Upload all models and assets for ary (latest)
Browse filesThis view is limited to 50 files because it contains too many changes.
See raw diff
- .gitattributes +1 -0
- README.md +215 -175
- models/embeddings/aligned/ary_128d.bin +3 -0
- models/embeddings/aligned/ary_128d.meta.json +1 -0
- models/embeddings/aligned/ary_128d.projection.npy +3 -0
- models/embeddings/aligned/ary_128d_metadata.json +8 -0
- models/embeddings/aligned/ary_32d.bin +3 -0
- models/embeddings/aligned/ary_32d.meta.json +1 -0
- models/embeddings/aligned/ary_32d.projection.npy +3 -0
- models/embeddings/aligned/ary_32d_metadata.json +8 -0
- models/embeddings/aligned/ary_64d.bin +3 -0
- models/embeddings/aligned/ary_64d.meta.json +1 -0
- models/embeddings/aligned/ary_64d.projection.npy +3 -0
- models/embeddings/aligned/ary_64d_metadata.json +8 -0
- models/embeddings/monolingual/ary_128d.bin +2 -2
- models/embeddings/monolingual/ary_128d_metadata.json +1 -1
- models/embeddings/monolingual/ary_32d.bin +2 -2
- models/embeddings/monolingual/ary_32d_metadata.json +1 -1
- models/embeddings/monolingual/ary_64d.bin +2 -2
- models/embeddings/monolingual/ary_64d_metadata.json +1 -1
- models/subword_markov/ary_markov_ctx1_subword.parquet +2 -2
- models/subword_markov/ary_markov_ctx1_subword_metadata.json +2 -2
- models/subword_markov/ary_markov_ctx2_subword.parquet +2 -2
- models/subword_markov/ary_markov_ctx2_subword_metadata.json +2 -2
- models/subword_markov/ary_markov_ctx3_subword.parquet +2 -2
- models/subword_markov/ary_markov_ctx3_subword_metadata.json +2 -2
- models/subword_markov/ary_markov_ctx4_subword.parquet +2 -2
- models/subword_markov/ary_markov_ctx4_subword_metadata.json +2 -2
- models/subword_ngram/ary_2gram_subword.parquet +2 -2
- models/subword_ngram/ary_2gram_subword_metadata.json +2 -2
- models/subword_ngram/ary_3gram_subword.parquet +2 -2
- models/subword_ngram/ary_3gram_subword_metadata.json +2 -2
- models/subword_ngram/ary_4gram_subword.parquet +2 -2
- models/subword_ngram/ary_4gram_subword_metadata.json +2 -2
- models/subword_ngram/ary_5gram_subword.parquet +3 -0
- models/subword_ngram/ary_5gram_subword_metadata.json +7 -0
- models/tokenizer/ary_tokenizer_16k.model +2 -2
- models/tokenizer/ary_tokenizer_16k.vocab +0 -0
- models/tokenizer/ary_tokenizer_32k.model +2 -2
- models/tokenizer/ary_tokenizer_32k.vocab +0 -0
- models/tokenizer/ary_tokenizer_64k.model +2 -2
- models/tokenizer/ary_tokenizer_64k.vocab +0 -0
- models/tokenizer/ary_tokenizer_8k.model +2 -2
- models/tokenizer/ary_tokenizer_8k.vocab +0 -0
- models/vocabulary/ary_vocabulary.parquet +2 -2
- models/vocabulary/ary_vocabulary_metadata.json +9 -9
- models/word_markov/ary_markov_ctx1_word.parquet +2 -2
- models/word_markov/ary_markov_ctx1_word_metadata.json +2 -2
- models/word_markov/ary_markov_ctx2_word.parquet +2 -2
- models/word_markov/ary_markov_ctx2_word_metadata.json +2 -2
.gitattributes
CHANGED
|
@@ -39,3 +39,4 @@ visualizations/tsne_sentences.png filter=lfs diff=lfs merge=lfs -text
|
|
| 39 |
visualizations/tsne_words.png filter=lfs diff=lfs merge=lfs -text
|
| 40 |
visualizations/zipf_law.png filter=lfs diff=lfs merge=lfs -text
|
| 41 |
visualizations/position_encoding_comparison.png filter=lfs diff=lfs merge=lfs -text
|
|
|
|
|
|
| 39 |
visualizations/tsne_words.png filter=lfs diff=lfs merge=lfs -text
|
| 40 |
visualizations/zipf_law.png filter=lfs diff=lfs merge=lfs -text
|
| 41 |
visualizations/position_encoding_comparison.png filter=lfs diff=lfs merge=lfs -text
|
| 42 |
+
visualizations/embedding_tsne_multilingual.png filter=lfs diff=lfs merge=lfs -text
|
README.md
CHANGED
|
@@ -10,11 +10,21 @@ tags:
|
|
| 10 |
- n-gram
|
| 11 |
- markov
|
| 12 |
- wikipedia
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 13 |
- monolingual
|
| 14 |
- family-arabic
|
| 15 |
license: mit
|
| 16 |
library_name: wikilangs
|
| 17 |
-
pipeline_tag:
|
| 18 |
datasets:
|
| 19 |
- omarkamali/wikipedia-monthly
|
| 20 |
dataset_info:
|
|
@@ -23,10 +33,10 @@ dataset_info:
|
|
| 23 |
metrics:
|
| 24 |
- name: best_compression_ratio
|
| 25 |
type: compression
|
| 26 |
-
value: 4.
|
| 27 |
- name: best_isotropy
|
| 28 |
type: isotropy
|
| 29 |
-
value: 0.
|
| 30 |
- name: vocabulary_size
|
| 31 |
type: vocab
|
| 32 |
value: 0
|
|
@@ -60,7 +70,7 @@ We analyze tokenizers, n-gram models, Markov chains, vocabulary statistics, and
|
|
| 60 |
- [3. Markov Chain Evaluation](#3-markov-chain-evaluation)
|
| 61 |
- [4. Vocabulary Analysis](#4-vocabulary-analysis)
|
| 62 |
- [5. Word Embeddings Evaluation](#5-word-embeddings-evaluation)
|
| 63 |
-
- [6. Morphological Analysis (Experimental)](#6
|
| 64 |
- [7. Summary & Recommendations](#7-summary--recommendations)
|
| 65 |
- [Metrics Glossary](#appendix-metrics-glossary--interpretation-guide)
|
| 66 |
- [Visualizations Index](#visualizations-index)
|
|
@@ -80,47 +90,47 @@ We analyze tokenizers, n-gram models, Markov chains, vocabulary statistics, and
|
|
| 80 |
|
| 81 |
| Vocab Size | Compression | Avg Token Len | UNK Rate | Total Tokens |
|
| 82 |
|------------|-------------|---------------|----------|--------------|
|
| 83 |
-
| **8k** | 3.
|
| 84 |
-
| **16k** | 3.
|
| 85 |
-
| **32k** |
|
| 86 |
-
| **64k** | 4.
|
| 87 |
|
| 88 |
### Tokenization Examples
|
| 89 |
|
| 90 |
Below are sample sentences tokenized with each vocabulary size:
|
| 91 |
|
| 92 |
-
**Sample 1:**
|
| 93 |
|
| 94 |
| Vocab | Tokens | Count |
|
| 95 |
|-------|--------|-------|
|
| 96 |
-
| 8k |
|
| 97 |
-
| 16k |
|
| 98 |
-
| 32k |
|
| 99 |
-
| 64k |
|
| 100 |
|
| 101 |
-
**Sample 2:**
|
| 102 |
|
| 103 |
| Vocab | Tokens | Count |
|
| 104 |
|-------|--------|-------|
|
| 105 |
-
| 8k |
|
| 106 |
-
| 16k |
|
| 107 |
-
| 32k |
|
| 108 |
-
| 64k |
|
| 109 |
|
| 110 |
-
**Sample 3:** `هادي صفحة د التوضيح، كلمة
|
| 111 |
|
| 112 |
| Vocab | Tokens | Count |
|
| 113 |
|-------|--------|-------|
|
| 114 |
-
| 8k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة
|
| 115 |
-
| 16k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة
|
| 116 |
-
| 32k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة
|
| 117 |
-
| 64k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة
|
| 118 |
|
| 119 |
|
| 120 |
### Key Findings
|
| 121 |
|
| 122 |
-
- **Best Compression:** 64k achieves 4.
|
| 123 |
-
- **Lowest UNK Rate:** 8k with 0.
