wikilangs/shn

Shan - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on Shan Wikipedia data. We analyze tokenizers, n-gram models, Markov chains, vocabulary statistics, and word embeddings.

📋 Repository Contents

Models & Assets

• Tokenizers (8k, 16k, 32k, 64k)
• N-gram models (2, 3, 4, 5-gram)
• Markov chains (context of 1, 2, 3, 4 and 5)
• Subword N-gram and Markov chains
• Embeddings in various sizes and dimensions (aligned and unaligned)
• Language Vocabulary
• Language Statistics

Analysis and Evaluation

• 1. Tokenizer Evaluation
• 2. N-gram Model Evaluation
• 3. Markov Chain Evaluation
• 4. Vocabulary Analysis
• 5. Word Embeddings Evaluation
• 6. Morphological Analysis (Experimental)
• 7. Summary & Recommendations
• Metrics Glossary
• Visualizations Index

---

1. Tokenizer Evaluation

Results

Vocab Size	Compression	Avg Token Len	UNK Rate	Total Tokens
8k	3.964x	3.97	1.0788%	1,015,636
16k	4.402x	4.40	1.1980%	914,601
32k	4.651x	4.65	1.2658%	865,595
64k	4.905x 🏆	4.91	1.3350%	820,755

Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

Sample 1: တႃႈႁိူဝ်းမိၼ် မိူင်းတူၼ် ၼႆႉ ပဵၼ်တႃႈႁိူဝ်းမိၼ် ဢၼ်မီးတီႈ ဝဵင်းမိူင်းတူၼ်၊ မိူင်း...

Vocab	Tokens	Count
8k	▁တႃႈႁိူဝ်းမိၼ် ▁မ ိူင်းတူၼ် ▁ၼႆႉ ▁ပဵၼ် တႃႈႁိူဝ်းမိၼ် ▁ဢၼ်မီးတီႈ ▁ဝဵင်းမ ိူင်းတူၼ် ၊ ... (+7 more)	17
16k	▁တႃႈႁိူဝ်းမိၼ် ▁မိူင်းတူၼ် ▁ၼႆႉ ▁ပဵၼ် တႃႈႁိူဝ်းမိၼ် ▁ဢၼ်မီးတီႈ ▁ဝဵင်းမ ိူင်းတူၼ်၊ ▁မိူင်းတႆး၊ ▁မိူင်းမျၢၼ်ႇမႃႇ ... (+4 more)	14
32k	▁တႃႈႁိူဝ်းမိၼ် ▁မိူင်းတူၼ် ▁ၼႆႉ ▁ပဵၼ် တႃႈႁိူဝ်းမိၼ် ▁ဢၼ်မီးတီႈ ▁ဝဵင်းမ ိူင်းတူၼ်၊ ▁မိူင်းတႆး၊ ▁မိူင်းမျၢၼ်ႇမႃႇ ... (+4 more)	14
64k	▁တႃႈႁိူဝ်းမိၼ် ▁မိူင်းတူၼ် ▁ၼႆႉ ▁ပဵၼ် တႃႈႁိူဝ်းမိၼ် ▁ဢၼ်မီးတီႈ ▁ဝဵင်းမ ိူင်းတူၼ်၊ ▁မိူင်းတႆး၊ ▁မိူင်းမျၢၼ်ႇမႃႇ ... (+4 more)	14

Sample 2: ၶႂ်ႈမၢႆထိုင်ဝႃႈ - တူဝ်ၼပ်ႉ 30 ၸိူဝ်းပဵၼ်ပီ ဢေႇတီႇ 30,

Vocab	Tokens	Count
8k	▁ၶႂ်ႈမၢႆထိုင်ဝႃႈ ▁- ▁တူဝ်ၼပ်ႉ ▁ 3 0 ▁ၸိူဝ်းပဵၼ်ပီ ▁ဢေႇတီႇ ▁ 3 ... (+2 more)	12
16k	▁ၶႂ်ႈမၢႆထိုင်ဝႃႈ ▁- ▁တူဝ်ၼပ်ႉ ▁ 3 0 ▁ၸိူဝ်းပဵၼ်ပီ ▁ဢေႇတီႇ ▁ 3 ... (+2 more)	12
32k	▁ၶႂ်ႈမၢႆထိုင်ဝႃႈ ▁- ▁တူဝ်ၼပ်ႉ ▁ 3 0 ▁ၸိူဝ်းပဵၼ်ပီ ▁ဢေႇတီႇ ▁ 3 ... (+2 more)	12
64k	▁ၶႂ်ႈမၢႆထိုင်ဝႃႈ ▁- ▁တူဝ်ၼပ်ႉ ▁ 3 0 ▁ၸိူဝ်းပဵၼ်ပီ ▁ဢေႇတီႇ ▁ 3 ... (+2 more)	12

