---
language: tcy
language_name: Tulu
language_family: dravidian_south
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-dravidian_south
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: 4.489
  - name: best_isotropy
    type: isotropy
    value: 0.9138
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-11
---

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

This repository contains NLP models trained and evaluated by Wikilangs, specifically on **Tulu** 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** | 3.480x | 3.48 | 0.1072% | 636,146 |
| **16k** | 3.878x | 3.88 | 0.1195% | 570,862 |
| **32k** | 4.194x | 4.19 | 0.1292% | 527,863 |
| **64k** | 4.489x 🏆 | 4.49 | 0.1383% | 493,153 |

### Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

**Sample 1:** `ವಿಶ್ವ ಸಂಸ್ಥೆಡು ಮಸ್ತು ಬೇಲೆ ಮಲ್ಪುನ ಅಂಗ ಪಂಡ ಭದ್ರತಾ ಮಂಡಳಿ. ಉಂದೆನ್ ವಿಶ್ವ ಸಂಸ್ಥೆ ದ ಕಾರ...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁ವಿಶ್ವ ▁ಸಂಸ್ಥೆ ಡು ▁ಮಸ್ತು ▁ಬೇಲೆ ▁ಮಲ್ಪುನ ▁ಅಂಗ ▁ಪಂಡ ▁ಭದ್ರ ತಾ ... (+12 more)` | 22 |
| 16k | `▁ವಿಶ್ವ ▁ಸಂಸ್ಥೆ ಡು ▁ಮಸ್ತು ▁ಬೇಲೆ ▁ಮಲ್ಪುನ ▁ಅಂಗ ▁ಪಂಡ ▁ಭದ್ರ ತಾ ... (+12 more)` | 22 |
| 32k | `▁ವಿಶ್ವ ▁ಸಂಸ್ಥೆ ಡು ▁ಮಸ್ತು ▁ಬೇಲೆ ▁ಮಲ್ಪುನ ▁ಅಂಗ ▁ಪಂಡ ▁ಭದ್ರ ತಾ ... (+12 more)` | 22 |
| 64k | `▁ವಿಶ್ವ ▁ಸಂಸ್ಥೆಡು ▁ಮಸ್ತು ▁ಬೇಲೆ ▁ಮಲ್ಪುನ ▁ಅಂಗ ▁ಪಂಡ ▁ಭದ್ರತಾ ▁ಮಂಡಳಿ . ... (+9 more)` | 19 |

**Sample 2:** `ಉಂದು ಉದ್ದೊ ಅಲಪುನ ಪರಂಗಿತಕುಲೆನ ಮಾನೊ. ಅಲತೆ ಉದ್ದೊ ಇಪ್ಪಿನೆಗ್ ಒಂಜಿ Furlong ಪನ್ಪೆರ್. ಉಂ...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁ಉಂದು ▁ಉದ್ದೊ ▁ಅಲಪುನ ▁ಪರ ಂಗಿ ತ ಕುಲೆನ ▁ಮಾನೊ . ▁ಅಲತೆ ... (+23 more)` | 33 |
| 16k | `▁ಉಂದು ▁ಉದ್ದೊ ▁ಅಲಪುನ ▁ಪರಂಗಿ ತ ಕುಲೆನ ▁ಮಾನೊ . ▁ಅಲತೆ ▁ಉದ್ದೊ ... (+20 more)` | 30 |
| 32k | `▁ಉಂದು ▁ಉದ್ದೊ ▁ಅಲಪುನ ▁ಪರಂಗಿ ತಕುಲೆನ ▁ಮಾನೊ . ▁ಅಲತೆ ▁ಉದ್ದೊ ▁ಇಪ್ಪಿನೆ ... (+18 more)` | 28 |
| 64k | `▁ಉಂದು ▁ಉದ್ದೊ ▁ಅಲಪುನ ▁ಪರಂಗಿ ತಕುಲೆನ ▁ಮಾನೊ . ▁ಅಲತೆ ▁ಉದ್ದೊ ▁ಇಪ್ಪಿನೆಗ್ ... (+13 more)` | 23 |

