---
language: su
language_name: Sundanese
language_family: austronesian_javanese
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-austronesian_javanese
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: 4.793
  - name: best_isotropy
    type: isotropy
    value: 0.7854
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-10
---

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

This repository contains NLP models trained and evaluated by Wikilangs, specifically on **Sundanese** 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** | 3.614x | 3.61 | 0.2895% | 1,045,476 |
| **16k** | 4.061x | 4.06 | 0.3254% | 930,202 |
| **32k** | 4.462x | 4.46 | 0.3575% | 846,599 |
| **64k** | 4.793x 🏆 | 4.79 | 0.3840% | 788,257 |

### Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

**Sample 1:** `Sukajaya nyaéta salah sahiji désa di kacamatan Ciséwu, Kabupatén Garut, Propinsi...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁suk ajaya ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁cis éw ... (+13 more)` | 23 |
| 16k | `▁sukajaya ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁cis éwu , ... (+11 more)` | 21 |
| 32k | `▁sukajaya ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁ciséwu , ▁kabupatén ... (+10 more)` | 20 |
| 64k | `▁sukajaya ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁ciséwu , ▁kabupatén ... (+10 more)` | 20 |

**Sample 2:** `Way Sindi nyaéta salah sahiji Désa di kacamatan Karya Penggawa, Kabupatén Pesisi...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁way ▁sin di ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁karya ... (+13 more)` | 23 |
| 16k | `▁way ▁sin di ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁karya ... (+13 more)` | 23 |
| 32k | `▁way ▁sin di ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁karya ... (+12 more)` | 22 |
| 64k | `▁way ▁sindi ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁karya ▁penggawa ... (+11 more)` | 21 |

**Sample 3:** `Linggamukti nyaéta salah sahiji désa di kacamatan Sucinaraja, Kabupatén Garut, P...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁lingg am ukti ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁su ... (+14 more)` | 24 |
| 16k | `▁lingg am ukti ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁su ... (+14 more)` | 24 |
| 32k | `▁lingg amukti ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁sucinaraja , ... (+11 more)` | 21 |
| 64k | `▁lingg amukti ▁nyaéta ▁salah ▁sahiji ▁désa ▁di ▁kacamatan ▁sucinaraja , ... (+11 more)` | 21 |


### Key Findings

- **Best Compression:** 64k achieves 4.793x compression
- **Lowest UNK Rate:** 8k with 0.2895% 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 | 8,615 | 13.07 | 119,237 | 36.6% | 51.4% |
| **2-gram** | Subword | 250 🏆 | 7.96 | 8,527 | 69.1% | 99.4% |
| **3-gram** | Word | 3,378 | 11.72 | 118,793 | 51.2% | 64.9% |
| **3-gram** | Subword | 2,021 | 10.98 | 49,956 | 27.1% | 75.5% |
| **4-gram** | Word | 3,002 | 11.55 | 162,065 | 53.7% | 67.2% |
| **4-gram** | Subword | 10,081 | 13.30 | 252,099 | 14.3% | 47.8% |
| **5-gram** | Word | 2,066 | 11.01 | 112,479 | 57.2% | 70.2% |
| **5-gram** | Subword | 31,527 | 14.94 | 709,433 | 10.6% | 36.5% |

### Top 5 N-grams by Size

**2-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `salah sahiji` | 29,861 |
| 2 | `astéroid ieu` | 29,850 |
| 3 | `ieu astéroid` | 29,850 |
| 4 | `nyaéta salah` | 26,619 |
| 5 | `di kacamatan` | 25,114 |

**3-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `nyaéta salah sahiji` | 26,442 |
| 2 | `désa di kacamatan` | 16,291 |
| 3 | `salah sahiji désa` | 15,457 |
| 4 | `sahiji désa di` | 15,449 |
| 5 | `rujukan tutumbu kaluar` | 14,998 |

**4-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `salah sahiji désa di` | 15,449 |
| 2 | `sahiji désa di kacamatan` | 15,446 |
| 3 | `nyaéta salah sahiji désa` | 15,429 |
| 4 | `the international astronomical union` | 14,930 |
| 5 | `astéroid kacatet gedéna 0` | 14,925 |

