ANN - Wikilangs Models
Comprehensive Research Report & Full Ablation Study
This repository contains NLP models trained and evaluated by Wikilangs, specifically on ANN Wikipedia data. We analyze tokenizers, n-gram models, Markov chains, vocabulary statistics, and word embeddings.
π Repository Contents
Models & Assets
- Tokenizers (8k, 16k, 32k, 64k)
- N-gram models (2, 3, 4-gram)
- Markov chains (context of 1, 2, 3 and 4)
- Subword N-gram and Markov chains
- Embeddings in various sizes and dimensions
- 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. Summary & Recommendations
- Metrics Glossary
- Visualizations Index
1. Tokenizer Evaluation
Results
| Vocab Size | Compression | Avg Token Len | UNK Rate | Total Tokens |
|---|---|---|---|---|
| 8k | 4.015x | 4.00 | 0.1279% | 136,774 |
| 16k | 4.268x π | 4.25 | 0.1360% | 128,655 |
Tokenization Examples
Below are sample sentences tokenized with each vocabulary size:
Sample 1: `Nα»wΓ¨ Γ¬re ido me Yurop.
150px|thumb|ImanΜ-ido Nα»wΓ¨
α»gbα»nΜ:Ido α»gbα»nΜ:Yurop`
| Vocab | Tokens | Count |
|---|---|---|
| 8k | βnα»wΓ¨ βΓ¬re βido βme βyurop . β 1 5 0 ... (+14 more) |
24 |
| 16k | βnα»wΓ¨ βΓ¬re βido βme βyurop . β 1 5 0 ... (+14 more) |
24 |
Sample 2: α»rα»n Γ¬re mkpulu-ija ge Γ²kjp me AganΜ Mkpulu Akwa Ibom me ido Naijiria. Mkpulu-ij...
| Vocab | Tokens | Count |
|---|---|---|
| 8k | βα»rα»n βΓ¬re βmkpulu - ija βge βΓ²k j p βme ... (+15 more) |
25 |
| 16k | βα»rα»n βΓ¬re βmkpulu - ija βge βΓ²kjp βme βaganΜ βmkpulu ... (+13 more) |
23 |
Sample 3: `Ikpα»ΜnΜ Γ¬re mfut uko Γ²kitibi mΓ¨ ito lek me emen ijα»nΜ. Γre mfut eyi acha ge.
α»...`
| Vocab | Tokens | Count |
|---|---|---|
| 8k | βikpα»ΜnΜ βΓ¬re βmfut βuko βΓ²kitibi βmΓ¨ βito βlek βme βemen ... (+13 more) |
23 |
| 16k | βikpα»ΜnΜ βΓ¬re βmfut βuko βΓ²kitibi βmΓ¨ βito βlek βme βemen ... (+13 more) |
23 |
Key Findings
- Best Compression: 16k achieves 4.268x compression
- Lowest UNK Rate: 8k with 0.1279% 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 | Perplexity | Entropy | Unique N-grams | Top-100 Coverage | Top-1000 Coverage |
|---|---|---|---|---|---|
| 2-gram | 1,256 π | 10.29 | 3,472 | 37.4% | 75.8% |
| 2-gram | 253 π | 7.98 | 1,421 | 67.5% | 99.4% |
| 3-gram | 2,655 | 11.37 | 5,764 | 25.2% | 58.7% |
| 3-gram | 1,454 | 10.51 | 7,986 | 32.8% | 79.6% |
| 4-gram | 4,959 | 12.28 | 9,244 | 18.4% | 44.6% |
| 4-gram | 4,906 | 12.26 | 25,888 | 20.8% | 56.1% |
Top 5 N-grams by Size
2-grams:
| Rank | N-gram | Count |
|---|---|---|
| 1 | agan Μ |
1,895 |
| 2 | me lek |
1,071 |
| 3 | me agan |
841 |
| 4 | me emen |
793 |
| 5 | ijα»n Μ |
694 |
3-grams:
| Rank | N-gram | Count |
|---|---|---|
| 1 | me agan Μ |
831 |
| 2 | Μ - mkpulu |
