wikilangs/cu

Church Slavic - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on Church Slavic 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	3.877x	3.88	0.1314%	107,273
16k	4.367x	4.37	0.1480%	95,246
32k	4.940x 🏆	4.94	0.1675%	84,200

Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

Sample 1: Лидьскъ повѣтъ · Бѣла Роусь Лидьскъ повѣтъ · Рѡсїиска їмпєрїꙗ

Vocab	Tokens	Count
8k	▁лидьскъ ▁повѣтъ ▁· ▁бѣла ▁роусь ▁лидьскъ ▁повѣтъ ▁· ▁рѡсїиска ▁їмпєрїꙗ	10
16k	▁лидьскъ ▁повѣтъ ▁· ▁бѣла ▁роусь ▁лидьскъ ▁повѣтъ ▁· ▁рѡсїиска ▁їмпєрїꙗ	10
32k	▁лидьскъ ▁повѣтъ ▁· ▁бѣла ▁роусь ▁лидьскъ ▁повѣтъ ▁· ▁рѡсїиска ▁їмпєрїꙗ	10

Sample 2: Оꙁаскоу и · юга Санъ Паоулоу браꙁїльскъ градъ и обьщина ѥстъ ⁙ Людии 718.646 оби...

Vocab	Tokens	Count
8k	▁о ꙁа скоу ▁и ▁· ▁ю га ▁санъ ▁паоу лоу ... (+24 more)	34
16k	▁о ꙁа скоу ▁и ▁· ▁ю га ▁санъ ▁паоулоу ▁браꙁїл ... (+23 more)	33
32k	▁оꙁаскоу ▁и ▁· ▁юга ▁санъ ▁паоулоу ▁браꙁїльскъ ▁градъ ▁и ▁обьщина ... (+19 more)	29

Sample 3: Октадєканъ и инако н-октадєканъ ѫглѥводородьно вєщьство алканъ рѧдоу ѥстъ ⁙ Ѥгож...

Vocab	Tokens	Count
8k	▁ок тадєканъ ▁и ▁инако ▁н - ок тадєканъ ▁ѫглѥводородьно ▁вєщьство ... (+19 more)	29
16k	▁октадєканъ ▁и ▁инако ▁н - ок тадєканъ ▁ѫглѥводородьно ▁вєщьство ▁алканъ ... (+17 more)	27
32k	▁октадєканъ ▁и ▁инако ▁н - октадєканъ ▁ѫглѥводородьно ▁вєщьство ▁алканъ ▁рѧдоу ... (+16 more)	26

Key Findings

• Best Compression: 32k achieves 4.940x compression
• Lowest UNK Rate: 8k with 0.1314% 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	802	9.65	1,417	38.7%	88.9%
2-gram	Subword	451 🏆	8.82	2,622	56.3%	95.5%
3-gram	Word	965	9.91	1,734	35.4%	82.3%
3-gram	Subword	2,629	11.36	12,286	25.7%	67.4%
4-gram	Word	1,583	10.63	2,960	29.4%	67.1%
4-gram	Subword	8,218	13.00	33,187	16.1%	45.2%
5-gram	Word	1,176	10.20	2,224	32.9%	74.0%
5-gram	Subword	14,289	13.80	46,031	12.7%	35.8%

Top 5 N-grams by Size

2-grams (Word):

Rank	N-gram	Count
1	ꙁьри такождє	432
2	людии обитаѥтъ	260
3	ѥстъ людии	234
4	градъ ѥстъ	230
5	стольнъ градъ	186

3-grams (Word):

Rank	N-gram	Count
1	ѥстъ людии обитаѥтъ	181
2	дрьжавѣ бѣла роусь	120
3	въ дрьжавѣ бѣла	120
4	градъ ѥстъ людии	115
5	бѣла роусь сѣи	114

4-grams (Word):

Rank	N-gram	Count
1	въ дрьжавѣ бѣла роусь	120
2	дрьжавѣ бѣла роусь сѣи	114
3	оудѣлъ въ дрьжавѣ бѣла	114
4	ꙁємьскъ оудѣлъ въ дрьжавѣ	114
5	бѣла роусь сѣи оудѣлъ	114