|
| 124 |
- **Trade-off:** Larger vocabularies improve compression but increase model size
|
| 125 |
- **Recommendation:** 32k vocabulary provides optimal balance for production use
|
| 126 |
|
|
@@ -137,12 +147,14 @@ Below are sample sentences tokenized with each vocabulary size:
|
|
| 137 |
|
| 138 |
| N-gram | Variant | Perplexity | Entropy | Unique N-grams | Top-100 Coverage | Top-1000 Coverage |
|
| 139 |
|--------|---------|------------|---------|----------------|------------------|-------------------|
|
| 140 |
-
| **2-gram** | Word |
|
| 141 |
-
| **2-gram** | Subword |
|
| 142 |
-
| **3-gram** | Word |
|
| 143 |
-
| **3-gram** | Subword | 3,
|
| 144 |
-
| **4-gram** | Word |
|
| 145 |
-
| **4-gram** | Subword |
|
|
|
|
|
|
|
| 146 |
|
| 147 |
### Top 5 N-grams by Size
|
| 148 |
|
|
@@ -152,8 +164,8 @@ Below are sample sentences tokenized with each vocabulary size:
|
|
| 152 |
|------|--------|-------|
|
| 153 |
| 1 | `واصلة ل` | 8,540 |
|
| 154 |
| 2 | `نسبة د` | 7,170 |
|
| 155 |
-
| 3 | `ف لمغريب` | 6,
|
| 156 |
-
| 4 | `ف إقليم` | 6,
|
| 157 |
| 5 | `ف نسبة` | 4,265 |
|
| 158 |
|
| 159 |
**3-grams (Word):**
|
|
@@ -164,7 +176,7 @@ Below are sample sentences tokenized with each vocabulary size:
|
|
| 164 |
| 2 | `فيها مصدر و` | 3,236 |
|
| 165 |
| 3 | `و نسبة د` | 2,894 |
|
| 166 |
| 4 | `مصدر و بايت` | 2,856 |
|
| 167 |
-
| 5 | `اللي خدامين ف` | 2,
|
| 168 |
|
| 169 |
**4-grams (Word):**
|
| 170 |
|
|
@@ -172,46 +184,66 @@ Below are sample sentences tokenized with each vocabulary size:
|
|
| 172 |
|------|--------|-------|
|
| 173 |
| 1 | `فيها مصدر و بايت` | 2,856 |
|
| 174 |
| 2 | `نسبة نّاس اللي خدامين` | 2,705 |
|
| 175 |
-
| 3 | `نّاس اللي خدامين ف` | 2,
|
| 176 |
| 4 | `على حساب لإحصاء الرسمي` | 2,501 |
|
| 177 |
-
| 5 |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 178 |
|
| 179 |
**2-grams (Subword):**
|
| 180 |
|
| 181 |
| Rank | N-gram | Count |
|
| 182 |
|------|--------|-------|
|
| 183 |
-
| 1 | `ا ل` |
|
| 184 |
-
| 2 | `_ ل` |
|
| 185 |
-
| 3 | `ة _` |
|
| 186 |
-
| 4 | `_ ا` |
|
| 187 |
-
| 5 | `_ م` |
|
| 188 |
|
| 189 |
**3-grams (Subword):**
|
| 190 |
|
| 191 |
| Rank | N-gram | Count |
|
| 192 |
|------|--------|-------|
|
| 193 |
-
| 1 | `_ ا ل` |
|
| 194 |
-
| 2 | `_ ف _` |
|
| 195 |
-
| 3 |
|
| 196 |
-
| 4 |
|
| 197 |
-
| 5 |
|
| 198 |
|
| 199 |
**4-grams (Subword):**
|
| 200 |
|
| 201 |
| Rank | N-gram | Count |
|
| 202 |
|------|--------|-------|
|
| 203 |
-
| 1 | `_ د ي ا` |
|
| 204 |
-
| 2 | `د ي ا ل` |
|
| 205 |
-
| 3 | `ي ا ل _` |
|
| 206 |
-
| 4 | `د _ ا ل` |
|
| 207 |
-
| 5 | `_ م ن _` |
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 208 |
|
| 209 |
|
| 210 |
### Key Findings
|
| 211 |
|
| 212 |
-
- **Best Perplexity:** 2-gram (subword) with
|
| 213 |
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
|
| 214 |
-
- **Coverage:** Top-1000 patterns cover ~
|
| 215 |
- **Recommendation:** 4-gram or 5-gram for best predictive performance
|
| 216 |
|
| 217 |
---
|
|
@@ -227,14 +259,14 @@ Below are sample sentences tokenized with each vocabulary size:
|
|
| 227 |
|
| 228 |
| Context | Variant | Avg Entropy | Perplexity | Branching Factor | Unique Contexts | Predictability |
|
| 229 |
|---------|---------|-------------|------------|------------------|-----------------|----------------|
|
| 230 |
-
| **1** | Word | 0.
|
| 231 |
-
| **1** | Subword | 1.
|
| 232 |
-
| **2** | Word | 0.
|
| 233 |
-
| **2** | Subword | 0.
|
| 234 |
-
| **3** | Word | 0.
|
| 235 |
-
| **3** | Subword | 0.
|
| 236 |
-
| **4** | Word | 0.
|
| 237 |
-
| **4** | Subword | 0.
|
| 238 |
|
| 239 |
### Generated Text Samples (Word-based)
|
| 240 |
|
|
@@ -242,27 +274,27 @@ Below are text samples generated from each word-based Markov chain model:
|
|
| 242 |
|
| 243 |
**Context Size 1:**
|
| 244 |
|
| 245 |
-
1. `ف
|
| 246 |
-
2.
|
| 247 |
-
3.
|
| 248 |
|
| 249 |
**Context Size 2:**
|
| 250 |
|
| 251 |
-
1. `واصلة ل
|
| 252 |
-
2. `نسبة د الناس النشيطين ف دوار
|
| 253 |
-
3. `ف لمغريب
|
| 254 |
|
| 255 |
**Context Size 3:**
|
| 256 |
|
| 257 |
-
1. `ف نسبة د الناس النشيطين ف دوار
|
| 258 |
-
2. `فيها مصدر و
|
| 259 |
-
3. `و نسبة د الشوماج واصلة ل
|
| 260 |
|
| 261 |
**Context Size 4:**
|
| 262 |
|
| 263 |
-
1. `نسبة نّاس اللي خدامين ف
|
| 264 |
-
2. `نّاس اللي خدامين ف
|
| 265 |
-
3. `على حساب لإحصاء الرسمي د عام
|
| 266 |
|
| 267 |
|
| 268 |
### Generated Text Samples (Subword-based)
|
|
@@ -271,34 +303,34 @@ Below are text samples generated from each subword-based Markov chain model:
|
|
| 271 |
|
| 272 |
**Context Size 1:**
|
| 273 |
|
| 274 |
-
1. `_
|
| 275 |
-
2.
|
| 276 |
-
3.
|
| 277 |
|
| 278 |
**Context Size 2:**
|
| 279 |
|
| 280 |
-
1.
|
| 281 |
-
2. `_
|
| 282 |
-
3. `ة_
|
| 283 |
|
| 284 |
**Context Size 3:**
|
| 285 |
|
| 286 |
-
1. `_
|
| 287 |
-
2. `_ف_
|
| 288 |
-
3.
|
| 289 |
|
| 290 |
**Context Size 4:**
|
| 291 |
|
| 292 |
-
1. `_ديال_
|
| 293 |
-
2.
|
| 294 |
-
3. `يال_
|
| 295 |
|
| 296 |
|
| 297 |
### Key Findings
|
| 298 |
|
| 299 |
- **Best Predictability:** Context-4 (word) with 97.9% predictability
|
| 300 |
- **Branching Factor:** Decreases with context size (more deterministic)
|
| 301 |
-
- **Memory Trade-off:** Larger contexts require more storage (
|
| 302 |
- **Recommendation:** Context-3 or Context-4 for text generation
|
| 303 |
|
| 304 |
---
|
|
@@ -314,64 +346,64 @@ Below are text samples generated from each subword-based Markov chain model:
|
|
| 314 |
|
| 315 |
| Metric | Value |
|
| 316 |
|--------|-------|
|
| 317 |
-
| Vocabulary Size |
|
| 318 |
-
| Total Tokens |
|
| 319 |
-
| Mean Frequency |
|
| 320 |
| Median Frequency | 4 |
|
| 321 |
-
| Frequency Std Dev |
|
| 322 |
|
| 323 |
### Most Common Words
|
| 324 |
|
| 325 |
| Rank | Word | Frequency |
|
| 326 |
|------|------|-----------|
|
| 327 |
-
| 1 | ف |
|
| 328 |
-
| 2 |
|
| 329 |
-
| 3 |
|
| 330 |
-
| 4 | ديال |
|
| 331 |
-
| 5 | من |
|
| 332 |
-
| 6 | ل | 23,
|
| 333 |
-
| 7 | على |
|
| 334 |
-
| 8 | لي |
|
| 335 |
-
| 9 |
|
| 336 |
-
| 10 | ب |
|
| 337 |
|
| 338 |
### Least Common Words (from vocabulary)
|
| 339 |
|
| 340 |
| Rank | Word | Frequency |
|
| 341 |
|------|------|-----------|
|
| 342 |
-
| 1 |
|
| 343 |
-
| 2 |
|
| 344 |
-
| 3 |
|
| 345 |
-
| 4 |
|
| 346 |
-
| 5 |
|
| 347 |
-
| 6 |
|
| 348 |
-
| 7 |
|
| 349 |
-
| 8 |
|
| 350 |
-
| 9 |
|
| 351 |
-
| 10 |
|
| 352 |
|
| 353 |
### Zipf's Law Analysis
|
| 354 |
|
| 355 |
| Metric | Value |
|
| 356 |
|--------|-------|
|
| 357 |
-
| Zipf Coefficient | 1.