**Sample 3:** `ಕಾಶಿ ಕ್ಶೇತ್ರೊಡ್ ಗ್ರಾಮ ದೇವತೆಯಾದಿತ್ತಿನ ಕಾಲಭೈರವೆ ಪನ್ಪಿನ ಶಿವ ಗಣ ಕದಿರೆದ ನಾಲ್ ಜೋಗಿ ಪುರ...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁ಕಾ ಶಿ ▁ಕ್ ಶೇ ತ್ರೊ ಡ್ ▁ಗ್ರಾಮ ▁ದೇವತೆ ಯಾ ದಿತ್ತಿನ ... (+30 more)` | 40 |
| 16k | `▁ಕಾಶಿ ▁ಕ್ಶೇ ತ್ರೊ ಡ್ ▁ಗ್ರಾಮ ▁ದೇವತೆ ಯಾದಿತ್ತಿನ ▁ಕಾಲ ಭೈರ ವೆ ... (+22 more)` | 32 |
| 32k | `▁ಕಾಶಿ ▁ಕ್ಶೇ ತ್ರೊಡ್ ▁ಗ್ರಾಮ ▁ದೇವತೆ ಯಾದಿತ್ತಿನ ▁ಕಾಲಭೈರ ವೆ ▁ಪನ್ಪಿನ ▁ಶಿವ ... (+18 more)` | 28 |
| 64k | `▁ಕಾಶಿ ▁ಕ್ಶೇತ್ರೊಡ್ ▁ಗ್ರಾಮ ▁ದೇವತೆ ಯಾದಿತ್ತಿನ ▁ಕಾಲಭೈರವೆ ▁ಪನ್ಪಿನ ▁ಶಿವ ▁ಗಣ ▁ಕದಿ ... (+13 more)` | 23 |


### Key Findings

- **Best Compression:** 64k achieves 4.489x compression
- **Lowest UNK Rate:** 8k with 0.1072% 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 | 8,933 | 13.12 | 13,353 | 9.7% | 32.3% |
| **2-gram** | Subword | 2,884 🏆 | 11.49 | 27,855 | 30.8% | 64.6% |
| **3-gram** | Word | 8,142 | 12.99 | 10,756 | 9.1% | 30.2% |
| **3-gram** | Subword | 24,830 | 14.60 | 135,097 | 10.1% | 29.9% |
| **4-gram** | Word | 26,886 | 14.71 | 31,900 | 4.4% | 14.2% |
| **4-gram** | Subword | 106,980 | 16.71 | 430,499 | 5.8% | 17.3% |
| **5-gram** | Word | 22,988 | 14.49 | 26,724 | 4.6% | 14.8% |
| **5-gram** | Subword | 191,997 | 17.55 | 551,532 | 4.4% | 12.7% |

### Top 5 N-grams by Size

**2-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `ಬೇತೆ ಬೇತೆ` | 1,021 |
| 2 | `ಸುರು ಮಲ್ತೆರ್` | 368 |
| 3 | `ಮಲ್ತ್ ದ್` | 344 |
| 4 | `ಕಿ ಮೀ` | 284 |
| 5 | `ಉಂಡು ಈ` | 276 |

**3-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `ಉಲ್ಲೇಕೊಲು ಬುಕ್ಕೊ ಜಾನಪದೊ` | 183 |
| 2 | `from the original` | 126 |
| 3 | `archived from the` | 125 |
| 4 | `the original on` | 117 |
| 5 | `ದಕ್ಷಿಣ ಕನ್ನಡ ಜಿಲ್ಲೆದ` | 114 |

**4-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `archived from the original` | 125 |
| 2 | `from the original on` | 117 |
| 3 | `ಲವ್ಸ್ ವಿಮೆನ್ ಸೌತ್ ಏಶಿಯಾ` | 102 |
| 4 | `ಬೇತೆ ಬಾಸೆಡ್ ಗೊಬ್ಬುದ ಪುದರ್` | 101 |
| 5 | `ಉಲ್ಲೇಕೊಲು ಬಾಸೆಲು ಬರವು ವಿಕಿಮೀಡಿಯನ್ಸ್` | 69 |

