omarkamali commited on
Commit
6e38dac
·
verified ·
1 Parent(s): a541b6f

Upload all models and assets for blk (latest)

Browse files
This view is limited to 50 files because it contains too many changes.   See raw diff
Files changed (50) hide show
  1. .gitattributes +2 -0
  2. README.md +202 -166
  3. models/embeddings/aligned/blk_128d.bin +3 -0
  4. models/embeddings/aligned/blk_128d.meta.json +1 -0
  5. models/embeddings/aligned/blk_128d.projection.npy +3 -0
  6. models/embeddings/aligned/blk_128d_metadata.json +8 -0
  7. models/embeddings/aligned/blk_32d.bin +3 -0
  8. models/embeddings/aligned/blk_32d.meta.json +1 -0
  9. models/embeddings/aligned/blk_32d.projection.npy +3 -0
  10. models/embeddings/aligned/blk_32d_metadata.json +8 -0
  11. models/embeddings/aligned/blk_64d.bin +3 -0
  12. models/embeddings/aligned/blk_64d.meta.json +1 -0
  13. models/embeddings/aligned/blk_64d.projection.npy +3 -0
  14. models/embeddings/aligned/blk_64d_metadata.json +8 -0
  15. models/embeddings/monolingual/blk_128d.bin +2 -2
  16. models/embeddings/monolingual/blk_128d_metadata.json +1 -1
  17. models/embeddings/monolingual/blk_32d.bin +2 -2
  18. models/embeddings/monolingual/blk_32d_metadata.json +1 -1
  19. models/embeddings/monolingual/blk_64d.bin +2 -2
  20. models/embeddings/monolingual/blk_64d_metadata.json +1 -1
  21. models/subword_markov/blk_markov_ctx1_subword.parquet +2 -2
  22. models/subword_markov/blk_markov_ctx1_subword_metadata.json +2 -2
  23. models/subword_markov/blk_markov_ctx2_subword.parquet +2 -2
  24. models/subword_markov/blk_markov_ctx2_subword_metadata.json +2 -2
  25. models/subword_markov/blk_markov_ctx3_subword.parquet +2 -2
  26. models/subword_markov/blk_markov_ctx3_subword_metadata.json +2 -2
  27. models/subword_markov/blk_markov_ctx4_subword.parquet +2 -2
  28. models/subword_markov/blk_markov_ctx4_subword_metadata.json +2 -2
  29. models/subword_ngram/blk_2gram_subword.parquet +2 -2
  30. models/subword_ngram/blk_2gram_subword_metadata.json +2 -2
  31. models/subword_ngram/blk_3gram_subword.parquet +2 -2
  32. models/subword_ngram/blk_3gram_subword_metadata.json +2 -2
  33. models/subword_ngram/blk_4gram_subword.parquet +2 -2
  34. models/subword_ngram/blk_4gram_subword_metadata.json +2 -2
  35. models/subword_ngram/blk_5gram_subword.parquet +3 -0
  36. models/subword_ngram/blk_5gram_subword_metadata.json +7 -0
  37. models/tokenizer/blk_tokenizer_16k.model +2 -2
  38. models/tokenizer/blk_tokenizer_16k.vocab +0 -0
  39. models/tokenizer/blk_tokenizer_32k.model +2 -2
  40. models/tokenizer/blk_tokenizer_32k.vocab +0 -0
  41. models/tokenizer/blk_tokenizer_64k.model +2 -2
  42. models/tokenizer/blk_tokenizer_64k.vocab +0 -0
  43. models/tokenizer/blk_tokenizer_8k.model +2 -2
  44. models/tokenizer/blk_tokenizer_8k.vocab +0 -0
  45. models/vocabulary/blk_vocabulary.parquet +2 -2
  46. models/vocabulary/blk_vocabulary_metadata.json +9 -9
  47. models/word_markov/blk_markov_ctx1_word.parquet +2 -2
  48. models/word_markov/blk_markov_ctx1_word_metadata.json +2 -2
  49. models/word_markov/blk_markov_ctx2_word.parquet +2 -2
  50. models/word_markov/blk_markov_ctx2_word_metadata.json +2 -2
.gitattributes CHANGED
@@ -39,3 +39,5 @@ visualizations/position_encoding_comparison.png filter=lfs diff=lfs merge=lfs -t
39
  visualizations/tsne_sentences.png filter=lfs diff=lfs merge=lfs -text
40
  visualizations/tsne_words.png filter=lfs diff=lfs merge=lfs -text
41
  visualizations/zipf_law.png filter=lfs diff=lfs merge=lfs -text
 
 
 
39
  visualizations/tsne_sentences.png filter=lfs diff=lfs merge=lfs -text
40
  visualizations/tsne_words.png filter=lfs diff=lfs merge=lfs -text
41
  visualizations/zipf_law.png filter=lfs diff=lfs merge=lfs -text
42
+ visualizations/embedding_tsne_multilingual.png filter=lfs diff=lfs merge=lfs -text
43
+ visualizations/ngram_coverage.png filter=lfs diff=lfs merge=lfs -text
README.md CHANGED
@@ -1,6 +1,6 @@
1
  ---
2
  language: blk
3
- language_name: BLK
4
  language_family: tibetoburman_other
5
  tags:
6
  - wikilangs
@@ -10,11 +10,21 @@ tags:
10
  - n-gram
11
  - markov
12
  - wikipedia
 
 
 
 
 
 
 
 
 
 
13
  - monolingual
14
  - family-tibetoburman_other
15
  license: mit
16
  library_name: wikilangs
17
- pipeline_tag: feature-extraction
18
  datasets:
19
  - omarkamali/wikipedia-monthly
20
  dataset_info:
@@ -23,20 +33,20 @@ dataset_info:
23
  metrics:
24
  - name: best_compression_ratio
25
  type: compression
26
- value: 4.845
27
  - name: best_isotropy
28
  type: isotropy
29
- value: 0.8617
30
  - name: vocabulary_size
31
  type: vocab
32
  value: 0
33
  generated: 2026-01-03
34
  ---
35
 
36
- # BLK - Wikilangs Models
37
  ## Comprehensive Research Report & Full Ablation Study
38
 
39
- This repository contains NLP models trained and evaluated by Wikilangs, specifically on **BLK** Wikipedia data.
40
  We analyze tokenizers, n-gram models, Markov chains, vocabulary statistics, and word embeddings.
41
 
42
  ## 📋 Repository Contents
@@ -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-morphological-analysis)
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** | 4.016x | 4.02 | 0.0510% | 1,061,065 |
84
- | **16k** | 4.425x | 4.43 | 0.0562% | 962,856 |
85
- | **32k** | 4.609x | 4.61 | 0.0585% | 924,505 |
86
- | **64k** | 4.845x 🏆 | 4.85 | 0.0615% | 879,406 |
87
 
88
  ### Tokenization Examples
89
 
90
  Below are sample sentences tokenized with each vocabulary size:
91
 
92
- **Sample 1:** `အမုဲင် ခမ်းထီ ကသှိုပ်စဒါႏ ငဝ်းလဝ်းနီꩻ ၃၅လာအို ၉၄ ထူႏတောမ်`
93
 
94
  | Vocab | Tokens | Count |
95
  |-------|--------|-------|
96
- | 8k | `▁အမုဲင် ▁ခမ်းထီ ▁က ှို ပ် စဒါႏ ▁ငဝ်း ဝ်း ... (+7 more)` | 17 |
97
- | 16k | `▁အမုဲင် ▁ခမ်းထီ ▁ကသ ှိုပ် စဒါႏ ▁ငဝ်း လဝ်း နီꩻ ▁၃၅ လာအို ... (+3 more)` | 13 |
98
- | 32k | `▁အမုဲင် ▁ခမ်းထီ ▁ကသှိုပ်စဒါႏ ▁ငဝ်း လဝ်း နီꩻ ▁၃၅လာအို ▁၉ ▁ထူႏတောမ်` | 10 |
99
- | 64k | `▁အမုဲင် ▁ခမ်းထီ ▁ကသှိုပ်စဒါႏ ▁ငဝ်းလဝ်းနီꩻ ▁၃၅လာအို ▁၉၄ ▁ထူႏတောမ်` | 7 |
100
 
101
- **Sample 2:** `the war is very bad!a website to summarise the war`
102
 
103
  | Vocab | Tokens | Count |
104
  |-------|--------|-------|
105
- | 8k | `▁the ▁w ar ▁is ▁ver y ▁b ad ! a ... (+11 more)` | 21 |
106
- | 16k | `▁the ▁war ▁is ▁ver y ▁b ad ! a ▁website ... (+6 more)` | 16 |
107
- | 32k | `▁the ▁war ▁is ▁very ▁b ad ! a ▁website ▁to ... (+3 more)` | 13 |
108
- | 64k | `▁the ▁war ▁is ▁very ▁bad ! a ▁website ▁to ▁summarise ... (+2 more)` | 12 |
109
 
110
- **Sample 3:** `မျန်မာခမ်းထီကိုယို ခမ်းနယ်ႏ အဝ်ႏ ( )ခမ်းနယ်ႏ နဝ်ꩻသွူ ။`
111
 
112
  | Vocab | Tokens | Count |
113
  |-------|--------|-------|
114
- | 8k | `▁မျန်မာခမ်းထီ ကိုယို ▁ခမ်းနယ်ႏ ▁အဝ်ႏ ▁( ▁၇ ▁) ခမ်းနယ်ႏ ▁နဝ်ꩻ သွူ ... (+1 more)` | 11 |
115
- | 16k | `▁မျန်မာခမ်းထီ ကိုယို ▁ခမ်းနယ်ႏ ▁အဝ်ႏ ▁( ▁၇ ▁) ခမ်းနယ်ႏ ▁နဝ်ꩻသွူ ▁။` | 10 |
116
- | 32k | `▁မျန်မာခမ်းထီ ကိုယို ▁ခမ်းနယ်ႏ ▁အဝ်ႏ ▁( ▁၇ ▁) ခမ်းနယ်ႏ ▁နဝ်ꩻသွူ ▁။` | 10 |
117
- | 64k | `▁မျန်မာခမ်းထီ ကိုယို ▁ခမ်းနယ်ႏ ▁အဝ်ႏ ▁( ▁၇ ▁) ခမ်းနယ်ႏ ▁နဝ်ꩻသွူ ▁။` | 10 |
118
 