|
| 358 |
-
| R² (Goodness of Fit) | 0.
|
| 359 |
| Adherence Quality | **excellent** |
|
| 360 |
|
| 361 |
### Coverage Analysis
|
| 362 |
|
| 363 |
| Top N Words | Coverage |
|
| 364 |
|-------------|----------|
|
| 365 |
-
| Top 100 |
|
| 366 |
-
| Top 1,000 |
|
| 367 |
-
| Top 5,000 |
|
| 368 |
-
| Top 10,000 |
|
| 369 |
|
| 370 |
### Key Findings
|
| 371 |
|
| 372 |
-
- **Zipf Compliance:** R²=0.
|
| 373 |
-
- **High Frequency Dominance:** Top 100 words cover
|
| 374 |
-
- **Long Tail:**
|
| 375 |
|
| 376 |
---
|
| 377 |
## 5. Word Embeddings Evaluation
|
|
@@ -387,37 +419,40 @@ Below are text samples generated from each subword-based Markov chain model:
|
|
| 387 |
|
| 388 |
### 5.1 Cross-Lingual Alignment
|
| 389 |
|
| 390 |
-
|
|
|
|
|
|
|
| 391 |
|
| 392 |
|
| 393 |
### 5.2 Model Comparison
|
| 394 |
|
| 395 |
| Model | Dimension | Isotropy | Semantic Density | Alignment R@1 | Alignment R@10 |
|
| 396 |
|-------|-----------|----------|------------------|---------------|----------------|
|
| 397 |
-
| **mono_32d** | 32 | 0.
|
| 398 |
-
| **mono_64d** | 64 | 0.
|
| 399 |
-
| **mono_128d** | 128 | 0.
|
|
|
|
|
|
|
|
|
|
| 400 |
|
| 401 |
### Key Findings
|
| 402 |
|
| 403 |
-
- **Best Isotropy:** mono_32d with 0.
|
| 404 |
-
- **Semantic Density:** Average pairwise similarity of 0.
|
| 405 |
-
- **Alignment Quality:**
|
| 406 |
- **Recommendation:** 128d aligned for best cross-lingual performance
|
| 407 |
|
| 408 |
---
|
| 409 |
## 6. Morphological Analysis (Experimental)
|
| 410 |
|
| 411 |
-
> ⚠️ **Warning:** This language shows low morphological productivity. The statistical signals used for this analysis may be noisy or less reliable than for morphologically rich languages.
|
| 412 |
-
|
| 413 |
This section presents an automated morphological analysis derived from the statistical divergence between word-level and subword-level models. By analyzing where subword predictability spikes and where word-level coverage fails, we can infer linguistic structures without supervised data.
|
| 414 |
|
| 415 |
### 6.1 Productivity & Complexity
|
| 416 |
|
| 417 |
| Metric | Value | Interpretation | Recommendation |
|
| 418 |
|--------|-------|----------------|----------------|
|
| 419 |
-
| Productivity Index | **
|
| 420 |
-
| Idiomaticity Gap |
|
| 421 |
|
| 422 |
### 6.2 Affix Inventory (Productive Units)
|
| 423 |
|
|
@@ -426,16 +461,17 @@ These are the most productive prefixes and suffixes identified by sampling the v
|
|
| 426 |
#### Productive Prefixes
|
| 427 |
| Prefix | Examples |
|
| 428 |
|--------|----------|
|
| 429 |
-
| `-ال` |
|
| 430 |
-
| `-لم` |
|
| 431 |
-
| `-كا` |
|
| 432 |
|
| 433 |
#### Productive Suffixes
|
| 434 |
| Suffix | Examples |
|
| 435 |
|--------|----------|
|
| 436 |
-
|
|
| 437 |
-
|
|
| 438 |
-
|
|
|
|
|
| 439 |
|
| 440 |
### 6.3 Bound Stems (Lexical Roots)
|
| 441 |
|
|
@@ -443,18 +479,18 @@ Bound stems are high-frequency subword units that are semantically cohesive but
|
|
| 443 |
|
| 444 |
| Stem | Cohesion | Substitutability | Examples |
|
| 445 |
|------|----------|------------------|----------|
|
| 446 |
-
|
|
| 447 |
-
|
|
| 448 |
-
|
|
| 449 |
-
|
|
| 450 |
-
|
|
| 451 |
-
| `النا` | 1.
|
| 452 |
-
|
|
| 453 |
-
| `مغري` | 2.
|
| 454 |
-
|
|
| 455 |
-
|
|
| 456 |
-
|
|
| 457 |
-
|
|
| 458 |
|
| 459 |
### 6.4 Affix Compatibility (Co-occurrence)
|
| 460 |
|
|
@@ -462,14 +498,16 @@ This table shows which prefixes and suffixes most frequently co-occur on the sam
|
|
| 462 |
|
| 463 |
| Prefix | Suffix | Frequency | Examples |
|
| 464 |
|--------|--------|-----------|----------|
|
| 465 |
-
| `-ال` |
|
| 466 |
-
| `-ال` | `-ات` |
|
| 467 |
-
| `-ال` |
|
| 468 |
-
|
|
| 469 |
-
| `-لم` |
|
| 470 |
-
| `-لم` |
|
| 471 |
-
|
|
| 472 |
-
|
|
|
|
|
|
|
|
| 473 |
|
| 474 |
### 6.5 Recursive Morpheme Segmentation
|
| 475 |
|
|
@@ -477,26 +515,28 @@ Using **Recursive Hierarchical Substitutability**, we decompose complex words in
|
|
| 477 |
|
| 478 |
| Word | Suggested Split | Confidence | Stem |
|
| 479 |
|------|-----------------|------------|------|
|
| 480 |
-
|
|
| 481 |
-
|
|
| 482 |
-
|
|
| 483 |
-
|
|
| 484 |
-
|
|
| 485 |
-
|
|
| 486 |
-
|
|
| 487 |
-
|
|
| 488 |
-
|
|
| 489 |
-
|
|
| 490 |
-
|
|
| 491 |
-
|
|
| 492 |
-
|
|
| 493 |
-
|
|
| 494 |
-
|
|
| 495 |
|
| 496 |
### 6.6 Linguistic Interpretation
|
| 497 |
|
| 498 |
> **Automated Insight:**
|
| 499 |
-
The language Moroccan Arabic
|
|
|
|
|
|
|
| 500 |
|
| 501 |
---
|
| 502 |
## 7. Summary & Recommendations
|
|
@@ -507,8 +547,8 @@ The language Moroccan Arabic appears to be more isolating or has a highly fixed
|
|
| 507 |
|
| 508 |
| Component | Recommended | Rationale |
|
| 509 |
|-----------|-------------|-----------|
|
| 510 |
-
| Tokenizer | **64k BPE** | Best compression (4.
|
| 511 |
-
| N-gram | **2-gram** | Lowest perplexity (
|
| 512 |
| Markov | **Context-4** | Highest predictability (97.9%) |
|
| 513 |
| Embeddings | **100d** | Balanced semantic capture and isotropy |
|
| 514 |
|
|
@@ -723,4 +763,4 @@ MIT License - Free for academic and commercial use.
