README.md

metadata

language: pnb
language_name: Western Panjabi
language_family: indoaryan_central
tags:
  - wikilangs
  - nlp
  - tokenizer
  - embeddings
  - n-gram
  - markov
  - wikipedia
  - feature-extraction
  - sentence-similarity
  - tokenization
  - n-grams
  - markov-chain
  - text-mining
  - fasttext
  - babelvec
  - vocabulous
  - vocabulary
  - monolingual
  - family-indoaryan_central
license: mit
library_name: wikilangs
pipeline_tag: text-generation
datasets:
  - omarkamali/wikipedia-monthly
dataset_info:
  name: wikipedia-monthly
  description: Monthly snapshots of Wikipedia articles across 300+ languages
metrics:
  - name: best_compression_ratio
    type: compression
    value: 3.987
  - name: best_isotropy
    type: isotropy
    value: 0.8211
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-10T00:00:00.000Z

Western Panjabi - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on Western Panjabi 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	3.374x	3.35	0.0495%	1,253,323
16k	3.663x	3.64	0.0537%	1,154,342
32k	3.861x	3.84	0.0566%	1,095,265
64k	3.987x 🏆	3.96	0.0585%	1,060,503

Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

Sample 1: <font size="+1" بلی size="1" : : : ناں : Pseudotriakis microdon size="1" تے پھرن...

Vocab	Tokens	Count
8k	▁< font ▁size ="+ 1 " ▁بلی ▁size =" 1 ... (+26 more)	36
16k	▁< font ▁size ="+ 1 " ▁بلی ▁size =" 1 ... (+25 more)	35
32k	▁< font ▁size ="+ 1 " ▁بلی ▁size =" 1 ... (+22 more)	32
64k	▁< font ▁size ="+ 1 " ▁بلی ▁size =" 1 ... (+22 more)	32

Sample 2: واقعے جم موت ہور دیکھو ہجری شمسی عیسوی کیلنڈر ہجری کیلنڈر حوالے باہرلےجوڑ ہجری ت...

Vocab	Tokens	Count
8k	▁واقعے ▁جم ▁موت ▁ہور ▁دیکھو ▁ہجری ▁شمسی ▁عیسوی ▁کیلنڈر ▁ہجری ... (+20 more)	30
16k	▁واقعے ▁جم ▁موت ▁ہور ▁دیکھو ▁ہجری ▁شمسی ▁عیسوی ▁کیلنڈر ▁ہجری ... (+20 more)	30
32k	▁واقعے ▁جم ▁موت ▁ہور ▁دیکھو ▁ہجری ▁شمسی ▁عیسوی ▁کیلنڈر ▁ہجری ... (+20 more)	30
64k	▁واقعے ▁جم ▁موت ▁ہور ▁دیکھو ▁ہجری ▁شمسی ▁عیسوی ▁کیلنڈر ▁ہجری ... (+20 more)	30

Sample 3: thumbnail یورپا مشتری پاندھی دا 6واں چند اے۔ ایہنوں 8 جنوری، وچ گلیلیو نے لبیا س...

Vocab	Tokens	Count
8k	▁thumbnail ▁یورپ ا ▁مشت ری ▁پاندھی ▁دا ▁ 6 واں ... (+26 more)	36
16k	▁thumbnail ▁یورپ ا ▁مشتری ▁پاندھی ▁دا ▁ 6 واں ▁چند ... (+23 more)	33
32k	▁thumbnail ▁یورپ ا ▁مشتری ▁پاندھی ▁دا ▁ 6 واں ▁چند ... (+22 more)	32
64k	▁thumbnail ▁یورپ ا ▁مشتری ▁پاندھی ▁دا ▁ 6 واں ▁چند ... (+21 more)	31