Sample 3: ၶႂ်ႈမၢႆထိုင်ဝႃႈ - တူဝ်ၼပ်ႉ 47 ၸိူဝ်းပဵၼ်ပီ ဢေႇတီႇ 47,

Vocab	Tokens	Count
8k	▁ၶႂ်ႈမၢႆထိုင်ဝႃႈ ▁- ▁တူဝ်ၼပ်ႉ ▁ 4 7 ▁ၸိူဝ်းပဵၼ်ပီ ▁ဢေႇတီႇ ▁ 4 ... (+2 more)	12
16k	▁ၶႂ်ႈမၢႆထိုင်ဝႃႈ ▁- ▁တူဝ်ၼပ်ႉ ▁ 4 7 ▁ၸိူဝ်းပဵၼ်ပီ ▁ဢေႇတီႇ ▁ 4 ... (+2 more)	12
32k	▁ၶႂ်ႈမၢႆထိုင်ဝႃႈ ▁- ▁တူဝ်ၼပ်ႉ ▁ 4 7 ▁ၸိူဝ်းပဵၼ်ပီ ▁ဢေႇတီႇ ▁ 4 ... (+2 more)	12
64k	▁ၶႂ်ႈမၢႆထိုင်ဝႃႈ ▁- ▁တူဝ်ၼပ်ႉ ▁ 4 7 ▁ၸိူဝ်းပဵၼ်ပီ ▁ဢေႇတီႇ ▁ 4 ... (+2 more)	12

Key Findings

• Best Compression: 64k achieves 4.905x compression
• Lowest UNK Rate: 8k with 1.0788% unknown tokens
• Trade-off: Larger vocabularies improve compression but increase model size
• Recommendation: 32k vocabulary provides optimal balance for production use

---

2. N-gram Model Evaluation

Results

N-gram	Variant	Perplexity	Entropy	Unique N-grams	Top-100 Coverage	Top-1000 Coverage
2-gram	Word	304 🏆	8.25	6,013	75.0%	92.0%
2-gram	Subword	774	9.60	13,675	49.8%	86.7%
3-gram	Word	430	8.75	11,217	69.6%	89.1%
3-gram	Subword	4,483	12.13	77,354	27.7%	59.2%
4-gram	Word	621	9.28	23,157	67.2%	84.2%
4-gram	Subword	15,378	13.91	268,593	20.2%	44.3%
5-gram	Word	620	9.28	22,270	68.2%	83.7%
5-gram	Subword	30,653	14.90	454,166	17.4%	39.0%

Top 5 N-grams by Size

2-grams (Word):

Rank	N-gram	Count
1	1 ဝၼ်း	30,342
2	ၼႆႉ မီးဝႆႉတီႈ	5,369
3	ပဵၼ် ယဝ်ႉ	5,138
4	ၸွမ်းလူၺ်ႈ သဵၼ်ႈမၢႆႁူဝ်ႁိူၼ်း	4,826
5	သေ ႁူဝ်ၼပ်ႉၵူၼ်း	4,818

3-grams (Word):

Rank	N-gram	Count
1	ယဝ်ႉ ၶူတ်ႉဢွင်ႈတီႈၼႆႉ ပဵၼ်	4,773
2	ၸႄႈတိူင်းတႃႈလိူဝ်ႇ ယဝ်ႉ ၶူတ်ႉဢွင်ႈတီႈၼႆႉ	4,773
3	သေ ႁူဝ်ၼပ်ႉၵူၼ်း ယူႇသဝ်း	4,741
4	သဵၼ်ႈမၢႆႁူဝ်ႁိူၼ်း သေ တီႈ	4,740
5	ယူႇသဝ်း ႁူမ်ႈ မီးယူႇ	4,740

4-grams (Word):

Rank	N-gram	Count
1	ၸႄႈတိူင်းတႃႈလိူဝ်ႇ ယဝ်ႉ ၶူတ်ႉဢွင်ႈတီႈၼႆႉ ပဵၼ်	4,773
2	ႁူဝ်ၼပ်ႉၵူၼ်း ယူႇသဝ်း ႁူမ်ႈ မီးယူႇ	4,740
3	သေ ႁူဝ်ၼပ်ႉၵူၼ်း ယူႇသဝ်း ႁူမ်ႈ	4,740
4	ၸွမ်းလူၺ်ႈ သဵၼ်ႈမၢႆႁူဝ်ႁိူၼ်း သေ တီႈ	4,735
5	ၵေႃႉ သေ ႁူဝ်ၼပ်ႉၵူၼ်း ယူႇသဝ်း	4,595