|
|
| 723 |
---
|
| 724 |
*Generated by Wikilangs Models Pipeline*
|
| 725 |
|
| 726 |
-
*Report Date: 2026-01-03
|
|
|
|
| 10 |
- n-gram
|
| 11 |
- markov
|
| 12 |
- wikipedia
|
| 13 |
+
- feature-extraction
|
| 14 |
+
- sentence-similarity
|
| 15 |
+
- tokenization
|
| 16 |
+
- n-grams
|
| 17 |
+
- markov-chain
|
| 18 |
+
- text-mining
|
| 19 |
+
- fasttext
|
| 20 |
+
- babelvec
|
| 21 |
+
- vocabulous
|
| 22 |
+
- vocabulary
|
| 23 |
- monolingual
|
| 24 |
- family-arabic
|
| 25 |
license: mit
|
| 26 |
library_name: wikilangs
|
| 27 |
+
pipeline_tag: text-generation
|
| 28 |
datasets:
|
| 29 |
- omarkamali/wikipedia-monthly
|
| 30 |
dataset_info:
|
|
|
|
| 33 |
metrics:
|
| 34 |
- name: best_compression_ratio
|
| 35 |
type: compression
|
| 36 |
+
value: 4.171
|
| 37 |
- name: best_isotropy
|
| 38 |
type: isotropy
|
| 39 |
+
value: 0.8303
|
| 40 |
- name: vocabulary_size
|
| 41 |
type: vocab
|
| 42 |
value: 0
|
|
|
|
| 70 |
- [3. Markov Chain Evaluation](#3-markov-chain-evaluation)
|
| 71 |
- [4. Vocabulary Analysis](#4-vocabulary-analysis)
|
| 72 |
- [5. Word Embeddings Evaluation](#5-word-embeddings-evaluation)
|
| 73 |
+
- [6. Morphological Analysis (Experimental)](#6--morphological-analysis-experimental)
|
| 74 |
- [7. Summary & Recommendations](#7-summary--recommendations)
|
| 75 |
- [Metrics Glossary](#appendix-metrics-glossary--interpretation-guide)
|
| 76 |
- [Visualizations Index](#visualizations-index)
|
|
|
|
| 90 |
|
| 91 |
| Vocab Size | Compression | Avg Token Len | UNK Rate | Total Tokens |
|
| 92 |
|------------|-------------|---------------|----------|--------------|
|
| 93 |
+
| **8k** | 3.480x | 3.48 | 0.0910% | 300,099 |
|
| 94 |
+
| **16k** | 3.753x | 3.76 | 0.0981% | 278,271 |
|
| 95 |
+
| **32k** | 3.983x | 3.99 | 0.1041% | 262,209 |
|
| 96 |
+
| **64k** | 4.171x 🏆 | 4.18 | 0.1090% | 250,397 |
|
| 97 |
|
| 98 |
### Tokenization Examples
|
| 99 |
|
| 100 |
Below are sample sentences tokenized with each vocabulary size:
|
| 101 |
|
| 102 |
+
**Sample 1:** `لجدوال ديال الترتيب شوف حتى بوطولا 1 بوطولا 2 لهيكلة لهرمية د لبوطولات ديال كورة...`
|
| 103 |
|
| 104 |
| Vocab | Tokens | Count |
|
| 105 |
|-------|--------|-------|
|
| 106 |
+
| 8k | `▁لجدوال ▁ديال ▁الترتيب ▁شوف ▁حتى ▁بوطولا ▁ 1 ▁بوطولا ▁ ... (+17 more)` | 27 |
|
| 107 |
+
| 16k | `▁لجدوال ▁ديال ▁الترتيب ▁شوف ▁حتى ▁بوطولا ▁ 1 ▁بوطولا ▁ ... (+17 more)` | 27 |
|
| 108 |
+
| 32k | `▁لجدوال ▁ديال ▁الترتيب ▁شوف ▁حتى ▁بوطولا ▁ 1 ▁بوطولا ▁ ... (+17 more)` | 27 |
|
| 109 |
+
| 64k | `▁لجدوال ▁ديال ▁الترتيب ▁شوف ▁حتى ▁بوطولا ▁ 1 ▁بوطولا ▁ ... (+17 more)` | 27 |
|
| 110 |
|
| 111 |
+
**Sample 2:** `هادي صفحة د التوضيح، كلمة أنفا يمكن يكونو عندها هاد لمعاني: مقاطعة أنفا: حي كاين...`
|
| 112 |
|
| 113 |
| Vocab | Tokens | Count |
|
| 114 |
|-------|--------|-------|
|
| 115 |
+
| 8k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة ▁أن فا ▁يمكن ▁يكونو ... (+27 more)` | 37 |
|
| 116 |
+
| 16k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة ▁أنفا ▁يمكن ▁يكونو ▁عندها ... (+23 more)` | 33 |
|
| 117 |
+
| 32k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة ▁أنفا ▁يمكن ▁يكونو ▁عندها ... (+23 more)` | 33 |
|
| 118 |
+
| 64k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة ▁أنفا ▁يمكن ▁يكونو ▁عندها ... (+23 more)` | 33 |
|
| 119 |
|
| 120 |
+
**Sample 3:** `هادي صفحة د التوضيح، كلمة منى يمكن يكونو عندها هاد لمعاني: منى صابر منى أمرشا من...`
|
| 121 |
|
| 122 |
| Vocab | Tokens | Count |
|
| 123 |
|-------|--------|-------|
|
| 124 |
+
| 8k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة ▁من ى ▁يمكن ▁يكونو ... (+17 more)` | 27 |
|
| 125 |
+
| 16k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة ▁منى ▁يمكن ▁يكونو ▁عندها ... (+13 more)` | 23 |
|
| 126 |
+
| 32k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة ▁منى ▁يمكن ▁يكونو ▁عندها ... (+12 more)` | 22 |
|
| 127 |
+
| 64k | `▁هادي ▁صفحة ▁د ▁التوضيح ، ▁كلمة ▁منى ▁يمكن ▁يكونو ▁عندها ... (+10 more)` | 20 |
|
| 128 |
|
| 129 |
|
| 130 |
### Key Findings
|
| 131 |
|
| 132 |
+
- **Best Compression:** 64k achieves 4.171x compression
|
| 133 |
+
- **Lowest UNK Rate:** 8k with 0.0910% unknown tokens
|
| 134 |
- **Trade-off:** Larger vocabularies improve compression but increase model size
|
| 135 |
- **Recommendation:** 32k vocabulary provides optimal balance for production use
|
| 136 |
|
|
|
|
| 147 |
|
| 148 |
| N-gram | Variant | Perplexity | Entropy | Unique N-grams | Top-100 Coverage | Top-1000 Coverage |
|
| 149 |
|--------|---------|------------|---------|----------------|------------------|-------------------|
|
| 150 |
+
| **2-gram** | Word | 7,228 | 12.82 | 39,512 | 23.0% | 50.8% |
|
| 151 |
+
| **2-gram** | Subword | 424 🏆 | 8.73 | 5,903 | 58.0% | 96.4% |
|
| 152 |
+
| **3-gram** | Word | 5,655 | 12.47 | 43,555 | 27.5% | 57.1% |
|
| 153 |
+
| **3-gram** | Subword | 3,784 | 11.89 | 44,651 | 23.1% | 60.7% |
|
| 154 |
+
| **4-gram** | Word | 7,985 | 12.96 | 70,559 | 27.5% | 53.6% |
|
| 155 |
+
| **4-gram** | Subword | 20,064 | 14.29 | 220,807 | 12.0% | 36.0% |
|
| 156 |
+
| **5-gram** | Word | 7,565 | 12.89 | 58,964 | 28.5% | 52.9% |
|
| 157 |
+
| **5-gram** | Subword | 62,379 | 15.93 | 527,725 | 7.3% | 25.0% |
|
| 158 |
|
| 159 |
### Top 5 N-grams by Size
|
| 160 |
|
|
|
|
| 164 |
|------|--------|-------|
|
| 165 |
| 1 | `واصلة ل` | 8,540 |
|
| 166 |
| 2 | `نسبة د` | 7,170 |
|
| 167 |
+
| 3 | `ف لمغريب` | 6,305 |
|
| 168 |
+
| 4 | `ف إقليم` | 6,018 |
|
| 169 |
| 5 | `ف نسبة` | 4,265 |
|
| 170 |
|
| 171 |
**3-grams (Word):**
|
|
|
|
| 176 |
| 2 | `فيها مصدر و` | 3,236 |
|
| 177 |
| 3 | `و نسبة د` | 2,894 |
|
| 178 |
| 4 | `مصدر و بايت` | 2,856 |
|
| 179 |
+
| 5 | `اللي خدامين ف` | 2,760 |
|
| 180 |
|
| 181 |
**4-grams (Word):**
|
| 182 |
|
|
|
|
| 184 |
|------|--------|-------|
|
| 185 |
| 1 | `فيها مصدر و بايت` | 2,856 |
|
| 186 |
| 2 | `نسبة نّاس اللي خدامين` | 2,705 |
|
| 187 |
+
| 3 | `نّاس اللي خدامين ف` | 2,594 |
|
| 188 |
| 4 | `على حساب لإحصاء الرسمي` | 2,501 |
|
| 189 |
+
| 5 | `لإحصاء الرسمي د عام` | 2,500 |
|
| 190 |
+
|
| 191 |
+
**5-grams (Word):**
|
| 192 |
+
|
| 193 |
+
| Rank | N-gram | Count |
|
| 194 |
+
|------|--------|-------|
|
| 195 |
+
| 1 | `نسبة نّاس اللي خدامين ف` | 2,593 |
|
| 196 |
+
| 2 | `هاد دّوار كينتامي ل مشيخة` | 2,500 |
|
| 197 |
+
| 3 | `حساب لإحصاء الرسمي د عام` | 2,500 |
|
| 198 |
+
| 4 | `لمغريب هاد دّوار كينتامي ل` | 2,500 |
|
| 199 |
+
| 5 | `ف لمغريب هاد دّوار كينتامي` | 2,500 |
|
| 200 |
|
| 201 |
**2-grams (Subword):**
|
| 202 |
|
| 203 |
| Rank | N-gram | Count |