Key Findings

• Best Compression: 64k achieves 3.987x compression
• Lowest UNK Rate: 8k with 0.0495% 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	75,798	16.21	740,485	13.0%	25.5%
2-gram	Subword	455 🏆	8.83	31,574	58.1%	95.3%
3-gram	Word	362,363	18.47	1,592,960	4.5%	12.9%
3-gram	Subword	4,157	12.02	200,891	24.0%	60.3%
4-gram	Word	1,268,078	20.27	3,340,374	2.5%	7.4%
4-gram	Subword	25,110	14.62	1,043,941	12.3%	32.8%
5-gram	Word	1,264,061	20.27	2,644,505	2.5%	7.3%
5-gram	Subword	106,026	16.69	3,000,396	7.2%	21.1%

Top 5 N-grams by Size

2-grams (Word):

Rank	N-gram	Count
1	د ی	636,166
2	تو ں	423,885
3	نو ں	352,128
4	ا ے	155,281
5	دے لئی	102,000

3-grams (Word):

Rank	N-gram	Count
1	اس د ی	29,046
2	انہاں د ی	27,692
3	انہاں نو ں	24,454
4	font size 1	24,402
5	د ی طرف	22,242

4-grams (Word):

Rank	N-gram	Count
1	ی وجہ تو ں	15,972
2	د ی وجہ تو	15,769
3	font size 1 size	9,010
4	size 1 color black	8,781
5	دے ناں تو ں	8,743

5-grams (Word):

Rank	N-gram	Count
1	د ی وجہ تو ں	15,758
2	font size 1 size 1	8,428
3	د ی طرف تو ں	7,772
4	size 1 size 1 color	6,657
5	1 size 1 color black	5,232

2-grams (Subword):

Rank	N-gram	Count
1	ے _	5,428,043
2	ی _	4,517,358
3	_ ا	4,456,935
4	_ د	3,754,809
5	ں _	3,049,226

3-grams (Subword):

Rank	N-gram	Count
1	د ے _	1,633,658
2	ا ں _	1,428,031
3	_ د ے	1,418,307
4	ت ے _	1,198,221
5	_ و چ	983,245

4-grams (Subword):

Rank	N-gram	Count
1	_ د ے _	1,415,850
2	_ و چ _	931,900
3	_ ت ے _	767,638
4	‏ ‏ ی _	616,950
5	د ‏ ‏ ی	612,667

5-grams (Subword):

Rank	N-gram	Count
1	_ د ‏ ‏ ی	612,245
2	د ‏ ‏ ی _	604,110
3	_ ت و ‏ ں	423,919
4	ت و ‏ ں _	421,873
5	و ‏ ‏ ں _	329,449

Key Findings

• Best Perplexity: 2-gram (subword) with 455
• Entropy Trend: Decreases with larger n-grams (more predictable)
• Coverage: Top-1000 patterns cover ~21% 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	0.7943	1.734	9.73	826,993	20.6%
1	Subword	0.7157	1.642	6.80	15,827	28.4%
2	Word	0.3933	1.313	2.41	8,042,106	60.7%
2	Subword	0.6609	1.581	4.50	107,498	33.9%
3	Word	0.1776	1.131	1.44	19,356,492	82.2%
3	Subword	0.6554	1.575	3.88	483,762	34.5%
4	Word	0.0843 🏆	1.060	1.16	27,821,467	91.6%
4	Subword	0.6318	1.550	3.14	1,876,646	36.8%

Generated Text Samples (Word-based)

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

Context Size 1:

1. دے مطابق لکڑی چھوٹی انتظامی ونڈ سمیں روچک کھیڈ کلب وغیرہ مسجد تعمیر کیتے اکبر خان
2. ں ختم ہوئے گئے جتھے تک نوبت وی کِسے نظام الدین جنرل دے لئی جو تپسیا
3. وچ ریک تو ں اپنے تعلق اُتے محمول منقول مشاہدہ کرنا چاہندا سی جو عورتاں د

Context Size 2:

1. د ی قدیم تریخ دی تریخ ہندستان دی ونڈ پیکنگ اوپیرا چین دا سارا دار و مدار
2. تو ں انہاں دا ناں خرزادہ سی اصل وچ کمانڈر کمیسار تے ممبر ملکاں د ی فیکٹری
3. نو ں جدید بناؤن لئی ورتے جاسکدے نیں کیلیفورنیا وچ اک آفریدی پشتون معاشرے دے رہنماواں تو

