Related Experiment Video
Updated: Jun 13, 2026
13:32
Recording Human Electrocorticographic (ECoG) Signals for Neuroscientific Research and Real-time Functional Cortical Mapping
Published on: June 26, 2012
DiagPat: An Explainable Language Detection Model Using EEG Signals
Tugce Keles1, Kubra Yildirim1, Dahiru Tanko2
1Department of Digital Forensics Engineering, College of Technology, Firat University, 23119 Elazig, Turkey.
Sensors (Basel, Switzerland)
|June 12, 2026
Summary
This study introduces DiagPat, an explainable feature engineering model for electroencephalography (EEG) language detection. DiagPat accurately classifies languages and tasks from brain activity, offering a lightweight and interpretable solution for EEG-based language identification.
Area of Science:
- Neuroscience
- Computational Linguistics
- Biomedical Engineering
Background:
- Electroencephalography (EEG) is a non-invasive, cost-effective tool for studying brain activity during language processing.
- Previous EEG language studies faced limitations: small datasets, focus on native speakers or speech units, limited experimental settings, and reliance on complex, uninterpretable deep learning models.
Purpose of the Study:
- To develop and validate an explainable feature engineering (XFE) model for accurate EEG-based language detection.
- To address limitations of prior studies by using a larger, diverse dataset and focusing on direct language and task mode classification.
- To create a computationally efficient and interpretable framework for analyzing brain activity related to language.
Main Methods:
- Curated a new EEG dataset from 346 participants (Arabic and Turkish) in reading and listening modes, yielding 6364 EEG segments.
- Proposed DiagPat, an XFE model using diagonal pattern analysis for feature extraction from EEG channels and signals.
- Integrated DiagPat with iterative neighborhood component analysis (INCA) for feature selection and a k-nearest neighbors (tkNN) classifier for prediction, utilizing Directed Lobish (DLob) for explainability.
Main Results:
- Achieved >90% accuracy across nine classification cases (language, mode, mixed), with accuracies ranging from 92.14% to 99.35% (10-fold CV).
- Generated case-specific cortical connectome diagrams for interpretable characterization of language- and mode-related brain activity.
- Reported subject-independent accuracies (LOSO CV) ranging from 29.75% to 83.50%, demonstrating generalization capabilities.
Conclusions:
- DiagPat offers an accurate, lightweight, and explainable framework for EEG-based language detection and task mode classification.
- The model's explainability through cortical connectome diagrams aids in understanding language-related brain activity.
- The study provides a robust methodology for advancing EEG applications in neurolinguistics and brain-computer interfaces.
