README.md

---
language: war
language_name: Waray
language_family: austronesian_philippine_central
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_philippine_central
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: 3.934
  - name: best_isotropy
    type: isotropy
    value: 0.8470
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-11
---

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

This repository contains NLP models trained and evaluated by Wikilangs, specifically on **Waray** 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** | 3.050x | 3.05 | 0.5424% | 356,557 |
| **16k** | 3.360x | 3.36 | 0.5975% | 323,705 |
| **32k** | 3.653x | 3.66 | 0.6496% | 297,739 |
| **64k** | 3.934x 🏆 | 3.94 | 0.6996% | 276,458 |

### Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

**Sample 1:** `An Condado han Philadelphia in uska condado ha estado han Estados Unidos nga Pen...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁an ▁condado ▁han ▁philadelphia ▁in ▁uska ▁condado ▁ha ▁estado ▁han ... (+9 more)` | 19 |
| 16k | `▁an ▁condado ▁han ▁philadelphia ▁in ▁uska ▁condado ▁ha ▁estado ▁han ... (+6 more)` | 16 |
| 32k | `▁an ▁condado ▁han ▁philadelphia ▁in ▁uska ▁condado ▁ha ▁estado ▁han ... (+5 more)` | 15 |
| 64k | `▁an ▁condado ▁han ▁philadelphia ▁in ▁uska ▁condado ▁ha ▁estado ▁han ... (+5 more)` | 15 |

**Sample 2:** `An Condado han Atchison in uska condado ha estado han Estados Unidos nga Missour...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁an ▁condado ▁han ▁at chis on ▁in ▁uska ▁condado ▁ha ... (+7 more)` | 17 |
| 16k | `▁an ▁condado ▁han ▁at chis on ▁in ▁uska ▁condado ▁ha ... (+7 more)` | 17 |
| 32k | `▁an ▁condado ▁han ▁at chis on ▁in ▁uska ▁condado ▁ha ... (+7 more)` | 17 |
| 64k | `▁an ▁condado ▁han ▁at chis on ▁in ▁uska ▁condado ▁ha ... (+7 more)` | 17 |

**Sample 3:** `An Briatexte amo in usa ka komyun ha departamento han Tarn ngan ha rehiyon han M...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁an ▁b ria tex te ▁amo ▁in ▁usa ▁ka ▁komyun ... (+21 more)` | 31 |
| 16k | `▁an ▁b ria tex te ▁amo ▁in ▁usa ▁ka ▁komyun ... (+19 more)` | 29 |
| 32k | `▁an ▁b ria tex te ▁amo ▁in ▁usa ▁ka ▁komyun ... (+19 more)` | 29 |
| 64k | `▁an ▁b ria tex te ▁amo ▁in ▁usa ▁ka ▁komyun ... (+19 more)` | 29 |


### Key Findings

- **Best Compression:** 64k achieves 3.934x compression
- **Lowest UNK Rate:** 8k with 0.5424% 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 | 6,771 | 12.73 | 587,286 | 39.6% | 56.6% |
| **2-gram** | Subword | 247 🏆 | 7.95 | 5,747 | 68.3% | 99.7% |
| **3-gram** | Word | 17,032 | 14.06 | 1,208,224 | 31.7% | 47.8% |
| **3-gram** | Subword | 1,598 | 10.64 | 47,440 | 34.4% | 76.9% |
| **4-gram** | Word | 50,245 | 15.62 | 3,168,479 | 23.5% | 38.5% |
| **4-gram** | Subword | 5,867 | 12.52 | 301,900 | 26.2% | 55.3% |
| **5-gram** | Word | 68,533 | 16.06 | 2,392,175 | 19.2% | 34.2% |
| **5-gram** | Subword | 14,203 | 13.79 | 1,156,846 | 23.4% | 47.8% |

### Top 5 N-grams by Size

**2-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `nahilalakip ha` | 1,196,502 |
| 2 | `in nahilalakip` | 1,196,444 |
| 3 | `in uska` | 1,161,092 |
| 4 | `mga kasarigan` | 1,140,811 |
| 5 | `familia nga` | 1,136,619 |

**3-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `in nahilalakip ha` | 1,196,442 |
| 2 | `ha genus nga` | 1,059,395 |
| 3 | `nahilalakip ha genus` | 1,059,391 |
| 4 | `uska species han` | 1,059,159 |
| 5 | `in uska species` | 1,059,158 |