**5-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `archived from the original on` | 117 |
| 2 | `ಆ ಇ ಈ ಉ ಊ` | 44 |
| 3 | `ಅ ಆ ಇ ಈ ಉ` | 44 |
| 4 | `ಈ ಗೊಬ್ಬುನು ಗೊಬ್ಬುವೆರ್ ಉಂದೊಂಜಿ ಜನಪದ` | 44 |
| 5 | `ಗೊಬ್ಬುನು ಗೊಬ್ಬುವೆರ್ ಉಂದೊಂಜಿ ಜನಪದ ಗೊಬ್ಬು` | 44 |

**2-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `. _` | 75,674 |
| 2 | `ನ _` | 60,829 |
| 3 | `ದ _` | 53,561 |
| 4 | `, _` | 46,858 |
| 5 | `_ ಅ` | 44,463 |

**3-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `ರ್ . _` | 22,026 |
| 2 | `_ ಮ ಲ್` | 19,290 |
| 3 | `_ ಬು ಕ್` | 17,803 |
| 4 | `ಬು ಕ್ ಕೊ` | 16,406 |
| 5 | `ಕ್ ಕೊ _` | 16,006 |

**4-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ ಬು ಕ್ ಕೊ` | 16,108 |
| 2 | `ಬು ಕ್ ಕೊ _` | 15,889 |
| 3 | `_ ಒಂ ಜಿ _` | 6,746 |
| 4 | `ತೆ ರ್ . _` | 5,683 |
| 5 | `ಪುಂ ಡು . _` | 5,602 |

**5-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ ಬು ಕ್ ಕೊ _` | 15,626 |
| 2 | `_ ಉಂ ಡು . _` | 4,227 |
| 3 | `_ ಮ ಲ್ ತೆ ರ್` | 2,964 |
| 4 | `_ ಉ ಲ್ ಲೇ ಕೊ` | 2,908 |
| 5 | `ಪು ವೆ ರ್ . _` | 2,871 |


### Key Findings

- **Best Perplexity:** 2-gram (subword) with 2,884
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
- **Coverage:** Top-1000 patterns cover ~13% 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 | 0.6655 | 1.586 | 3.96 | 194,606 | 33.5% |
| **1** | Subword | 1.3289 | 2.512 | 22.51 | 3,088 | 0.0% |
| **2** | Word | 0.1252 | 1.091 | 1.21 | 768,732 | 87.5% |
| **2** | Subword | 0.8840 | 1.845 | 5.37 | 69,504 | 11.6% |
| **3** | Word | 0.0244 | 1.017 | 1.03 | 929,219 | 97.6% |
| **3** | Subword | 0.5546 | 1.469 | 2.95 | 372,985 | 44.5% |
| **4** | Word | 0.0078 🏆 | 1.005 | 1.01 | 956,167 | 99.2% |
| **4** | Subword | 0.3581 | 1.282 | 1.82 | 1,099,032 | 64.2% |

### Generated Text Samples (Word-based)

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

**Context Size 1:**

1. `ಬುಕ್ಕೊ ಭಗವಾನ್ ಕ್ರಷ್ಣ ಜನ್ಮಾಷ್ಟಮಿನ್ ಊರುದ ಭೂತೊಲೆನ ಬದಿಮಾಡೊ ಮಾತ ಯಾದವೆರ್ ಆಗಸ್ಟ್ ಮೇರ್ ಇಹಲೋಕ ಬುಡ್ಪುನೆಟ ಅಮೇರಿ...`
2. `ಈ ಪರಮಾಣು ಇಂದು ನಿರ್ಮಾನ ಮಾಲ್ತೆರ್ ಅಯಿನ್ ಸಪ್ಪು ಸೊಪ್ಪು ಪನ್ಪೇರ್ ಆಂಡಲಾ ಉಂದು ಪುರತಾನಡು ಕರಿಂಜೆ ಬೈಲ್ ಇಂದ್ ಪಣ್ಪೆ...`
3. `ಒಂಜಿ ಮುಖ್ಯವಾಯಿನ ಅಂಶೊಲೆನ್ ಸಮಾಧಾನ ಪಡಿಸಾದ್ ಬೇತೆ ಬೇತೆ ಪುದರುಲು ಫೆಬ್ರುವರಿಸೌತ್‌ ಏಷ್ಯನ್‌ ಕ್ರೀಡಾಕೂಟ ಭಾರತ ರಾಷ್...`

