MoL-2025-21: Lonke, Dorielle (2025) A Linguistically Grounded Evaluation of Anthropomorphic Language Detection in AI Research. [Report]
![[thumbnail of MoL-2025-21.text.pdf]](https://eprints.illc.uva.nl/style/images/fileicons/text.png) |
Text
MoL-2025-21.text.pdf - Published Version Download (790kB) |
Abstract
This thesis presents a conceptual framework of anthropomorphic language in AI research that serves as a theoretical baseline for evaluating existing approaches for its detection. Drawing on existing work, a taxonomy of human-like attributes frequently ascribed to AI systems is outlined. Relying on linguistic theory of grammatical animacy, as well as thematic proto-role theory and frame semantics, a linguistic model is constructed, representing the various ways in which anthropomorphic descriptions are expressed in natural language. The linguistic model serves as a baseline for the construction of a challenge set that is used towards the evaluation of the state-of-the-art approaches to anthropomorphic language detection. The evaluation shows that the current unsupervised approaches are not suitable for all types of linguistic structures, suggesting the need for alternative methods. Furthermore, due to the increasingly anthropomorphic language employed in body of work that pertains to AI technologies, the question arises whether unsupervised approaches relying on large language models trained on this type of data are a reliable method for detecting this phenomenon.
| Item Type: | Report |
|---|---|
| Report Nr: | MoL-2025-21 |
| Series Name: | Master of Logic Thesis (MoL) Series |
| Year: | 2025 |
| Subjects: | Language Logic |
| Depositing User: | Dr Marco Vervoort |
| Date Deposited: | 16 Oct 2025 23:06 |
| Last Modified: | 16 Oct 2025 23:06 |
| URI: | https://eprints.illc.uva.nl/id/eprint/2392 |
Actions (login required)
 |
View Item |
