# Generated Reference Token BiLSTM-CRF

This model is the token-level baseline trained from the analyzer-derived weak labels.

## Split Summary

- Train sentences: 6719
- Dev sentences: 840
- Test sentences: 840
- Train tokens: 26590
- Dev tokens: 3352
- Test tokens: 3332

## Metrics

- Dev semantic accuracy: 0.9192
- Dev syntactic accuracy: 0.9186
- Dev pragmatic accuracy: 0.9436
- Dev joint sequence accuracy: 0.7512
- Test semantic accuracy: 0.9214
- Test syntactic accuracy: 0.9208
- Test pragmatic accuracy: 0.9469
- Test joint sequence accuracy: 0.7738

## Label Maps

### semantic

| ID | Label |
|---|---|
| 0 | `حالة` |
| 1 | `زمان` |
| 2 | `متقبل` |
| 3 | `محمول` |
| 4 | `مكان` |
| 5 | `منادى` |
| 6 | `منفذ` |
| 7 | `—` |

### syntactic

| ID | Label |
|---|---|
| 0 | `اسم_استفهام` |
| 1 | `اسم_كان` |
| 2 | `ظرف` |
| 3 | `فاعل` |
| 4 | `فعل` |
| 5 | `فعل_ناقص` |
| 6 | `مفعول_به` |
| 7 | `منادى` |
| 8 | `—` |

### pragmatic

| ID | Label |
|---|---|
| 0 | `بؤرة_جديد` |
| 1 | `بؤرة_مقابلة` |
| 2 | `تمهيد` |
| 3 | `ذيل` |
| 4 | `مبتدأ` |
| 5 | `محور` |
| 6 | `منادى` |
| 7 | `موضوع` |
| 8 | `—` |

## History

| Epoch | Train loss | Dev loss | Dev sem | Dev syn | Dev prag | Dev joint |
|---|---:|---:|---:|---:|---:|---:|
| 1 | 2.759 | 1.3413 | 0.8699 | 0.8699 | 0.9045 | 0.575 |
| 2 | 1.1177 | 1.0195 | 0.8965 | 0.8974 | 0.9266 | 0.6738 |
| 3 | 0.7277 | 0.9254 | 0.9081 | 0.9078 | 0.9382 | 0.7131 |
| 4 | 0.4847 | 0.9633 | 0.9126 | 0.9132 | 0.9382 | 0.7238 |
| 5 | 0.3237 | 0.9584 | 0.9147 | 0.915 | 0.9394 | 0.7429 |
| 6 | 0.2203 | 1.0079 | 0.9192 | 0.9186 | 0.9436 | 0.7512 |

## Notes

- This model uses weak token supervision derived from the current analyzer.
- Structural labels remain rule-derived and are not trained here.
- This is the first token-level publication-ready baseline for the current corpus state.