**Context Size 2:**

1. `ಬೇತೆ ಬೇತೆ ಪ್ರಕಾರೊದಂಚಿನ ಬೇರುಲು ತೋಜಿದ್ ಬತ್ತ್ಂಚಡ್ ಕಾವ್ಯೊ ಮಹಾಕಾವ್ಯೊ ಕಿನ್ಯ ಕತೆ ಕಾದಂಬರಿ ಬಂಗಾರ್‌ದಂಗಿದ ಕತೆ ನ...`
2. `ಸುರು ಮಲ್ತೆರ್ ಅಲ್ಪ ಕಲ್ಕತ್ತಾ ವಿಶ್ವವಿದ್ಯಾಲಯೊಗು ಪ್ರವೇಶ ಪರೀಕ್ಷೆಡ್ ಪಾಸ್ ಆಯೆರೆ ಸಾಧ್ಯ ಉಂಡು ಸರಕಾರೊಗು ಜಾಸ್ತಿ ತ...`
3. `ಮಲ್ತ್ ದ್ ಮೈಸೂರು ವಿಶ್ವವಿದ್ಯಾನಿಲಯದ್ ಪಿಎಚ್ ಡಿ ನಿಬಂಧ ತುಳು ಜನಪದ ಕಾವ್ಯಗಳಲ್ಲಿ ಸಮಾನ ಆಶಯಗಳು ತೌಲನಿಕ ಅಧ್ಯಯನ ಉದ್...`

**Context Size 3:**

1. `from the original on 16 june retrieved 16 june 15 ನೇ ವರ್ಷ ಉಪ್ಪುನಗನೆ ಆಂಕರ್ ಆದ್ ಪಾದಾರ್ಪಣೆ ಮಂತಿನ ಮೋಝಿ`
2. `archived from the original on 28 january govind mishra gets saraswati samman the hindu 12 february a...`
3. `the original on h e schapiro s j farah i hau j use of primates in the eu`

**Context Size 4:**

1. `archived from the original on 25 september retrieved ಈ ಕ್ರಮೊನ್ ಡೇವಿಡ್ ಲೀನ್ ಬುಕ್ಕೊ ಇಂಗ್ಮರ್ ಬರ್ಗ್ಮನ್ ಆ...`
2. `from the original on retrieved ವೃತ್ತಿಜೀವನೊ ಶಬರಿಮಲೆ ಸ್ವಾಮಿ ಚಿತ್ರೊದ ಶ್ರೀಹರಿ ಮಾಯೆಯ ಅವತಾರ ಪನ್ಪಿನ ಪದ್ಯಗ್ ...`
3. `ಬೇತೆ ಬಾಸೆಡ್ ಗೊಬ್ಬುದ ಪುದರ್ ನೀರಿನಲ್ಲಿ ಎಣಿಕೆಯ ಆಟ ಕನ್ನಡಡ್ ಉಲ್ಲೇಕೊಲು ಗೊಬ್ಬುಲು ಆಟಿ ತಿಂಗೊಲು`


### Generated Text Samples (Subword-based)

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

**Context Size 1:**

1. `_೨೯೦೦_ಬರ್._mba)_ಮುಂ`
2. `ರ್._ಪುನಡ್_ಹೋಲಿ_ಗುಳು-_ಸುತ್`
3. `ನ_ಕ್ಲಾ_ತನ್ನುಕೂರುಚಿಗೆಲು_ಒರಿ`

