README.md

---
language: trv
language_name: Taroko
language_family: austronesian_formosan
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_formosan
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: 3.923
  - name: best_isotropy
    type: isotropy
    value: 0.7817
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-11
---

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

This repository contains NLP models trained and evaluated by Wikilangs, specifically on **Taroko** 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** | 3.409x | 3.41 | 0.1717% | 804,137 |
| **16k** | 3.644x | 3.65 | 0.1835% | 752,396 |
| **32k** | 3.786x | 3.79 | 0.1907% | 724,248 |
| **64k** | 3.923x 🏆 | 3.92 | 0.1976% | 698,951 |

### Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

**Sample 1:** `Nlixan （丟棄的線） EX:smeli naq ware puto sneqic nlixan bubu na ka laqi mqedin. Pnyah...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁n lixan ▁( 丟 棄 的 線 ) ▁ex : ... (+22 more)` | 32 |
| 16k | `▁nlixan ▁( 丟 棄 的 線 ) ▁ex : sme ... (+19 more)` | 29 |
| 32k | `▁nlixan ▁( 丟 棄 的 線 ) ▁ex : smeli ... (+17 more)` | 27 |
| 64k | `▁nlixan ▁( 丟棄的線 ) ▁ex : smeli ▁naq ▁ware ▁puto ... (+14 more)` | 24 |

**Sample 2:** `Empprngaw kari(溝通、談話) Yaku ni bubu mu, empprngaw kari han!`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁empprngaw ▁kari ( 溝 通 、 談 話 ) ▁yaku ... (+8 more)` | 18 |
| 16k | `▁empprngaw ▁kari ( 溝 通 、 談話 ) ▁yaku ▁ni ... (+7 more)` | 17 |
| 32k | `▁empprngaw ▁kari ( 溝 通 、 談話 ) ▁yaku ▁ni ... (+7 more)` | 17 |
| 64k | `▁empprngaw ▁kari ( 溝通 、 談話 ) ▁yaku ▁ni ▁bubu ... (+6 more)` | 16 |

**Sample 3:** `縮圖|Reynhekwo , Switzerland Reynhekwo / Renhokuo (聯合國): 193個國`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁縮圖 | reyn he kwo ▁, ▁s wit zer land ... (+19 more)` | 29 |
| 16k | `▁縮圖 | reyn hekwo ▁, ▁switzerland ▁reyn hekwo ▁/ ▁ren ... (+11 more)` | 21 |
| 32k | `▁縮圖 | reyn hekwo ▁, ▁switzerland ▁reynhekwo ▁/ ▁ren hokuo ... (+9 more)` | 19 |
| 64k | `▁縮圖 | reynhekwo ▁, ▁switzerland ▁reynhekwo ▁/ ▁renhokuo ▁( 聯合國 ... (+6 more)` | 16 |


### Key Findings

- **Best Compression:** 64k achieves 3.923x compression
- **Lowest UNK Rate:** 8k with 0.1717% 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 | 7,668 | 12.90 | 19,784 | 17.2% | 43.1% |
| **2-gram** | Subword | 262 🏆 | 8.03 | 4,696 | 69.8% | 98.6% |
| **3-gram** | Word | 8,123 | 12.99 | 21,493 | 20.5% | 40.6% |
| **3-gram** | Subword | 1,934 | 10.92 | 22,856 | 31.5% | 73.9% |
| **4-gram** | Word | 13,253 | 13.69 | 36,605 | 20.8% | 34.4% |
| **4-gram** | Subword | 9,420 | 13.20 | 96,381 | 16.1% | 45.8% |
| **5-gram** | Word | 9,241 | 13.17 | 26,596 | 23.7% | 37.8% |
| **5-gram** | Subword | 28,374 | 14.79 | 203,086 | 10.5% | 30.9% |

### Top 5 N-grams by Size

**2-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `kiya ka` | 2,292 |
| 2 | `kana ka` | 1,899 |
| 3 | `seejiq o` | 1,657 |
| 4 | `tnpusu seejiq` | 1,508 |
| 5 | `o mangal` | 1,468 |

