South Azerbaijani - Wikilangs Models
Comprehensive Research Report & Full Ablation Study
This repository contains NLP models trained and evaluated by Wikilangs, specifically on South Azerbaijani 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 | 3.140x | 3.14 | 0.4916% | 361,073 |
| 16k | 3.514x | 3.52 | 0.5501% | 322,683 |
| 32k | 3.859x | 3.86 | 0.6041% | 293,823 |
| 64k | 4.154x 🏆 | 4.16 | 0.6502% | 272,975 |
Tokenization Examples
Below are sample sentences tokenized with each vocabulary size:
Sample 1: پوچتوووی ( ) روسیه اؤلکهسینده یئر آلان بیر کند دیر و آرخانقلسک اوبلاستیندا یئرل...
| Vocab | Tokens | Count |
|---|---|---|
| 8k | ▁پو چ ت وو وی ▁( ▁) ▁روسیه ▁اؤلکه ▁سینده ... (+12 more) |
22 |
| 16k | ▁پو چ ت وو وی ▁( ▁) ▁روسیه ▁اؤلکه ▁سینده ... (+12 more) |
22 |
| 32k | ▁پو چت وووی ▁( ▁) ▁روسیه ▁اؤلکه ▁سینده ▁یئر ▁آلان ... (+10 more) |
20 |
| 64k | ▁پو چت وووی ▁( ▁) ▁روسیه ▁اؤلکه ▁سینده ▁یئر ▁آلان ... (+10 more) |
20 |
Sample 2: هیندوستان اؤلکهسینین کارناتاکا ایالتینده بیر کند دیر. بۇ کنده کانادا دیلی دانیش...
| Vocab | Tokens | Count |
|---|---|---|
| 8k | ▁هیندوستان ▁اؤلکه ▁سینین ▁کارناتاکا ▁ایالتینده ▁بیر ▁کند ▁دیر . ▁بۇ ... (+7 more) |
17 |
| 16k | ▁هیندوستان ▁اؤلکه ▁سینین ▁کارناتاکا ▁ایالتینده ▁بیر ▁کند ▁دیر . ▁بۇ ... (+7 more) |
17 |
| 32k | ▁هیندوستان ▁اؤلکه ▁سینین ▁کارناتاکا ▁ایالتینده ▁بیر ▁کند ▁دیر . ▁بۇ ... (+7 more) |
17 |
| 64k | ▁هیندوستان ▁اؤلکه ▁سینین ▁کارناتاکا ▁ایالتینده ▁بیر ▁کند ▁دیر . ▁بۇ ... (+7 more) |
17 |
Sample 3: پیایو، روسیه ( ) روسیه اؤلکهسینده یئر آلان بیر کند دیر و مورمانسک اوبلاستیندا ی...
| Vocab | Tokens | Count |
|---|---|---|
| 8k | ▁پی ای و ، ▁روسیه ▁( ▁) ▁روسیه ▁اؤلکه ▁سینده ... (+14 more) |
24 |
| 16k | ▁پی ای و ، ▁روسیه ▁( ▁) ▁روسیه ▁اؤلکه ▁سینده ... (+12 more) |
22 |
| 32k | ▁پی ایو ، ▁روسیه ▁( ▁) ▁روسیه ▁اؤلکه ▁سینده ▁یئر ... (+11 more) |
21 |
| 64k | ▁پی ایو ، ▁روسیه ▁( ▁) ▁روسیه ▁اؤلکه ▁سینده ▁یئر ... (+11 more) |
21 |
Key Findings
- Best Compression: 64k achieves 4.154x compression
- Lowest UNK Rate: 8k with 0.4916% 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 | 8,053 | 12.98 | 158,968 | 25.7% | 56.1% |
| 2-gram | Subword | 528 🏆 | 9.04 | 12,667 | 51.7% | 95.7% |
| 3-gram | Word | 10,252 | 13.32 | 236,817 | 22.6% | 53.6% |
| 3-gram | Subword | 3,765 | 11.88 | 106,797 | 23.2% | 62.4% |
| 4-gram | Word | 17,203 | 14.07 | 427,241 | 19.0% | 47.9% |
| 4-gram | Subword | 15,109 | 13.88 | 582,040 | 14.6% | 44.8% |
| 5-gram | Word | 19,665 | 14.26 | 390,296 | 17.2% | 45.0% |
| 5-gram | Subword | 37,921 | 15.21 | 1,605,166 | 11.7% | 37.8% |
Top 5 N-grams by Size
2-grams (Word):
| Rank | N-gram | Count |
|---|---|---|
| 1 | ایشلدنلری طرفیندن |
75,586 |
| 2 | مقالهسیندن گؤتورولوبدور |
75,505 |
| 3 | ویکیپدیاسینین ایشلدنلری |
73,736 |
| 4 | اینگیلیسجه ویکیپدیاسینین |
71,134 |
| 5 | قایناقلار اینگیلیسجه |
70,887 |
3-grams (Word):