119
 
120
  ### Key Findings
121
 
122
- - **Best Compression:** 64k achieves 4.845x compression
123
- - **Lowest UNK Rate:** 8k with 0.0510% unknown tokens
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 | 2,554 | 11.32 | 4,328 | 21.2% | 57.8% |
141
- | **2-gram** | Subword | 1,405 🏆 | 10.46 | 24,351 | 42.7% | 76.9% |
142
- | **3-gram** | Word | 3,876 | 11.92 | 6,558 | 18.8% | 47.3% |
143
- | **3-gram** | Subword | 11,360 | 13.47 | 129,980 | 18.9% | 45.0% |
144
- | **4-gram** | Word | 16,945 | 14.05 | 23,380 | 8.9% | 21.9% |
145
- | **4-gram** | Subword | 54,384 | 15.73 | 407,218 | 10.1% | 25.7% |
 
 
146
 
147
  ### Top 5 N-grams by Size
148
 
@@ -150,9 +162,9 @@ Below are sample sentences tokenized with each vocabulary size:
150
 
151
  | Rank | N-gram | Count |
152
  |------|--------|-------|
153
- | 1 | `နဝ်ꩻ အဝ်ႏဒျာႏ` | 718 |
154
  | 2 | `အဝ်ႏဒျာႏ မျန်မာခမ်းထီ` | 691 |
155
- | 3 | `ခရိစ်နေင်ႏ ဗာႏ` | 404 |
156
  | 4 | `ဗာႏ စာႏရင်ꩻအလꩻ` | 320 |
157
  | 5 | `မျန်မာခမ်းထီ အခဝ်ထာႏဝ` | 295 |
158
 
@@ -172,46 +184,66 @@ Below are sample sentences tokenized with each vocabulary size:
172
  |------|--------|-------|
173
  | 1 | `နဝ်ꩻ အဝ်ႏဒျာႏ မျန်မာခမ်းထီ အခဝ်ထာႏဝ` | 282 |
174
  | 2 | `ခရိစ်နေင်ႏ ဗာႏ စာႏရင်ꩻအလꩻ ဝေင်ꩻကိုနဝ်ꩻ` | 161 |
175
- | 3 | `သီမားသားဖုံႏ မွူးနီꩻအုံပဆားနီꩻဖုံႏတောမ်ႏ အထွတ်အမျတ်မွူးနီꩻဖုံႏ အာႏကွိုꩻ` | 153 |
176
- | 4 | `လွူးဖွာꩻသားဖုံႏ သီမားသားဖုံႏ မွူးနီꩻအုံပဆားနီꩻဖုံႏတောမ်ႏ အထွတ်အမျတ်မွူးနီꩻဖုံႏ` | 153 |
177
  | 5 | `ထာꩻထွာဖုံႏ လွူးဖွာꩻသားဖုံႏ သီမားသားဖုံႏ မွူးနီꩻအုံပဆားနီꩻဖုံႏတောမ်ႏ` | 153 |
178
 
 
 
 
 
 
 
 
 
 
 
179
  **2-grams (Subword):**
180
 
181
  | Rank | N-gram | Count |
182
  |------|--------|-------|
183
- | 1 | `ာ ႏ` | 142,556 |
184
- | 2 | `၊ _` | 135,431 |
185
- | 3 | `ꩻ _` | 126,463 |
186
- | 4 | `ဝ် ꩻ` | 102,686 |
187
- | 5 | `င် ꩻ` | 97,010 |
188
 
189
  **3-grams (Subword):**
190
 
191
  | Rank | N-gram | Count |
192
  |------|--------|-------|
193
- | 1 | `န ဝ် ꩻ` | 77,017 |
194
- | 2 | `ဝ် ꩻ _` | 57,560 |
195
- | 3 | `ꩻ ၊ _` | 31,807 |
196
- | 4 | `သွူ ။ _` | 31,585 |
197
- | 5 | `ႏ ၊ _` | 30,939 |
198
 
199
  **4-grams (Subword):**
200
 
201
  | Rank | N-gram | Count |
202
  |------|--------|-------|
203
- | 1 | `န ဝ် ꩻ _` | 45,458 |
204
- | 2 | `နေ ာ ဝ် ꩻ` | 23,540 |
205
- | 3 | `ꩻ သွူ ။ _` | 18,995 |
206
- | 4 | `ꩻ န ဝ် ꩻ` | 18,028 |
207
- | 5 | `ႏ န ဝ် ꩻ` | 17,062 |
 
 
 
 
 
 
 
 
 
 
208
 
209
 
210
  ### Key Findings
211
 
212
- - **Best Perplexity:** 2-gram (subword) with 1,405
213
  - **Entropy Trend:** Decreases with larger n-grams (more predictable)
214
- - **Coverage:** Top-1000 patterns cover ~26% of corpus
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 | 0.2313 | 1.174 | 1.60 | 382,155 | 76.9% |
231
- | **1** | Subword | 1.2242 | 2.336 | 21.07 | 2,910 | 0.0% |
232
- | **2** | Word | 0.0413 | 1.029 | 1.06 | 611,948 | 95.9% |
233
- | **2** | Subword | 0.7533 | 1.686 | 5.49 | 61,297 | 24.7% |
234
- | **3** | Word | 0.0155 | 1.011 | 1.02 | 648,373 | 98.5% |
235
- | **3** | Subword | 0.4736 | 1.389 | 2.77 | 336,631 | 52.6% |
236
- | **4** | Word | 0.0088 🏆 | 1.006 | 1.01 | 660,080 | 99.1% |
237
- | **4** | Subword | 0.3161 | 1.245 | 1.90 | 934,101 | 68.4% |
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. `၂ ဖြားတန်မောင်ငွေ ဥက္ကဋ္ဌ နာယက ဗွေႏဗွန်ကျောင်ꩻဝေင်ꩻသီႏသဲင်ႏ အကျိုꩻဆောင်ႏနဝ်ꩻ ဗွေႏဗွန်လုံးဖိုးကျောင်ꩻ...`
246
- 2. `၃ နီꩻကို နုဲင်ႏငံႏတောႏအစိုႏရ ကတဲမ်းထွို့ꩻဒါႏ ဗုဲင်းရတ်သ်ပုဂ္ဂိုလ်ႏယင်ဟန်ႏသားနဝ်ꩻ ဝွေꩻသီး အွဉ်မာꩻချွမ...`
247
- 3. `၁ အဟွိုန်အထီႏ မဉ်ႏအာနောဝ်ꩻ ဗွေႏမျတ်ဘုရာꩻ ကဟော်ꩻသေꩻခါꩻအတွိုင်ꩻသွူ ပိဋကတ်ပြန်ႏပအိုဝ်ႏစွိုꩻကထေသေꩻခါꩻ ပြ...`
248
 
249
  **Context Size 2:**
250
 
251
- 1. `နဝ်ꩻ အဝ်ႏဒျာႏ မျန်မာခမ်းထီ အခဝ်ထာႏဝ မကွေးတွိုင်ꩻဒေႏသတန် ချောက်ခရဲင်ႏ ဝေင်ꩻနယ်ႏချောက်ကို ကအဝ်ႏဒါႏ ဧရာ...`
252
- 2. `အဝ်ႏဒျာႏ မျန်မာခမ်းထီ ဖြဝ်ꩻခမ်းနယ်ႏအခဝ်နဝ် ကလော်ꩻခရဲင်ႏ ဝေင်ꩻနယ်ႏညောင်ႏရွီႏကို ကအဝ်ႏဒါႏ ဝေင်ꩻတဝေင်ꩻဒ...`
253
- 3. `ခရိစ်နေင်ႏ ဗာႏ မွူးကွို့ꩻနာꩻစာႏရင်ꩻအလꩻ အဝ်ႏ ဖြာꩻသွူ အွိုင်ႏလွယ်အွန်ႏကို လိုꩻဖြာꩻခြွဉ်းနဝ်ꩻ နေင်ႏ ဗာႏ...`
254
 
255
  **Context Size 3:**
256
 
257
- 1. `နဝ်ꩻ အဝ်ႏဒျာႏ မျန်မာခမ်းထီ အခဝ်ခိုႏ သကုဲင်းတွိုင်ꩻဒေႏသတန် အခဝ်ကွဉ်ႏထင်ꩻ ကတာခရဲင်ႏ ဝေင်ꩻနယ်ႏအိဉ်းတော်...`
258
- 2. `အဝ်ႏဒျာႏ မျန်မာခမ်းထီ အခဝ်ထာႏဝ မန္တလေꩻတွိုင်ꩻဒေႏသတန် ဝေင်ꩻနယ်ႏနာထိုးစီးကို ကအဝ်ႏဒါႏ ဝေင်ꩻနယ်ႏရွုမ်ꩻ ...`
259
- 3. `ခရိစ်နေင်ႏ ဗာႏ စာႏရင်ꩻအလꩻ ဝေင်ꩻကိုနဝ်ꩻ လိုꩻဖြာꩻခြွဉ်းအဝ်ႏ ၂၅၂ ဖြာꩻသွူ အာႏကွိုꩻ`
260
 
261
  **Context Size 4:**
262
 
263
- 1. `နဝ်ꩻ အဝ်ႏဒျာႏ မျန်မာခမ်းထီ အခဝ်ထာႏဝ မန်ႏတလေꩻတွိုင်ꩻဒေႏသတန် မိဉ်းဆျန်ခရဲင်ႏကို ကအဝ်ႏပါသော့ꩻဒါႏ ဝေင်ꩻန...`
264
- 2. `ခရိစ်နေင်ႏ ဗာႏ စာႏရင်ꩻအလꩻ ဝေင်ꩻကိုနဝ်ꩻ လိုꩻဖြာꩻခြွဉ်းအဝ်ႏ ၅၈၇ ရပ်သွူ အာႏကွိုꩻ`
265
- 3. `လွူးဖွာꩻသားဖုံႏ သီမားသားဖုံႏ မွူးနီꩻအုံပဆားနီꩻဖုံႏတောမ်ႏ အထွတ်အမျတ်မွူးနီꩻဖုံႏ အာႏကွိုꩻ`
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. `_လွေꩻလꩻဗာꩻ_၈-_နာတ်ꩻ`
275
- 2. `ꩻငဝ်း၊_ထွာယ်ဆည်းရွယ်ႏန`
276
- 3. `ႏ_သမ်းပလောက်ထာႏတသီးထ`
277
 
278
  **Context Size 2:**
279
 
280
- 1. `ာႏရေꩻတွယ်ႏတယယ်ꩻထင်ႏနောဝ်`
281
- 2. `၊_နာႏလွယ်၊_ထွာဆေ့ꩻရွစ်ဟော`
282
- 3. `ꩻ__ကတ္တရာꩻနုဲင်းယိုသွူ-_သ`
283
 