5-grams (Word):

Rank	N-gram	Count
1	သေ ႁူဝ်ၼပ်ႉၵူၼ်း ယူႇသဝ်း ႁူမ်ႈ မီးယူႇ	4,740
2	ၵေႃႉ သေ ႁူဝ်ၼပ်ႉၵူၼ်း ယူႇသဝ်း ႁူမ်ႈ	4,595
3	ၸႄႈတိူင်းတႃႈလိူဝ်ႇ ယဝ်ႉ ၶူတ်ႉဢွင်ႈတီႈၼႆႉ ပဵၼ် ယဝ်ႉ	4,586
4	ယဝ်ႉ ၸွမ်းလူၺ်ႈ သဵၼ်ႈမၢႆႁူဝ်ႁိူၼ်း သေ တီႈ	4,549
5	ယဝ်ႉ ၶူတ်ႉဢွင်ႈတီႈၼႆႉ ပဵၼ် ယဝ်ႉ ၸွမ်းလူၺ်ႈ	4,548

2-grams (Subword):

Rank	N-gram	Count
1	ၼ် း	202,089
2	း _	191,313
3	) _	136,283
4	_ (	136,166
5	င် း	128,103

3-grams (Subword):

Rank	N-gram	Count
1	ဝ ၼ် း	122,686
2	_ ဝ ၼ်	119,537
3	) _ ဝ	116,929
4	ၼ် း _	111,432
5	း _ (	90,718

4-grams (Subword):

Rank	N-gram	Count
1	_ ဝ ၼ် း	119,517
2	) _ ဝ ၼ်	116,755
3	ဝ ၼ် း _	89,164
4	ၼ် း _ (	86,035
5	ယ ဝ် ႉ ။	44,872

5-grams (Subword):

Rank	N-gram	Count
1	) _ ဝ ၼ် း	116,755
2	_ ဝ ၼ် း _	88,818
3	ဝ ၼ် း _ (	85,077
4	ယ ဝ် ႉ ။ _	44,169
5	1 ) _ ဝ ၼ်	38,381

Key Findings

• Best Perplexity: 2-gram (word) with 304
• Entropy Trend: Decreases with larger n-grams (more predictable)
• Coverage: Top-1000 patterns cover ~39% of corpus
• Recommendation: 4-gram or 5-gram for best predictive performance

---

3. Markov Chain Evaluation

Results

Context	Variant	Avg Entropy	Perplexity	Branching Factor	Unique Contexts	Predictability
1	Word	0.2330	1.175	1.77	288,426	76.7%
1	Subword	0.1366	1.099	3.35	23,531	86.3%
2	Word	0.0481	1.034	1.09	510,797	95.2%
2	Subword	0.3263	1.254	2.92	78,767	67.4%
3	Word	0.0147	1.010	1.03	554,509	98.5%
3	Subword	0.4202	1.338	2.67	229,767	58.0%
4	Word	0.0059 🏆	1.004	1.01	566,193	99.4%
4	Subword	0.3421	1.268	2.01	614,095	65.8%

Generated Text Samples (Word-based)

Below are text samples generated from each word-based Markov chain model:

Context Size 1:

1. ဝၼ်း 27 ဝၼ်း 9 ၸုမ်းၼၼ်ႉ ဢွၼ်ၵၼ်ၶပ်ႉယႆပႆၸွမ်း သဵၼ်ႈတၢင်းပၢင်းပိတၵၢတ်ႈလႄႈ ၵူၼ်းသမ်ႉပေႃးတဵမ်ၵဵဝ်ႇတဵမ်တ...
2. 1 ဝၼ်း လိူၼ်သႅပ်ႇထႅမ်ႇပႃႇ 1 ဝၼ်း 6 ဝၼ်း 7 ဝၼ်း 28 ဝၼ်း 19 ဝၼ်း 5 ၶိုၼ်းယဝ်ႉ မိၼ်းယႄးၵျေႃႇၸႂႃႇၵေႃႈ
3. ယဝ်ႉ ၸွမ်းလူၺ်ႈ သဵၼ်ႈမၢႆႁူဝ်ႁိူၼ်း သေ ပဵၼ်မႃး ငဝ်းမၢပ်ႈႁိူဝ်ႈလႄႈ တီႈလွၵ်းသီမၢပ်ႈႁိူဝ်ႈၼႆႉ တေလႆႈဢဝ်သီ...