|
| 204 |
|------|--------|-------|
|
| 205 |
+
| 1 | `ا ل` | 347,466 |
|
| 206 |
+
| 2 | `_ ل` | 278,371 |
|
| 207 |
+
| 3 | `ة _` | 229,442 |
|
| 208 |
+
| 4 | `_ ا` | 220,960 |
|
| 209 |
+
| 5 | `_ م` | 156,801 |
|
| 210 |
|
| 211 |
**3-grams (Subword):**
|
| 212 |
|
| 213 |
| Rank | N-gram | Count |
|
| 214 |
|------|--------|-------|
|
| 215 |
+
| 1 | `_ ا ل` | 216,048 |
|
| 216 |
+
| 2 | `_ ف _` | 83,146 |
|
| 217 |
+
| 3 | `ا ت _` | 63,800 |
|
| 218 |
+
| 4 | `ي ة _` | 60,271 |
|
| 219 |
+
| 5 | `_ د _` | 59,563 |
|
| 220 |
|
| 221 |
**4-grams (Subword):**
|
| 222 |
|
| 223 |
| Rank | N-gram | Count |
|
| 224 |
|------|--------|-------|
|
| 225 |
+
| 1 | `_ د ي ا` | 47,798 |
|
| 226 |
+
| 2 | `د ي ا ل` | 47,559 |
|
| 227 |
+
| 3 | `ي ا ل _` | 33,039 |
|
| 228 |
+
| 4 | `د _ ا ل` | 32,831 |
|
| 229 |
+
| 5 | `_ م ن _` | 28,909 |
|
| 230 |
+
|
| 231 |
+
**5-grams (Subword):**
|
| 232 |
+
|
| 233 |
+
| Rank | N-gram | Count |
|
| 234 |
+
|------|--------|-------|
|
| 235 |
+
| 1 | `_ د ي ا ل` | 47,427 |
|
| 236 |
+
| 2 | `د ي ا ل _` | 32,608 |
|
| 237 |
+
| 3 | `_ ع ل ى _` | 19,473 |
|
| 238 |
+
| 4 | `_ ا ل ل ي` | 18,967 |
|
| 239 |
+
| 5 | `ا ل ل ي _` | 18,744 |
|
| 240 |
|
| 241 |
|
| 242 |
### Key Findings
|
| 243 |
|
| 244 |
+
- **Best Perplexity:** 2-gram (subword) with 424
|
| 245 |
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
|
| 246 |
+
- **Coverage:** Top-1000 patterns cover ~25% of corpus
|
| 247 |
- **Recommendation:** 4-gram or 5-gram for best predictive performance
|
| 248 |
|
| 249 |
---
|
|
|
|
| 259 |
|
| 260 |
| Context | Variant | Avg Entropy | Perplexity | Branching Factor | Unique Contexts | Predictability |
|
| 261 |
|---------|---------|-------------|------------|------------------|-----------------|----------------|
|
| 262 |
+
| **1** | Word | 0.8561 | 1.810 | 5.38 | 178,865 | 14.4% |
|
| 263 |
+
| **1** | Subword | 1.1236 | 2.179 | 8.36 | 2,156 | 0.0% |
|
| 264 |
+
| **2** | Word | 0.2259 | 1.169 | 1.49 | 962,233 | 77.4% |
|
| 265 |
+
| **2** | Subword | 0.8160 | 1.761 | 5.10 | 18,029 | 18.4% |
|
| 266 |
+
| **3** | Word | 0.0618 | 1.044 | 1.10 | 1,431,084 | 93.8% |
|
| 267 |
+
| **3** | Subword | 0.8022 | 1.744 | 4.13 | 91,858 | 19.8% |
|
| 268 |
+
| **4** | Word | 0.0208 🏆 | 1.015 | 1.04 | 1,574,083 | 97.9% |
|
| 269 |
+
| **4** | Subword | 0.6604 | 1.581 | 2.86 | 379,445 | 34.0% |
|
| 270 |
|
| 271 |
### Generated Text Samples (Word-based)
|
| 272 |
|
|
|
|
| 274 |
|
| 275 |
**Context Size 1:**
|
| 276 |
|
| 277 |
+
1. `ف دور السفير اللول تبعوه كتر من أبسط تعريف من chinese medicinal herbs plants biological reviews`
|
| 278 |
+
2. `و واخا تايقولو بلي مفكرين وصحافيين من الريف الشرقي د الناس والقرع بفلوسو لخاصة د تقويم`
|
| 279 |
+
3. `د لمنتجات د الناس اللي كتب بزاف ديال عوام كيوافق 676 233 1 نسبة د لأمية`
|
| 280 |
|
| 281 |
**Context Size 2:**
|
| 282 |
|
| 283 |
+
1. `واصلة ل 40 1 و نسبة د لأمية واصلة ل 43 43 25 39 عام 25 83`
|
| 284 |
+
2. `نسبة د الناس النشيطين ف دوار اكرنو معاد تزاد ب 25 6 و نسبة د الناس النشيطين`
|
| 285 |
+
3. `ف لمغريب هاد دّوار كينتامي ل مشيخة أيت قضني لي كتضم 7 د دّواور لعاداد د سّكان`
|
| 286 |
|
| 287 |
**Context Size 3:**
|
| 288 |
|
| 289 |
+
1. `ف نسبة د الناس النشيطين ف دوار أيت بلقاس واصلة ل 39 06 و نسبة د الشوماج واصلة`
|
| 290 |
+
2. `فيها مصدر و بايت زادهوم داريجابوت مسكونين ف إقليم سيدي قاسم جهة رّباط سلا قنيطرة ساكنين فيها واحد`
|
| 291 |
+
3. `و نسبة د الشوماج واصلة ل 10 45 نوطات مصادر ف لمغريب ف إقليم تارودانت زادهوم داريجابوت`
|
| 292 |
|
| 293 |
**Context Size 4:**
|
| 294 |
|
| 295 |
+
1. `نسبة نّاس اللي خدامين ف مصادر درعة تافيلالت قروية ف إقليم ميدلت مسكونين ف إقليم ميدلت قروية ف إقليم`
|
| 296 |
+
2. `نّاس اللي خدامين ف لپريڤي 64 5 مصادر درعة تافيلالت قروية ف إقليم تينغير مسكونين ف إقليم تينغير قروية`
|
| 297 |
+
3. `على حساب لإحصاء الرسمي د عام نوطات مصادر ف لمغريب ف إقليم تارودانت زادهوم داريجابوت`
|
| 298 |
|
| 299 |
|
| 300 |
### Generated Text Samples (Subword-based)
|
|
|
|
| 303 |
|
| 304 |
**Context Size 1:**
|
| 305 |
|
| 306 |
+
1. `_-_دو،_ب_خبّقصوان`
|
| 307 |
+
2. `انزالتسوبومشية_ف`
|
| 308 |
+
3. `لإف_كمة_داللوغر_`
|
| 309 |
|
| 310 |
**Context Size 2:**
|
| 311 |
|
| 312 |
+
1. `الصحيزية_نّاسة_:_4`
|
| 313 |
+
2. `_لبيات_ف_حيواح_بز`
|
| 314 |
+
3. `ة_عاواع_كتسمة_تحيّ`
|
| 315 |
|
| 316 |
**Context Size 3:**
|
| 317 |
|
| 318 |
+
1. `_الروما_ؤروپ_د_إيز`
|
| 319 |
+
2. `_ف_لعالمغريب_دوا_ك`
|
| 320 |
+
3. `ات_عام._اللي_ل_لما`
|
| 321 |
|
| 322 |
**Context Size 4:**
|
| 323 |
|
| 324 |
+
1. `_ديال_أفلام_مكبّس)._`
|
| 325 |
+
2. `ديال_الزايير_ديال_ت`
|
| 326 |
+
3. `يال_شي_قضيب_على_الط`
|
| 327 |
|
| 328 |
|
| 329 |
### Key Findings
|
| 330 |
|
| 331 |
- **Best Predictability:** Context-4 (word) with 97.9% predictability
|
| 332 |
- **Branching Factor:** Decreases with context size (more deterministic)
|
| 333 |
+
- **Memory Trade-off:** Larger contexts require more storage (379,445 contexts)
|
| 334 |
- **Recommendation:** Context-3 or Context-4 for text generation
|
| 335 |
|
| 336 |
---
|
|
|
|
| 346 |
|
| 347 |
| Metric | Value |
|
| 348 |
|--------|-------|
|
| 349 |
+
| Vocabulary Size | 78,779 |
|
| 350 |
+
| Total Tokens | 2,032,841 |
|
| 351 |
+
| Mean Frequency | 25.80 |
|
| 352 |
| Median Frequency | 4 |
|
| 353 |
+
| Frequency Std Dev | 515.92 |
|
| 354 |
|
| 355 |
### Most Common Words
|
| 356 |
|
| 357 |
| Rank | Word | Frequency |
|
| 358 |
|------|------|-----------|
|
| 359 |
+
| 1 | ف | 83,458 |
|
| 360 |
+
| 2 | و | 59,829 |
|
| 361 |
+
| 3 | د | 59,731 |
|
| 362 |
+
| 4 | ديال | 32,565 |
|
| 363 |
+
| 5 | من | 29,236 |
|
| 364 |
+
| 6 | ل | 23,572 |
|
| 365 |
+
| 7 | على | 19,570 |
|
| 366 |
+
| 8 | لي | 18,402 |
|
| 367 |
+
| 9 | اللي | 17,442 |
|
| 368 |
+
| 10 | ب | 17,233 |
|
| 369 |
|
| 370 |
### Least Common Words (from vocabulary)
|
| 371 |
|
| 372 |
| Rank | Word | Frequency |
|
| 373 |
|------|------|-----------|
|
| 374 |
+
| 1 | بوفوار | 2 |
|
| 375 |
+
| 2 | بيتسي | 2 |
|
| 376 |
+
| 3 | وصانعي | 2 |
|
| 377 |
+
| 4 | وأهميتها | 2 |
|
| 378 |
+
| 5 | بورديو | 2 |
|
| 379 |
+
| 6 | بلومر | 2 |
|
| 380 |
+
| 7 | مقترحة | 2 |
|
| 381 |
+
| 8 | anchor | 2 |
|
| 382 |
+
| 9 | بعصبة | 2 |
|
| 383 |
+
| 10 | ماڭي | 2 |
|
| 384 |
|
| 385 |
### Zipf's Law Analysis
|
| 386 |
|
| 387 |
| Metric | Value |
|
| 388 |
|--------|-------|
|
| 389 |
+
| Zipf Coefficient | 1.0213 |
|
| 390 |
+
| R² (Goodness of Fit) | 0.998918 |
|
| 391 |
| Adherence Quality | **excellent** |
|
| 392 |
|
| 393 |
### Coverage Analysis
|
| 394 |
|
| 395 |