**3-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `tnpusu seejiq o` | 1,449 |
| 2 | `seejiq o mangal` | 1,444 |
| 3 | `pnyahan pnatas 參考資料` | 1,005 |
| 4 | `hiyi ka kana` | 723 |
| 5 | `sapah ka kneegu` | 722 |

**4-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `tnpusu seejiq o mangal` | 1,443 |
| 2 | `hiyi tnpusu seejiq o` | 722 |
| 3 | `na hiyi tnpusu seejiq` | 722 |
| 4 | `sapah ka kneegu na` | 722 |
| 5 | `ka kneegu na sapah` | 722 |

**5-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `ka kana knhbragan na hiyi` | 722 |
| 2 | `sapah ka kneegu na sapah` | 722 |
| 3 | `kana knhbragan na hiyi tnpusu` | 722 |
| 4 | `knhbragan na hiyi tnpusu seejiq` | 722 |
| 5 | `na hiyi tnpusu seejiq o` | 722 |

**2-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `a _` | 159,040 |
| 2 | `a n` | 151,698 |
| 3 | `_ k` | 114,078 |
| 4 | `n g` | 106,714 |
| 5 | `n _` | 93,857 |

**3-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `a n _` | 72,949 |
| 2 | `_ k a` | 53,488 |
| 3 | `k a _` | 48,276 |
| 4 | `a n g` | 38,914 |
| 5 | `n g _` | 36,466 |

**4-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ k a _` | 38,804 |
| 2 | `a n g _` | 18,270 |
| 3 | `g a n _` | 15,177 |
| 4 | `_ n a _` | 14,291 |
| 5 | `a l a n` | 13,651 |

**5-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `a l a n g` | 11,226 |
| 2 | `i q a n _` | 10,533 |
| 3 | `n i q a n` | 10,125 |
| 4 | `k a w a s` | 10,012 |
| 5 | `l a n g _` | 9,914 |


### Key Findings

- **Best Perplexity:** 2-gram (subword) with 262
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
- **Coverage:** Top-1000 patterns cover ~31% 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 | 0.7260 | 1.654 | 5.46 | 66,965 | 27.4% |
| **1** | Subword | 1.3366 | 2.526 | 8.22 | 3,648 | 0.0% |
| **2** | Word | 0.2882 | 1.221 | 1.68 | 365,251 | 71.2% |
| **2** | Subword | 0.4317 | 1.349 | 2.61 | 29,967 | 56.8% |
| **3** | Word | 0.0872 | 1.062 | 1.14 | 612,144 | 91.3% |
| **3** | Subword | 0.4820 | 1.397 | 2.65 | 78,261 | 51.8% |
| **4** | Word | 0.0266 🏆 | 1.019 | 1.04 | 698,380 | 97.3% |
| **4** | Subword | 0.4748 | 1.390 | 2.26 | 207,257 | 52.5% |

### Generated Text Samples (Word-based)

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

**Context Size 1:**

1. `ka tucay cungcen di ririh tan paah baraw o kndadax cu prajil pusa kari qpruhan nii`
2. `na lxanan waya mi kingal hngkawas na sin ing wen hwa 文化 pusu nniqan hiya han`
3. `o nirih na bukung klwaan cing ci pnaah hngkawas mnda kingal alang icil so niyi bungka`

**Context Size 2:**

1. `kiya ka kiya ni nii lhbun bi dgiyaq kana ki wada paru bale qqtaun quri kesun yisu`
2. `kana ka snluan ruwan klwaan dnii ga ida niqan ka sediq kiya knkana dapa lmiqu mi ccamac`
3. `seejiq o mangal 2 niqan 2 paru nniqan rnaaw ni ungat bi knsyangan ni niqan kingal ka`

**Context Size 3:**

1. `tnpusu seejiq o mangal 88 niqan 2 609 hiyi sp rahuq na uxay tnpusu seejiq o mangal 80`
2. `seejiq o mangal 83 niqan 1 347 hiyi koia kana ka kleegan seejiq ga ni rahuq na o4`
3. `pnyahan pnatas 參考資料 內政部戶政司全球資訊網 原住民族委員會全球資訊網統計資料 hangan alang 部落名稱 alang qnagan tukubeycu na alang 部...`