**5-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `salah sahiji désa di kacamatan` | 15,446 |
| 2 | `nyaéta salah sahiji désa di` | 15,429 |
| 3 | `minangka beubeulahan planétisimal objék di` | 14,925 |
| 4 | `asteroid téh bagéan tina astéroid` | 14,925 |
| 5 | `nganjrek deukeut jeung marcapada ékséntrisitas` | 14,925 |

**2-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `a n` | 1,250,483 |
| 2 | `a _` | 1,066,804 |
| 3 | `n _` | 801,241 |
| 4 | `n g` | 770,939 |
| 5 | `k a` | 571,201 |

**3-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `a n _` | 417,933 |
| 2 | `_ k a` | 355,900 |
| 3 | `n a _` | 318,266 |
| 4 | `_ d i` | 307,852 |
| 5 | `a n g` | 284,934 |

**4-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `e u n _` | 144,400 |
| 2 | `k e u n` | 135,792 |
| 3 | `i n a _` | 133,616 |
| 4 | `_ d i _` | 127,925 |
| 5 | `_ a s t` | 120,933 |

**5-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `k e u n _` | 129,890 |
| 2 | `s t é r o` | 89,884 |
| 3 | `é r o i d` | 89,804 |
| 4 | `t é r o i` | 89,803 |
| 5 | `_ a s t é` | 89,744 |


### Key Findings

- **Best Perplexity:** 2-gram (subword) with 250
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
- **Coverage:** Top-1000 patterns cover ~37% 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 | 0.9632 | 1.950 | 8.46 | 260,446 | 3.7% |
| **1** | Subword | 1.1518 | 2.222 | 7.12 | 4,969 | 0.0% |
| **2** | Word | 0.2938 | 1.226 | 1.70 | 2,198,896 | 70.6% |
| **2** | Subword | 0.6319 | 1.550 | 3.75 | 35,377 | 36.8% |
| **3** | Word | 0.0779 | 1.055 | 1.13 | 3,734,334 | 92.2% |
| **3** | Subword | 0.6394 | 1.558 | 3.52 | 132,696 | 36.1% |
| **4** | Word | 0.0225 🏆 | 1.016 | 1.03 | 4,192,253 | 97.7% |
| **4** | Subword | 0.6390 | 1.557 | 3.00 | 466,876 | 36.1% |

### Generated Text Samples (Word-based)

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

**Context Size 1:**

1. `di handap dipaké pikeun ngajéntrékeun pamuka pikeun rahayatna dipaksa néken perjangjian anu dirojong...`
2. `nu kahiji smp rayudin guru lagu kahijina ka tukang balap tim mclaren mercedes benz e300 kakayaanna`
3. `astéroid amor the iceman winona ryder edgar allan poé 335 sedengkeun magnitudo mutlakna 22 23 3`

**Context Size 2:**

1. `salah sahiji désa di kacamatan idi tunong kabupatén aceh tamiang propinsi acéh indonésia manyak paye...`
2. `ieu astéroid kacatet gedéna 0 482 sedengkeun magnitudo mutlakna 26 9 ari nu jadi référénsina mah nya...`
3. `astéroid ieu asteroid téh bagéan tina astéroid amor anu nganjrek deukeut jeung marcapada ékséntrisit...`

**Context Size 3:**

1. `nyaéta salah sahiji désa di kacamatan tano tombangan angkola kabupatén tapanuli kidul propinsi sumat...`
2. `désa di kacamatan jujuhan kabupatén bungo propinsi jambi indonésia renah mendaluh renah mendaluh`
3. `salah sahiji désa di kacamatan bantarujeg kabupatén majalengka propinsi jawa barat anggota mpr fkp d...`

**Context Size 4:**

1. `salah sahiji désa di kacamatan hantara kabupatén kuningan propinsi jawa barat indonésia beusi mangru...`
2. `sahiji désa di kacamatan bangun purba kabupatén deli serdang propinsi sumatra kalér indonésia hinai ...`
3. `nyaéta salah sahiji désa di kacamatan pesisir bukit kota sungai penuh propinsi jambi indonésia pesis...`


