/or interpreting references to objects, which is a simple but fundamental use of language. The agents are implemented as artificial neural networks. Artificial neural networks not only dominate machine learning research but are also popular as models of cognitive functions. We use the case studies to learn about cognitive mechanisms related to language as well as to suggest ways in which artificial neural networks may benefit from integrating such mechanisms. Case study 1 shows how reasoning about the speaker’s intention and the context can help children learn the meanings of new words. Case study 2 demonstrates how considerations about the context can lead to the emergence of object references at different levels of specificity (as in "Fido", "dalmatian", "dog'', "animal"). Case study 3 models aspects of the bidirectional influence between visual perception and (emergent) language. In general, I argue that the combination of communication games and artificial neural networks generates a versatile framework for studying language-cognition interfaces. A holistic approach to modeling language will increase our understanding of how humans learn and use language as well as our ability to emulate this behavior in machines. |
| URL: | https://doi.org/10.48693/379 https://osnadocs.ub.uni-osnabrueck.de/handle/ds-202308049542 |
| Subject Keywords: | artificial intelligence; language evolution; word learning; deep learning; agent-based simulation |
| Issue Date: | 4-Aug-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_ohmer.pdf | Präsentationsformat | 7,69 MB | Adobe PDF | thesis_ohmer.pdf  View/Open |
This item is licensed under a
Creative Commons License