**Context Size 4:**

1. `tnpusu seejiq o mangal 6 niqan 7 hiyi koia kana ka kleegan seejiq ga ni rahuq na o1 pusupnyahan`
2. `hiyi tnpusu seejiq o mangal 97 niqan 331 hiyi sp rahuq na uxay tnpusu seejiq o mangal 41 niqan`
3. `hiyi ka kana knhbragan na hiyi tnpusu seejiq o mangal 75 niqan 779 hiyi koia kana ka kleegan seejiq`


### Generated Text Samples (Subword-based)

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

**Context Size 1:**

1. `_syrísna_po,_kng`
2. `ac_msun_musey_2_`
3. `n_c_mtax.】_hmiy-`

**Context Size 2:**

1. `a_mri_mdada_tru.s`
2. `angcin),_mqnhban_`
3. `_ki_mi_kapah_do_2`

**Context Size 3:**

1. `an_hiya_mpdaun_seu`
2. `_kanana_bale_meran`
3. `ka_uri,_beran_riyu`

**Context Size 4:**

1. `_ka_hmrinas_ka_daw,`
2. `ang_mkbrnux_na_skde`
3. `gan_kasi_ka_waso_ni`


### Key Findings

- **Best Predictability:** Context-4 (word) with 97.3% predictability
- **Branching Factor:** Decreases with context size (more deterministic)
- **Memory Trade-off:** Larger contexts require more storage (207,257 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 | 26,300 |
| Total Tokens | 761,987 |
| Mean Frequency | 28.97 |
| Median Frequency | 3 |
| Frequency Std Dev | 343.10 |

### Most Common Words

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | ka | 39,083 |
| 2 | na | 16,339 |
| 3 | o | 12,805 |
| 4 | alang | 9,788 |
| 5 | ni | 8,476 |
| 6 | u | 8,051 |
| 7 | niqan | 7,350 |
| 8 | mi | 6,845 |
| 9 | kiya | 6,666 |
| 10 | dha | 6,542 |

### Least Common Words (from vocabulary)

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | ptaqi | 2 |
| 2 | kiyang | 2 |
| 3 | skyidaw | 2 |
| 4 | qbrus | 2 |
| 5 | mnurax | 2 |
| 6 | kmawah | 2 |
| 7 | beydat | 2 |
| 8 | mjilux | 2 |
| 9 | 衣物等 | 2 |
| 10 | mpggaalu | 2 |

### Zipf's Law Analysis

| Metric | Value |
|--------|-------|
| Zipf Coefficient | 1.2169 |
| R² (Goodness of Fit) | 0.992292 |
| Adherence Quality | **excellent** |

### Coverage Analysis

| Top N Words | Coverage |
|-------------|----------|
| Top 100 | 43.4% |
| Top 1,000 | 73.8% |
| Top 5,000 | 89.7% |
| Top 10,000 | 94.4% |

### Key Findings

- **Zipf Compliance:** R²=0.9923 indicates excellent adherence to Zipf's law
- **High Frequency Dominance:** Top 100 words cover 43.4% of corpus
- **Long Tail:** 16,300 words needed for remaining 5.6% 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 | 0.7817 | 0.3299 | N/A | N/A |
| **mono_64d** | 64 | 0.5768 | 0.2955 | N/A | N/A |
| **mono_128d** | 128 | 0.1326 | 0.2814 | N/A | N/A |
| **aligned_32d** | 32 | 0.7817 🏆 | 0.3225 | 0.0220 | 0.1500 |
| **aligned_64d** | 64 | 0.5768 | 0.2983 | 0.0320 | 0.2400 |
| **aligned_128d** | 128 | 0.1326 | 0.2761 | 0.0640 | 0.2760 |

### Key Findings

- **Best Isotropy:** aligned_32d with 0.7817 (more uniform distribution)
- **Semantic Density:** Average pairwise similarity of 0.3006. Lower values indicate better semantic separation.
- **Alignment Quality:** Aligned models achieve up to 6.4% 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 | **0.223** | 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` | syawswocya, ssikun, sulu |
| `-m` | mbomou, mhiyang, mrunu |
| `-p` | psnaqun, philippine, psuung |
| `-t` | tyencucyaw, tmbawa, taha |
| `-k` | kayi, kntruma, kwose |
| `-c` | cyapiar, cyupin, cyuan |
| `-h` | hngakan, hwami, hnridan |
| `-n` | ncyaropihay, nga, nrihan |

