Model card
Assessment model
Glot estimates a CEFR-oriented merged proficiency band for German learner writing and provides structured linguistic feedback to support reflection, placement, and teacher review.
Overview
- Model name
- 20260529_115718_C2b_RandomForest
- Modelling approach
- Linguistic feature analysis with machine learning
- Algorithm
- RandomForest
- Label scheme
- Merged bands
- Feature count
- 32
Output labels
The model predicts one of three merged CEFR-oriented proficiency bands. It does not distinguish exact A1–C2 levels.
Elementary
Assessed band
Elementary (A1–A2-oriented evidence)
Basic user
Elementary
Intermediate
Advanced
Intermediate
Assessed band
Intermediate (B1–B2-oriented evidence)
Independent user
Elementary
Intermediate
Advanced
Advanced
Assessed band
Advanced (C1–C2-oriented evidence)
Proficient user
Elementary
Intermediate
Advanced
Performance
Metrics from held-out test data. These reflect the model's performance under the training conditions and may not generalise to all learner populations.
Accuracy
92.3%
Macro F1
92.3%
| Band | Precision | Recall | F1 |
|---|---|---|---|
| Elementary | 95.5% | 91.5% | 93.5% |
| Intermediate | 88.4% | 88.4% | 88.4% |
| Advanced | 93.0% | 97.0% | 94.9% |
Limitations
- Trained on written German texts; not validated on speech transcripts.
- Predicts merged proficiency bands only (Elementary, Intermediate, Advanced). It does not distinguish A1 from A2, B1 from B2, or C1 from C2.
- Confidence scores are model probabilities, not calibrated certainty estimates. They reflect the classifier's internal decision boundary, not real-world measurement precision.
- Performance varies by proficiency level and text length. Short texts and mixed-level writing may produce less reliable estimates.
Glot does not provide official CEFR certification or exact A1–C2 classification. Automated assessment supports reflection and placement decisions, but it should not replace expert human evaluation.