README.md

---
language: ast
language_name: Asturian
language_family: romance_iberian
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-romance_iberian
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: 4.429
  - name: best_isotropy
    type: isotropy
    value: 0.7932
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-04
---

# Asturian - Wikilangs Models
## Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on **Asturian** 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

![Performance Dashboard](visualizations/performance_dashboard.png)

### Analysis and Evaluation

- [1. Tokenizer Evaluation](#1-tokenizer-evaluation)
- [2. N-gram Model Evaluation](#2-n-gram-model-evaluation)
- [3. Markov Chain Evaluation](#3-markov-chain-evaluation)
- [4. Vocabulary Analysis](#4-vocabulary-analysis)
- [5. Word Embeddings Evaluation](#5-word-embeddings-evaluation)
- [6. Morphological Analysis (Experimental)](#6--morphological-analysis-experimental)
- [7. Summary & Recommendations](#7-summary--recommendations)
- [Metrics Glossary](#appendix-metrics-glossary--interpretation-guide)
- [Visualizations Index](#visualizations-index)

---
## 1. Tokenizer Evaluation

![Tokenizer Compression](visualizations/tokenizer_compression.png)

![Tokenizer Fertility](visualizations/tokenizer_fertility.png)

![Tokenizer OOV](visualizations/tokenizer_oov.png)

![Total Tokens](visualizations/tokenizer_total_tokens.png)

### Results

| Vocab Size | Compression | Avg Token Len | UNK Rate | Total Tokens |
|------------|-------------|---------------|----------|--------------|
| **8k** | 3.571x | 3.57 | 0.0264% | 863,429 |
| **16k** | 3.921x | 3.92 | 0.0290% | 786,292 |
| **32k** | 4.205x | 4.21 | 0.0311% | 733,251 |
| **64k** | 4.429x 🏆 | 4.43 | 0.0327% | 696,255 |

### Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

**Sample 1:** `Pol nome de Pedru'l Grande conocemos a dos monarques europeos: Pedru III d'Aragó...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁pol ▁nome ▁de ▁ped ru ' l ▁grande ▁cono ce ... (+21 more)` | 31 |
| 16k | `▁pol ▁nome ▁de ▁pedru ' l ▁grande ▁cono ce mos ... (+18 more)` | 28 |
| 32k | `▁pol ▁nome ▁de ▁pedru ' l ▁grande ▁conocemos ▁a ▁dos ... (+15 more)` | 25 |
| 64k | `▁pol ▁nome ▁de ▁pedru ' l ▁grande ▁conocemos ▁a ▁dos ... (+15 more)` | 25 |

**Sample 2:** `Yuki Ohashi (, ) ye un futbolista xaponés. Clubes Referencies Enllaces esternos ...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁yu ki ▁oh as hi ▁(, ▁) ▁ye ▁un ▁futbolista ... (+14 more)` | 24 |
| 16k | `▁yu ki ▁oh ashi ▁(, ▁) ▁ye ▁un ▁futbolista ▁xaponés ... (+12 more)` | 22 |
| 32k | `▁yuki ▁oh ashi ▁(, ▁) ▁ye ▁un ▁futbolista ▁xaponés . ... (+11 more)` | 21 |
| 64k | `▁yuki ▁oh ashi ▁(, ▁) ▁ye ▁un ▁futbolista ▁xaponés . ... (+11 more)` | 21 |

**Sample 3:** `Fechos Nacencies Muertes Referencies Enllaces esternos V e.C.`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁fechos ▁nacencies ▁muertes ▁referencies ▁enllaces ▁esternos ▁v ▁e . c ... (+1 more)` | 11 |
| 16k | `▁fechos ▁nacencies ▁muertes ▁referencies ▁enllaces ▁esternos ▁v ▁e . c ... (+1 more)` | 11 |
| 32k | `▁fechos ▁nacencies ▁muertes ▁referencies ▁enllaces ▁esternos ▁v ▁e . c ... (+1 more)` | 11 |
| 64k | `▁fechos ▁nacencies ▁muertes ▁referencies ▁enllaces ▁esternos ▁v ▁e . c ... (+1 more)` | 11 |