**Context Size 2:**

1. `._ಗೆಲ್_ಸೇಂಟ್_ಗೆಂದಿನೆಡ್ದಾವರಪುನ`
2. `ನ_ಪಂಚದರ_ರಾ_ಅವೇ_ಮುಲ್ಪ_ಬ`
3. `ದ_ಮಲ್ಪುವೆರ್._ಪನ್ಪೆರ್._ಚೆಪ್`

**Context Size 3:**

1. `ರ್._ಉಂದು_ಗಳಸೊಂದೆರ್._ಮುರಳಿ_ಮೋ`
2. `_ಮಲ್ಪುನೆನ್_ಎಸ್.ಆರ್.ಟಿ.ರಾಮರಾ`
3. `_ಬುಕ್ಕೊ_ಬುಕ್ಕೊ_‘ಸಮಾಜಶಾಸ್ತ್ರ,`

**Context Size 4:**

1. `_ಬುಕ್ಕೊ_ವೃತ್ತಿಜೀವನೊನು,_ಝೀ_ಕನ್ನ`
2. `ಬುಕ್ಕೊ_ಎದುರು_ಮಾತಾ_ಬಾರಿ_ತಮಿಳು_ಚಿ`
3. `_ಒಂಜಿ_ಸಂವತ್ಸರೊಗು_ಸಂಬಂಧ_ಪಟ್ಟಿ_`


### Key Findings

- **Best Predictability:** Context-4 (word) with 99.2% predictability
- **Branching Factor:** Decreases with context size (more deterministic)
- **Memory Trade-off:** Larger contexts require more storage (1,099,032 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 | 69,521 |
| Total Tokens | 891,538 |
| Mean Frequency | 12.82 |
| Median Frequency | 3 |
| Frequency Std Dev | 98.43 |

### Most Common Words

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | ಬುಕ್ಕೊ | 16,006 |
| 2 | ಈ | 8,118 |
| 3 | ಒಂಜಿ | 7,047 |
| 4 | ಉಂಡು | 5,603 |
| 5 | ಉಂದು | 3,570 |
| 6 | ಬೇತೆ | 3,318 |
| 7 | ಡ್ | 3,231 |
| 8 | ದ | 2,967 |
| 9 | ಮಲ್ತೆರ್ | 2,955 |
| 10 | ಉಲ್ಲೇಕೊಲು | 2,788 |

### Least Common Words (from vocabulary)

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | ನಿಯೋಕಾರ್ಟೆಕ್ಸ್ | 2 |
| 2 | ಆನೆಡ್ದ್ | 2 |
| 3 | ಆನೆಲೆಡ್ | 2 |
| 4 | ಆನೆಲೆಡ್ದ್ | 2 |
| 5 | ಅರಿವಿನ | 2 |
| 6 | ಸಪ್ಸತಾನ | 2 |
| 7 | ದುಸಿಟ್ | 2 |
| 8 | ಕಿಯಾಕ್ಕಾಯಿ | 2 |
| 9 | ಸಪಸತಾನ | 2 |
| 10 | ಬ್ಯೂರೋಗ್ | 2 |

### Zipf's Law Analysis

| Metric | Value |
|--------|-------|
| Zipf Coefficient | 0.8974 |
| R² (Goodness of Fit) | 0.993056 |
| Adherence Quality | **excellent** |

### Coverage Analysis

| Top N Words | Coverage |
|-------------|----------|
| Top 100 | 18.1% |
| Top 1,000 | 42.4% |
| Top 5,000 | 64.8% |
| Top 10,000 | 74.7% |

### Key Findings

- **Zipf Compliance:** R²=0.9931 indicates excellent adherence to Zipf's law
- **High Frequency Dominance:** Top 100 words cover 18.1% of corpus
- **Long Tail:** 59,521 words needed for remaining 25.3% 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 | 0.9138 🏆 | 0.2830 | N/A | N/A |
| **mono_64d** | 64 | 0.8541 | 0.2149 | N/A | N/A |
| **mono_128d** | 128 | 0.4366 | 0.1887 | N/A | N/A |
| **aligned_32d** | 32 | 0.9138 | 0.2860 | 0.0120 | 0.0520 |
| **aligned_64d** | 64 | 0.8541 | 0.2203 | 0.0040 | 0.0900 |
| **aligned_128d** | 128 | 0.4366 | 0.1899 | 0.0240 | 0.1360 |