| Top N Words | Coverage |
|
| 396 |
|-------------|----------|
|
| 397 |
+
| Top 100 | 38.6% |
|
| 398 |
+
| Top 1,000 | 62.9% |
|
| 399 |
+
| Top 5,000 | 77.8% |
|
| 400 |
+
| Top 10,000 | 84.2% |
|
| 401 |
|
| 402 |
### Key Findings
|
| 403 |
|
| 404 |
+
- **Zipf Compliance:** R²=0.9989 indicates excellent adherence to Zipf's law
|
| 405 |
+
- **High Frequency Dominance:** Top 100 words cover 38.6% of corpus
|
| 406 |
+
- **Long Tail:** 68,779 words needed for remaining 15.8% coverage
|
| 407 |
|
| 408 |
---
|
| 409 |
## 5. Word Embeddings Evaluation
|
|
|
|
| 419 |
|
| 420 |
### 5.1 Cross-Lingual Alignment
|
| 421 |
|
| 422 |
+

|
| 423 |
+
|
| 424 |
+

|
| 425 |
|
| 426 |
|
| 427 |
### 5.2 Model Comparison
|
| 428 |
|
| 429 |
| Model | Dimension | Isotropy | Semantic Density | Alignment R@1 | Alignment R@10 |
|
| 430 |
|-------|-----------|----------|------------------|---------------|----------------|
|
| 431 |
+
| **mono_32d** | 32 | 0.8303 🏆 | 0.3306 | N/A | N/A |
|
| 432 |
+
| **mono_64d** | 64 | 0.8186 | 0.2546 | N/A | N/A |
|
| 433 |
+
| **mono_128d** | 128 | 0.6893 | 0.2062 | N/A | N/A |
|
| 434 |
+
| **aligned_32d** | 32 | 0.8303 | 0.3293 | 0.0120 | 0.1380 |
|
| 435 |
+
| **aligned_64d** | 64 | 0.8186 | 0.2507 | 0.0360 | 0.1920 |
|
| 436 |
+
| **aligned_128d** | 128 | 0.6893 | 0.2101 | 0.0580 | 0.2760 |
|
| 437 |
|
| 438 |
### Key Findings
|
| 439 |
|
| 440 |
+
- **Best Isotropy:** mono_32d with 0.8303 (more uniform distribution)
|
| 441 |
+
- **Semantic Density:** Average pairwise similarity of 0.2636. Lower values indicate better semantic separation.
|
| 442 |
+
- **Alignment Quality:** Aligned models achieve up to 5.8% R@1 in cross-lingual retrieval.
|
| 443 |
- **Recommendation:** 128d aligned for best cross-lingual performance
|
| 444 |
|
| 445 |
---
|
| 446 |
## 6. Morphological Analysis (Experimental)
|
| 447 |
|
|
|
|
|
|
|
| 448 |
This section presents an automated morphological analysis derived from the statistical divergence between word-level and subword-level models. By analyzing where subword predictability spikes and where word-level coverage fails, we can infer linguistic structures without supervised data.
|
| 449 |
|
| 450 |
### 6.1 Productivity & Complexity
|
| 451 |
|
| 452 |
| Metric | Value | Interpretation | Recommendation |
|
| 453 |
|--------|-------|----------------|----------------|
|
| 454 |
+
| Productivity Index | **5.000** | High morphological productivity | Reliable analysis |
|
| 455 |
+
| Idiomaticity Gap | **1.114** | High formulaic/idiomatic content | - |
|
| 456 |
|
| 457 |
### 6.2 Affix Inventory (Productive Units)
|
| 458 |
|
|
|
|
| 461 |
#### Productive Prefixes
|
| 462 |
| Prefix | Examples |
|
| 463 |
|--------|----------|
|
| 464 |
+
| `-ال` | العزابة, التيستات, البخارية |
|
| 465 |
+
| `-لم` | لمهرجان, لمدارس, لموناخ |
|
| 466 |
+
| `-كا` | كاليدونيا, كايتعلّقو, كاتنفخ |
|
| 467 |
|
| 468 |
#### Productive Suffixes
|
| 469 |
| Suffix | Examples |
|
| 470 |
|--------|----------|
|
| 471 |
+
| `-ة` | العزابة, البخارية, صيفية |
|
| 472 |
+
| `-ات` | بلافوايديات, التيستات, طرات |
|
| 473 |
+
| `-ية` | البخارية, صيفية, الشقرونية |
|
| 474 |
+
| `-ين` | احساين, للعين, الأوكسجين |
|
| 475 |
|
| 476 |
### 6.3 Bound Stems (Lexical Roots)
|
| 477 |
|
|
|
|
| 479 |
|
| 480 |
| Stem | Cohesion | Substitutability | Examples |
|
| 481 |
|------|----------|------------------|----------|
|
| 482 |
+
| `اللو` | 1.86x | 61 contexts | اللوز, اللور, اللول |
|
| 483 |
+
| `انية` | 1.80x | 68 contexts | كانية, سانية, دانية |
|
| 484 |
+
| `الات` | 1.71x | 65 contexts | سالات, صالات, حالات |
|
| 485 |
+
| `جماع` | 1.94x | 38 contexts | جماعي, إجماع, تجماع |
|
| 486 |
+
| `لمغر` | 1.94x | 30 contexts | لمغرب, لمغربي, فلمغرب |
|
| 487 |
+
| `النا` | 1.58x | 63 contexts | الناي, الناس, النار |
|
| 488 |
+
| `حصاء` | 2.26x | 14 contexts | إحصاء, ليحصاء, لإحصاء |
|
| 489 |
+
| `مغري` | 2.07x | 18 contexts | مغرية, مغريب, لمغريب |
|
| 490 |
+
| `دهوم` | 2.15x | 16 contexts | ضدهوم, بعدهوم, زادهوم |
|
| 491 |
+
| `إحصا` | 2.07x | 17 contexts | إحصاء, لإحصا, إحصائي |
|
| 492 |
+
| `لجوا` | 1.81x | 26 contexts | الجوا, لجواد, لجواب |
|
| 493 |
+
| `قليم` | 2.06x | 17 contexts | إقليم, اقليم, فقليم |
|
| 494 |
|
| 495 |
### 6.4 Affix Compatibility (Co-occurrence)
|
| 496 |
|
|
|
|
| 498 |
|
| 499 |
| Prefix | Suffix | Frequency | Examples |
|
| 500 |
|--------|--------|-----------|----------|
|
| 501 |
+
| `-ال` | `-ة` | 275 words | المقبرة, السيارة |
|
| 502 |
+
| `-ال` | `-ات` | 133 words | الفقريات, الزمانات |
|
| 503 |
+
| `-ال` | `-ية` | 132 words | الأوروپية, الجنية |
|
| 504 |
+
| `-ال` | `-ين` | 73 words | السلاڤيين, النيوزيلانضيين |
|
| 505 |
+
| `-لم` | `-ة` | 57 words | لممكنة, لمناسبة |
|
| 506 |
+
| `-لم` | `-ات` | 30 words | لماوات, لمغريبيات |
|
| 507 |
+
| `-لم` | `-ين` | 29 words | لمحمّلين, لمغنّيين |
|
| 508 |
+
| `-لم` | `-ية` | 22 words | لمورفولوجية, لمنصورية |
|
| 509 |
+
| `-كا` | `-ات` | 1 words | كاربونات, كائنات |
|
| 510 |
+
| `-كا` | `-ين` | 1 words | كاترين, كالكيريين |
|
| 511 |
|
| 512 |
### 6.5 Recursive Morpheme Segmentation
|
| 513 |
|
|
|
|
| 515 |
|
| 516 |
| Word | Suggested Split | Confidence | Stem |
|
| 517 |
|------|-----------------|------------|------|
|
| 518 |
+
| المباشرين | **`ال-مباشر-ين`** | 6.0 | `مباشر` |
|
| 519 |
+
| السلاڤيين | **`ال-سلاڤي-ين`** | 6.0 | `سلاڤي` |
|
| 520 |
+
| لمرتابطين | **`لم-رتابط-ين`** | 6.0 | `رتابط` |
|
| 521 |
+
| المساهمات | **`ال-مساهم-ات`** | 6.0 | `مساهم` |
|
| 522 |
+
| الطاكسونات | **`ال-طاكسون-ات`** | 6.0 | `طاكسون` |
|
| 523 |
+
| المتفوقين | **`ال-متفوق-ين`** | 6.0 | `متفوق` |
|
| 524 |
+
| الإتفاقية | **`ال-إتفاق-ية`** | 6.0 | `إتفاق` |
|
| 525 |
+
| النتيهازيين | **`ال-نتيهازي-ين`** | 6.0 | `نتيهازي` |
|
| 526 |
+
| النهاريين | **`ال-نهاري-ين`** | 6.0 | `نهاري` |
|
| 527 |
+
| الجتيماعية | **`ال-جتيماع-ية`** | 6.0 | `جتيماع` |
|
| 528 |
+
| المستقبلية | **`ال-مستقبل-ية`** | 6.0 | `مستقبل` |
|
| 529 |
+
| السبيطارات | **`ال-سبيطار-ات`** | 6.0 | `سبيطار` |
|
| 530 |
+
| اللولانيات | **`ال-لولاني-ات`** | 6.0 | `لولاني` |
|
| 531 |
+
| السيناريوات | **`ال-سيناريو-ات`** | 6.0 | `سيناريو` |
|
| 532 |
+
| المستخدمين | **`ال-مستخدم-ين`** | 6.0 | `مستخدم` |
|
| 533 |
|
| 534 |
### 6.6 Linguistic Interpretation
|
| 535 |
|
| 536 |
> **Automated Insight:**
|
| 537 |
+
The language Moroccan Arabic shows high morphological productivity. The subword models are significantly more efficient than word models, suggesting a rich system of affixation or compounding.