284
  **Context Size 3:**
285
 
286
- 1. `နဝ်ꩻတဲ့_စူꩻအကိုသူꩻ_ထွာ_ပသေဒဲ`
287
- 2. `ဝ်ꩻ_ကဖေႏအံႏ_ခမ်းနေႏရာႏသွု`
288
- 3. `ꩻ၊_၃။_ဖြော်ချာ_လူကြိုက်များရ`
289
 
290
  **Context Size 4:**
291
 
292
- 1. `နဝ်ꩻ_လိုꩻဖြာꩻ၊_နောဝ်ထင်ꩻ_ဖုံ`
293
- 2. `နောဝ်ꩻ_အွဉ်အကိုရေꩻ_(၂၂၂)။_`
294
- 3. `ꩻသွူ။_အားထွုတ်"ဗုဒ္ဓါနုဿတိကမ္မဋ္ဌ`
295
 
296
 
297
  ### Key Findings
298
 
299
  - **Best Predictability:** Context-4 (word) with 99.1% predictability
300
  - **Branching Factor:** Decreases with context size (more deterministic)
301
- - **Memory Trade-off:** Larger contexts require more storage (934,101 contexts)
302
  - **Recommendation:** Context-3 or Context-4 for text generation
303
 
304
  ---
@@ -314,55 +346,55 @@ Below are text samples generated from each subword-based Markov chain model:
314
 
315
  | Metric | Value |
316
  |--------|-------|
317
- | Vocabulary Size | 68,078 |
318
- | Total Tokens | 398,630 |
319
- | Mean Frequency | 5.86 |
320
  | Median Frequency | 2 |
321
- | Frequency Std Dev | 39.89 |
322
 
323
  ### Most Common Words
324
 
325
  | Rank | Word | Frequency |
326
  |------|------|-----------|
327
- | 1 | ၂ | 3,802 |
328
- | 2 | ၃ | 3,377 |
329
- | 3 | ၁ | 3,336 |
330
  | 4 | အာႏကွိုꩻ | 3,141 |
331
- | 5 | နဝ်ꩻ | 2,713 |
332
- | 6 | ၄ | 2,610 |
333
- | 7 | ၅ | 2,059 |
334
- | 8 | ထွာဒျာႏ | 1,628 |
335
- | 9 | ၆ | 1,583 |
336
- | 10 | အဝ်ႏဒျာႏ | 1,493 |
337
 
338
  ### Least Common Words (from vocabulary)
339
 
340
  | Rank | Word | Frequency |
341
  |------|------|-----------|
342
- | 1 | ယံဇိတံ | 2 |
343
- | 2 | ထာꩻအောင်ႏယမျိုꩻနွ်ꩻ | 2 |
344
- | 3 | လိုꩻစင်ꩻတဖဲ့ꩻသား | 2 |
345
- | 4 | ပစ္စာဟရိဿာမိ | 2 |
346
- | 5 | အဖွန်ႏအခွမ်ꩻတွယ်ꩻမုꩻ | 2 |
347
- | 6 | တပုဲင်ႏဗွာ | 2 |
348
- | 7 | nashi | 2 |
349
- | 8 | ညီႏလာႏခံႏအကို | 2 |
350
- | 9 | ပြဲႏထောင်ႏစု | 2 |
351
- | 10 | ဆောင်ႏရွတ်ဖေႏ | 2 |
352
 
353
  ### Zipf's Law Analysis
354
 
355
  | Metric | Value |
356
  |--------|-------|
357
- | Zipf Coefficient | 0.7925 |
358
- | R² (Goodness of Fit) | 0.997962 |
359
  | Adherence Quality | **excellent** |
360
 
361
  ### Coverage Analysis
362
 
363
  | Top N Words | Coverage |
364
  |-------------|----------|
365
- | Top 100 | 18.0% |
366
  | Top 1,000 | 34.4% |
367
  | Top 5,000 | 51.9% |
368
  | Top 10,000 | 61.5% |
@@ -370,8 +402,8 @@ Below are text samples generated from each subword-based Markov chain model:
370
  ### Key Findings
371
 
372
  - **Zipf Compliance:** R²=0.9980 indicates excellent adherence to Zipf's law
373
- - **High Frequency Dominance:** Top 100 words cover 18.0% of corpus
374
- - **Long Tail:** 58,078 words needed for remaining 38.5% coverage
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
- > *Note: Multilingual alignment visualization not available for this language.*
 
 
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 | 0.8617 🏆 | 0.3357 | N/A | N/A |
398
- | **mono_64d** | 64 | 0.8600 | 0.2769 | N/A | N/A |
399
- | **mono_128d** | 128 | 0.6775 | 0.2411 | N/A | N/A |
 
 
 
400
 
401
  ### Key Findings
402
 
403
- - **Best Isotropy:** mono_32d with 0.8617 (more uniform distribution)
404
- - **Semantic Density:** Average pairwise similarity of 0.2846. Lower values indicate better semantic separation.
405
- - **Alignment Quality:** No aligned models evaluated in this run.
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 | **0.000** | Low morphological productivity | ⚠️ Likely unreliable |
420
- | Idiomaticity Gap | **-1.000** | Low formulaic content | - |
421
 
422
  ### 6.2 Affix Inventory (Productive Units)
423
 
@@ -426,21 +461,20 @@ These are the most productive prefixes and suffixes identified by sampling the v
426
  #### Productive Prefixes
427
  | Prefix | Examples |
428
  |--------|----------|
429
- | `-လိ` | လိတ်ပအိုဝ်ႏဗဟိုႏသွဉ်တန်ꩻ, လိုꩻမျတ်ဖုံႏ, လိတ်ပအိုဝ်ႏစောင်ႏကို |
430
- | `-လို` | လိုꩻမျတ်ဖုံႏ, လိုꩻသီးဖုံႏယို, လိုꩻသီးယိုနဝ်ꩻ |
431
- | `-လိုꩻ` | လိုꩻမျတ်ဖုံႏ, လိုꩻသီးဖုံႏယို, လိုꩻသီးယိုနဝ်ꩻ |
432
 
433
  #### Productive Suffixes
434
  | Suffix | Examples |
435
  |--------|----------|
436
- | `-ꩻ` | လိတ်ပအိုဝ်ႏဗဟိုႏသွဉ်တန်ꩻ, ခိုမူႏခွန်နင်ꩻ, အလင်နဝ်ꩻ |
437
- | `-ႏ` | ပအိုဝ်ႏတာႏ, ကျင်ꩻလွဉ်ꩻမဲဉ်ႏမဲဉ်ႏဒျာႏ, ခမ်းထီနဲင်ႏငန်ႏ |
438
- | `-်ꩻ` | လိတ်ပအိုဝ်ႏဗဟိုႏသွဉ်တန်ꩻ, ခိုမူႏခွန်နင်ꩻ, အလင်နဝ်ꩻ |
439
- | `-ဝ်ꩻ` | အလင်နဝ်ꩻ, အမတ်ဖုံႏနောဝ်ꩻ, ပါꩻမုဲင်ꩻနဝ်ꩻ |
440
- | `-နဝ်ꩻ` | အလင်နဝ်ꩻ, ပါꩻမုဲင်ꩻနဝ်ꩻ, ခွန်ထွန်းအောင်နဝ်ꩻ |
441
- | `-်း` | လုမ်းလုမ်း, ဘဝလိုꩻခမ်း, အုံတပန်း |
442
- | `-ာႏ` | ပအိုဝ်ႏတာႏ, ကျင်ꩻလွဉ်ꩻမဲဉ်ႏမဲဉ်ႏဒျာႏ, ကိုꩻကွယ်ႏဆရာႏမာႏ |
443
- | `-်ႏ` | ခမ်းထီနဲင်ႏငန်ႏ, ကမ္ဘာႏဟမ်ႏ, သကဒါဂါမိဖိုလ်ႏ |
444
 
445
  ### 6.3 Bound Stems (Lexical Roots)
446
 
@@ -455,16 +489,16 @@ This table shows which prefixes and suffixes most frequently co-occur on the sam
455
 
456
  | Prefix | Suffix | Frequency | Examples |
457
  |--------|--------|-----------|----------|
458
- | `-လိ` | `-ꩻ` | 82 words | လိုꩻရွိုင်ꩻ, လိုꩻဘဝခြွေနယ်ꩻ |
459
- | `-လိ` | `-ႏ` | 54 words | လိုႏဖေႏအာႏငါႏ, လိုꩻနွို့လိုꩻထန်ႏ |
460
- | `-လိ` | `-်ꩻ` | 50 words | လိုꩻရွိုင်ꩻ, လိုꩻဘဝခြွေနယ်ꩻ |
461
- | `-လိ` | `-ဝ်ꩻ` | 38 words | လိတ်လုဲင်ꩻပညာႏသျင်ႏသီးနဝ်ꩻ, လိတ်မွူးတွယ်ꩻနဝ်ꩻ |
462
- | `-လိ` | `-နဝ်ꩻ` | 30 words | လိတ်လုဲင်ꩻပညာႏသျင်ႏသီးနဝ်ꩻ, လိတ်မွူးတွယ်ꩻနဝ်ꩻ |
463
- | `-လိ` | `-ို` | 24 words | လိုႏသော့ꩻလိတ်မွူးကို, လိုꩻတဟဝ်တဝ်းယို |
464
- | `-လိ` | `-်ႏ` | 18 words | လိုꩻနွို့လိုꩻထန်ႏ, လိုꩻဖြာꩻခြွဉ်းအောဝ်ႏ |
465
- | `-လိ` | `-်း` | 17 words | လိုꩻတဲ့ယဝ်း, လိုꩻတသေတဝ်း |
466
- | `-လိ` | `-ာႏ` | 16 words | လိုꩻသꩻရာႏ, လိုꩻခြွေလိုꩻခြာႏ |
467
- | `-လိ` | `-ွူ` | 7 words | လိုꩻမျိုꩻဖုံႏဒျာႏသွူ, လိုꩻအွူးဟွူ |
468
 
469
  ### 6.5 Recursive Morpheme Segmentation
470
 
@@ -472,26 +506,28 @@ Using **Recursive Hierarchical Substitutability**, we decompose complex words in
472
 