Context Size 2:

1. 1 ဝၼ်း လိူၼ်ၼူဝ်ႇဝႅမ်ႇပႃႇ 1 ဝၼ်း လိူၼ်ၾႅပ်ႇဝႃႇရီႇ 1 ဝၼ်း သႅပ်ႇထႅမ်ႇပႃႇ ၁ ဝၼ်း ၂ ဝၼ်း ၃ ဝၼ်း ၄ ဝၼ်း
2. ၼႆႉ မီးဝႆႉတီႈ ဢိူင်ႇသတဵင်ႇ ၸႄႈဝဵင်းꧤူၵ်ႉပျိၼ်း ၸႄႈတွၼ်ႈၵေႃႉတွင်း ၸႄႈတိူင်းတၼိၼ်းတႃႇယီႇ ယဝ်ႉ ၶူတ်ႉဢွင...
3. ပဵၼ် ယဝ်ႉ ၸွမ်းလူၺ်ႈ သဵၼ်ႈမၢႆႁူဝ်ႁိူၼ်း သေ တီႈ ဢိူင်ႇဢၢႆႇတူဝ်ႇၼႆႉ မီးဝႆႉ ၸၢႆး ၵေႃႉ ယိင်း ၵေႃႉ သေ ႁူဝ...

Context Size 3:

1. ၸႄႈတိူင်းတႃႈလိူဝ်ႇ ယဝ်ႉ ၶူတ်ႉဢွင်ႈတီႈၼႆႉ ပဵၼ် ယဝ်ႉ ၸွမ်းလူၺ်ႈ သဵၼ်ႈမၢႆႁူဝ်ႁိူၼ်း သေ တီႈ ဢိူင်ႇတယႅတ်ႉ...
2. ယဝ်ႉ ၶူတ်ႉဢွင်ႈတီႈၼႆႉ ပဵၼ် ယဝ်ႉ ၸွမ်းလူၺ်ႈ သဵၼ်ႈမၢႆႁူဝ်ႁိူၼ်း သေ တီႈ ဢိူင်ႇတလၢႆးသျေႇ ၸၢၼ်း ၼႆႉ မီးဝႆ...
3. သေ ႁူဝ်ၼပ်ႉၵူၼ်း ယူႇသဝ်း ႁူမ်ႈ မီးယူႇ ၵေႃႉယဝ်ႉ ၽိုၼ်ဢိင် ၸႄႈဝဵင်းငၸုၼ်ႇ

Context Size 4:

1. ၸႄႈတိူင်းတႃႈလိူဝ်ႇ ယဝ်ႉ ၶူတ်ႉဢွင်ႈတီႈၼႆႉ ပဵၼ် ယဝ်ႉ ၸွမ်းလူၺ်ႈ သဵၼ်ႈမၢႆႁူဝ်ႁိူၼ်း သေ တီႈ ဢိူင်ႇၵျူဝ်ႉ...
2. ႁူဝ်ၼပ်ႉၵူၼ်း ယူႇသဝ်း ႁူမ်ႈ မီးယူႇ ၵေႃႉယဝ်ႉ ၽိုၼ်ဢိင် ၸႄႈဝဵင်းမတယႃႇ
3. သေ ႁူဝ်ၼပ်ႉၵူၼ်း ယူႇသဝ်း ႁူမ်ႈ မီးယူႇ ၵေႃႉယဝ်ႉ ၽိုၼ်ဢိင် ၸႄႈဝဵင်းပျေႃႇပွႆႇ

Generated Text Samples (Subword-based)

Below are text samples generated from each subword-based Markov chain model:

Context Size 1:

1. _ပၢၵ်ႈပၼ်း_သီႁဝ်_။_တွ
2. းၼၼ်းၸူဝ်း_5)_ဝ၊_ဢေႃ
3. ၼ်ပၼ်_၊_ၵေႃႇ_ယိင်ၸိူဝ်းလ

Context Size 2:

1. ၼ်း။_ၸွမ်_ၵူၼ်းၸုၵျီႇ_(5)
2. း_(29)_ဝၼ်း_(1)_ဝၼ်
3. )_ဝၢၼ်_ၸၢႆး_(14)_ဝၼ်

Context Size 3:

1. ဝၼ်း_(2)_ဝၼ်း_ၽၢႆႇတူၵ်း
2. _ဝၼ်းဢွၵ်ႇၼႆ_လဝ်ႈထိုင်တႃႇ_
3. )_ဝၼ်း_(12)_ဝၼ်း_(28

Context Size 4:

1. _ဝၼ်း_(21)_ဝၼ်း_(16)_
2. )_ဝၼ်း_(20)_ဝၼ်း။_လိူၼ်သႅ
3. ဝၼ်း_(24)_ဝၼ်း_(20)_ဝ

Key Findings

• Best Predictability: Context-4 (word) with 99.4% predictability
• Branching Factor: Decreases with context size (more deterministic)
• Memory Trade-off: Larger contexts require more storage (614,095 contexts)
• Recommendation: Context-3 or Context-4 for text generation