### Key Findings

- **Best Isotropy:** mono_32d with 0.9138 (more uniform distribution)
- **Semantic Density:** Average pairwise similarity of 0.2305. Lower values indicate better semantic separation.
- **Alignment Quality:** Aligned models achieve up to 2.4% R@1 in cross-lingual retrieval.
- **Recommendation:** 128d aligned for best cross-lingual performance

---
## 6.  Morphological Analysis (Experimental)

This section presents an automated morphological analysis derived from the statistical divergence between word-level and subword-level models. By analyzing where subword predictability spikes and where word-level coverage fails, we can infer linguistic structures without supervised data.

### 6.1 Productivity & Complexity

| Metric | Value | Interpretation | Recommendation |
|--------|-------|----------------|----------------|
| Productivity Index | **5.000** | High morphological productivity | Reliable analysis |
| Idiomaticity Gap | **0.167** | 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 |
|--------|----------|
| `-ಕ` | ಕುದಿಕೆ, ಕುಡಿಯ, ಕೊಕ್ಕೋ |
| `-ಸ` | ಸ್ಪಿನ್ನರ್, ಸಾಲದಾಕುಳೆಗ್, ಸುಂದರ |
| `-ಮ` | ಮಳೆಕಾಡು, ಮಾಣಿ, ಮುಟ್ಟೊ |
| `-ಪ` | ಪಬ್ಲಿಷಿಂಗ್, ಪುಟ್ಟ್‌ದಿಪ್ಪೊಡು, ಪುಟ್ಟವೊಂಡಿನ |
| `-ಬ` | ಬುಕ್ಕೊಪಟ್ಣ, ಬೇಲೆದಕುಲೆಗ್, ಬಳಲೊಂದು |
| `-ಅ` | ಅರೆತ್ತ್, ಅಶ್ವಿನಿನ, ಅಪ್ನಿ |
| `-ನ` | ನೆಂಪಾ, ನೆಂಪು, ನಡಪಾಬಿ |
| `-ತ` | ತಾಲ್ದ್‌ಂಡ್ಂದ್, ತುಳುನಾಡ್‍ದ, ತಪಾಸಣೆ |

#### Productive Suffixes
| Suffix | Examples |
|--------|----------|
| `-ನ` | ಪುಟ್ಟವೊಂಡಿನ, ರೂಪಿಸಾಯಿನ, ಶೆಟ್ರೆನ |
| `-ದ` | ತುಳುನಾಡ್‍ದ, ಎರಡನೆದ, ವಿಲೀನೊದ |
| `-ರ` | ಸುಂದರ, ಸೊರ, ಸುಶ್ರೀಂದ್ರ |
| `-ಯ` | ಶ್ರೀಕಂಠಯ್ಯ, ಕುಡಿಯ, ಕ್ರಿಟೇಷಿಯ |
| `-ತ` | ಶಕ್ತಿತ, ಅಮುರ್ತ, ನುಗತ್ತ |
| `-ಕ` | ವೈಶೇಷಿಕ, ಸ್ತಾಪಕ, ಪಾರಂಪರಿಕ |
| `-s` | magnets, rights, mers |
| `-ಲ` | ಗೊತ್ತಿಲ್ಲ, ದಾದಂಡಲ, ಪೊರ್ಲ |

### 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` | 3.03x | 9 contexts | action, nation, nations |
| `atio` | 3.02x | 6 contexts | nation, nations, national |