473
  | Word | Suggested Split | Confidence | Stem |
474
  |------|-----------------|------------|------|
475
- | ဥပဇ္ဈာယ်ႏ | **`ဥပဇ္ဈာယ-်ႏ`** | 4.5 | `ဥပဇ္ဈာယ` |
476
- | ပငါပရာꩻဖုံႏနဝ်ꩻ | **`ပငါပရာꩻဖုံႏ-နဝ်ꩻ`** | 4.5 | `ပငါပရာꩻဖုံႏ` |
477
- | အနမ်းနဝ်ꩻ | **`အနမ်း-နဝ်ꩻ`** | 4.5 | `အနမ်း` |
478
- | မွူးရဝ်ꩻနီꩻနဝ်ꩻ | **`မွူးရဝ်ꩻနီꩻ-နဝ်ꩻ`** | 4.5 | `မွူးရဝ်ꩻနီꩻ` |
479
- | ရဟန္တာႏသီးနဝ်ꩻ | **`ရဟန္တာႏသီး-နဝ်ꩻ`** | 4.5 | `ရဟန္တာႏသီး` |
480
- | ယိုခါနဝ်ꩻ | **`ယိုခါ-နဝ်ꩻ`** | 4.5 | `ယိုခါ` |
481
- | စဲ့ꩻရေꩻနဝ်ꩻ | **`စဲ့ꩻရေꩻ-နဝ်ꩻ`** | 4.5 | `စဲ့ꩻရေꩻ` |
482
- | လိတ်ကရိုꩻယိုနဝ်ꩻ | **`လိ-တ်ကရိုꩻယ-ို-နဝ်ꩻ`** | 4.5 | `တ်ကရိုꩻယ` |
483
- | ကုဲင်ထိုꩻနဝ်ꩻ | **`ကုဲင်ထိုꩻ-နဝ်ꩻ`** | 4.5 | `ကုဲင်ထိုꩻ` |
484
- | လိုꩻသꩻရာႏ | **`လိုꩻ-သꩻရာႏ`** | 4.5 | `သꩻရာႏ` |
485
- | လိက်ဖြိုင်ႏ | **`လိ-က်ဖြိုင-်ႏ`** | 3.0 | `က်ဖြိုင` |
486
- | လိုꩻအဆင်ႏအရန်း | **`လိုꩻ-အဆင်ႏအရန-်း`** | 3.0 | `အဆင်ႏအရန` |
487
- | လိတ်ကျမ်ꩻ | **`လိ-တ်ကျမ-်ꩻ`** | 3.0 | `တ်ကျမ` |
488
- | လိုꩻသဒ္ဓါႏအဝ်ႏ | **`လိုꩻ-သဒ္ဓါႏအဝ-်ႏ`** | 3.0 | `သဒ္ဓါႏအဝ` |
489
- | ဘာႏဝနာႏနဝ်ꩻ | **`ဘာႏဝန-ာႏ-နဝ်ꩻ`** | 3.0 | `ဘာႏဝန` |
490
 
491
  ### 6.6 Linguistic Interpretation
492
 
493
  > **Automated Insight:**
494
- The language BLK appears to be more isolating or has a highly fixed vocabulary. Word-level models perform nearly as well as subword models, indicating fewer productive morphological processes.
 
 
495
 
496
  ---
497
  ## 7. Summary & Recommendations
@@ -503,7 +539,7 @@ The language BLK appears to be more isolating or has a highly fixed vocabulary.
503
  | Component | Recommended | Rationale |
504
  |-----------|-------------|-----------|
505
  | Tokenizer | **64k BPE** | Best compression (4.85x) |
506
- | N-gram | **2-gram** | Lowest perplexity (1,405) |
507
  | Markov | **Context-4** | Highest predictability (99.1%) |
508
  | Embeddings | **100d** | Balanced semantic capture and isotropy |
509
 
@@ -718,4 +754,4 @@ MIT License - Free for academic and commercial use.
718
  ---
719
  *Generated by Wikilangs Models Pipeline*
720
 
721
- *Report Date: 2026-01-03 07:25:42*
 
1
  ---
2
  language: blk
3
+ language_name: Pa'o Karen
4
  language_family: tibetoburman_other
5
  tags:
6
  - wikilangs
 
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-tibetoburman_other
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: 4.848
37
  - name: best_isotropy
38
  type: isotropy
39
+ value: 0.8632
40
  - name: vocabulary_size
41
  type: vocab
42
  value: 0
43
  generated: 2026-01-03
44
  ---
45
 
46
+ # Pa'o Karen - Wikilangs Models
47
  ## Comprehensive Research Report & Full Ablation Study
48
 
49
+ This repository contains NLP models trained and evaluated by Wikilangs, specifically on **Pa'o Karen** Wikipedia data.
50
  We analyze tokenizers, n-gram models, Markov chains, vocabulary statistics, and word embeddings.
51
 
52
  ## 📋 Repository Contents
 
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** | 4.022x | 4.02 | 0.0580% | 1,056,850 |
94
+ | **16k** | 4.430x | 4.43 | 0.0639% | 959,541 |
95
+ | **32k** | 4.613x | 4.61 | 0.0665% | 921,415 |
96
+ | **64k** | 4.848x 🏆 | 4.85 | 0.0699% | 876,870 |
97
 
98
  ### Tokenization Examples
99
 
100
  Below are sample sentences tokenized with each vocabulary size:
101
 
102
+ **Sample 1:** `မျန်မာခမ်းထီကိုယို တွိုင်ꩻဒေႏသတန် အဝ်ႏ ( )တွိုင်ꩻ နဝ်ꩻသွူ ။`
103
 
104
  | Vocab | Tokens | Count |
105
  |-------|--------|-------|
106
+ | 8k | `▁မျန်မာခမ်းထီ ကိုယို ▁တွိုင်ꩻဒေႏသတန် ▁အဝ်ႏ ▁( ▁၇ ▁) တွိုင်ꩻ ▁နဝ်ꩻ သွူ ... (+1 more)` | 11 |
107
+ | 16k | `▁မျန်မာခမ်းထီ ကိုယို ▁တွိုင်ꩻဒေႏသတန် ▁အဝ်ႏ ▁( ▁၇ ▁) တွိုင်ꩻ ▁နဝ်ꩻသွူ ▁။` | 10 |
108
+ | 32k | `▁မျန်မာခမ်းထီ ကိုယို ▁တွိုင်ꩻဒေႏသတန် ▁အဝ်ႏ ▁( ▁၇ ▁) တွိုင်ꩻ ▁နဝ်ꩻသွူ ▁။` | 10 |
109
+ | 64k | `▁မျန်မာခမ်းထီ ကိုယို ▁တွိုင်ꩻဒေႏသတန် ▁အဝ်ႏ ▁( ▁၇ ▁) တွိုင်ꩻ ▁နဝ်ꩻသွူ ▁။` | 10 |
110
 
111
+ **Sample 2:** `ဝေင်ꩻနောင်ꩻတရားယိုနဝ်ꩻ အဝ်ႏဒျာႏ မျန်မာခမ်းထီ ဖြဝ်ꩻခမ်းနယ်ႏအခဝ်နဝ်၊ တောင်ႏကီꩻခရ...`
112
 
113
  | Vocab | Tokens | Count |
114
  |-------|--------|-------|
115
+ | 8k | `▁ဝေင်ꩻန ောင်ꩻ တရား ယိုနဝ်ꩻ ▁အဝ်ႏဒျာႏ ▁မျန်မာခမ်းထီ ▁၊ ▁ဖြဝ်ꩻခမ်းနယ်ႏ အခဝ်နဝ်၊ ▁တောင်ႏကီꩻခရဲင်ႏ ... (+8 more)` | 18 |
116
+ | 16k | `▁ဝေင်ꩻန ောင်ꩻ တရား ယိုနဝ်ꩻ ▁အဝ်ႏဒျာႏ ▁မျန်မာခမ်းထီ ▁၊ ▁ဖြဝ်ꩻခမ်းနယ်ႏ အခဝ်နဝ်၊ ▁တောင်ႏကီꩻခရဲင်ႏ ... (+8 more)` | 18 |
117
+ | 32k | `▁ဝေင်ꩻန ောင်ꩻ တရား ယိုနဝ်ꩻ ▁အဝ်ႏဒျာႏ ▁မျန်မာခမ်းထီ ▁၊ ▁ဖြဝ်ꩻခမ်းနယ်ႏ အခဝ်နဝ်၊ ▁တောင်ႏကီꩻခရဲင်ႏ ... (+8 more)` | 18 |
118
+ | 64k | `▁ဝေင်ꩻနောင်ꩻ တရားယိုနဝ်ꩻ ▁အဝ်ႏဒျာႏ ▁မျန်မာခမ်းထီ ▁၊ ▁ဖြဝ်ꩻခမ်းနယ်ႏ အခဝ်နဝ်၊ ▁တောင်ႏကီꩻခရဲင်ႏ ▁၊ ▁ဝေင်ꩻနယ်ႏပ ... (+6 more)` | 16 |
119
 
120
+ **Sample 3:** `အမုဲင် ခမ်းထီ ကသှိုပ်စဒါႏ ငဝ်းလဝ်းနီꩻ ၃၅လာအို ၉၄ ထူႏတောမ်`
121
 
122
  | Vocab | Tokens | Count |
123
  |-------|--------|-------|
124
+ | 8k | `▁အမုဲင် ▁ခမ်းထီ ▁က ှို ပ် စဒါႏ ▁ငဝ်း ဝ်း ... (+6 more)` | 16 |
125
+ | 16k | `▁အမုဲင် ▁ခမ်းထီ ▁ကသ ှိုပ် စဒါႏ ▁ငဝ်း လဝ်း နီꩻ ▁၃၅ လာအို ... (+3 more)` | 13 |
126
+ | 32k | `▁အမုဲင် ▁ခမ်းထီ ▁ကသှိုပ်စဒါႏ ▁ငဝ်းလဝ်းနီꩻ ▁၃၅လာအို ▁၉ ▁ထူႏတောမ်` | 8 |
127
+ | 64k | `▁အမုဲင် ▁ခမ်းထီ ▁ကသှိုပ်စဒါႏ ▁ငဝ်းလဝ်းနီꩻ ▁၃၅လာအို ▁၉၄ ▁ထူႏတောမ်` | 7 |
128
 
129
 
130
  ### Key Findings
131
 
132
+ - **Best Compression:** 64k achieves 4.848x compression
133
+ - **Lowest UNK Rate:** 8k with 0.0580% 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 | 2,539 | 11.31 | 4,306 | 21.2% | 57.9% |
151
+ | **2-gram** | Subword | 1,398 🏆 | 10.45 | 24,285 | 42.8% | 77.0% |
152
+ | **3-gram** | Word | 3,862 | 11.92 | 6,537 | 18.8% | 47.3% |
153
+ | **3-gram** | Subword | 11,299 | 13.46 | 129,572 | 19.0% | 45.1% |
154
+ | **4-gram** | Word | 16,871 | 14.04 | 23,296 | 9.0% | 22.0% |
155
+ | **4-gram** | Subword | 54,089 | 15.72 | 405,489 | 10.1% | 25.8% |
156
+ | **5-gram** | Word | 15,317 | 13.90 | 19,946 | 8.7% | 21.0% |
157
+ | **5-gram** | Subword | 138,288 | 17.08 | 617,898 | 5.8% | 16.6% |
158
 