### Key Findings

- **Best Compression:** 64k achieves 4.429x compression
- **Lowest UNK Rate:** 8k with 0.0264% 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

![N-gram Perplexity](visualizations/ngram_perplexity.png)

![N-gram Unique](visualizations/ngram_unique.png)

![N-gram Coverage](visualizations/ngram_coverage.png)

### Results

| N-gram | Variant | Perplexity | Entropy | Unique N-grams | Top-100 Coverage | Top-1000 Coverage |
|--------|---------|------------|---------|----------------|------------------|-------------------|
| **2-gram** | Word | 132,138 | 17.01 | 1,341,882 | 9.8% | 21.7% |
| **2-gram** | Subword | 260 🏆 | 8.02 | 19,027 | 69.7% | 99.1% |
| **3-gram** | Word | 640,312 | 19.29 | 2,878,367 | 4.2% | 10.7% |
| **3-gram** | Subword | 2,218 | 11.12 | 138,526 | 28.0% | 72.3% |
| **4-gram** | Word | 1,536,908 | 20.55 | 4,654,291 | 3.3% | 7.6% |
| **4-gram** | Subword | 13,337 | 13.70 | 787,142 | 13.9% | 39.3% |
| **5-gram** | Word | 1,050,558 | 20.00 | 2,949,427 | 4.8% | 9.6% |
| **5-gram** | Subword | 57,630 | 15.81 | 2,701,102 | 7.8% | 23.5% |

### Top 5 N-grams by Size

**2-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `de la` | 877,001 |
| 2 | `de los` | 325,167 |
| 3 | `la so` | 218,605 |
| 4 | `a la` | 213,098 |
| 5 | `de les` | 205,401 |

**3-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `referencies enllaces esternos` | 102,198 |
| 2 | `de la so` | 48,437 |
| 3 | `d estaos xuníos` | 34,372 |
| 4 | `enllaces esternos de` | 33,442 |
| 5 | `una población de` | 30,281 |

**4-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `referencies enllaces esternos de` | 32,439 |
| 2 | `tien una población de` | 26,725 |
| 3 | `una población de y` | 19,595 |
| 4 | `y una superficie de` | 19,554 |
| 5 | `población de y una` | 19,514 |

**5-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `tien una población de y` | 19,555 |
| 2 | `una población de y una` | 19,513 |
| 3 | `de y una superficie de` | 19,492 |
| 4 | `población de y una superficie` | 19,490 |
| 5 | `y una superficie de km` | 19,254 |

**2-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `a _` | 12,223,314 |
| 2 | `e _` | 10,169,137 |
| 3 | `s _` | 9,980,231 |
| 4 | `_ d` | 9,749,761 |
| 5 | `e s` | 9,339,123 |

**3-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ d e` | 7,125,386 |
| 2 | `d e _` | 5,278,423 |
| 3 | `e s _` | 4,734,999 |
| 4 | `o s _` | 3,881,527 |
| 5 | `l a _` | 3,034,851 |

**4-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ d e _` | 4,909,705 |
| 2 | `_ l a _` | 2,443,055 |
| 3 | `d e _ l` | 1,642,151 |
| 4 | `a _ d e` | 1,399,483 |
| 5 | `s _ d e` | 1,367,031 |

**5-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ d e _ l` | 1,593,336 |
| 2 | `e _ l a _` | 1,090,094 |
| 3 | `_ d e l _` | 1,070,352 |
| 4 | `s _ d e _` | 1,000,253 |
| 5 | `a _ d e _` | 970,617 |


### Key Findings

- **Best Perplexity:** 2-gram (subword) with 260
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
- **Coverage:** Top-1000 patterns cover ~23% of corpus
- **Recommendation:** 4-gram or 5-gram for best predictive performance

---
## 3. Markov Chain Evaluation

![Markov Entropy](visualizations/markov_entropy.png)