### 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 |
|--------|--------|-----------|----------|
| `-ಸ` | `-ದ` | 48 words | ಸುತ್ತುಮುತ್ತುದ, ಸಂಬರೊದ |
| `-ಪ` | `-ನ` | 43 words | ಪುರುಷೆರೆನ, ಪುಡಾಯಿನ |
| `-ಕ` | `-ದ` | 38 words | ಕಾಸರಗೋಡುದ, ಕೋರ್ಟ್‌ದ |
| `-ಮ` | `-ದ` | 37 words | ಮುದೆಲ್‍ದ, ಮೇಘನಾದ |
| `-ಸ` | `-ನ` | 37 words | ಸರ್ವಜ್ಞನ, ಸೃಷ್ಟಿಯಾಯಿನ |
| `-ಕ` | `-ನ` | 36 words | ಕಣೊಕುಲೆನ, ಕಡ್ತೊಂದಿನ |
| `-ಮ` | `-ನ` | 36 words | ಮಂಜಿನ, ಮಂಥನ |
| `-ಬ` | `-ದ` | 35 words | ಬೆರಸ್‍ದ, ಬೇರದ |
| `-ಪ` | `-ದ` | 32 words | ಪುದರ್‍ದ, ಪರಿಚ್ಚೇದೊದ |
| `-ಬ` | `-ನ` | 30 words | ಬದಲ್ಪುನ, ಬೂರಿನ |

### 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 |
|------|-----------------|------------|------|
| ಏಕಪಾತ್ರಾಭಿನಯ | **`ಏಕಪಾತ್ರಾಭಿ-ನ-ಯ`** | 7.5 | `ನ` |
| ಸಮಾಜೊಲೆನ್ | **`ಸ-ಮ-ಾಜೊಲೆನ್`** | 4.5 | `ಾಜೊಲೆನ್` |
| ಇಪ್ಪುವೇರ್ | **`ಇ-ಪ-್ಪುವೇರ್`** | 4.5 | `್ಪುವೇರ್` |
| ಜಪಾನಿಯೆರೆಗ್ | **`ಜ-ಪ-ಾನಿಯೆರೆಗ್`** | 4.5 | `ಾನಿಯೆರೆಗ್` |
| ಮಗುಪ್ಪೊಡು | **`ಮ-ಗ-ುಪ್ಪೊಡು`** | 4.5 | `ುಪ್ಪೊಡು` |
| ಎಲ್ಯಪನಿರ್ದ್ | **`ಎ-ಲ-್ಯಪನಿರ್ದ್`** | 4.5 | `್ಯಪನಿರ್ದ್` |
| ಇನ್ನಿಂಗ್ಸ್ದ | **`ಇನ್ನಿಂಗ್ಸ್-ದ`** | 4.5 | `ಇನ್ನಿಂಗ್ಸ್` |
| ಪತ್ರೊಲೆಡ್ | **`ಪ-ತ-್ರೊಲೆಡ್`** | 4.5 | `್ರೊಲೆಡ್` |
| ಉಡುಪಿಡುಪ್ಪುನ | **`ಉಡುಪಿಡುಪ್ಪು-ನ`** | 4.5 | `ಉಡುಪಿಡುಪ್ಪು` |
| ಆದಿಪ್ಪುಂದು | **`ಆ-ದ-ಿಪ್ಪುಂದು`** | 4.5 | `ಿಪ್ಪುಂದು` |
| ವ್ಯಾಪಾರೊದ | **`ವ್ಯಾಪಾರೊ-ದ`** | 4.5 | `ವ್ಯಾಪಾರೊ` |
| ಜತ್ತಿನಾರ್ | **`ಜ-ತ-್ತಿನಾರ್`** | 4.5 | `್ತಿನಾರ್` |
| traditions | **`tradition-s`** | 4.5 | `tradition` |
| ಅಮಿಲಾಯಿಡ್ | **`ಅ-ಮ-ಿಲಾಯಿಡ್`** | 4.5 | `ಿಲಾಯಿಡ್` |
| ಶ್ರೀಹರಿಕೋಟಾದ | **`ಶ್ರೀಹರಿಕೋಟಾ-ದ`** | 4.5 | `ಶ್ರೀಹರಿಕೋಟಾ` |

### 6.6 Linguistic Interpretation

> **Automated Insight:**
The language Tulu 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 (4.49x) |
| N-gram | **2-gram** | Lowest perplexity (2,884) |
| Markov | **Context-4** | Highest predictability (99.2%) |
| 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: 2026-01-11 00:33:22*