159
  ### Top 5 N-grams by Size
160
 
 
162
 
163
  | Rank | N-gram | Count |
164
  |------|--------|-------|
165
+ | 1 | `နဝ်ꩻ အဝ်ႏဒျာႏ` | 719 |
166
  | 2 | `အဝ်ႏဒျာႏ မျန်မာခမ်းထီ` | 691 |
167
+ | 3 | `ခရိစ်နေင်ႏ ဗာႏ` | 403 |
168
  | 4 | `ဗာႏ စာႏရင်ꩻအလꩻ` | 320 |
169
  | 5 | `မျန်မာခမ်းထီ အခဝ်ထာႏဝ` | 295 |
170
 
 
184
  |------|--------|-------|
185
  | 1 | `နဝ်ꩻ အဝ်ႏဒျာႏ မျန်မာခမ်းထီ အခဝ်ထာႏဝ` | 282 |
186
  | 2 | `ခရိစ်နေင်ႏ ဗာႏ စာႏရင်ꩻအလꩻ ဝေင်ꩻကိုနဝ်ꩻ` | 161 |
187
+ | 3 | `လွူးဖွာꩻသားဖုံႏ သီမားသားဖုံႏ မွူးနီꩻအုံပဆားနီꩻဖုံႏတောမ်ႏ အထွတ်အမျတ်မွူးနီꩻဖုံႏ` | 153 |
188
+ | 4 | `သီမားသားဖုံႏ မွူးနီꩻအုံပဆားနီꩻဖုံႏတောမ်ႏ အထွတ်အမျတ်မွူးနီꩻဖုံႏ အာႏကွိုꩻ` | 153 |
189
  | 5 | `ထာꩻထွာဖုံႏ လွူးဖွာꩻသားဖုံႏ သီမားသားဖုံႏ မွူးနီꩻအုံပဆားနီꩻဖုံႏတောမ်ႏ` | 153 |
190
 
191
+ **5-grams (Word):**
192
+
193
+ | Rank | N-gram | Count |
194
+ |------|--------|-------|
195
+ | 1 | `လွူးဖွာꩻသားဖုံႏ သီမားသားဖုံႏ မွူးနီꩻအုံပဆားနီꩻဖုံႏတောမ်ႏ အထွတ်အမျတ်မွူးနီꩻဖုံႏ အာႏကွိုꩻ` | 153 |
196
+ | 2 | `ထာꩻထွာဖုံႏ လွူးဖွာꩻသားဖုံႏ သီမားသားဖုံႏ မွူးနီꩻအုံပဆားနီꩻဖုံႏတောမ်ႏ အထွတ်အမျတ်မွူးနီꩻဖုံႏ` | 153 |
197
+ | 3 | `သွူ ထာꩻထွာဖုံႏ လွူးဖွာꩻသားဖုံႏ သီမားသားဖုံႏ မွူးနီꩻအုံပဆားနီꩻဖုံႏတောမ်ႏ` | 151 |
198
+ | 4 | `ခရိစ်နေင်ႏ ဗာႏ စာႏရင်ꩻအလꩻ ဝေင်ꩻကိုနဝ်ꩻ လိုꩻဖြာꩻခြွဉ်းအဝ်ႏ` | 131 |
199
+ | 5 | `အဝ်ႏသော့ꩻနဝ်ꩻသွူ ခရိစ်နေင်ႏ ဗာႏ စာႏရင်ꩻအလꩻ ဝေင်ꩻကိုနဝ်ꩻ` | 111 |
200
+
201
  **2-grams (Subword):**
202
 
203
  | Rank | N-gram | Count |
204
  |------|--------|-------|
205
+ | 1 | `ာ ႏ` | 142,384 |
206
+ | 2 | `၊ _` | 135,380 |
207
+ | 3 | `ꩻ _` | 126,353 |
208
+ | 4 | `ဝ် ꩻ` | 102,695 |
209
+ | 5 | `င် ꩻ` | 96,805 |
210
 
211
  **3-grams (Subword):**
212
 
213
  | Rank | N-gram | Count |
214
  |------|--------|-------|
215
+ | 1 | `န ဝ် ꩻ` | 77,014 |
216
+ | 2 | `ဝ် ꩻ _` | 57,567 |
217
+ | 3 | `ꩻ ၊ _` | 31,811 |
218
+ | 4 | `သွူ ။ _` | 31,570 |
219
+ | 5 | `ႏ ၊ _` | 30,928 |
220
 
221
  **4-grams (Subword):**
222
 
223
  | Rank | N-gram | Count |
224
  |------|--------|-------|
225
+ | 1 | `န ဝ် ꩻ _` | 45,450 |
226
+ | 2 | `နေ ာ ဝ် ꩻ` | 23,553 |
227
+ | 3 | `ꩻ သွူ ။ _` | 18,993 |
228
+ | 4 | `ꩻ န ဝ် ꩻ` | 18,023 |
229
+ | 5 | `ႏ န ဝ် ꩻ` | 17,057 |
230
+
231
+ **5-grams (Subword):**
232
+
233
+ | Rank | N-gram | Count |
234
+ |------|--------|-------|
235
+ | 1 | `ဝ် ꩻ သွူ ။ _` | 15,761 |
236
+ | 2 | `ꩻ န ဝ် ꩻ _` | 12,522 |
237
+ | 3 | `နေ ာ ဝ် ꩻ _` | 11,865 |
238
+ | 4 | `ႏ န ဝ် ꩻ _` | 10,503 |
239
+ | 5 | `န ဝ် ꩻ သွူ ။` | 10,311 |
240
 
241
 
242
  ### Key Findings
243
 
244
+ - **Best Perplexity:** 2-gram (subword) with 1,398
245
  - **Entropy Trend:** Decreases with larger n-grams (more predictable)
246
+ - **Coverage:** Top-1000 patterns cover ~17% 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 | 0.2308 | 1.173 | 1.60 | 381,069 | 76.9% |
263
+ | **1** | Subword | 1.2202 | 2.330 | 20.98 | 2,909 | 0.0% |
264
+ | **2** | Word | 0.0412 | 1.029 | 1.06 | 609,269 | 95.9% |
265
+ | **2** | Subword | 0.7534 | 1.686 | 5.49 | 61,020 | 24.7% |
266
+ | **3** | Word | 0.0155 | 1.011 | 1.02 | 645,305 | 98.5% |
267
+ | **3** | Subword | 0.4733 | 1.388 | 2.77 | 335,231 | 52.7% |
268
+ | **4** | Word | 0.0088 🏆 | 1.006 | 1.01 | 656,933 | 99.1% |
269
+ | **4** | Subword | 0.3156 | 1.245 | 1.90 | 930,014 | 68.4% |
270
 
271
  ### Generated Text Samples (Word-based)
272
 
 
274
 
275
  **Context Size 1:**
276
 
277
+ 1. `၂ ဖြုံႏလဲ့ အဝ်ႏသွူ ခမ်းတွူးကောင်ꩻယို အမိဉ်ꩻနဝ်ꩻ ဖန်ဖေႏ စဲဉ်ႏဖေႏဒျာႏလွဉ်းလွဉ်းသွူ ယိုလွုမ်ꩻမကာႏ ဗွေႏဗ...`
278
+ 2. `၃ ပွုမ်ႏယိုသွူ က အဟံ ခွေနဝ်ꩻ ကောလက္ခံႏသား ၂ ၃ ပေါႏပါႏဠိဒျာႏနဝ်ꩻ သော့ꩻတောဝ်းအမုဲင် ဟော်ꩻဖတ်ဗော့ꩻ ပါႏဠ...`
279
+ 3. `၁ ခြပ် စီ သွံဆီသူ တနတ်တလီꩻ air combat information management unit mimu ဝေင်ꩻနယ်ႏရွုမ်ꩻဖုံႏနဝ်ꩻ အဝ်ႏဒ...`
280
 
281
  **Context Size 2:**
282
 
283
+ 1. `နဝ်ꩻ အဝ်ႏဒျာႏ မျန်မာခမ်းထီ ဖြဝ်ꩻခမ်းနယ်ႏအခဝ်ကွဉ်ႏ မွိုင်ꩻတုံခရဲင်ႏ ဝေင်ꩻနယ်ႏမွိုင်ꩻတုံကို ကပါဒါႏ ဝေင...`
284
+ 2. `အဝ်ႏဒျာႏ မျန်မာခမ်းထီ အခဝ်ထာႏဝ ပဂိုꩻတွိုင်ꩻဒေႏသတန် အခဝ်ကွဉ်ႏထင်ꩻ တောင်ႏအူခရဲင်ႏ ဝေင်ꩻနယ်ႏဖျူးကို ကပါ...`
285
+ 3. `ခရိစ်နေင်ႏ ဗာႏ စဲ့ꩻအစိုႏရစိုးကို ကဗွောင်လွေꩻဒါႏ ခမ်းလင်လစ်ꩻခမ်းတောမ်ႏ ဖြေꩻစာကွန်ႏ လွယ်စယ်ခမ်းကူဂဲတ်လ...`
286
 
287
  **Context Size 3:**
288
 
289
+ 1. `နဝ်ꩻ အဝ်ႏဒျာႏ မျန်မာခမ်းထီ အခဝ်ကွဉ်ႏထင်ꩻ ဖြဝ်ꩻခမ်းနယ်ႏ အခဝ်ထင်ꩻ ဟိုပန်ခရဲင်ႏ ဝနမ်းပဲင်ႏအိုပ်ချုတ်ခွင...`
290
+ 2. `အဝ်ႏဒျာႏ မျန်မာခမ်းထီ အခဝ်ထာႏဝ နေပီဒေါ်ခမ်းခြွဉ်းဗူႏဟံႏနယ်ႏ လယ်ဝွေးခရဲင်ႏကို ကအဝ်ႏပါသော့ꩻဒါႏ ဝေင်ꩻနယ...`
291
+ 3. `ခရိစ်နေင်ႏ ဗာႏ စာႏရင်ꩻအလꩻ ဝေင်ꩻကိုနဝ်ꩻ လိုꩻဖြာꩻအဝ်ႏ ၁၁ ၃၀၅ ဖြာꩻသွူ အဝ်ႏဒျာႏ ထာဝယ် မေက် ကာꩻတဖူꩻတန်လော...`
292
 