373 |
| 3 | agan Μ - |
372 |
| 4 | Μ a me |
371 |
| 5 | α»gbα»n Μ : |
339 |
4-grams:
| Rank | N-gram | Count |
|---|---|---|
| 1 | agan Μ - mkpulu |
364 |
| 2 | agan Μ inyα»n Μ |
302 |
| 3 | Μ ichep - ura |
255 |
| 4 | Μ mbum - ura |
235 |
| 5 | . α»gbα»n Μ : |
233 |
Key Findings
- Best Perplexity: 2-gram with 253
- Entropy Trend: Decreases with larger n-grams (more predictable)
- Coverage: Top-1000 patterns cover ~56% of corpus
- Recommendation: 4-gram or 5-gram for best predictive performance
3. Markov Chain Evaluation
Results
| Context | Avg Entropy | Perplexity | Branching Factor | Unique Contexts | Predictability |
|---|---|---|---|---|---|
| 1 | 0.6614 | 1.582 | 4.53 | 10,250 | 33.9% |
| 1 | 1.2658 | 2.405 | 9.65 | 293 | 0.0% |
| 2 | 0.3064 | 1.237 | 1.78 | 46,339 | 69.4% |
| 2 | 1.1141 | 2.165 | 5.65 | 2,827 | 0.0% |
| 3 | 0.1431 | 1.104 | 1.26 | 82,397 | 85.7% |
| 3 | 0.7678 | 1.703 | 3.09 | 15,977 | 23.2% |
| 4 | 0.0688 π | 1.049 | 1.11 | 104,013 | 93.1% |
| 4 | 0.4695 π | 1.385 | 1.96 | 49,379 | 53.0% |
Generated Text Samples
Below are text samples generated from each Markov chain model:
Context Size 1:
Μ mbum - paeonia mΓ¨ ebot mΓ¨ Γ¬re anam oron - enenwaan Μ sa me achame lek keya serum ; mΓ¨ ikaan Μ inu ike ikpa ebi kΓ¨ ido ya me. erieen Μ igege inu kire ama ebilene ; α»mα» Γ¬luk me < emirate ] eyi
Context Size 2:
agan Μ osiki mΓ¨ ichep - ura kan Μ , ike ekpabe me ibebene emen 1990 chame lek otu - ifuk ebi Γ¬luk me lek Γ¨wΓͺ sayara me afirika agan Μ osiki mbumme agan Μ - mkpulu 36 cha α»mα» ore ama ibot irΓ’n ire bagidadi . α»gbα»n Μ
Context Size 3:
me agan Μ Γ²siki . emen - awaji atilantik , emen - awaji eyi india , emen -Μ - mkpulu yi Γ¬nan Μ a me ujit fo ulom , mΓ¨ Γ¬yaka Γ¬bα»kα» si mkpukpe Γ²miminagan Μ - mkpulu yi obenbe Γ¬re 9 , 251 kmΒ² , sΓ otu - ifuk ifit ya
Context Size 4:
agan Μ - mkpulu yi Γ¬re okwaan Μ imo , Γ²kilibi iraka me okike ijα»n Μ kan Μ .agan Μ inyα»n Μ mΓ¨ α»fα»kα» agan Μ osiki . iman Μ yi Γ¬libi inan Μ a me ofifiΜ ichep - ura , bawuchi mΓ¨ gombe me agan Μ osiki , mΓ¨ kogi me agan Μ inyα»n
Key Findings
- Best Predictability: Context-4 with 93.1% predictability
- Branching Factor: Decreases with context size (more deterministic)
- Memory Trade-off: Larger contexts require more storage (49,379 contexts)
- Recommendation: Context-3 or Context-4 for text generation
4. Vocabulary Analysis
Statistics
| Metric | Value |
|---|---|
| Vocabulary Size | 4,397 |
| Total Tokens | 93,822 |
| Mean Frequency | 21.34 |
| Median Frequency | 4 |
| Frequency Std Dev | 149.48 |
Most Common Words
| Rank | Word | Frequency |
|---|---|---|
| 1 | me | 7,521 |
| 2 | mè | 2,898 |
| 3 | agan | 1,911 |