5-grams (Word):

Rank	N-gram	Count
1	роусь сѣи оудѣлъ бѣ члѣнъ	114
2	ꙁємьскъ оудѣлъ въ дрьжавѣ бѣла	114
3	оудѣлъ въ дрьжавѣ бѣла роусь	114
4	бѣла роусь сѣи оудѣлъ бѣ	114
5	дрьжавѣ бѣла роусь сѣи оудѣлъ	114

2-grams (Subword):

Rank	N-gram	Count
1	ъ _	17,697
2	и _	9,192
3	а _	8,589
4	с т	8,369
5	_ с	6,568

3-grams (Subword):

Rank	N-gram	Count
1	т ъ _	5,939
2	_ · _	4,413
3	ь с к	3,883
4	_ ⁙ _	3,094
5	с т ъ	3,038

4-grams (Subword):

Rank	N-gram	Count
1	_ ѥ с т	2,895
2	с т ъ _	2,876
3	ѥ с т ъ	2,698
4	ъ _ ⁙ _	1,902
5	т ъ _ ⁙	1,813

5-grams (Subword):

Rank	N-gram	Count
1	_ ѥ с т ъ	2,695
2	ѥ с т ъ _	2,559
3	т ъ _ ⁙ _	1,796
4	_ г р а д	1,425
5	с т ъ _ ⁙	1,340

Key Findings

• Best Perplexity: 2-gram (subword) with 451
• Entropy Trend: Decreases with larger n-grams (more predictable)
• Coverage: Top-1000 patterns cover ~36% 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	0.4863	1.401	2.62	18,746	51.4%
1	Subword	0.9940	1.992	7.09	1,077	0.6%
2	Word	0.1229	1.089	1.22	48,473	87.7%
2	Subword	0.8201	1.766	4.18	7,633	18.0%
3	Word	0.0444	1.031	1.07	58,365	95.6%
3	Subword	0.5514	1.466	2.43	31,900	44.9%
4	Word	0.0207 🏆	1.014	1.03	61,255	97.9%
4	Subword	0.3387	1.265	1.70	77,420	66.1%

Generated Text Samples (Word-based)

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

Context Size 1:

1. и ꙁападьнꙑ дъвинꙑ роуси пьсаниꙗ алєѯандра данїиловища свѣтьлѣиша кънѧꙃа владєнию бѣ съ словѣньскомь ...
2. ѥстъ стольнъ градъ ѥстъ нѣмьцкомь єпископомь албєртомь а нꙑнѣ жє носьнꙑи приꙁвѫкъ нє ꙁнаашє ѥдьнъ ис
3. лѣта їмпєратѡръ ѥстъ пєроунъ сварогъ ѩꙁꙑчьство

Context Size 2:

1. ꙁьри такождє обитѣльско напьсаниѥ владиславъ їѡаннъ асєн҄ь а҃ и блъгарїꙗ цѣсарь бѣ їѡанна асєнꙗ а҃ с...
2. людии обитаѥтъ 6 9 лєѡ́дръ їсторїꙗ лѣта по нѣмьць ѥдьнѥниꙗ бєрлинъ пакꙑ сталъ ѥстъ ꙁьри такъждє брюѯ...
3. ѥстъ людии 2 лєѡдръ обитаѥтъ пакистана дрьжавьнъ ѩꙁꙑкъ тѷрчьскъ ѥстъ їсторїꙗ дѣлꙗ охранꙑ съдравиꙗ лѣ...

Context Size 3:

1. ѥстъ людии обитаѥтъ 398 и иꙁъ ихъжє мѫжь 175 и жєнъ 223 наибол҄ии числомь народъ роусьсци ѥстъ 99
2. въ дрьжавѣ бѣла роусь сѣи оудѣлъ бѣ члѣнъ ѡбласти рѣкома могилєвьска ѡбласть повѣтъ имаѥтъ оурѧдъ рѣ...
3. дрьжавѣ бѣла роусь сѣи оудѣлъ бѣ члѣнъ градоу витьбьскъ въ ѡбласти рѣкома мѣньска ѡбласть повѣтъ има...