293
  **Context Size 4:**
294
 
295
+ 1. `နဝ်ꩻ အဝ်ႏဒျာႏ မျန်မာခမ်းထီ အခဝ်ထာႏဝ ဧရာႏဝတီႏတွိုင်ꩻဒေႏသတန် မအူပိဉ်ခရဲင်ႏ ကို ကပါဒါႏ ဝေင်ꩻနယ်ႏတဖြုံႏဒ...`
296
+ 2. `ခရိစ်နေင်ႏ ဗာႏ စာႏရင်ꩻအလꩻ ဝေင်ꩻကိုနဝ်ꩻ လိုꩻဖြာꩻအဝ်ႏ ၁၀ ၄၄၃ ဖြာꩻသွူ အဉ်းမယို ကရီးခါနဝ်ꩻ ထွာဒျာႏ ဒုံအဉ...`
297
+ 3. `ထာꩻထွာဖုံႏ လွူးဖွာꩻသားဖုံႏ သီမားသားဖုံႏ မွူးနီꩻအုံပဆားနီꩻဖုံႏတောမ်ႏ အထွတ်အမျတ်မွူးနီꩻဖုံႏ အာႏကွိုꩻ`
298
 
299
 
300
  ### Generated Text Samples (Subword-based)
 
303
 
304
  **Context Size 1:**
305
 
306
+ 1. `_တသီႏပေႏစမုံးဝါꩻစွဉ်းထဲ`
307
+ 2. `ꩻနဝ်ႏပုဂ္ဂိုလ်ႏ_ဇာဝ်ꩻသား`
308
+ 3. `ႏအခြာဏ်ႏတောယိုတဲ့_ဟောႏရာ`
309
 
310
  **Context Size 2:**
311
 
312
+ 1. `ာႏနဝ်ꩻ_အောဝ်ꩻသွူကျောင်ႏဒျာ`
313
+ 2. `၊_တွိုက်_ကြွဲႏ_ဖန်_သွူ။_ဓမ္မပ`
314
+ 3. `ꩻ_ခင်ႏငံႏ_မန်း"ကို_ကတဲမ်`
315
 
316
  **Context Size 3:**
317
 
318
+ 1. `နဝ်ꩻသွူ။_နီကွဉ်ကꩻမွိုန်း။_၂။`
319
+ 2. `ဝ်ꩻ_အံႏဖြာꩻနောဝ်ꩻ_ပအိုဝ်ႏယို`
320
+ 3. `ꩻ၊_မဲ့သျင်ႏကျင်ꩻ။_နီလိတ်_အဝ်`
321
 
322
  **Context Size 4:**
323
 
324
+ 1. `နဝ်ꩻ_ဟဲ့ꩻဗာႏသꩻ_ကွား_ကွန်ပေ`
325
+ 2. `နောဝ်ꩻ_ဘဝပေါင်ꩻ_ရွဉ်ခန်ဗီႏ_`
326
+ 3. `ꩻသွူ။_ပအိုဝ်ႏစွိုးခွိုꩻသီး_သွိုန်ႏသ`
327
 
328
 
329
  ### Key Findings
330
 
331
  - **Best Predictability:** Context-4 (word) with 99.1% predictability
332
  - **Branching Factor:** Decreases with context size (more deterministic)
333
+ - **Memory Trade-off:** Larger contexts require more storage (930,014 contexts)
334
  - **Recommendation:** Context-3 or Context-4 for text generation
335
 
336
  ---
 
346
 
347
  | Metric | Value |
348
  |--------|-------|
349
+ | Vocabulary Size | 67,819 |
350
+ | Total Tokens | 396,228 |
351
+ | Mean Frequency | 5.84 |
352
  | Median Frequency | 2 |
353
+ | Frequency Std Dev | 39.85 |
354
 
355
  ### Most Common Words
356
 
357
  | Rank | Word | Frequency |
358
  |------|------|-----------|
359
+ | 1 | ၂ | 3,796 |
360
+ | 2 | ၃ | 3,380 |
361
+ | 3 | ၁ | 3,330 |
362
  | 4 | အာႏကွိုꩻ | 3,141 |
363
+ | 5 | နဝ်ꩻ | 2,717 |
364
+ | 6 | ၄ | 2,608 |
365
+ | 7 | ၅ | 2,058 |
366
+ | 8 | ထွာဒျာႏ | 1,623 |
367
+ | 9 | ၆ | 1,585 |
368
+ | 10 | အဝ်ႏဒျာႏ | 1,494 |
369
 
370
  ### Least Common Words (from vocabulary)
371
 
372
  | Rank | Word | Frequency |
373
  |------|------|-----------|
374
+ | 1 | တထာနမ်းနောဝ်ꩻ | 2 |
375
+ | 2 | တထာဖြွီꩻဖုံႏ | 2 |
376
+ | 3 | antihistamine | 2 |
377
+ | 4 | ပထမခွိုꩻ | 2 |
378
+ | 5 | ဒုတိယခွိုꩻ | 2 |
379
+ | 6 | histamine | 2 |
380
+ | 7 | တနယ်ႏလိုမ်းဆဲင်ႏရာꩻ | 2 |
381
+ | 8 | အခြေပြုမူလတန်ꩻ | 2 |
382
+ | 9 | ပထမကြီးတန်ꩻတွမ်ႏ | 2 |
383
+ | 10 | ရန်ႏကုန်ႏတုံး | 2 |
384
 
385
  ### Zipf's Law Analysis
386
 
387
  | Metric | Value |
388
  |--------|-------|
389
+ | Zipf Coefficient | 0.7916 |
390
+ | R² (Goodness of Fit) | 0.998007 |
391
  | Adherence Quality | **excellent** |
392
 
393
  ### Coverage Analysis
394
 
395
  | Top N Words | Coverage |
396
  |-------------|----------|
397
+ | Top 100 | 17.9% |
398
  | Top 1,000 | 34.4% |
399
  | Top 5,000 | 51.9% |
400
  | Top 10,000 | 61.5% |
 
402
  ### Key Findings
403
 
404
  - **Zipf Compliance:** R²=0.9980 indicates excellent adherence to Zipf's law
405
+ - **High Frequency Dominance:** Top 100 words cover 17.9% of corpus
406
+ - **Long Tail:** 57,819 words needed for remaining 38.5% coverage
407
 
408
  ---
409
  ## 5. Word Embeddings Evaluation
 
419
 
420
  ### 5.1 Cross-Lingual Alignment
421
 
422
+ ![Alignment Quality](visualizations/embedding_alignment_quality.png)
423
+
424
+ ![Multilingual t-SNE](visualizations/embedding_tsne_multilingual.png)
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 | 0.8632 🏆 | 0.3270 | N/A | N/A |
432
+ | **mono_64d** | 64 | 0.8595 | 0.2722 | N/A | N/A |
433
+ | **mono_128d** | 128 | 0.6854 | 0.2261 | N/A | N/A |
434
+ | **aligned_32d** | 32 | 0.8632 | 0.3317 | 0.0135 | 0.1716 |
435
+ | **aligned_64d** | 64 | 0.8595 | 0.2717 | 0.0745 | 0.2844 |
436
+ | **aligned_128d** | 128 | 0.6854 | 0.2281 | 0.1625 | 0.3386 |
437
 
438
  ### Key Findings
439
 
440
+ - **Best Isotropy:** mono_32d with 0.8632 (more uniform distribution)
441
+ - **Semantic Density:** Average pairwise similarity of 0.2762. Lower values indicate better semantic separation.
442
+ - **Alignment Quality:** Aligned models achieve up to 16.3% 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 | **0.267** | High formulaic/idiomatic content | - |
456
 
457
  ### 6.2 Affix Inventory (Productive Units)
458
 
 
461
  #### Productive Prefixes
462
  | Prefix | Examples |
463
  |--------|----------|
464
+ | `-လိ` | လိက်ပအိုဝ်ႏ, လိုꩻစားယိုဖုံႏနဝ်ꩻ, လိတ်လုံးကို |
465
+ | `-လို` | လိုꩻစားယိုဖုံႏနဝ်ꩻ, လိုꩻခိုဖုံႏလဲ့, လိုꩻမွိုက်နဝ်ꩻ |
 
466
 
467
  #### Productive Suffixes
468
  | Suffix | Examples |
469
  |--------|----------|
470
+ | `-ꩻ` | ဝန်ႏကိုတွော့ꩻ, ရခဲင်ႏခွန်ဟော်ခံꩻ, သဗ္ဗညုဘုရာꩻ |
471
+ | `-ႏ` | လဲဉ်အံႏ, အလင်္ကာႏ, မာꩻသွော့ကုသိုလ်ႏ |
472
+ | `-်ꩻ` | လာအိုခမ်းထီနဝ်ꩻ, တသီႏအံႏနယ်ꩻနဝ်ꩻ, ယိုသွံပါꩻထွာနဝ်ꩻ |
473
+ | `-း` | အောဝ်ႏဟမ်ႏခမ်ႏဖာႏလောင်း, လူထုအလောင်း, ဉာဏ်ႏတောႏဆꩻချာလွဉ်းလွဉ်း |
474
+ | `-ဝ်ꩻ` | လာအိုခမ်းထီနဝ်ꩻ, တသီႏအံႏနယ်ꩻနဝ်ꩻ, ယိုသွံပါꩻထွာနဝ်ꩻ |
475
+ | `-်း` | အောဝ်ႏဟမ်ႏခမ်ႏဖာႏလောင်း, လူထုအလောင်း, ဉာဏ်ႏတောႏဆꩻချာလွဉ်းလွဉ်း |
476
+ | `-နဝ်ꩻ` | လာအိုခမ်းထီနဝ်ꩻ, တသီႏအံႏနယ်ꩻနဝ်ꩻ, ယိုသွံပါꩻထွာနဝ်ꩻ |
477
+ | `-ာႏ` | အလင်္ကာႏ, ဖန်ဆင်ꩻမာꩻခါꩻဒျာႏ, ကိုꩻကွယ်ႏသားအာဗာႏ |
478
 