---

4. Vocabulary Analysis

Statistics

Metric	Value
Vocabulary Size	47,353
Total Tokens	767,152
Mean Frequency	16.20
Median Frequency	3
Frequency Std Dev	582.68

Most Common Words

Rank	Word	Frequency
1	ဝၼ်း	116,548
2	1	32,050
3	ယဝ်ႉ	11,719
4	ၽိုၼ်ဢိင်	11,655
5	သေ	10,963
6	ၵေႃႉ	9,578
7	ၼႆႉ	8,785
8	ပဵၼ်	7,402
9	တီႈ	5,950
10	မီးဝႆႉ	5,835

Least Common Words (from vocabulary)

Rank	Word	Frequency
1	ၽိုၼ်မိူၼ်	2
2	copies	2
3	ပဵၼ်တႆးၽိဝ်း	2
4	ၸၢႆးဢၢၼ်းတႆး	2
5	ၼၢင်းယွတ်ႈၼု	2
6	ၸၢႆးၵျီး	2
7	မိူင်းယႆယဝ်ႉ	2
8	ပူဝ်ႇမျႃႉ	2
9	ၶုၼ်ၵျေႃႉၶႅင်ႇ	2
10	လွင်ႈငမ်းယဵၼ်မိူင်းတႆး	2

Zipf's Law Analysis

Metric	Value
Zipf Coefficient	0.9775
R² (Goodness of Fit)	0.985701
Adherence Quality	excellent

Coverage Analysis

Top N Words	Coverage
Top 100	58.5%
Top 1,000	73.3%
Top 5,000	82.5%
Top 10,000	87.2%

Key Findings

• Zipf Compliance: R²=0.9857 indicates excellent adherence to Zipf's law
• High Frequency Dominance: Top 100 words cover 58.5% of corpus
• Long Tail: 37,353 words needed for remaining 12.8% coverage

---

5. Word Embeddings Evaluation

5.1 Cross-Lingual Alignment

5.2 Model Comparison

Model	Dimension	Isotropy	Semantic Density	Alignment R@1	Alignment R@10
mono_32d	32	0.7537 🏆	0.3337	N/A	N/A
mono_64d	64	0.3939	0.2857	N/A	N/A
mono_128d	128	0.0610	0.2919	N/A	N/A
aligned_32d	32	0.7537	0.3194	0.0180	0.1380
aligned_64d	64	0.3939	0.2880	0.0300	0.1900
aligned_128d	128	0.0610	0.2969	0.0420	0.2220

Key Findings

• Best Isotropy: mono_32d with 0.7537 (more uniform distribution)
• Semantic Density: Average pairwise similarity of 0.3026. Lower values indicate better semantic separation.
• Alignment Quality: Aligned models achieve up to 4.2% R@1 in cross-lingual retrieval.
• Recommendation: 128d aligned for best cross-lingual performance

---

6. Morphological Analysis (Experimental)

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.

6.1 Productivity & Complexity

Metric	Value	Interpretation	Recommendation
Productivity Index	5.000	High morphological productivity	Reliable analysis
Idiomaticity Gap	1.149	High formulaic/idiomatic content	-

6.2 Affix Inventory (Productive Units)

These are the most productive prefixes and suffixes identified by sampling the vocabulary for global substitutability patterns. A unit is considered an affix if stripping it leaves a valid stem that appears in other contexts.

Productive Prefixes

Prefix	Examples
-ပ	ပၢၼ်လဵဝ်ၵၼ်တင်း, ပူႇတၢၼ်ႇ, ပုဂံ
-တ	တီႈၼႂ်းၸႄႈဝဵင်းယိၼ်းမႃႇပိၼ်ႇၼႆႉ, တွင်ႇလိူဝ်း, တီႈၼၼ်ႈၵေႃႈ
-မ	မိူင်းလႂ်မိူင်းၼၼ်ႉ, မိူင်းမျၢၼ်ႇမႃႇယဝ်ႉ, မရဵင်း
-သ	သမ်ႉၸၢင်ႈ, သင်ႇဝႆႉဝႃႈ, သေဢမ်ႇဝႃႈ
-ဢ	ဢၼ်တႄႉ, ဢိူင်ႇသီးၵူင်းၼႆႉ, ဢဝ်ပရိၼိပ်ႉပၢၼ်ႇယဝ်ႉ
-ၵ	ၵေႃလိၵ်ႈလၢႆးလႄႈ, ၵႂႃႇဢႅဝ်ႇ, ၵၢၼ်ၸိူင်ႉႁိုဝ်လႄႈသင်
-လ	လုမ်းတႅၼ်းၽွင်းသၢင်ႈၾိင်ႈမိူင်း, လိူင်ႇၼမ်ယဝ်ႉ, လိၵ်ႈတႆးလူင်မိူဝ်ႈပၢၼ်ၵဵင်းဝီ
-ဢၼ	ဢၼ်တႄႉ, ဢၼ်ႁဵတ်းၵၢၼ်တႆး, ဢၼ်ဢွၼ်ႇၸၢႆး