### Generated Text Samples (Subword-based)

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

**Context Size 1:**

1. `as)_neugeukinua_`
2. `_dil_dértapiswi_`
3. `n_pleukeuloral_g`

**Context Size 2:**

1. `an_teun_(ter._ama`
2. `a_muh_so._–_lo_na`
3. `n_to_ta_bangkoti_`

**Context Size 3:**

1. `an_cijelia,_saratu`
2. `_kalén_biblanda_ny`
3. `na_jeunakeun_baria`

**Context Size 4:**

1. `eun_ngritic_swedish`
2. `keun_yén_anu_anu_ja`
3. `ina_katematika_bebe`


### Key Findings

- **Best Predictability:** Context-4 (word) with 97.7% predictability
- **Branching Factor:** Decreases with context size (more deterministic)
- **Memory Trade-off:** Larger contexts require more storage (466,876 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 | 116,875 |
| Total Tokens | 6,065,431 |
| Mean Frequency | 51.90 |
| Median Frequency | 4 |
| Frequency Std Dev | 952.21 |

### Most Common Words

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | di | 128,510 |
| 2 | nu | 90,309 |
| 3 | astéroid | 89,739 |
| 4 | jeung | 83,019 |
| 5 | anu | 78,713 |
| 6 | nyaéta | 74,994 |
| 7 | ieu | 72,373 |
| 8 | dina | 59,209 |
| 9 | the | 54,138 |
| 10 | tina | 45,336 |

### Least Common Words (from vocabulary)

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | éksométéorologi | 2 |
| 2 | kejut | 2 |
| 3 | advektif | 2 |
| 4 | sirkulasina | 2 |
| 5 | pamelajaran | 2 |
| 6 | méchain | 2 |
| 7 | reflektor | 2 |
| 8 | spiralna | 2 |
| 9 | sombréro | 2 |
| 10 | halona | 2 |

### Zipf's Law Analysis

| Metric | Value |
|--------|-------|
| Zipf Coefficient | 1.0758 |
| R² (Goodness of Fit) | 0.997896 |
| Adherence Quality | **excellent** |

### Coverage Analysis

| Top N Words | Coverage |
|-------------|----------|
| Top 100 | 40.3% |
| Top 1,000 | 65.1% |
| Top 5,000 | 80.6% |
| Top 10,000 | 86.6% |

### Key Findings

- **Zipf Compliance:** R²=0.9979 indicates excellent adherence to Zipf's law
- **High Frequency Dominance:** Top 100 words cover 40.3% of corpus
- **Long Tail:** 106,875 words needed for remaining 13.4% 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 | 0.7778 | 0.3399 | N/A | N/A |
| **mono_64d** | 64 | 0.7854 | 0.2837 | N/A | N/A |
| **mono_128d** | 128 | 0.7675 | 0.2154 | N/A | N/A |
| **aligned_32d** | 32 | 0.7778 | 0.3496 | 0.0800 | 0.3720 |
| **aligned_64d** | 64 | 0.7854 🏆 | 0.2975 | 0.1840 | 0.5560 |
| **aligned_128d** | 128 | 0.7675 | 0.2138 | 0.2800 | 0.6620 |

### Key Findings

- **Best Isotropy:** aligned_64d with 0.7854 (more uniform distribution)
- **Semantic Density:** Average pairwise similarity of 0.2833. Lower values indicate better semantic separation.
- **Alignment Quality:** Aligned models achieve up to 28.0% 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 | **3.692** | High morphological productivity | Reliable analysis |
| Idiomaticity Gap | **0.922** | 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 |
|--------|----------|
| `-s` | supaya, sayonara, saimbangna |
| `-di` | diriku, diandih, diinterprétasi |
| `-ka` | kaisaryah, kasuburan, kamilil |
| `-a` | amorp, adjective, a1 |
| `-pa` | parki, pangngoranna, pasiapan |
| `-ma` | mahesa, matsukata, markedly |
| `-k` | kaisaryah, kustomisasi, ketumbar |
| `-sa` | sayonara, saimbangna, sacrifice |