479
  ### 6.3 Bound Stems (Lexical Roots)
480
 
 
489
 
490
  | Prefix | Suffix | Frequency | Examples |
491
  |--------|--------|-----------|----------|
492
+ | `-လိ` | `-ꩻ` | 83 words | လိုꩻမဉ်ꩻ, လိုꩻနမ်းအကိုအထန်ႏနီဖဲ့ꩻ |
493
+ | `-လိ` | `-ႏ` | 64 words | လိုꩻစွဲဉ်ႏ, လိုꩻမုရေꩻအစွိုꩻအဗူႏဖုံႏ |
494
+ | `-လိ` | `-်ꩻ` | 61 words | လိုꩻမဉ်ꩻ, လိုꩻယုက်နဝ်ꩻ |
495
+ | `-လိ` | `-ဝ်ꩻ` | 45 words | လိုꩻယုက်နဝ်ꩻ, လိတ်မွူးပအိုဝ်ႏယိုခါနဝ်ꩻ |
496
+ | `-လိ` | `-နဝ်ꩻ` | 37 words | လိုꩻယုက်နဝ်ꩻ, လိတ်မွူးပအိုဝ်ႏယိုခါနဝ်ꩻ |
497
+ | `-လိ` | `-း` | 36 words | လိုꩻခမ်း, လိုႏတဝ်း |
498
+ | `-လိ` | `-်ႏ` | 23 words | လိုꩻစွဲဉ်ႏ, လိုꩻသွုန်ႏထီဓာတ်တွမ်ႏ |
499
+ | `-လိ` | `-်း` | 19 words | လိုꩻခမ်း, လိုႏတဝ်း |
500
+ | `-လိ` | `-ာႏ` | 15 words | လိုꩻမျိုꩻတွမ်ႏခမ်းထီအတာႏ, လိတ်လုဲင်ꩻတွမ်ႏအနုပညာႏ |
501
+ | `-လိ` | `-ွူ` | 5 words | လိုꩻမဉ်အံႏနွောင်ꩻနိစ်စက်ဒါႏဝင်ꩻဖုံႏနဝ်ꩻသွူ, လိုႏမာꩻထူႏလွလဲဉ်းဒျာႏနောဝ်ꩻသွူ |
502
 
503
  ### 6.5 Recursive Morpheme Segmentation
504
 
 
506
 
507
  | Word | Suggested Split | Confidence | Stem |
508
  |------|-----------------|------------|------|
509
+ | ကွဲညညနဝ်ꩻ | **`ကွဲညည-နဝ်ꩻ`** | 4.5 | `က���ဲညည` |
510
+ | သꩻကိုနဝ်ꩻ | **`သꩻကို-နဝ်ꩻ`** | 4.5 | `သꩻကို` |
511
+ | လိုꩻယင်ဟန်ႏနဝ်ꩻ | **`လို-ꩻယင်ဟန-်ႏ-နဝ်ꩻ`** | 4.5 | `ꩻယင်ဟန` |
512
+ | နင်ꩻသုမနာနဝ်ꩻ | **`နင်ꩻသုမနာ-နဝ်ꩻ`** | 4.5 | `နင်ꩻသုမနာ` |
513
+ | ပုဏ္ဏာꩻနဝ်ꩻ | **`ပုဏ္ဏာꩻ-နဝ်ꩻ`** | 4.5 | `ပုဏ္ဏာꩻ` |
514
+ | နာꩻတဲ့နဝ်ꩻ | **`နာꩻတဲ့-နဝ်ꩻ`** | 4.5 | `နာꩻတဲ့` |
515
+ | ခန္ဓာႏတန်ယိုနဝ်ꩻ | **`ခန္ဓာႏတန်ယို-နဝ်ꩻ`** | 4.5 | `ခန္ဓာႏတန်ယို` |
516
+ | ရောင်ထာꩻနဝ်ꩻ | **`ရောင်ထာꩻ-နဝ်ꩻ`** | 4.5 | `ရောင်ထာꩻ` |
517
+ | ခယ်ႏမူႏနဝ်ꩻ | **`ခယ်ႏမူႏ-နဝ်ꩻ`** | 4.5 | `ခယ်ႏမူႏ` |
518
+ | အနာႏဂတ်နဝ်ꩻ | **`အနာႏဂတ်-နဝ်ꩻ`** | 4.5 | `အနာႏဂတ်` |
519
+ | ရဟန်ꩻသာႏမဏေႏနဝ်ꩻ | **`ရဟန်ꩻသာႏမဏေႏ-နဝ်ꩻ`** | 4.5 | `ရဟန်ꩻသာႏမဏေႏ` |
520
+ | ထွို့ꩻစွဲႏနဝ်ꩻ | **`ထွို့ꩻစွဲႏ-နဝ်ꩻ`** | 4.5 | `ထွို့ꩻစွဲႏ` |
521
+ | စူမွူးနဝ်ꩻ | **`စူမွူး-နဝ်ꩻ`** | 4.5 | `စူမွူး` |
522
+ | ပွိုးနဝ်ꩻ | **`ပွိုး-နဝ်ꩻ`** | 4.5 | `ပွိုး` |
523
+ | သင်္ဃာႏတောႏနဝ်ꩻ | **`သင်္ဃာႏတေ-ာႏ-နဝ်ꩻ`** | 3.0 | `သင်္ဃာႏတေ` |
524
 
525
  ### 6.6 Linguistic Interpretation
526
 
527
  > **Automated Insight:**
528
+ The language Pa'o Karen shows high morphological productivity. The subword models are significantly more efficient than word models, suggesting a rich system of affixation or compounding.
529
+
530
+ > **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.
531
 
532
  ---
533
  ## 7. Summary & Recommendations
 
539
  | Component | Recommended | Rationale |
540
  |-----------|-------------|-----------|
541
  | Tokenizer | **64k BPE** | Best compression (4.85x) |
542
+ | N-gram | **2-gram** | Lowest perplexity (1,398) |
543
  | Markov | **Context-4** | Highest predictability (99.1%) |
544
  | Embeddings | **100d** | Balanced semantic capture and isotropy |
545
 
 
754
  ---
755
  *Generated by Wikilangs Models Pipeline*
756
 
757
+ *Report Date: 2026-01-03 19:13:44*
models/embeddings/aligned/blk_128d.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:403700a4e9862340aaa7a24e865cde86065373f52217722bf81e7cfa17087900
3
+ size 1035782926
models/embeddings/aligned/blk_128d.meta.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"lang": "blk", "dim": 128, "max_seq_len": 512, "is_aligned": true}
models/embeddings/aligned/blk_128d.projection.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2788b6ffae44eec975960edafea55b75c0d04f5dc4c549d8dd1ee9e6e68a6082
3
+ size 65664
models/embeddings/aligned/blk_128d_metadata.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "language": "blk",
3
+ "dimension": 128,
4
+ "version": "aligned",
5
+ "hub_language": "en",
6
+ "seed_vocab_size": 443,
7
+ "vocab_size": 11047
8
+ }
models/embeddings/aligned/blk_32d.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c4dfad7bd3baa1b474eab1507ac372f77e770d7754062f51acd866cdaa048fa
3
+ size 259298830
models/embeddings/aligned/blk_32d.meta.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"lang": "blk", "dim": 32, "max_seq_len": 512, "is_aligned": true}
models/embeddings/aligned/blk_32d.projection.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7d02f3aa633829607bafb710514b3bdef5e180bb6d6c5e529057276d14df54a1
3
+ size 4224
models/embeddings/aligned/blk_32d_metadata.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "language": "blk",
3
+ "dimension": 32,
4
+ "version": "aligned",
5
+ "hub_language": "en",
6
+ "seed_vocab_size": 443,
7
+ "vocab_size": 11047
8
+ }
models/embeddings/aligned/blk_64d.bin ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0fa083255cd98cbc2b60e9b5a608ab16bbf3ed3c3691498c7c74c9b713e1a565
3
+ size 518126862
models/embeddings/aligned/blk_64d.meta.json ADDED
@@ -0,0 +1 @@
 
 
1
+ {"lang": "blk", "dim": 64, "max_seq_len": 512, "is_aligned": true}
models/embeddings/aligned/blk_64d.projection.npy ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1c54425017a59dd8729ea29c9822d943c1550022b477b13069132755a49f81f6
3
+ size 16512
models/embeddings/aligned/blk_64d_metadata.json ADDED
@@ -0,0 +1,8 @@
 
 
 
 
 
 
 
 
 
1
+ {
2
+ "language": "blk",
3
+ "dimension": 64,
4
+ "version": "aligned",
5
+ "hub_language": "en",
6
+ "seed_vocab_size": 443,
7
+ "vocab_size": 11047
8
+ }
models/embeddings/monolingual/blk_128d.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:849c0bf818b84aff6ad56616057a43cf1087071bb06730ea49f4f4fd35fc3106
3
- size 1035828932
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:403700a4e9862340aaa7a24e865cde86065373f52217722bf81e7cfa17087900
3
+ size 1035782926
models/embeddings/monolingual/blk_128d_metadata.json CHANGED
@@ -11,5 +11,5 @@
11
  "encoding_method": "rope",
12
  "dim": 128
13
  },
14
- "vocab_size": 11091
15
  }
 
11
  "encoding_method": "rope",
12
  "dim": 128
13
  },
14
+ "vocab_size": 11047
15
  }
models/embeddings/monolingual/blk_32d.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:de9eeb9d196eece4ed3965d511a1314e0f2cf59efc97054b08865446875f3bbd
3
- size 259311044
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:3c4dfad7bd3baa1b474eab1507ac372f77e770d7754062f51acd866cdaa048fa
3
+ size 259298830
models/embeddings/monolingual/blk_32d_metadata.json CHANGED
@@ -11,5 +11,5 @@
11
  "encoding_method": "rope",
12
  "dim": 32
13
  },
14
- "vocab_size": 11091
15
  }
 
11
  "encoding_method": "rope",
12
  "dim": 32
13
  },
14
+ "vocab_size": 11047
15
  }
models/embeddings/monolingual/blk_64d.bin CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:27b17305d78322b33cc4baaad0b9e0e0fdb42f39380214b5f5965697b4cad3ae
3
- size 518150340
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0fa083255cd98cbc2b60e9b5a608ab16bbf3ed3c3691498c7c74c9b713e1a565
3
+ size 518126862
models/embeddings/monolingual/blk_64d_metadata.json CHANGED
@@ -11,5 +11,5 @@
11
  "encoding_method": "rope",
12
  "dim": 64
13
  },
14
- "vocab_size": 11091
15
  }
 
11
  "encoding_method": "rope",
12
  "dim": 64
13
  },
14
+ "vocab_size": 11047
15
  }
models/subword_markov/blk_markov_ctx1_subword.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f3b2f1848d241b73c04054b4240c8432860a0e2355d8a3823c68d65c2387e9f3
3
- size 370202
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:8408df6165f248626fd4210eb323ab4d4b42675200c9aa45aa1c12b12a79d849
3
+ size 369981
models/subword_markov/blk_markov_ctx1_subword_metadata.json CHANGED
@@ -2,6 +2,6 @@
2
  "context_size": 1,
3
  "variant": "subword",
4
  "language": "blk",
5
- "unique_contexts": 2910,
6
- "total_transitions": 6938576
7
  }
 