Productive Suffixes

Suffix	Examples
-s	classes, dress, layouts
-n	foreign, christian, berlin
-e	give, aubange, lifestyle
-d	passed, afraid, ဝၢၼ်ႈလူင်တွင်းgad
-on	migration, opinion, xenophon
-ng	achang, trading, zhejiang
-y	day, modernity, turkey
-t	east, recordsost, crescent

6.3 Bound Stems (Lexical Roots)

Bound stems are high-frequency subword units that are semantically cohesive but rarely appear as standalone words. These often correspond to the 'core' of a word that requires inflection or derivation to be valid.

Stem	Cohesion	Substitutability	Examples
tion	2.53x	13 contexts	action, nation, options
atio	2.48x	11 contexts	nation, nations, station

6.4 Affix Compatibility (Co-occurrence)

This table shows which prefixes and suffixes most frequently co-occur on the same stems, revealing the 'stacking' rules of the language's morphology.

Prefix	Suffix	Frequency	Examples
-s	-s	8 words	scales, shows
-s	-t	6 words	scoot, significant
-s	-d	5 words	statehood, switzerland
-s	-y	5 words	study, slowly
-s	-n	4 words	sangken, sovereign
-s	-e	3 words	spike, shwe
-s	-ed	3 words	supported, specialized
-s	-ng	2 words	shandong, sung
-s	-g	2 words	shandong, sung
-s	-on	2 words	simpson, scorpion

6.5 Recursive Morpheme Segmentation

Using Recursive Hierarchical Substitutability, we decompose complex words into their constituent morphemes. This approach handles nested affixes (e.g., prefix-prefix-root-suffix).

Word	Suggested Split	Confidence	Stem
operations	operation-s	4.5	operation
ႁၵ်ႉမိူင်း	ႁ-ၵ-်ႉမိူင်း	4.5	်ႉမိူင်း
လၵ်းမိူင်းၼႆႉ	လ-ၵ-်းမိူင်းၼႆႉ	4.5	်းမိူင်းၼႆႉ
လၵ်းမိူင်း	လ-ၵ-်းမိူင်း	4.5	်းမိူင်း
လဝ်ႈထိုင်	လ-ဝ-်ႈထိုင်	3.0	်ႈထိုင်
တီႈလူႇတၢၼ်း	တ-ီႈလူႇတၢၼ်း	1.5	ီႈလူႇတၢၼ်း
expressway	expresswa-y	1.5	expresswa
လိူၼ်ႁူၵ်း	လ-ိူၼ်ႁူၵ်း	1.5	ိူၼ်ႁူၵ်း
ဢဝ်ငဝ်းလႅင်း	ဢဝ-်ငဝ်းလႅင်း	1.5	်ငဝ်းလႅင်း
ၶဝ်တွၼ်းလိူဝ်သေ	ၶဝ-်တွၼ်းလိူဝ်သေ	1.5	်တွၼ်းလိူဝ်သေ
ဢေႃးၽႃႇမင်ႇၵလႃႇ	ဢ-ေႃးၽႃႇမင်ႇၵလႃႇ	1.5	ေႃးၽႃႇမင်ႇၵလႃႇ
ဢမ်ႇလီလိုမ်း	ဢ-မ်ႇလီလိုမ်း	1.5	မ်ႇလီလိုမ်း
မိူင်းဢႃႇဝႃႉၵေႃႈ	မ-ိူင်းဢႃႇဝႃႉၵေႃႈ	1.5	ိူင်းဢႃႇဝႃႉၵေႃႈ
ဢၼ်မီးၵုင်ႇမုၼ်	ဢၼ-်မီးၵုင်ႇမုၼ်	1.5	်မီးၵုင်ႇမုၼ်
ဢိင်ၼိူဝ်လူၺ်ႈ	ဢ-ိင်ၼိူဝ်လူၺ်ႈ	1.5	ိင်ၼိူဝ်လူၺ်ႈ

6.6 Linguistic Interpretation

Automated Insight: The language Shan shows high morphological productivity. The subword models are significantly more efficient than word models, suggesting a rich system of affixation or compounding.

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.

