Semantic Feature Extraction for Task-Agnostic and Cross-Lingual Dementia Detection

Abstract

Nowadays, with dementia being a prevalent condition in the older population, research has turned towards the analysis of speech by leveraging natural language processing methods and applying them to text and audio data as a potential method for an efficient and non-invasive detection. 

Commonly used tests in dementia diagnosis are picture description tasks, such as the "Cookie Theft" test, in which patients are asked to describe what they see in the images. These interviews are often conducted in English, resulting in a majority of commonly used datasets and subsequent work being centered around the English language, leaving a largely unexplored gap regarding language diversity. Furthermore, the use of these picture description tasks also limits the textual domain that these models work with to the contents displayed in the images. Therefore, this thesis aims to extract semantic features from these text transcripts with an emphasis on analyzing their effectiveness and transferability to other domains and languages, specifically German.

Research Questions:
 

1. How can NLP methods be leveraged to analyse semantic patterns in texts to aid in Alzheimer’sDisease detection?
2. How well can these captured patterns be utilized outside of the Cookie Theft Task and what are the key challenges in doing so?
3. What methods can be used to transfer these semantic patterns to other languages and to what extent?

---

Attribute	Value
Title (de)	Syntaktische und Semantische Featureextraktion für Aufgaben- und Sprachagnostische Demenzerkennung
Title (en)	Syntactic and Semantic Feature Extraction for Task- and Language-agnostic Dementia Detection
Project
Type	Master's Thesis
Status	started
Student	Peter Yi Shan
Advisor	Joshua Oehms
Supervisor	Prof. Dr. Florian Matthes
Start Date	15.12.2025
Sebis Contributor Agreement signed on
Checklist filled	No
Submission date	15.06.2026