**4-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `nahilalakip ha genus nga` | 1,059,391 |
| 2 | `in nahilalakip ha genus` | 1,059,349 |
| 3 | `in uska species han` | 1,059,153 |
| 4 | `hini subspecies nga nakalista` | 1,022,924 |
| 5 | `waray hini subspecies nga` | 1,022,924 |

**5-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `in nahilalakip ha genus nga` | 1,059,349 |
| 2 | `waray hini subspecies nga nakalista` | 1,022,924 |
| 3 | `hini subspecies nga nakalista mga` | 1,022,834 |
| 4 | `subspecies nga nakalista mga kasarigan` | 1,022,831 |
| 5 | `in uska species han magnoliopsida` | 202,211 |

**2-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `a _` | 15,944,152 |
| 2 | `n _` | 13,013,735 |
| 3 | `a n` | 11,288,391 |
| 4 | `_ n` | 11,062,347 |
| 5 | `g a` | 9,641,404 |

**3-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `a n _` | 7,717,416 |
| 2 | `n g a` | 6,387,666 |
| 3 | `_ n g` | 6,209,737 |
| 4 | `g a _` | 6,146,693 |
| 5 | `_ h a` | 4,483,798 |

**4-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ n g a` | 6,206,771 |
| 2 | `n g a _` | 4,713,126 |
| 3 | `g a n _` | 2,648,745 |
| 4 | `_ i n _` | 2,593,266 |
| 5 | `s p e c` | 2,524,428 |

**5-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ n g a _` | 4,702,772 |
| 2 | `s p e c i` | 2,510,965 |
| 3 | `p e c i e` | 2,471,786 |
| 4 | `e c i e s` | 2,467,835 |
| 5 | `c i e s _` | 2,396,660 |


### Key Findings

- **Best Perplexity:** 2-gram (subword) with 247
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
- **Coverage:** Top-1000 patterns cover ~48% 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 | 0.9524 | 1.935 | 6.58 | 763,014 | 4.8% |
| **1** | Subword | 0.8828 | 1.844 | 5.77 | 3,095 | 11.7% |
| **2** | Word | 0.3392 | 1.265 | 1.87 | 4,989,422 | 66.1% |
| **2** | Subword | 0.6967 | 1.621 | 4.99 | 17,724 | 30.3% |
| **3** | Word | 0.2062 | 1.154 | 1.43 | 9,222,656 | 79.4% |
| **3** | Subword | 0.7365 | 1.666 | 4.85 | 88,254 | 26.3% |
| **4** | Word | 0.1336 🏆 | 1.097 | 1.25 | 13,040,247 | 86.6% |
| **4** | Subword | 0.7108 | 1.637 | 3.87 | 427,478 | 28.9% |

### Generated Text Samples (Word-based)

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

**Context Size 1:**

1. `nga nakalista mga sumpay ha familia nga ginhulagway ni léon dufour hadton an notocelis ngan familia`
2. `in nahilalakip ha genus nga nakalista mga kasarigan website zoonomen zoological nomenclature and ter...`
3. `an leptomastix tanasijtshuki in uska species han bryozoa world checklist of the global lepidoptera n...`

**Context Size 2:**

1. `nahilalakip ha genus nga tonnoiriella ngan familia nga terebellidae waray hini subspecies nga nakali...`
2. `in nahilalakip ha genus nga clitoria ngan familia nga rubiaceae waray hini subspecies nga nakalista ...`
3. `in uska species han diptera nga ginhulagway ni henri ernest baillon ngan ginhatag han pagkayana nga ...`

**Context Size 3:**

1. `in nahilalakip ha genus nga buxus ngan familia nga cheliferidae waray hini subspecies nga nakalista ...`
2. `ha genus nga exorista ngan familia nga tetranychidae waray hini subspecies nga nakalista mga kasarig...`
3. `nahilalakip ha genus nga miostauropus ngan familia nga rutelidae waray hini subspecies nga nakalista...`

**Context Size 4:**

1. `nahilalakip ha genus nga euprosopia ngan familia nga platystomatidae mabibilngan ini ha sri lanka wa...`
2. `in nahilalakip ha genus nga haploops ngan familia nga ampeliscidae waray hini subspecies nga nakalis...`
3. `in uska species han liliopsida nga ginhulagway ni john hutchinson ngan john mcewan dalziel an bertie...`