#### Productive Suffixes
| Suffix | Examples |
|--------|----------|
| `-n` | peladjaran, citizen, lampahan |
| `-a` | supaya, neringa, sayonara |
| `-an` | peladjaran, lampahan, kasuburan |
| `-na` | saimbangna, tajukna, polipropiléna |
| `-s` | closures, liabilities, standards |
| `-un` | nginebkeun, impun, ngagerakkeun |
| `-ng` | mgōng, gedang, stemming |
| `-i` | parki, kustomisasi, diinterprétasi |

### 6.3 Bound Stems (Lexical Roots)

Bound stems are high-frequency subword units that are semantically cohesive but rarely appear as standalone words. These often correspond to the 'core' of a word that requires inflection or derivation to be valid.

| Stem | Cohesion | Substitutability | Examples |
|------|----------|------------------|----------|
| `tion` | 2.79x | 59 contexts | tiong, notion, lotion |
| `angk` | 1.64x | 309 contexts | angké, angke, angka |
| `ngka` | 1.65x | 215 contexts | ingka, angka, ingkah |
| `ukan` | 1.83x | 73 contexts | bukan, sukan, kukang |
| `ikeu` | 2.22x | 30 contexts | ikeun, pikeu, pikeun |
| `engk` | 1.62x | 106 contexts | engké, engke, engkos |
| `entu` | 1.83x | 49 contexts | tentu, hentu, centum |
| `sahi` | 2.47x | 15 contexts | sahii, sahid, sahih |
| `ropi` | 2.15x | 20 contexts | ropin, tropi, propil |
| `ndon` | 1.76x | 37 contexts | london, condon, bondon |
| `stér` | 2.63x | 10 contexts | stéril, stérol, stéréo |
| `roid` | 2.34x | 12 contexts | viroid, tiroid, toroid |

### 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 |
|--------|--------|-----------|----------|
| `-di` | `-n` | 171 words | diasumsikeun, diiringan |
| `-s` | `-a` | 132 words | suriawiria, senjatana |
| `-ka` | `-n` | 118 words | kadéwasaan, kacamtan |
| `-pa` | `-n` | 116 words | payen, paragon |
| `-ka` | `-an` | 106 words | kadéwasaan, kacamtan |
| `-p` | `-n` | 105 words | payen, paragon |
| `-di` | `-un` | 103 words | diasumsikeun, direalisasikeun |
| `-pa` | `-an` | 99 words | panyusuhan, panyocokan |
| `-s` | `-n` | 80 words | satupun, sakapeun |
| `-p` | `-an` | 80 words | panyusuhan, panyocokan |

### 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 |
|------|-----------------|------------|------|
| pengajian | **`pengaj-i-an`** | 7.5 | `i` |
| impianana | **`impia-na-na`** | 7.5 | `na` |
| electricians | **`electrici-an-s`** | 7.5 | `an` |
| panghitungan | **`panghitu-ng-an`** | 7.5 | `ng` |
| heulaanan | **`heula-an-an`** | 7.5 | `an` |
| perdananya | **`perdan-an-ya`** | 7.5 | `an` |
| deukeuteunana | **`deukeuteu-na-na`** | 7.5 | `na` |
| kotakulon | **`ko-ta-kulon`** | 7.5 | `kulon` |
| valenciennes | **`valencien-n-es`** | 7.5 | `n` |
| brisingidae | **`brisingid-a-e`** | 7.5 | `a` |
| intermittent | **`intermitte-n-t`** | 7.5 | `n` |
| palestinians | **`palestini-an-s`** | 7.5 | `an` |
| ngawurukanana | **`ngawuruka-na-na`** | 7.5 | `na` |
| dicangkokkeun | **`dicangkokk-e-un`** | 7.5 | `e` |
| andelfingen | **`andelfi-ng-en`** | 7.5 | `ng` |

### 6.6 Linguistic Interpretation

> **Automated Insight:**
The language Sundanese 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

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

### Production Recommendations

| Component | Recommended | Rationale |
|-----------|-------------|-----------|
| Tokenizer | **64k BPE** | Best compression (4.79x) |
| N-gram | **2-gram** | Lowest perplexity (250) |
| Markov | **Context-4** | Highest predictability (97.7%) |
| 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: 2026-01-10 23:25:18*