| Rank | N-gram | Count |
|---|---|---|
| 1 | ویکیپدیاسینین ایشلدنلری طرفیندن |
73,736 |
| 2 | اینگیلیسجه ویکیپدیاسینین ایشلدنلری |
71,134 |
| 3 | قایناقلار اینگیلیسجه ویکیپدیاسینین |
70,813 |
| 4 | بیر یاشاییش منطقهسیدیر |
40,398 |
| 5 | بیر کند دیر |
30,448 |
4-grams (Word):
| Rank | N-gram | Count |
|---|---|---|
| 1 | اینگیلیسجه ویکیپدیاسینین ایشلدنلری طرفیندن |
71,134 |
| 2 | قایناقلار اینگیلیسجه ویکیپدیاسینین ایشلدنلری |
70,813 |
| 3 | سوْن نۆفوس ساییمی اساسيندا |
24,567 |
| 4 | شهرلرین لیستی قایناقلار اینگیلیسجه |
22,936 |
| 5 | لیستی قایناقلار اینگیلیسجه ویکیپدیاسینین |
22,936 |
5-grams (Word):
| Rank | N-gram | Count |
|---|---|---|
| 1 | قایناقلار اینگیلیسجه ویکیپدیاسینین ایشلدنلری طرفیندن |
70,813 |
| 2 | شهرلرین لیستی قایناقلار اینگیلیسجه ویکیپدیاسینین |
22,936 |
| 3 | لیستی قایناقلار اینگیلیسجه ویکیپدیاسینین ایشلدنلری |
22,936 |
| 4 | گؤتورولوبدور ۸ آقوست تاریخینده یوْخلانیلیبدیر |
17,804 |
| 5 | مقالهسیندن گؤتورولوبدور ۸ آقوست تاریخینده |
17,804 |
2-grams (Subword):
| Rank | N-gram | Count |
|---|---|---|
| 1 | ی ن |
1,870,834 |
| 2 | _ ا |
1,661,040 |
| 3 | ی _ |
1,438,867 |
| 4 | ا ی |
1,394,440 |
| 5 | ن _ |
1,218,001 |
3-grams (Subword):
| Rank | N-gram | Count |
|---|---|---|
| 1 | _ ا ی |
718,299 |
| 2 | ی ن د |
659,203 |
| 3 | د ه _ |
586,025 |
| 4 | ل ا ر |
581,126 |
| 5 | ا ی ن |
470,804 |
4-grams (Subword):
| Rank | N-gram | Count |
|---|---|---|
| 1 | ن د ه _ |
347,491 |
| 2 | ل ا ر _ |
329,654 |
| 3 | ی ن د ه |
321,093 |
| 4 | _ ب ی ر |
258,934 |
| 5 | ن ی ن _ |
257,847 |
5-grams (Subword):
| Rank | N-gram | Count |
|---|---|---|
| 1 | ی ن د ه _ |
319,618 |
| 2 | ق ا ی ن ا |
236,232 |
| 3 | _ ق ا ی ن |
235,936 |
| 4 | ی ن د ن _ |
199,409 |
| 5 | ی ن گ ی ل |
172,619 |
Key Findings
- Best Perplexity: 2-gram (subword) with 528
- Entropy Trend: Decreases with larger n-grams (more predictable)
- Coverage: Top-1000 patterns cover ~38% 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 | 0.6633 | 1.584 | 5.08 | 728,851 | 33.7% |
| 1 | Subword | 1.0599 | 2.085 | 9.06 | 3,419 | 0.0% |
| 2 | Word | 0.1969 | 1.146 | 1.48 | 3,698,118 | 80.3% |
| 2 | Subword | 0.9299 | 1.905 | 6.57 | 30,973 | 7.0% |
| 3 | Word | 0.0689 | 1.049 | 1.14 | 5,453,316 | 93.1% |
| 3 | Subword | 0.8407 | 1.791 | 4.70 | 203,342 | 15.9% |
| 4 | Word | 0.0340 🏆 | 1.024 | 1.07 | 6,184,673 | 96.6% |
| 4 | Subword | 0.6999 | 1.624 | 3.22 | 955,119 | 30.0% |
Generated Text Samples (Word-based)
Below are text samples generated from each word-based Markov chain model:
Context Size 1:
و اؤلوملر باخیر دؤولت موزیک دوْغوملار ۲ مئی eamon gilmore shooter young artist awardbest breakthroug...بیر فوتبالیست هوجومچو موقعیتینده اوْیناییب قایناقلار ایلده آمریکالی سیاستچیلر میلادی ایلده آذربایجان...اینگیلیسجه phyllis and the new york آمریکانین نبراسکا ایالتینده بیر شهردیر و باتی آجورلو قصبهسینده ...
