Update README.md

README.md

@@ -28,8 +28,7 @@ Unique Words** subset against three competitive attention baselines under a
 controlled multi-seed protocol.
 
 The repository includes four base architectures (Baseline, Luong, MHSA, and
-the proposed FAA) and three FAA checkpoints fine-tuned for cross-domain
-adaptation to Kurdish Person, Place, and Month name subsets.
+the proposed FAA).
 
 ## Repository Structure
 
@@ -44,9 +43,6 @@ KHWR/
 ├── Baseline-Word-Model/             # CRNN without attention (seed 42)
 ├── Luong-Word-Model/                # Luong multiplicative attention (seed 42)
 ├── MHSA-Word-Model/                 # Multi-Head Self-Attention (seed 42)
-├── FAA-Person-Names-Adapted/        # FAA fine-tuned on Person Names (100%)
-├── FAA-Place-Names-Adapted/         # FAA fine-tuned on Place Names (100%)
-├── FAA-Month-Names-Adapted/         # FAA fine-tuned on Month Names (100%)
 ├── Scripts/
 │   ├── train.py
 │   └── inference.py
@@ -90,24 +86,6 @@ Test set of 8,036 word images, five random seeds (42, 7, 123, 456, 789):
 | MHSA | 0.0403 ± 0.0023 | 0.1709 ± 0.0079 |
 | **FAA (proposed)** | **0.0358 ± 0.0013** | **0.1426 ± 0.0044** |
 
-The improvements of FAA over Luong and MHSA are statistically significant
-under both the paired bootstrap (B = 100,000) and the seed-level Wilcoxon
-signed-rank test (p = 0.031 for each), and borderline against the Baseline
-(p = 0.063) where FAA wins on four of five seeds and ties on the fifth.
-
-### CTC Beam Search with Character-Level Language Model Rescoring
-
-Seed-42 FAA checkpoint, NVIDIA RTX 4060:
-
-| Decoding Strategy | CER | WER | Time (ms) |
-|-------------------|:-:|:-:|:-:|
-| Greedy CTC | 0.0373 | 0.1480 | 0.7 |
-| Beam search (k = 10) | 0.0369 | 0.1470 | 11.4 |
-| Beam-10 + 5-gram LM (w = 0.4) | 0.0334 | 0.1279 | 11.5 |
-| **Beam-20 + 5-gram LM (w = 0.4)** | **0.0332** | **0.1273** | **42.4** |
-
-The best configuration reduces the CER by 11.0% relative to greedy decoding.
-
 ### Few-Shot Cross-Domain Adaptation
 
 FAA seed-42 checkpoint adapted to three additional DASTNUS subsets:
@@ -197,9 +175,6 @@ python Scripts/train.py \
 | `Baseline-Word-Model/` | CRNN (no attention) | 0.0380 | 0.1544 |
 | `Luong-Word-Model/` | CRNN + Luong attention | 0.0391 | 0.1663 |
 | `MHSA-Word-Model/` | CRNN + Multi-Head Self-Attention | 0.0383 | 0.1655 |
-| `FAA-Person-Names-Adapted/` | FAA fine-tuned (Person Names, 100%) | 0.0380 | 0.1543 |
-| `FAA-Place-Names-Adapted/` | FAA fine-tuned (Place Names, 100%) | 0.0268 | 0.1230 |
-| `FAA-Month-Names-Adapted/` | FAA fine-tuned (Month Names, 100%) | 0.0087 | 0.0527 |
 
 All values reported on the held-out test split of the corresponding subset
 under greedy CTC decoding, seed 42.
@@ -212,22 +187,11 @@ text dataset. Relevant statistics:
 | Subset | Samples | Unique Words | Vocabulary |
 |--------|:-:|:-:|:-:|
 | Unique Words | 54,191 | 2,750 | 21,796 (full DASTNUS) |
-| Person Names | DASTNUS subset | — | — |
-| Place Names | DASTNUS subset | — | — |
-| Month Names | DASTNUS subset | 12 | 12 |
-
-Writer-disjoint 70 / 15 / 15 train / validation / test splits.
 
 ## Citation
 
 ```bibtex
-@article{KHWR2026,
-  title   = {Text-Based Kurdish Handwritten Word Recognition with
-             Frequency-Adaptive Attention on the DASTNUS Dataset},
-  author  = {[Author list to be added]},
-  journal = {[Target venue]},
-  year    = {2026}
-}
+
 ```
 
 ## License