---

7. Summary & Recommendations

Production Recommendations

Component	Recommended	Rationale
Tokenizer	64k BPE	Best compression (4.91x)
N-gram	2-gram	Lowest perplexity (304)
Markov	Context-4	Highest predictability (99.4%)
Embeddings	100d	Balanced semantic capture and isotropy

---

Appendix: Metrics Glossary & Interpretation Guide

This section provides definitions, intuitions, and guidance for interpreting the metrics used throughout this report.

Tokenizer Metrics

Compression Ratio

Definition: The ratio of characters to tokens (chars/token). Measures how efficiently the tokenizer represents text.

Intuition: Higher compression means fewer tokens needed to represent the same text, reducing sequence lengths for downstream models. A 3x compression means ~3 characters per token on average.

What to seek: Higher is generally better for efficiency, but extremely high compression may indicate overly aggressive merging that loses morphological information.

Average Token Length (Fertility)

Definition: Mean number of characters per token produced by the tokenizer.

Intuition: Reflects the granularity of tokenization. Longer tokens capture more context but may struggle with rare words; shorter tokens are more flexible but increase sequence length.

What to seek: Balance between 2-5 characters for most languages. Arabic/morphologically-rich languages may benefit from slightly longer tokens.

Unknown Token Rate (OOV Rate)

Definition: Percentage of tokens that map to the unknown/UNK token, indicating words the tokenizer cannot represent.

Intuition: Lower OOV means better vocabulary coverage. High OOV indicates the tokenizer encounters many unseen character sequences.

What to seek: Below 1% is excellent; below 5% is acceptable. BPE tokenizers typically achieve very low OOV due to subword fallback.

N-gram Model Metrics

Perplexity

Definition: Measures how "surprised" the model is by test data. Mathematically: 2^(cross-entropy). Lower values indicate better prediction.

Intuition: If perplexity is 100, the model is as uncertain as if choosing uniformly among 100 options at each step. A perplexity of 10 means effectively choosing among 10 equally likely options.

What to seek: Lower is better. Perplexity decreases with larger n-grams (more context). Values vary widely by language and corpus size.

Entropy

Definition: Average information content (in bits) needed to encode the next token given the context. Related to perplexity: perplexity = 2^entropy.

Intuition: High entropy means high uncertainty/randomness; low entropy means predictable patterns. Natural language typically has entropy between 1-4 bits per character.

What to seek: Lower entropy indicates more predictable text patterns. Entropy should decrease as n-gram size increases.

Coverage (Top-K)

Definition: Percentage of corpus occurrences explained by the top K most frequent n-grams.

Intuition: High coverage with few patterns indicates repetitive/formulaic text; low coverage suggests diverse vocabulary usage.

What to seek: Depends on use case. For language modeling, moderate coverage (40-60% with top-1000) is typical for natural text.

Markov Chain Metrics

Average Entropy

Definition: Mean entropy across all contexts, measuring average uncertainty in next-word prediction.

Intuition: Lower entropy means the model is more confident about what comes next. Context-1 has high entropy (many possible next words); Context-4 has low entropy (few likely continuations).

What to seek: Decreasing entropy with larger context sizes. Very low entropy (<0.1) indicates highly deterministic transitions.

Branching Factor

Definition: Average number of unique next tokens observed for each context.

Intuition: High branching = many possible continuations (flexible but uncertain); low branching = few options (predictable but potentially repetitive).

What to seek: Branching factor should decrease with context size. Values near 1.0 indicate nearly deterministic chains.

Predictability

Definition: Derived metric: (1 - normalized_entropy) × 100%. Indicates how deterministic the model's predictions are.

Intuition: 100% predictability means the next word is always certain; 0% means completely random. Real text falls between these extremes.

What to seek: Higher predictability for text generation quality, but too high (>98%) may produce repetitive output.

Vocabulary & Zipf's Law Metrics

Zipf's Coefficient

Definition: The slope of the log-log plot of word frequency vs. rank. Zipf's law predicts this should be approximately -1.

Intuition: A coefficient near -1 indicates the corpus follows natural language patterns where a few words are very common and most words are rare.

What to seek: Values between -0.8 and -1.2 indicate healthy natural language distribution. Deviations may suggest domain-specific or artificial text.

R² (Coefficient of Determination)

Definition: Measures how well the linear fit explains the frequency-rank relationship. Ranges from 0 to 1.

Intuition: R² near 1.0 means the data closely follows Zipf's law; lower values indicate deviation from expected word frequency patterns.

What to seek: R² > 0.95 is excellent; > 0.99 indicates near-perfect Zipf adherence typical of large natural corpora.

Vocabulary Coverage

Definition: Cumulative percentage of corpus tokens accounted for by the top N words.