Context Size 2:
ایشلدنلری طرفیندن ballard مقالهسیندن گؤتورولوبدور ۳۰ نوْوامبر تاریخینده رالی قوزئی کارولینا ایالتین...مقالهسیندن گؤتورولوبدور ۲۲ آقوست تاریخینده یوْخلانیلیبدیر ایالتین شهرلری آمریکا بیرلشمیش ایالتلری ک...ویکیپدیاسینین ایشلدنلری طرفیندن piguet مقالهسیندن گؤتورولوبدور ۱۹ جولای یوْخلانیلیبدیر شهرلری en ...
Context Size 3:
ویکیپدیاسینین ایشلدنلری طرفیندن phosphate مقالهسیندن گؤتورولوبدور ۳۰ نوْوامبر تاریخینده یوْخلانیل...اینگیلیسجه ویکیپدیاسینین ایشلدنلری طرفیندن mała مقالهسیندن گؤتورولوبدور ۱۲ آقوست تاریخینده یوْخلا...قایناقلار اینگیلیسجه ویکیپدیاسینین ایشلدنلری طرفیندن nigra مقالهسیندن گؤتورولوبدور ۲۷ جولای تاری...
Context Size 4:
اینگیلیسجه ویکیپدیاسینین ایشلدنلری طرفیندن tachov district مقالهسیندن گؤتورولوبدور ۱۹ جولای یوْخل...قایناقلار اینگیلیسجه ویکیپدیاسینین ایشلدنلری طرفیندن reed مقالهسیندن گؤتورولوبدور ۲۲ ژانویه تاری...سوْن نۆفوس ساییمی اساسيندا نفر ایمیش قایناقلار جومهوریتینین شهرلری en bədəlan
Generated Text Samples (Subword-based)
Below are text samples generated from each subword-based Markov chain model:
Context Size 1:
_طین_اینالرشه_میینینویره_ب.st_آذاؤلیلده_s_مول_کل
Context Size 2:
ینی_اوبونیرلرین_ش_این_چاری_اوربّع_دی_حؤکواءنینهسیناق
Context Size 3:
_ایشتیرامبر_charahینده_یئرلشیرکت_()_ده_یوْخلو_"_the_ism
Context Size 4:
نده_یئرلشیر._بۇ_شهرلار_اینسانی._۲۴_آقوینده_چیخماق_شکیلات)
Key Findings
- Best Predictability: Context-4 (word) with 96.6% predictability
- Branching Factor: Decreases with context size (more deterministic)
- Memory Trade-off: Larger contexts require more storage (955,119 contexts)
- Recommendation: Context-3 or Context-4 for text generation
4. Vocabulary Analysis
Statistics
| Metric | Value |
|---|---|
| Vocabulary Size | 271,726 |
| Total Tokens | 12,485,100 |
| Mean Frequency | 45.95 |
| Median Frequency | 3 |
| Frequency Std Dev | 1144.86 |
Most Common Words
| Rank | Word | Frequency |
|---|---|---|
| 1 | و | 284,866 |
| 2 | بیر | 169,436 |
| 3 | اینگیلیسجه | 149,744 |
| 4 | قایناقلار | 142,037 |
| 5 | the | 114,223 |
| 6 | تاریخینده | 92,091 |
| 7 | قایناقلار | 90,964 |
| 8 | ایلده | 83,776 |
| 9 | شهرلری | 81,908 |
| 10 | طرفیندن | 80,193 |
Least Common Words (from vocabulary)
| Rank | Word | Frequency |
|---|---|---|
| 1 | ائششکین | 2 |
| 2 | لابیسی | 2 |
| 3 | آذریها | 2 |
| 4 | داشناکلارلا | 2 |
| 5 | قۇلان | 2 |
| 6 | آسینۇس | 2 |
| 7 | ائششهیینین | 2 |
| 8 | تاپؽلمیشدیر | 2 |
| 9 | kulan | 2 |
| 10 | کسا | 2 |
Zipf's Law Analysis
| Metric | Value |
|---|---|
| Zipf Coefficient | 1.1608 |
| R² (Goodness of Fit) | 0.995522 |
| Adherence Quality | excellent |
Coverage Analysis
| Top N Words | Coverage |
|---|---|
| Top 100 | 34.4% |
| Top 1,000 | 64.8% |
| Top 5,000 | 79.6% |
| Top 10,000 | 84.6% |
Key Findings
- Zipf Compliance: R²=0.9955 indicates excellent adherence to Zipf's law
- High Frequency Dominance: Top 100 words cover 34.4% of corpus
- Long Tail: 261,726 words needed for remaining 15.4% 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 | 0.8266 🏆 | 0.3562 | N/A | N/A |
| mono_64d | 64 | 0.7978 | 0.2932 | N/A | N/A |
| mono_128d | 128 | 0.7560 | 0.2495 | N/A | N/A |
| aligned_32d | 32 | 0.8266 | 0.3594 | 0.0580 | 0.2760 |
| aligned_64d | 64 | 0.7978 | 0.3041 | 0.1220 | 0.4360 |
| aligned_128d | 128 | 0.7560 | 0.2442 | 0.2380 | 0.6200 |
Key Findings
- Best Isotropy: mono_32d with 0.8266 (more uniform distribution)
- Semantic Density: Average pairwise similarity of 0.3011. Lower values indicate better semantic separation.