| 4 | ebi | 1,731 |
| 5 | ido | 1,637 |
| 6 | Γ¬re | 1,613 |
| 7 | lek | 1,578 |
| 8 | eyi | 1,242 |
| 9 | ya | 1,175 |
| 10 | emen | 1,068 |
Least Common Words (from vocabulary)
| Rank | Word | Frequency |
|---|---|---|
| 1 | 3500 | 2 |
| 2 | iyaak | 2 |
| 3 | medvedev | 2 |
| 4 | race | 2 |
| 5 | lenin | 2 |
| 6 | α»kα»lα»ba | 2 |
| 7 | ΗΉkα»α»n | 2 |
| 8 | edeh | 2 |
| 9 | ogwuile | 2 |
| 10 | bruxelles | 2 |
Zipf's Law Analysis
| Metric | Value |
|---|---|
| Zipf Coefficient | 1.1589 |
| RΒ² (Goodness of Fit) | 0.990794 |
| Adherence Quality | excellent |
Coverage Analysis
| Top N Words | Coverage |
|---|---|
| Top 100 | 58.8% |
| Top 1,000 | 87.2% |
| Top 5,000 | 0.0% |
| Top 10,000 | 0.0% |
Key Findings
- Zipf Compliance: RΒ²=0.9908 indicates excellent adherence to Zipf's law
- High Frequency Dominance: Top 100 words cover 58.8% of corpus
- Long Tail: -5,603 words needed for remaining 100.0% coverage
5. Word Embeddings Evaluation
Model Comparison
| Model | Vocab Size | Dimension | Avg Norm | Std Norm | Isotropy |
|---|---|---|---|---|---|
| mono_32d | 1,947 | 32 | 2.277 | 0.449 | 0.1631 π |
| mono_64d | 1,947 | 64 | 2.234 | 0.444 | 0.0371 |
| mono_128d | 1,947 | 128 | 2.248 | 0.444 | 0.0066 |
| embeddings_enhanced | 0 | 0 | 0.000 | 0.000 | 0.0000 |
Key Findings
- Best Isotropy: mono_32d with 0.1631 (more uniform distribution)
- Dimension Trade-off: Higher dimensions capture more semantics but reduce isotropy
- Vocabulary Coverage: All models cover 1,947 words
- Recommendation: 100d for balanced semantic capture and efficiency
6. Summary & Recommendations
Production Recommendations
| Component | Recommended | Rationale |
|---|---|---|
| Tokenizer | 32k BPE | Best compression (4.27x) with low UNK rate |
| N-gram | 5-gram | Lowest perplexity (253) |
| Markov | Context-4 | Highest predictability (93.1%) |
| 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) Γ 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.
RΒ² (Coefficient of Determination)
Definition: Measures how well the linear fit explains the frequency-rank relationship. Ranges from 0 to 1.
Intuition: RΒ² near 1.0 means the data closely follows Zipf's law; lower values indicate deviation from expected word frequency patterns.
What to seek: RΒ² > 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},
publisher = {HuggingFace},
url = {https://huggingface.co/wikilangs}
institution = {Omneity Labs}
}
License
MIT License - Free for academic and commercial use.
Links
- π Website: wikilangs.org
- π€ Models: huggingface.co/wikilangs
- π Data: wikipedia-monthly
- π€ Author: Omar Kamali
Generated by Wikilangs Models Pipeline
Report Date: 2025-12-27 06:05:57