![Markov Contexts](visualizations/markov_contexts.png)

![Markov Branching](visualizations/markov_branching.png)

### Results

| Context | Variant | Avg Entropy | Perplexity | Branching Factor | Unique Contexts | Predictability |
|---------|---------|-------------|------------|------------------|-----------------|----------------|
| **1** | Word | 1.0362 | 2.051 | 12.93 | 1,199,957 | 0.0% |
| **1** | Subword | 1.1986 | 2.295 | 7.97 | 10,438 | 0.0% |
| **2** | Word | 0.4189 | 1.337 | 2.57 | 15,504,920 | 58.1% |
| **2** | Subword | 0.6561 | 1.576 | 4.28 | 83,238 | 34.4% |
| **3** | Word | 0.1863 | 1.138 | 1.44 | 39,817,744 | 81.4% |
| **3** | Subword | 0.6835 | 1.606 | 4.02 | 356,042 | 31.6% |
| **4** | Word | 0.0788 🏆 | 1.056 | 1.14 | 57,235,451 | 92.1% |
| **4** | Subword | 0.6840 | 1.607 | 3.51 | 1,432,910 | 31.6% |

### Generated Text Samples (Word-based)

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

**Context Size 1:**

1. `de teniente xeneral del planeta mientres la realidá quiciabes d ochobre foi escritu por aciu una`
2. `la cual el propósitu un estilu y hornsby consiguieron 31 d alabama intentó nun tour a`
3. `y derechos humanos ta estremada en determinóse que caltener la so home l minsiterio de candela`

**Context Size 2:**

1. `de la cocina nos años y escuchar música dende la edá kim young chae sbs jumpmbc nonstop`
2. `de los fundadores de los cinco principales epítetos y títulos descriptivos de los chola fueron movío...`
3. `la so bona contrarreló calteniendo a dellos decretos prohibiendo la llibre asociación como ye l cuan...`

**Context Size 3:**

1. `referencies enllaces esternos green breasted mangu english wikipedia consultáu l 2 de marzu de estab...`
2. `de la so política d esclusión nel sieglu xx en que camudó de nome los líderes del movimientu`
3. `enllaces esternos de xapón de la prefeutura de hyogo llocalización con una superficie de km ver tami...`

**Context Size 4:**

1. `referencies enllaces esternos de piloña de piloña`
2. `tien una población de y una superficie de km y una población de referencies enllaces esternos de xap...`
3. `una población de y una superficie de km referencies enllaces esternos d aquila`


### Generated Text Samples (Subword-based)

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

**Context Size 1:**

1. `_eral_r_s_de_prm`
2. `el_untodesopay_c`
3. `armbra_a_wozall_`

**Context Size 2:**

1. `a_dada_d'alicu_de`
2. `e_al_crein_ings._`
3. `s_agu_pobres_saos`

**Context Size 3:**

1. `_de_s'atroxina_pa_`
2. `de_los_nuevu._fíos`
3. `es_deste_-_frivaes`

**Context Size 4:**

1. `_de_mouther_de_fort`
2. `_la_cada_y_márquist`
3. `de_la_sociedá_nacio`


### Key Findings

- **Best Predictability:** Context-4 (word) with 92.1% predictability
- **Branching Factor:** Decreases with context size (more deterministic)
- **Memory Trade-off:** Larger contexts require more storage (1,432,910 contexts)
- **Recommendation:** Context-3 or Context-4 for text generation

---
## 4. Vocabulary Analysis

![Zipf's Law](visualizations/zipf_law.png)

![Top Words](visualizations/top20_words.png)

![Coverage Curve](visualizations/vocab_coverage.png)

### Statistics

| Metric | Value |
|--------|-------|
| Vocabulary Size | 552,425 |
| Total Tokens | 74,325,511 |
| Mean Frequency | 134.54 |
| Median Frequency | 4 |
| Frequency Std Dev | 9254.05 |