Intuition: Shows how concentrated word usage is. If top-100 words cover 50% of text, the corpus relies heavily on common words.

What to seek: Top-100 covering 30-50% is typical. Higher coverage indicates more repetitive text; lower suggests richer vocabulary.

Word Embedding Metrics

Isotropy

Definition: Measures how uniformly distributed vectors are in the embedding space. Computed as the ratio of minimum to maximum singular values.

Intuition: High isotropy (near 1.0) means vectors spread evenly in all directions; low isotropy means vectors cluster in certain directions, reducing expressiveness.

What to seek: Higher isotropy generally indicates better-quality embeddings. Values > 0.1 are reasonable; > 0.3 is good. Lower-dimensional embeddings tend to have higher isotropy.

Average Norm

Definition: Mean magnitude (L2 norm) of word vectors in the embedding space.

Intuition: Indicates the typical "length" of vectors. Consistent norms suggest stable training; high variance may indicate some words are undertrained.

What to seek: Relatively consistent norms across models. The absolute value matters less than consistency (low std deviation).

Cosine Similarity

Definition: Measures angular similarity between vectors, ranging from -1 (opposite) to 1 (identical direction).

Intuition: Words with similar meanings should have high cosine similarity. This is the standard metric for semantic relatedness in embeddings.

What to seek: Semantically related words should score > 0.5; unrelated words should be near 0. Synonyms often score > 0.7.

t-SNE Visualization

Definition: t-Distributed Stochastic Neighbor Embedding - a dimensionality reduction technique that preserves local structure for visualization.

Intuition: Clusters in t-SNE plots indicate groups of semantically related words. Spread indicates vocabulary diversity; tight clusters suggest semantic coherence.

What to seek: Meaningful clusters (e.g., numbers together, verbs together). Avoid over-interpreting distances - t-SNE preserves local, not global, structure.

General Interpretation Guidelines

1. Compare within model families: Metrics are most meaningful when comparing models of the same type (e.g., 8k vs 64k tokenizer).
2. Consider trade-offs: Better performance on one metric often comes at the cost of another (e.g., compression vs. OOV rate).
3. Context matters: Optimal values depend on downstream tasks. Text generation may prioritize different metrics than classification.
4. Corpus influence: All metrics are influenced by corpus characteristics. Wikipedia text differs from social media or literature.
5. Language-specific patterns: Morphologically rich languages (like Arabic) may show different optimal ranges than analytic languages.

Visualizations Index

Visualization	Description
Tokenizer Compression	Compression ratios by vocabulary size
Tokenizer Fertility	Average token length by vocabulary
Tokenizer OOV	Unknown token rates
Tokenizer Total Tokens	Total tokens by vocabulary
N-gram Perplexity	Perplexity by n-gram size
N-gram Entropy	Entropy by n-gram size
N-gram Coverage	Top pattern coverage
N-gram Unique	Unique n-gram counts
Markov Entropy	Entropy by context size
Markov Branching	Branching factor by context
Markov Contexts	Unique context counts
Zipf's Law	Frequency-rank distribution with fit
Vocab Frequency	Word frequency distribution
Top 20 Words	Most frequent words
Vocab Coverage	Cumulative coverage curve
Embedding Isotropy	Vector space uniformity
Embedding Norms	Vector magnitude distribution
Embedding Similarity	Word similarity heatmap
Nearest Neighbors	Similar words for key terms
t-SNE Words	2D word embedding visualization
t-SNE Sentences	2D sentence embedding visualization
Position Encoding	Encoding method comparison
Model Sizes	Storage requirements
Performance Dashboard	Comprehensive performance overview

---

About This Project

Data Source

Models trained on wikipedia-monthly - a monthly snapshot of Wikipedia articles across 300+ languages.

Project

A project by Wikilangs - Open-source NLP models for every Wikipedia language.

Maintainer

Omar Kamali - Omneity Labs

Citation

If you use these models in your research, please cite:

@misc{wikilangs2025,
  author = {Kamali, Omar},
  title = {Wikilangs: Open NLP Models for Wikipedia Languages},
  year = {2025},
  doi = {10.5281/zenodo.18073153},
  publisher = {Zenodo},
  url = {https://huggingface.co/wikilangs}
  institution = {Omneity Labs}
}

License

MIT License - Free for academic and commercial use.

Links

• 🌐 Website: wikilangs.org
• 🤗 Models: huggingface.co/wikilangs
• 📊 Data: wikipedia-monthly
• 👤 Author: Omar Kamali
• 🤝 Sponsor: Featherless AI

---

Generated by Wikilangs Models Pipeline

Report Date: 2026-01-10 20:12:17