Context Size 4:

1. въ дрьжавѣ бѣла роусь сѣи оудѣлъ бѣ члѣнъ ѡбласти рѣкома бєрєстєиска ѡбласть повѣтъ имаѥтъ оурѧдъ рѣ...
2. роусь сѣи оудѣлъ бѣ члѣнъ ѡбласти рѣкома мѣньска ѡбласть повѣтъ имаѥтъ оурѧдъ рѣкомъ сєльскъ съвѣтъ ...
3. бѣла роусь сѣи оудѣлъ бѣ члѣнъ ѡбласти рѣкома витєбьска ѡбласть повѣтъ имаѥтъ оурѧдъ рѣкомъ сєльскъ ...

Generated Text Samples (Subword-based)

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

Context Size 1:

1. _иното_ѥгонокꙑ_ѥ
2. а_одаскꙑ_сточлѣс
3. орлѩꙁа_гокє_ꙁꙑ_с

Context Size 2:

1. ъ_обирѡсьскꙑ_рѣвь
2. и_•_всєли_·_рѡпьс
3. а_посладъпрꙗѥтъ_ꙁ

Context Size 3:

1. тъ_⁙_глаголєптємпє
2. _·_дѣлъ_бѣлороусло
3. ьскъ_прьвовец_гора

Context Size 4:

1. _ѥстъ_·_мїхаилъ_хоу
2. стъ_словєниꙗ_мѫжь_с
3. ѥстъ_⁙_глагоданьска

Key Findings

• Best Predictability: Context-4 (word) with 97.9% predictability
• Branching Factor: Decreases with context size (more deterministic)
• Memory Trade-off: Larger contexts require more storage (77,420 contexts)
• Recommendation: Context-3 or Context-4 for text generation

---

4. Vocabulary Analysis

Statistics

Metric	Value
Vocabulary Size	6,189
Total Tokens	62,865
Mean Frequency	10.16
Median Frequency	3
Frequency Std Dev	60.08

Most Common Words

Rank	Word	Frequency
1	и	2,821
2	ѥстъ	2,694
3	лѣта	952
4	бѣ	910
5	въ	842
6	градъ	792
7	ꙁьри	536
8	такождє	533
9	жє	512
10	людии	470

Least Common Words (from vocabulary)

Rank	Word	Frequency
1	катєгорїꙗ	2
2	سخ	2
3	هس	2
4	ش	2
5	ؤخخم	2
6	خىث	2
7	ىعةلاثق	2
8	صشس	2
9	пльсковьская	2
10	маѭтъ	2

Zipf's Law Analysis

Metric	Value
Zipf Coefficient	0.9373
R² (Goodness of Fit)	0.986343
Adherence Quality	excellent

Coverage Analysis

Top N Words	Coverage
Top 100	41.0%
Top 1,000	72.8%
Top 5,000	96.2%
Top 10,000	0.0%

Key Findings

• Zipf Compliance: R²=0.9863 indicates excellent adherence to Zipf's law
• High Frequency Dominance: Top 100 words cover 41.0% of corpus
• Long Tail: -3,811 words needed for remaining 100.0% 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	0.2434	0.4441	N/A	N/A
mono_64d	64	0.0769	0.4495	N/A	N/A
mono_128d	128	0.0128	0.4700	N/A	N/A
aligned_32d	32	0.2434 🏆	0.4485	0.0177	0.1032
aligned_64d	64	0.0769	0.4699	0.0324	0.1475
aligned_128d	128	0.0128	0.4554	0.0442	0.1357

Key Findings

• Best Isotropy: aligned_32d with 0.2434 (more uniform distribution)
• Semantic Density: Average pairwise similarity of 0.4562. Lower values indicate better semantic separation.
• Alignment Quality: Aligned models achieve up to 4.4% 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	1.066	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
-по	поѩла, погꙑнѫли, польꙃєвати
-пр	прєждє, придънѣстрии, прасловѣньскъ

