Research projects
In this section, you can find a quick overview of the ongoing research projects that I (co)supervise.
Most of the listed projects are conducted at the Department of Artificial Intelligence and Computational Cognitive Science at the University of Groningen.
​
Additionally, there are previous projects initiated at the Faculty of Arts (RUG) and international collaborations.
Robotics
WOW Bilingual Robot Game: Enhancing English Learning in Dutch Children
This study investigates whether the WOW bilingual robot game, delivered via the Alpha Mini robot, enhances English vocabulary learning in Dutch 7–8-year-old children. The primary outcome is word recall (production), not pronunciation. A between-subjects design ensures each child experiences only one condition, avoiding carryover effects.
Singing to Remember: The Impact of AI-Generated Music on L2 Retention and Recall on Children
This study examines how AI-generated music and social robot interaction affect second language (L2) vocabulary retention and recall in children. It explores whether LLM-generated songs delivered by a social robot improve L2 learning compared to traditional or single-modality methods.
Aphasia & AI
The Predictive Power of Syntactic and Semantic Markers of Speech in Schizophrenia Spectrum Disorder and Wernicke’s Aphasia
​​​
​
This study explores whether syntactic, semantic, and acoustic speech markers can distinguish between Schizophrenia Spectrum Disorders (SSD), Wernicke’s aphasia, and healthy controls. Using NLP and audio analysis on patient interview data, the goal is to identify predictive features for diagnosis. Results may inform automated diagnostic tools to support clinical decision-making.
Utilizing Automatic Speech Recognition for the Detection of Speech Errors to Classify Aphasia Types
This study uses Automatic Speech Recognition (ASR) to identify error patterns in aphasic speech, such as substitutions and deletions. By analyzing ASR output, a classifier could predict aphasia based on error rates, providing a potential quick diagnostic tool (e.g., flagging possible aphasia if WER exceeds a threshold).
Utilizing LLMs to Classify Aphasia Types in Greek
In this study, a Multilayer Perceptron (MLP) neural network was developed to classify transcribed speech data from Greek native speakers with aphasia, specifically distinguishing between ​fluent and non-fluent speech​. This approach aims to address the complexities inherent in aphasia classification by leveraging machine learning techniques to identify distinguishing linguistic patterns within the Greek-speaking population.
Multilingual Processing
Multilingual Language Processing of Speakers of Hindi and Sindhi
This project investigates how individuals process sentences in Hindi, focusing on comprehension during listening. The study aims to better understand how language experience shapes sentence processing, particularly in bilingual contexts where the language is not formally learned.
Bilingual Processing of Evidentiality in Bulgarian
This study investigates how L1 and L2 Bulgarian speakers process evidentiality, a grammatical marker indicating information source. Using a self-paced reading task, results showed differences in reading speed between evidential types, but no significant group differences or accuracy effects.
The Role of Language Dominance in Cross-Linguistic Influence: The Case of Suret
This study examines whether language dominance predicts Cross-Linguistic Influence (CLI) in Suret-Dutch/Swedish bilinguals. Two generations completed a question-formation task, but no significant effect of dominance was found. Results suggest language dominance is not a reliable predictor of CLI.
Processing of grammatical evidentiality in Turkish:
An ERP study
This study investigates the neural processing of evidentiality in Turkish using EEG. Evidentiality, marked by two obligatory suffixes in Turkish, indicates whether information is directly or indirectly acquired. Native speakers are tested with grammatical and ungrammatical sentences. It is hypothesized that violations will elicit a P600 and possibly an N400, reflecting both morphosyntactic and pragmatic processing.
Improving Transfer Learning for NER for Indian Languages
This project investigates cross-lingual transfer learning for Named-Entity Recognition (NER) across ~40 Indian language pairs, each combining a high-resource and low-resource language. It explores how linguistic relatedness—based on syntactic, phonological, and genetic distances—affects model performance. The study also compares monolingual vs. multilingual models and zero-shot vs. few-shot transfer.
Design Proposal, Implementation and Evaluation of an SRS-Driven Agentic Conversational Chatbot
This study motivates the design and evaluates the effectiveness of an agentic architecture that leverages state-of-the-art conversational LLMs to maximize vocabulary acquisition and optimize retrieval pathways for fluent conversation. The overall architecture integrates spaced repetition memory research to enhance memory entrenchment and employs conversational interactions to strengthen retrieval pathways, enabling comprehensive integration of new vocabulary.