#### Productive Suffixes
| Suffix | Examples |
|--------|----------|
| `-n` | qhdin, yican, hngakan |
| `-an` | yican, hngakan, dmatan |
| `-ng` | mhiyang, mkgarang, 1alang |
| `-g` | mhiyang, mkgarang, 1alang |
| `-a` | nga, syawswocya, tmbawa |
| `-u` | mbomou, mrunu, sulu |
| `-y` | ncyaropihay, aripay, amnesty |
| `-i` | kayi, yami, hwami |

### 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 |
|------|----------|------------------|----------|
| `uwan` | 1.92x | 113 contexts | tuwan, luwan, kuwan |
| `iyan` | 1.68x | 106 contexts | siyan, kiyan, diyan |
| `atas` | 2.25x | 22 contexts | matas, patas, natas |
| `inga` | 1.58x | 78 contexts | ingal, kinga, pingan |
| `eeji` | 2.41x | 16 contexts | seeji, seejia, seejiq |
| `ngal` | 1.55x | 74 contexts | mngal, ingal, ngala |
| `anga` | 1.34x | 137 contexts | manga, hanga, angal |
| `ahan` | 1.42x | 95 contexts | tahan, qahan, wahan |
| `seej` | 2.41x | 13 contexts | seeji, seejia, seejiq |
| `alay` | 1.96x | 22 contexts | balay, malay, lalay |
| `waan` | 2.00x | 20 contexts | rwaan, hwaan, kwaan |
| `lwaa` | 2.31x | 11 contexts | klwaam, klwaan, qlwaan |

### 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 |
|--------|--------|-----------|----------|
| `-p` | `-n` | 226 words | prilan, ptasun |
| `-s` | `-n` | 170 words | snhian, snluun |
| `-p` | `-an` | 170 words | prilan, ppaan |
| `-k` | `-n` | 135 words | kalibuan, kyrgyazstan |
| `-k` | `-an` | 108 words | kalibuan, kyrgyazstan |
| `-s` | `-an` | 107 words | snhian, snyusan |
| `-t` | `-n` | 107 words | tnegjyalan, tetun |
| `-c` | `-n` | 90 words | cangmyeyn, cungcgn |
| `-c` | `-ng` | 82 words | cucngtang, cinghung |
| `-c` | `-g` | 82 words | cucngtang, cinghung |

### 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 |
|------|-----------------|------------|------|
| syeyncing | **`syeync-i-ng`** | 7.5 | `i` |
| taypinyan | **`taypin-y-an`** | 7.5 | `y` |
| kongciyun | **`kongci-y-un`** | 7.5 | `y` |
| phdeyngki | **`p-h-deyngki`** | 7.5 | `deyngki` |
| sunghosay | **`sungho-s-ay`** | 7.5 | `s` |
| niyawcwey | **`niyawc-w-ey`** | 7.5 | `w` |
| mingcutan | **`mingcu-t-an`** | 7.5 | `t` |
| tyeynsing | **`tyeyns-i-ng`** | 7.5 | `i` |
| pncubuwan | **`pn-cu-buwan`** | 7.5 | `buwan` |
| mincucuyi | **`mincucu-y-i`** | 7.5 | `y` |
| teynckung | **`teynck-u-ng`** | 7.5 | `u` |
| yueynsuay | **`yueyns-u-ay`** | 7.5 | `u` |
| pnkbrihan | **`pn-k-brihan`** | 7.5 | `brihan` |
| hwangcuyey | **`hwangcu-y-ey`** | 7.5 | `y` |
| peyruskeni | **`peyruske-n-i`** | 7.5 | `n` |

### 6.6 Linguistic Interpretation

> **Automated Insight:**
The language Taroko 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 (3.92x) |
| N-gram | **2-gram** | Lowest perplexity (262) |
| Markov | **Context-4** | Highest predictability (97.3%) |
| 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 01:38:25*