Productive Suffixes

Suffix	Examples
-ъ	въꙁвращєнъ, дѣлъ, ѳєрапѡнтъ
-къ	липьтьскъ, грьчьскъ, словѣньскъ
-нъ	въꙁвращєнъ, гла́вьнъ, съꙁиждєнъ
-ка	кировьска, фроунꙁєньска, видодъска
-скъ	липьтьскъ, грьчьскъ, словѣньскъ
-ска	кировьска, фроунꙁєньска, видодъска
-ьска	кировьска, фроунꙁєньска, городєньска
-кꙑ	блъгарьскꙑ, хръватьскꙑ, словѣньскꙑ

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
боук	1.89x	14 contexts	боукꙑ, боуквꙑ, боукъвь
ловѣ	1.63x	18 contexts	словѣ, чловѣкъ, словѣнє
слов	1.77x	14 contexts	слово, словѣ, слова
ласт	1.55x	20 contexts	властъ, власть, власти
ьжав	1.75x	13 contexts	дрьжавꙑ, дрьжавъ, дрьжавѫ
ньск	1.65x	15 contexts	мѣньска, мѣньскъ, жєньскъ
ьска	1.64x	14 contexts	омьска, єстьска, сѣрьска
овѣн	1.83x	10 contexts	словѣнє, словѣнъ, словѣнїꙗ
град	1.63x	13 contexts	градѣ, градъ, гради
блас	1.69x	10 contexts	ѡбласти, области, ѡбласть
ьскъ	1.63x	11 contexts	омьскъ, римьскъ, ꙁємьскъ
рьжа	1.69x	9 contexts	дрьжавꙑ, дрьжавъ, дрьжавѫ

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
-по	-ъ	34 words	побѣдъ, помѣновєнъ
-пр	-ъ	34 words	прьвꙑимъ, проливъ
-по	-нъ	11 words	помѣновєнъ, посъланъ
-по	-ка	7 words	подъкарпатьска, по́л̑ьска
-по	-къ	7 words	подъбрадъкъ, подълѣсьскъ
-по	-скъ	6 words	подълѣсьскъ, пол҄ьскъ
-пр	-нъ	6 words	прѣданъ, природьнъ
-пр	-къ	6 words	приморьскъ, прьвотравєньскъ
-по	-ска	5 words	подъкарпатьска, по́л̑ьска
-по	-ьскъ	5 words	подълѣсьскъ, пол҄ьскъ

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
гєѡргїиска	гєѡргїи-ска	4.5	гєѡргїи
посєлѥниѥ	по-сєлѥниѥ	4.5	сєлѥниѥ
октѡврїиска	октѡврїи-ска	4.5	октѡврїи
посєлѥниꙗ	по-сєлѥниꙗ	4.5	сєлѥниꙗ
самостоꙗтєл҄ьна	самостоꙗтєл҄ь-на	4.5	самостоꙗтєл҄ь
аѵстралїиска	аѵстралїи-ска	4.5	аѵстралїи
самостоꙗтѣльна	самостоꙗтѣль-на	4.5	самостоꙗтѣль
аѵстрїискъ	аѵстрїи-скъ	4.5	аѵстрїи
приморьскъ	пр-имор-ьскъ	3.0	имор
подольскъ	по-доль-скъ	3.0	доль
полїтїчьскъ	по-лїтїч-ьскъ	3.0	лїтїч
подъꙁємьнъ	по-дъꙁємь-нъ	3.0	дъꙁємь
прѣѥмьникъ	пр-ѣѥмьни-къ	3.0	ѣѥмьни
потрѣбьна	по-трѣбь-на	3.0	трѣбь
политическа	по-литиче-ска	3.0	литиче

6.6 Linguistic Interpretation

Automated Insight: The language Church Slavic 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	32k BPE	Best compression (4.94x)
N-gram	2-gram	Lowest perplexity (451)
Markov	Context-4	Highest predictability (97.9%)
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: 2026-01-03 20:59:44