2
  "context_size": 1,
3
  "variant": "subword",
4
  "language": "blk",
5
+ "unique_contexts": 2909,
6
+ "total_transitions": 6915482
7
  }
models/subword_markov/blk_markov_ctx2_subword.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:8d89d3ad1a647826ea8221d87d4f188b1de1a7d698456940763c0aed02388fdf
3
- size 2224824
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a5e324d6710acdab35813a0690df45b4bb82a1f338fefeb1dfb302cb862feffb
3
+ size 2207526
models/subword_markov/blk_markov_ctx2_subword_metadata.json CHANGED
@@ -2,6 +2,6 @@
2
  "context_size": 2,
3
  "variant": "subword",
4
  "language": "blk",
5
- "unique_contexts": 61297,
6
- "total_transitions": 6935314
7
  }
 
2
  "context_size": 2,
3
  "variant": "subword",
4
  "language": "blk",
5
+ "unique_contexts": 61020,
6
+ "total_transitions": 6912217
7
  }
models/subword_markov/blk_markov_ctx3_subword.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:e7a1aadc96aaeb58a6546d15275e93c39e8ee61c836927fd4300c346d76a391b
3
- size 8382547
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:cc2951517ed22f32a80efd5e9e298688441e727a479bf113c80c825cfdecccec
3
+ size 8349591
models/subword_markov/blk_markov_ctx3_subword_metadata.json CHANGED
@@ -2,6 +2,6 @@
2
  "context_size": 3,
3
  "variant": "subword",
4
  "language": "blk",
5
- "unique_contexts": 336631,
6
- "total_transitions": 6932052
7
  }
 
2
  "context_size": 3,
3
  "variant": "subword",
4
  "language": "blk",
5
+ "unique_contexts": 335231,
6
+ "total_transitions": 6908952
7
  }
models/subword_markov/blk_markov_ctx4_subword.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ea3c67743f31a2c851e8f3f3a6592c350caec10a77d9fea4ac375a862f6be6ae
3
- size 20618538
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a88ac23056b5d8ff51043e29ec4fb0dad72fa159e6eb03a7ae5ac83b04eebeca
3
+ size 20532781
models/subword_markov/blk_markov_ctx4_subword_metadata.json CHANGED
@@ -2,6 +2,6 @@
2
  "context_size": 4,
3
  "variant": "subword",
4
  "language": "blk",
5
- "unique_contexts": 934101,
6
- "total_transitions": 6928790
7
  }
 
2
  "context_size": 4,
3
  "variant": "subword",
4
  "language": "blk",
5
+ "unique_contexts": 930014,
6
+ "total_transitions": 6905687
7
  }
models/subword_ngram/blk_2gram_subword.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:aa3fd33203c185a7173b0a19a097c225f98847fe4b5e8d2470493098adeffc4f
3
- size 361985
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ddd57f72086d45f08c72496387c4748bf0eed17d21967adf6aa6cbc8aaa904f6
3
+ size 360618
models/subword_ngram/blk_2gram_subword_metadata.json CHANGED
@@ -2,6 +2,6 @@
2
  "n": 2,
3
  "variant": "subword",
4
  "language": "blk",
5
- "unique_ngrams": 24351,
6
- "total_ngrams": 6938576
7
  }
 
2
  "n": 2,
3
  "variant": "subword",
4
  "language": "blk",
5
+ "unique_ngrams": 24285,
6
+ "total_ngrams": 6915482
7
  }
models/subword_ngram/blk_3gram_subword.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:34a673bb8ff0f8d0ce8a38c20764c99b533e5d3b05e2b0409cec35a4dc63f108
3
- size 1910510
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:7a261c56eaf693966bbbe141ad39a52625a255d363b681f0c73cfbab8fd56fb7
3
+ size 1911098
models/subword_ngram/blk_3gram_subword_metadata.json CHANGED
@@ -2,6 +2,6 @@
2
  "n": 3,
3
  "variant": "subword",
4
  "language": "blk",
5
- "unique_ngrams": 129980,
6
- "total_ngrams": 6935314
7
  }
 
2
  "n": 3,
3
  "variant": "subword",
4
  "language": "blk",
5
+ "unique_ngrams": 129572,
6
+ "total_ngrams": 6912217
7
  }
models/subword_ngram/blk_4gram_subword.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:ad489f9217fc89e1f791b12a3222263cddcb0e0633e8c10a4c0c47aedfdc9570
3
- size 6363654
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:be032f1abb64e62b3c61e15dfdb12233c96f20d0e7cdaf4e3d9963642d73c6a2
3
+ size 6340669
models/subword_ngram/blk_4gram_subword_metadata.json CHANGED
@@ -2,6 +2,6 @@
2
  "n": 4,
3
  "variant": "subword",
4
  "language": "blk",
5
- "unique_ngrams": 407218,
6
- "total_ngrams": 6932052
7
  }
 
2
  "n": 4,
3
  "variant": "subword",
4
  "language": "blk",
5
+ "unique_ngrams": 405489,
6
+ "total_ngrams": 6908952
7
  }
models/subword_ngram/blk_5gram_subword.parquet ADDED
@@ -0,0 +1,3 @@
 
 
 
 
1
+ version https://git-lfs.github.com/spec/v1
2
+ oid sha256:779c130b39e10df04bd2e6b93a828523b14d28ebb8b83b15eec8f0ec66c7c017
3
+ size 10402193
models/subword_ngram/blk_5gram_subword_metadata.json ADDED
@@ -0,0 +1,7 @@
 
 
 
 
 
 
 
 
1
+ {
2
+ "n": 5,
3
+ "variant": "subword",
4
+ "language": "blk",
5
+ "unique_ngrams": 617898,
6
+ "total_ngrams": 6905687
7
+ }
models/tokenizer/blk_tokenizer_16k.model CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:cc0e9e63fac0ed90decd12841053888549e74d29340b045b453d1d22af1d7ddc
3
- size 681362
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:30eb488c8e0f35b91a2678a2a9d2078827de56f06c0d91649abd05d522350e02
3
+ size 682843
models/tokenizer/blk_tokenizer_16k.vocab CHANGED
The diff for this file is too large to render. See raw diff
 
models/tokenizer/blk_tokenizer_32k.model CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:f126ac43be0e6ce337bb595e3f2c105d53ace6fe352dcb131fd8f674b76159cf
3
- size 1258797
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:1988f84f7a26b3f64cce0a6218fee9d6621004dd26480800f9e27ff7cba05a55
3
+ size 1262243
models/tokenizer/blk_tokenizer_32k.vocab CHANGED
The diff for this file is too large to render. See raw diff
 
models/tokenizer/blk_tokenizer_64k.model CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:167575152f3e5342b9a090a90d5e343cfd3cdd0597adfa44421d24c67af92969
3
- size 2501030
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:10e2f514fe94c1a565806a68b496d89bc55964e9c23541ad67bb6c042b1fd56f
3
+ size 2509455
models/tokenizer/blk_tokenizer_64k.vocab CHANGED
The diff for this file is too large to render. See raw diff
 
models/tokenizer/blk_tokenizer_8k.model CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:7b0d3dcf17c7a21318f97a63127193ca8986b12cc4621950f17ca81225afe7fa
3
- size 458144
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:ab465d697427d54b9749f77fdbe4485162e186bd196a241169b52118e40872c7
3
+ size 458800
models/tokenizer/blk_tokenizer_8k.vocab CHANGED
The diff for this file is too large to render. See raw diff
 
models/vocabulary/blk_vocabulary.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:55be2c429edb280fe0b2070ded305af1d3f07559a14b5a9bac8cf0766d0ff793
3
- size 1555201
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:2da353abc73b25a9c1b4a97a0d21ff4dfc3e247ecbc23be8aedcb6270aece732
3
+ size 1545076
models/vocabulary/blk_vocabulary_metadata.json CHANGED
@@ -1,17 +1,17 @@
1
  {
2
  "language": "blk",
3
- "vocabulary_size": 68078,
4
  "variant": "full",
5
  "statistics": {
6
- "type_token_ratio": 0.5362379026940269,
7
  "coverage": {
8
- "top_100": 0.10057494647582223,
9
- "top_1000": 0.19264495020778324,
10
- "top_5000": 0.2903072834971561,
11
- "top_10000": 0.34396334260243955
12
  },
13
- "hapax_count": 314132,
14
- "hapax_ratio": 0.8218832578948746,
15
- "total_documents": 3262
16
  }
17
  }
 
1
  {
2
  "language": "blk",
3
+ "vocabulary_size": 67819,
4
  "variant": "full",
5
  "statistics": {
6
+ "type_token_ratio": 0.5371489778516916,
7
  "coverage": {
8
+ "top_100": 0.10017123890999034,
9
+ "top_1000": 0.1920032133721381,
10
+ "top_5000": 0.28973059820868596,
11
+ "top_10000": 0.3434249191371814
12
  },
13
+ "hapax_count": 313307,
14
+ "hapax_ratio": 0.8220562228764241,
15
+ "total_documents": 3265
16
  }
17
  }
models/word_markov/blk_markov_ctx1_word.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:cdc7fcee57f6aee4ec698d86445b90105b2a0598b31b2c1cfe8e2a3d113bedf3
3
- size 21335834
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:a51f3825d5a92f3a490fc29ce0872293d964a2e1e6d05e98aa89e181cf1b79cf
3
+ size 21280155
models/word_markov/blk_markov_ctx1_word_metadata.json CHANGED
@@ -2,6 +2,6 @@
2
  "context_size": 1,
3
  "variant": "word",
4
  "language": "blk",
5
- "unique_contexts": 382155,
6
- "total_transitions": 709500
7
  }
 
2
  "context_size": 1,
3
  "variant": "word",
4
  "language": "blk",
5
+ "unique_contexts": 381069,
6
+ "total_transitions": 706270
7
  }
models/word_markov/blk_markov_ctx2_word.parquet CHANGED
@@ -1,3 +1,3 @@
1
  version https://git-lfs.github.com/spec/v1
2
- oid sha256:0a68a819a624f91e8cbe47de3432324bf5c1e928d4c5aa0ccbbbf3f6b3680065
3
- size 29817330
 
1
  version https://git-lfs.github.com/spec/v1
2
+ oid sha256:0ee560c8743ec212f1ab8718c192d6d770ae27a5c3bb622e626bd6d8d9558d66
3
+ size 29714039
models/word_markov/blk_markov_ctx2_word_metadata.json CHANGED
@@ -2,6 +2,6 @@
2
  "context_size": 2,
3
  "variant": "word",
4
  "language": "blk",
5
- "unique_contexts": 611948,
6
- "total_transitions": 706238
7
  }
 
2
  "context_size": 2,
3
  "variant": "word",
4
  "language": "blk",
5
+ "unique_contexts": 609269,
6
+ "total_transitions": 703005
7
  }