- Alignment Quality: Aligned models achieve up to 23.8% 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 | 0.190 | 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 |
|---|
Productive Suffixes
| Suffix | Examples |
|---|---|
-ین |
کومیتهسینین, قورانین, پاقلئنین |
-ان |
قافقازدان, آتاسیندان, تیتانلاردان |
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 |
|---|---|---|---|
رلری |
2.05x | 205 contexts | یرلری, ارلری, خطرلری |
اقلا |
1.90x | 131 contexts | ناقلا, اقلایس, داقلاس |
یبدی |
2.33x | 31 contexts | ییبدیر, گلیبدی, آلیبدی |
قلار |
2.01x | 54 contexts | حقلاری, لیقلار, حاقلار |
اریخ |
2.11x | 41 contexts | تاریخ, تاریخ, تواریخ |
ولوب |
1.85x | 60 contexts | اولوب, قولوب, بولوب |
تیند |
1.73x | 73 contexts | تینده, تیندل, تیندال |
یناق |
2.07x | 27 contexts | ایناق, قیناق, سیناق |
ئرلش |
2.13x | 24 contexts | یئرلشن, يئرلشن, یئرلشه |
ریخی |
2.00x | 22 contexts | مریخی, ریخین, مریخین |
قاین |
2.31x | 14 contexts | قاینا, قاینی, قاینز |
هرلر |
2.14x | 17 contexts | شهرلر, شهرلري, شهرلری |
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.
No significant affix co-occurrences detected.
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 |
|---|---|---|---|
| دۆکانلارینین | دۆکانلار-ین-ین |
6.0 | دۆکانلار |
| سیلاحینین | سیلاح-ین-ین |
6.0 | سیلاح |
| باشچیلارینین | باشچیلار-ین-ین |
6.0 | باشچیلار |
| دوخالارین | دوخالار-ین |
4.5 | دوخالار |
| آمارلارین | آمارلار-ین |
4.5 | آمارلار |
| تاریخچیلرین | تاریخچیلر-ین |
4.5 | تاریخچیلر |
| اؤدوللرین | اؤدوللر-ین |
4.5 | اؤدوللر |
| شکیلچیلرین | شکیلچیلر-ین |
4.5 | شکیلچیلر |
| کوْمونیستلرین | کوْمونیستلر-ین |
4.5 | کوْمونیستلر |
| بیوفیزیکین | بیوفیزیک-ین |
4.5 | بیوفیزیک |
| تاپینتیلارین | تاپینتیلار-ین |
4.5 | تاپینتیلار |
| میکروبلارین | میکروبلار-ین |
4.5 | میکروبلار |
| تیکیلینین | تیکیل-ین-ین |
3.0 | تیکیل |
| نفتالینین | نفتال-ین-ین |
3.0 | نفتال |
| والنتاینین | والنتا-ین-ین |
3.0 | والنتا |
6.6 Linguistic Interpretation
Automated Insight: The language South Azerbaijani 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 (4.15x) |
| N-gram | 2-gram | Lowest perplexity (528) |
| Markov | Context-4 | Highest predictability (96.6%) |
| 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
- Compare within model families: Metrics are most meaningful when comparing models of the same type (e.g., 8k vs 64k tokenizer).
- Consider trade-offs: Better performance on one metric often comes at the cost of another (e.g., compression vs. OOV rate).
- Context matters: Optimal values depend on downstream tasks. Text generation may prioritize different metrics than classification.
- Corpus influence: All metrics are influenced by corpus characteristics. Wikipedia text differs from social media or literature.
- 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
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 19:16:27