SGlot

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%

BandPrecisionRecallF1
Elementary95.5%91.5%93.5%
Intermediate88.4%88.4%88.4%
Advanced93.0%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.