### Most Common Words

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | de | 4,928,261 |
| 2 | la | 2,485,426 |
| 3 | y | 2,042,239 |
| 4 | d | 1,169,053 |
| 5 | a | 1,155,083 |
| 6 | del | 1,074,281 |
| 7 | en | 1,055,986 |
| 8 | que | 1,007,870 |
| 9 | los | 957,887 |
| 10 | l | 950,908 |

### Least Common Words (from vocabulary)

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | leptafeke | 2 |
| 2 | haua | 2 |
| 3 | küzdoblani | 2 |
| 4 | contrarrellatu | 2 |
| 5 | semilleru | 2 |
| 6 | bisterca | 2 |
| 7 | šafarsko | 2 |
| 8 | vyfalu | 2 |
| 9 | ribich | 2 |
| 10 | lacos | 2 |

### Zipf's Law Analysis

| Metric | Value |
|--------|-------|
| Zipf Coefficient | 0.9990 |
| R² (Goodness of Fit) | 0.995611 |
| Adherence Quality | **excellent** |

### Coverage Analysis

| Top N Words | Coverage |
|-------------|----------|
| Top 100 | 41.7% |
| Top 1,000 | 60.8% |
| Top 5,000 | 76.8% |
| Top 10,000 | 83.1% |

### Key Findings

- **Zipf Compliance:** R²=0.9956 indicates excellent adherence to Zipf's law
- **High Frequency Dominance:** Top 100 words cover 41.7% of corpus
- **Long Tail:** 542,425 words needed for remaining 16.9% coverage

---
## 5. Word Embeddings Evaluation

![Embedding Isotropy](visualizations/embedding_isotropy.png)

![Similarity Matrix](visualizations/embedding_similarity.png)

![t-SNE Words](visualizations/tsne_words.png)

![t-SNE Sentences](visualizations/tsne_sentences.png)


### 5.1 Cross-Lingual Alignment

![Alignment Quality](visualizations/embedding_alignment_quality.png)

![Multilingual t-SNE](visualizations/embedding_tsne_multilingual.png)


### 5.2 Model Comparison

| Model | Dimension | Isotropy | Semantic Density | Alignment R@1 | Alignment R@10 |
|-------|-----------|----------|------------------|---------------|----------------|
| **mono_32d** | 32 | 0.7932 | 0.3820 | N/A | N/A |
| **mono_64d** | 64 | 0.7818 | 0.2979 | N/A | N/A |
| **mono_128d** | 128 | 0.7210 | 0.2388 | N/A | N/A |
| **aligned_32d** | 32 | 0.7932 🏆 | 0.3922 | 0.3820 | 0.7300 |
| **aligned_64d** | 64 | 0.7818 | 0.3048 | 0.5840 | 0.8840 |
| **aligned_128d** | 128 | 0.7210 | 0.2380 | 0.7080 | 0.9240 |

### Key Findings

- **Best Isotropy:** aligned_32d with 0.7932 (more uniform distribution)
- **Semantic Density:** Average pairwise similarity of 0.3090. Lower values indicate better semantic separation.
- **Alignment Quality:** Aligned models achieve up to 70.8% 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 | **-0.591** | 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 |
|--------|----------|
| `-co` | control, coetzee, conversión |
| `-ma` | manifiéstase, marinel, matraqueo |
| `-re` | rehnskiöld, rendimientos, reichholf |

#### Productive Suffixes
| Suffix | Examples |
|--------|----------|
| `-s` | narganes, supracaudales, señálennos |
| `-a` | carga, balsámica, trueba |
| `-es` | narganes, supracaudales, rastres |
| `-os` | señálennos, sabéivos, visos |
| `-se` | escapóse, ñublense, manifiéstase |
| `-as` | tankas, aleutas, ḥechas |