Context Size 3:

1. اس د ی معرفت کہیا گیا سی کہ علم مثلثات کوریاضی دے اک علیحدہ موضوع دے طورپرمتعارف کروائے
2. انہاں د ی نظر تو ں مضمون دے مسودہ نگاراں وچو ں اک برج د ی شناخت دا
3. انہاں نو ں عذاب خدا دے ذریعہ سزا یافتہ افراد وچو ں چار ملکہ بطور ملکہ نيں پہلی

Context Size 4:

1. ی وجہ تو ں غیر واضح نيں حالاں کہ تبت دے ناں نو ں شری حمیرہ لکھیا گیا سی
2. د ی وجہ تو ں قیدیاں نو ں قتل کر دتا فرانسیسی گورنر ڈوپلے نے مظفر جنگ کيت ی
3. font size 1 size 1 color black lonoke county arkansas font 250px دیس صوبہ ساؤتھ ڈیکوٹا راجکعر کلیر ل...

Generated Text Samples (Subword-based)

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

Context Size 1:

1. _راٹی،_اٹہہد_حمی
2. اوچ_وہنے_منی_و‏ں
3. یالأنیدے_آشدھ_مب

Context Size 2:

1. ے_علیسانہار_کيتاں
2. ی_اک_صد_نے_آخری_ش
3. _اپنے_بھ_کيتائنگ_

Context Size 3:

1. دے_لفظاں_وچ_انہ_مظ
2. اں_دار_تے_اس_گئی۔_
3. _دے_نيں_کہ_ہوری_بی

Context Size 4:

1. _دے_خیبر_پانی_ھود_ک
2. _وچ_سرکارڈ_،_क्रिस_कुलथा
3. _تے_انہاں_دے_ہور_اے

Key Findings

• Best Predictability: Context-4 (word) with 91.6% predictability
• Branching Factor: Decreases with context size (more deterministic)
• Memory Trade-off: Larger contexts require more storage (1,876,646 contexts)
• Recommendation: Context-3 or Context-4 for text generation

---

4. Vocabulary Analysis

Statistics

Metric	Value
Vocabulary Size	354,441
Total Tokens	38,365,731
Mean Frequency	108.24
Median Frequency	4
Frequency Std Dev	4606.26

Most Common Words

Rank	Word	Frequency
1	دے	1,417,871
2	ں	946,354
3	وچ	938,439
4	تے	775,429
5	ی	685,094
6	د	647,998
7	دا	502,834
8	نے	448,856
9	اے	445,649
10	تو	435,054

Least Common Words (from vocabulary)

Rank	Word	Frequency
1	گوکلے	2
2	gokula	2
3	سہورا	2
4	سنسنوار	2
5	کٹھمبر	2
6	آغر	2
7	انیردھ	2
8	imadus	2
9	چورامان	2
10	بُندیل	2

Zipf's Law Analysis

Metric	Value
Zipf Coefficient	1.1062
R² (Goodness of Fit)	0.989961
Adherence Quality	excellent

Coverage Analysis

Top N Words	Coverage
Top 100	39.9%
Top 1,000	64.4%
Top 5,000	82.1%
Top 10,000	87.9%

Key Findings

• Zipf Compliance: R²=0.9900 indicates excellent adherence to Zipf's law
• High Frequency Dominance: Top 100 words cover 39.9% of corpus
• Long Tail: 344,441 words needed for remaining 12.1% 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	0.8211 🏆	0.4072	N/A	N/A
mono_64d	64	0.8095	0.3302	N/A	N/A
mono_128d	128	0.7605	0.2826	N/A	N/A
aligned_32d	32	0.8211	0.3992	0.0680	0.2880
aligned_64d	64	0.8095	0.3176	0.1360	0.4980
aligned_128d	128	0.7605	0.2618	0.2180	0.6080