|
| 538 |
+
|
| 539 |
+
> **Note on Idiomaticity:** The high Idiomaticity Gap suggests a large number of frequent multi-word expressions or formulaic sequences that are statistically distinct from their component parts.
|
| 540 |
|
| 541 |
---
|
| 542 |
## 7. Summary & Recommendations
|
|
|
|
| 547 |
|
| 548 |
| Component | Recommended | Rationale |
|
| 549 |
|-----------|-------------|-----------|
|
| 550 |
+
| Tokenizer | **64k BPE** | Best compression (4.17x) |
|
| 551 |
+
| N-gram | **2-gram** | Lowest perplexity (424) |
|
| 552 |
| Markov | **Context-4** | Highest predictability (97.9%) |
|
| 553 |
| Embeddings | **100d** | Balanced semantic capture and isotropy |
|
| 554 |
|
|
|
|
| 763 |
---
|
| 764 |
*Generated by Wikilangs Models Pipeline*
|
| 765 |
|
| 766 |
+
*Report Date: 2026-01-03 14:22:25*
|
models/embeddings/aligned/ary_128d.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bbfeb46b5666c8d64122bd217a3ab907c0c21862e928add63b11988411f58bd1
|
| 3 |
+
size 1060912662
|
models/embeddings/aligned/ary_128d.meta.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"lang": "ary", "dim": 128, "max_seq_len": 512, "is_aligned": true}
|
models/embeddings/aligned/ary_128d.projection.npy
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:02fef87bfc59c1f707a33d4e4f466b2425514a6aae0c6378fe22046746622e86
|
| 3 |
+
size 65664
|
models/embeddings/aligned/ary_128d_metadata.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"language": "ary",
|
| 3 |
+
"dimension": 128,
|
| 4 |
+
"version": "aligned",
|
| 5 |
+
"hub_language": "en",
|
| 6 |
+
"seed_vocab_size": 3796,
|
| 7 |
+
"vocab_size": 35328
|
| 8 |
+
}
|
models/embeddings/aligned/ary_32d.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9a06bb9ccf33412e88b0c6b296b5b85b834eee4b8d499102c47bd83e67febc5c
|
| 3 |
+
size 265780758
|
models/embeddings/aligned/ary_32d.meta.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"lang": "ary", "dim": 32, "max_seq_len": 512, "is_aligned": true}
|
models/embeddings/aligned/ary_32d.projection.npy
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:eb886167c81ee7c8dc1d46d989d761b84ea11a83290c26d66b2e1dc1aac5fbac
|
| 3 |
+
size 4224
|
models/embeddings/aligned/ary_32d_metadata.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"language": "ary",
|
| 3 |
+
"dimension": 32,
|
| 4 |
+
"version": "aligned",
|
| 5 |
+
"hub_language": "en",
|
| 6 |
+
"seed_vocab_size": 3796,
|
| 7 |
+
"vocab_size": 35328
|
| 8 |
+
}
|
models/embeddings/aligned/ary_64d.bin
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d8ea049910ca1cbe494ecfa4887bb1fe2c54a8fedd0e092181cf1830e949581c
|
| 3 |
+
size 530824726
|
models/embeddings/aligned/ary_64d.meta.json
ADDED
|
@@ -0,0 +1 @@
|
|
|
|
|
|
|
| 1 |
+
{"lang": "ary", "dim": 64, "max_seq_len": 512, "is_aligned": true}
|
models/embeddings/aligned/ary_64d.projection.npy
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8ddcb0a862b77d47360178cb10c0adf1ebea1b53686716f62119200c0028beff
|
| 3 |
+
size 16512
|
models/embeddings/aligned/ary_64d_metadata.json
ADDED
|
@@ -0,0 +1,8 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"language": "ary",
|
| 3 |
+
"dimension": 64,
|
| 4 |
+
"version": "aligned",
|
| 5 |
+
"hub_language": "en",
|
| 6 |
+
"seed_vocab_size": 3796,
|
| 7 |
+
"vocab_size": 35328
|
| 8 |
+
}
|
models/embeddings/monolingual/ary_128d.bin
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bbfeb46b5666c8d64122bd217a3ab907c0c21862e928add63b11988411f58bd1
|
| 3 |
+
size 1060912662
|
models/embeddings/monolingual/ary_128d_metadata.json
CHANGED
|
@@ -11,5 +11,5 @@
|
|
| 11 |
"encoding_method": "rope",
|
| 12 |
"dim": 128
|
| 13 |
},
|
| 14 |
-
"vocab_size":
|
| 15 |
}
|
|
|
|
| 11 |
"encoding_method": "rope",
|
| 12 |
"dim": 128
|
| 13 |
},
|
| 14 |
+
"vocab_size": 35328
|
| 15 |
}
|
models/embeddings/monolingual/ary_32d.bin
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:9a06bb9ccf33412e88b0c6b296b5b85b834eee4b8d499102c47bd83e67febc5c
|
| 3 |
+
size 265780758
|
models/embeddings/monolingual/ary_32d_metadata.json
CHANGED
|
@@ -11,5 +11,5 @@
|
|
| 11 |
"encoding_method": "rope",
|
| 12 |
"dim": 32
|
| 13 |
},
|
| 14 |
-
"vocab_size":
|
| 15 |
}
|
|
|
|
| 11 |
"encoding_method": "rope",
|
| 12 |
"dim": 32
|
| 13 |
},
|
| 14 |
+
"vocab_size": 35328
|
| 15 |
}
|
models/embeddings/monolingual/ary_64d.bin
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:d8ea049910ca1cbe494ecfa4887bb1fe2c54a8fedd0e092181cf1830e949581c
|
| 3 |
+
size 530824726
|
models/embeddings/monolingual/ary_64d_metadata.json
CHANGED
|
@@ -11,5 +11,5 @@
|
|
| 11 |
"encoding_method": "rope",
|
| 12 |
"dim": 64
|
| 13 |
},
|
| 14 |
-
"vocab_size":
|
| 15 |
}
|
|
|
|
| 11 |
"encoding_method": "rope",
|
| 12 |
"dim": 64
|
| 13 |
},
|
| 14 |
+
"vocab_size": 35328
|
| 15 |
}
|
models/subword_markov/ary_markov_ctx1_subword.parquet
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:55514cae207966a8d7941bc6bb156381590fb5450034766506d063720af59bbd
|
| 3 |
+
size 139719
|
models/subword_markov/ary_markov_ctx1_subword_metadata.json
CHANGED
|
@@ -2,6 +2,6 @@
|
|
| 2 |
"context_size": 1,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
-
"unique_contexts":
|
| 6 |
-
"total_transitions":
|
| 7 |
}
|
|
|
|
| 2 |
"context_size": 1,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
+
"unique_contexts": 2156,
|
| 6 |
+
"total_transitions": 11674521
|
| 7 |
}
|
models/subword_markov/ary_markov_ctx2_subword.parquet
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bdab4267cc7d278b8413758b9477e5cf1a89f1d231f8007c5be62d51c38175bf
|
| 3 |
+
size 800589
|
models/subword_markov/ary_markov_ctx2_subword_metadata.json
CHANGED
|
@@ -2,6 +2,6 @@
|
|
| 2 |
"context_size": 2,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
-
"unique_contexts":
|
| 6 |
-
"total_transitions":
|
| 7 |
}
|
|
|
|
| 2 |
"context_size": 2,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
+
"unique_contexts": 18029,
|
| 6 |
+
"total_transitions": 11663613
|
| 7 |
}
|
models/subword_markov/ary_markov_ctx3_subword.parquet
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:ab02a96e6f353bae3406898971d59327277b5f77ca5ecaa7af3558d3d4267e33
|
| 3 |
+
size 2871565
|
models/subword_markov/ary_markov_ctx3_subword_metadata.json
CHANGED
|
@@ -2,6 +2,6 @@
|
|
| 2 |
"context_size": 3,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
-
"unique_contexts":
|
| 6 |
-
"total_transitions":
|
| 7 |
}
|
|
|
|
| 2 |
"context_size": 3,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
+