### 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 |
|------|----------|------------------|----------|
| `iend` | 1.75x | 206 contexts | fiend, iendo, rienda |
| `ació` | 1.96x | 92 contexts | ñació, lació, xació |
| `ogra` | 1.57x | 189 contexts | logra, bogra, sogra |
| `ient` | 1.46x | 273 contexts | iente, cient, aient |
| `acio` | 1.55x | 167 contexts | bacio, facio, macio |
| `renc` | 1.71x | 99 contexts | frenc, lorenc, trench |
| `ntes` | 1.56x | 144 contexts | antes, entes, entesa |
| `enci` | 1.35x | 261 contexts | encia, cenci, venci |
| `efer` | 1.63x | 86 contexts | refer, defer, sefer |
| `ntos` | 1.72x | 67 contexts | antos, entos, tantos |
| `raci` | 1.41x | 164 contexts | racib, racio, iraci |
| `ontr` | 1.50x | 117 contexts | contr, kontra, lontra |

### 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 |
|--------|--------|-----------|----------|
| `-co` | `-s` | 55 words | consentimientos, correllaciones |
| `-ma` | `-a` | 44 words | maniobraba, marra |
| `-ma` | `-s` | 40 words | macromicetes, maorís |
| `-re` | `-a` | 39 words | reflorestada, respondida |
| `-co` | `-a` | 37 words | comitia, cornigera |
| `-re` | `-s` | 33 words | refundiándoles, reprogramables |
| `-re` | `-se` | 27 words | reproducense, retomándose |
| `-co` | `-es` | 23 words | correllaciones, coeditores |
| `-co` | `-se` | 22 words | confiándose, comercializábense |
| `-re` | `-es` | 20 words | refundiándoles, reprogramables |

### 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 |
|------|-----------------|------------|------|
| clamorosos | **`clamor-os-os`** | 6.0 | `clamor` |
| doloroses | **`dolor-os-es`** | 6.0 | `dolor` |
| velenoses | **`velen-os-es`** | 6.0 | `velen` |
| escribiríase | **`escribiría-se`** | 4.5 | `escribiría` |
| mundiales | **`mundial-es`** | 4.5 | `mundial` |
| desgraciaos | **`desgracia-os`** | 4.5 | `desgracia` |
| alfayates | **`alfayat-es`** | 4.5 | `alfayat` |
| cristalizase | **`cristaliza-se`** | 4.5 | `cristaliza` |
| remensura | **`re-mensura`** | 4.5 | `mensura` |
| desequilibraos | **`desequilibra-os`** | 4.5 | `desequilibra` |
| decretase | **`decreta-se`** | 4.5 | `decreta` |
| coartífice | **`co-artífice`** | 4.5 | `artífice` |
| declaráse | **`declará-se`** | 4.5 | `declará` |
| reordenar | **`re-ordenar`** | 4.5 | `ordenar` |
| pediatres | **`pediatr-es`** | 4.5 | `pediatr` |

### 6.6 Linguistic Interpretation

> **Automated Insight:**
The language Asturian 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

![Performance Dashboard](visualizations/performance_dashboard.png)

### Production Recommendations

| Component | Recommended | Rationale |
|-----------|-------------|-----------|
| Tokenizer | **64k BPE** | Best compression (4.43x) |
| N-gram | **2-gram** | Lowest perplexity (260) |
| Markov | **Context-4** | Highest predictability (92.1%) |
| 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) × 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](https://huggingface.co/datasets/omarkamali/wikipedia-monthly) - a monthly snapshot of Wikipedia articles across 300+ languages.

### Project

A project by **[Wikilangs](https://wikilangs.org)** - Open-source NLP models for every Wikipedia language.

### Maintainer

[Omar Kamali](https://omarkamali.com) - [Omneity Labs](https://omneitylabs.com)

### Citation

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

```bibtex
@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](https://wikilangs.org)
- 🤗 Models: [huggingface.co/wikilangs](https://huggingface.co/wikilangs)
- 📊 Data: [wikipedia-monthly](https://huggingface.co/datasets/omarkamali/wikipedia-monthly)
- 👤 Author: [Omar Kamali](https://huggingface.co/omarkamali)
- 🤝 Sponsor: [Featherless AI](https://featherless.ai)
---
*Generated by Wikilangs Models Pipeline*

*Report Date: 2026-01-04 02:53:18*