Key Findings

• Best Isotropy: mono_32d with 0.8211 (more uniform distribution)
• Semantic Density: Average pairwise similarity of 0.3331. Lower values indicate better semantic separation.
• Alignment Quality: Aligned models achieve up to 21.8% 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	-0.655	Low formulaic 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
-ال	الحرکۃ, الرُّکنِ, الجارود
-ا	اثرہويا, اورسرکشی, انورؔ
-م	مستنگ, مولاناعبدالرؤف, مرحمت
-ب	بیشکتاش, بیوکس, بانسری
-ک	کومچ, کاراگنڈا, کیبی
-س	سفین, سپردگی, سامع
-و	والصلۃ, ویلفئیر, وطواط
-پ	پرفائزتھے, پیچیدگى, پستاں

Productive Suffixes

Suffix	Examples
-ی	گھمری, کیبی, کوتای
-ں	دواخاناں, پستاں, تکبراں
-ا	کاراگنڈا, شانامتا, اثرہويا
-ن	ڈینوبیئن, سفین, ٹراجن
-اں	دواخاناں, پستاں, تکبراں
-s	uvs, hylocereus, sectors
-ر	جَور, نذير, فچنر
-ہ	آئنہ, تےحملہ, ریاضشہزادہ

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	3.07x	58 contexts	tiong, action, kition
ادشا	2.58x	40 contexts	پادشا, ادشاہ, بادشا
بادش	2.73x	27 contexts	بادشا, بادشان, بادشاہ
ھارت	2.32x	48 contexts	طھارت, دھارت, مھارت
یتاں	1.94x	74 contexts	حیتاں, گیتاں, جیتاں
مریک	2.32x	35 contexts	امریک, مریکل, مریکہ
لاقے	3.13x	12 contexts	غلاقے, علاقے, علاقےِ
ردار	1.66x	119 contexts	كردار, قردار, کردار
کومت	2.34x	28 contexts	حکومت, کومتے, ہکومت
حکوم	2.07x	43 contexts	حکومت, حکومٹ, حکومۃ
سلطن	2.35x	26 contexts	سلطنت, سلطنة, سلطنتِ
ستعم	2.21x	26 contexts	مستعمل, استعمی, ستعمال

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
-ا	-ی	59 words	ابچلی, البیرنی
-ال	-ی	41 words	البیرنی, السلیمی
-ا	-ں	40 words	ایواناں, اخواندیاں
-ا	-ا	37 words	اڈاندا, اینٹونیا
-ا	-اں	35 words	ایواناں, اخواندیاں
-م	-ی	33 words	مائکرونیشی, مرزاجانی
-ک	-ی	32 words	کابلی, کوریری
-س	-ی	32 words	سرکھائی, سنگتراشی
-ک	-ا	28 words	کانازاوا, کيتاگیاتھا
-ا	-ن	27 words	اوزگین, اکورگان

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
انقلابیان	انقلاب-ی-ان	7.5	ی
حبیریمنیا	حبیریم-ن-یا	7.5	ن
منموہنیاں	منموہن-ی-اں	7.5	ی
بناچاہندے	ب-نا-چاہندے	7.5	چاہندے
والزیارات	و-ال-زیارات	6.0	زیارات
یونانیدیس	یونانی-دی-س	6.0	یونانی
protestants	protestant-s	4.5	protestant
destinations	destination-s	4.5	destination
والانحطاط	و-الانحطاط	4.5	الانحطاط
reprinted	reprint-ed	4.5	reprint
ناخوشگوار	نا-خوشگوار	4.5	خوشگوار
بازنطینیاں	بازنطینی-اں	4.5	بازنطینی
اسماعیلاں	اسماعیل-اں	4.5	اسماعیل
respected	respect-ed	4.5	respect
اندازاًجنوب	ان-د-ازاًجنوب	4.5	ازاًجنوب

6.6 Linguistic Interpretation

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

---

7. Summary & Recommendations

Production Recommendations

Component	Recommended	Rationale
Tokenizer	64k BPE	Best compression (3.99x)
N-gram	2-gram	Lowest perplexity (455)
Markov	Context-4	Highest predictability (91.6%)
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 21:07:05