"unique_contexts": 91858,
|
| 6 |
+
"total_transitions": 11652705
|
| 7 |
}
|
models/subword_markov/ary_markov_ctx4_subword.parquet
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:2ea7ed36e08017627aac4cc8835eea513a82ec11c7be675a5928dd3f8519dc4a
|
| 3 |
+
size 9275191
|
models/subword_markov/ary_markov_ctx4_subword_metadata.json
CHANGED
|
@@ -2,6 +2,6 @@
|
|
| 2 |
"context_size": 4,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
-
"unique_contexts":
|
| 6 |
-
"total_transitions":
|
| 7 |
}
|
|
|
|
| 2 |
"context_size": 4,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
+
"unique_contexts": 379445,
|
| 6 |
+
"total_transitions": 11641797
|
| 7 |
}
|
models/subword_ngram/ary_2gram_subword.parquet
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7298aadecd73d04e9b7a4e3152c422afe282ef437eb416072ff772fc3c1dd431
|
| 3 |
+
size 83975
|
models/subword_ngram/ary_2gram_subword_metadata.json
CHANGED
|
@@ -2,6 +2,6 @@
|
|
| 2 |
"n": 2,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
-
"unique_ngrams":
|
| 6 |
-
"total_ngrams":
|
| 7 |
}
|
|
|
|
| 2 |
"n": 2,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
+
"unique_ngrams": 5903,
|
| 6 |
+
"total_ngrams": 11674521
|
| 7 |
}
|
models/subword_ngram/ary_3gram_subword.parquet
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7e04cb0f3f2a3712d267658842c70d1c1627ee29126f4da41b291573c910be00
|
| 3 |
+
size 604438
|
models/subword_ngram/ary_3gram_subword_metadata.json
CHANGED
|
@@ -2,6 +2,6 @@
|
|
| 2 |
"n": 3,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
-
"unique_ngrams":
|
| 6 |
-
"total_ngrams":
|
| 7 |
}
|
|
|
|
| 2 |
"n": 3,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
+
"unique_ngrams": 44651,
|
| 6 |
+
"total_ngrams": 11663613
|
| 7 |
}
|
models/subword_ngram/ary_4gram_subword.parquet
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:bea49a185c99f23829a987f8811565f84a2bf7bb939dd49da26296a7f133d3e0
|
| 3 |
+
size 2817173
|
models/subword_ngram/ary_4gram_subword_metadata.json
CHANGED
|
@@ -2,6 +2,6 @@
|
|
| 2 |
"n": 4,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
-
"unique_ngrams":
|
| 6 |
-
"total_ngrams":
|
| 7 |
}
|
|
|
|
| 2 |
"n": 4,
|
| 3 |
"variant": "subword",
|
| 4 |
"language": "ary",
|
| 5 |
+
"unique_ngrams": 220807,
|
| 6 |
+
"total_ngrams": 11652705
|
| 7 |
}
|
models/subword_ngram/ary_5gram_subword.parquet
ADDED
|
@@ -0,0 +1,3 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:83d654927713562a4a35f18a74435eddd9a4bdb3826398b5c97645d1d54e5586
|
| 3 |
+
size 7203019
|
models/subword_ngram/ary_5gram_subword_metadata.json
ADDED
|
@@ -0,0 +1,7 @@
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
|
| 1 |
+
{
|
| 2 |
+
"n": 5,
|
| 3 |
+
"variant": "subword",
|
| 4 |
+
"language": "ary",
|
| 5 |
+
"unique_ngrams": 527725,
|
| 6 |
+
"total_ngrams": 11641797
|
| 7 |
+
}
|
models/tokenizer/ary_tokenizer_16k.model
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:166276d42d772567739af90c69d0c0c3d3bfb356b5dc2ec54a6cfd604bc980ba
|
| 3 |
+
size 559287
|
models/tokenizer/ary_tokenizer_16k.vocab
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
models/tokenizer/ary_tokenizer_32k.model
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8cf5ee359c6c39652a55a46a80cdb073224a2ef37eb4ba1e16b72ea8f0a28c93
|
| 3 |
+
size 891778
|
models/tokenizer/ary_tokenizer_32k.vocab
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
models/tokenizer/ary_tokenizer_64k.model
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:62198000b383c4624da36e526abdec55d4335749afe544b9c02a5a2e5005311c
|
| 3 |
+
size 1592960
|
models/tokenizer/ary_tokenizer_64k.vocab
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
models/tokenizer/ary_tokenizer_8k.model
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:b311f5925653a8b93f64537886e776038c09e252ffaf68d3c837d169fc2ce5ca
|
| 3 |
+
size 396519
|
models/tokenizer/ary_tokenizer_8k.vocab
CHANGED
|
The diff for this file is too large to render.
See raw diff
|
|
|
models/vocabulary/ary_vocabulary.parquet
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:06972d02fe5edd7023b3c7da816f69a17e6b3e79183317908fa5f435bcd3fe0b
|
| 3 |
+
size 1396990
|
models/vocabulary/ary_vocabulary_metadata.json
CHANGED
|
@@ -1,17 +1,17 @@
|
|
| 1 |
{
|
| 2 |
"language": "ary",
|
| 3 |
-
"vocabulary_size":
|
| 4 |
"variant": "full",
|
| 5 |
"statistics": {
|
| 6 |
-
"type_token_ratio": 0.
|
| 7 |
"coverage": {
|
| 8 |
-
"top_100": 0.
|
| 9 |
-
"top_1000": 0.
|
| 10 |
-
"top_5000": 0.
|
| 11 |
-
"top_10000": 0.
|
| 12 |
},
|
| 13 |
-
"hapax_count":
|
| 14 |
-
"hapax_ratio": 0.
|
| 15 |
-
"total_documents":
|
| 16 |
}
|
| 17 |
}
|
|
|
|
| 1 |
{
|
| 2 |
"language": "ary",
|
| 3 |
+
"vocabulary_size": 78779,
|
| 4 |
"variant": "full",
|
| 5 |
"statistics": {
|
| 6 |
+
"type_token_ratio": 0.0838713972533892,
|
| 7 |
"coverage": {
|
| 8 |
+
"top_100": 0.36742576968301266,
|
| 9 |
+
"top_1000": 0.5995843327288514,
|
| 10 |
+
"top_5000": 0.7413866015051412,
|
| 11 |
+
"top_10000": 0.8022598684642346
|
| 12 |
},
|
| 13 |
+
"hapax_count": 100115,
|
| 14 |
+
"hapax_ratio": 0.5596330788064441,
|
| 15 |
+
"total_documents": 10908
|
| 16 |
}
|
| 17 |
}
|
models/word_markov/ary_markov_ctx1_word.parquet
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:7e482ecf3d5961f6adc3f0517998d923d016dd4049af100ca43da1d4f5714507
|
| 3 |
+
size 10606246
|
models/word_markov/ary_markov_ctx1_word_metadata.json
CHANGED
|
@@ -2,6 +2,6 @@
|
|
| 2 |
"context_size": 1,
|
| 3 |
"variant": "word",
|
| 4 |
"language": "ary",
|
| 5 |
-
"unique_contexts":
|
| 6 |
-
"total_transitions":
|
| 7 |
}
|
|
|
|
| 2 |
"context_size": 1,
|
| 3 |
"variant": "word",
|
| 4 |
"language": "ary",
|
| 5 |
+
"unique_contexts": 178865,
|
| 6 |
+
"total_transitions": 2122048
|
| 7 |
}
|
models/word_markov/ary_markov_ctx2_word.parquet
CHANGED
|
@@ -1,3 +1,3 @@
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
-
oid sha256:
|
| 3 |
-
size
|
|
|
|
| 1 |
version https://git-lfs.github.com/spec/v1
|
| 2 |
+
oid sha256:8099e9348247ef07140e6a4819679ee44e17e70727a77d4336c59f14eac55086
|
| 3 |
+
size 26663466
|
models/word_markov/ary_markov_ctx2_word_metadata.json
CHANGED
|
@@ -2,6 +2,6 @@
|
|
| 2 |
"context_size": 2,
|
| 3 |
"variant": "word",
|
| 4 |
"language": "ary",
|
| 5 |
-
"unique_contexts":
|
| 6 |
-
"total_transitions":
|
| 7 |
}
|
|
|
|
| 2 |
"context_size": 2,
|
| 3 |
"variant": "word",
|
| 4 |
"language": "ary",
|
| 5 |
+
"unique_contexts": 962233,
|
| 6 |
+
"total_transitions": 2111140
|
| 7 |
}
|