### Generated Text Samples (Subword-based)

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

**Context Size 1:**

1. `_doni_n_his_wano`
2. `abs_ithonce_ndul`
3. `ngspysogakid_and`

**Context Size 2:**

1. `a_ini_wcspectic_t`
2. `n_in_390._ara).,_`
3. `anterk_nganthopte`

**Context Size 3:**

1. `an_pindros_the_e._`
2. `nga_cera_nga_syste`
3. `_nga_sipalmeestrit`

**Context Size 4:**

1. `_nga_syahan_diaphen`
2. `nga_ginhulagway_ni_`
3. `gan_familia_nga_nak`


### Key Findings

- **Best Predictability:** Context-4 (word) with 86.6% predictability
- **Branching Factor:** Decreases with context size (more deterministic)
- **Memory Trade-off:** Larger contexts require more storage (427,478 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 | 548,741 |
| Total Tokens | 79,294,042 |
| Mean Frequency | 144.50 |
| Median Frequency | 4 |
| Frequency Std Dev | 11269.34 |

### Most Common Words

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | nga | 4,703,546 |
| 2 | in | 2,921,835 |
| 3 | an | 2,589,624 |
| 4 | ha | 1,893,436 |
| 5 | han | 1,625,103 |
| 6 | ngan | 1,415,641 |
| 7 | mga | 1,401,419 |
| 8 | species | 1,374,922 |
| 9 | of | 1,347,178 |
| 10 | nahilalakip | 1,196,509 |

### Least Common Words (from vocabulary)

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | sermersooq | 2 |
| 2 | kulusuk | 2 |
| 3 | sporcle | 2 |
| 4 | loocnon | 2 |
| 5 | estadyum | 2 |
| 6 | katalyst | 2 |
| 7 | haluk | 2 |
| 8 | bilginer | 2 |
| 9 | squibb | 2 |
| 10 | paca | 2 |

### Zipf's Law Analysis

| Metric | Value |
|--------|-------|
| Zipf Coefficient | 1.1284 |
| R² (Goodness of Fit) | 0.998865 |
| Adherence Quality | **excellent** |

### Coverage Analysis

| Top N Words | Coverage |
|-------------|----------|
| Top 100 | 58.7% |
| Top 1,000 | 76.4% |
| Top 5,000 | 85.5% |
| Top 10,000 | 89.1% |

### Key Findings

- **Zipf Compliance:** R²=0.9989 indicates excellent adherence to Zipf's law
- **High Frequency Dominance:** Top 100 words cover 58.7% of corpus
- **Long Tail:** 538,741 words needed for remaining 10.9% 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 | 0.8470 | 0.3349 | N/A | N/A |
| **mono_64d** | 64 | 0.8029 | 0.3103 | N/A | N/A |
| **mono_128d** | 128 | 0.7779 | 0.2306 | N/A | N/A |
| **aligned_32d** | 32 | 0.8470 🏆 | 0.3453 | 0.0380 | 0.2940 |
| **aligned_64d** | 64 | 0.8029 | 0.3113 | 0.1440 | 0.4780 |
| **aligned_128d** | 128 | 0.7779 | 0.2362 | 0.2780 | 0.6400 |

### Key Findings

- **Best Isotropy:** aligned_32d with 0.8470 (more uniform distribution)
- **Semantic Density:** Average pairwise similarity of 0.2947. Lower values indicate better semantic separation.
- **Alignment Quality:** Aligned models achieve up to 27.8% 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 | **-0.284** | 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 |
|--------|----------|
| `-a` | ameziane, alucao, astroma |
| `-ma` | majerorum, maracasi, marsa |
| `-s` | sustansiya, szijji, subtriflora |
| `-pa` | parle, pasiphimus, paradontophora |
| `-ca` | callites, callochiton, cabanisii |
| `-p` | parle, psdamyc, polystylata |
| `-ba` | basilewskyanus, bathyprion, balangigan |
| `-b` | biselinifera, basilewskyanus, bicolora |

#### Productive Suffixes
| Suffix | Examples |
|--------|----------|
| `-s` | basilewskyanus, bomareoides, pasiphimus |
| `-a` | sustansiya, wolynia, biselinifera |
| `-us` | basilewskyanus, pasiphimus, hesperotychus |
| `-is` | niasensis, moniliventris, admotalis |
| `-e` | ocellate, ameziane, parle |
| `-um` | majerorum, curvinervium, tapinelytrum |
| `-i` | kuehhasii, minjujuŭi, maracasi |
| `-es` | bomareoides, callites, pedes |

### 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 |
|------|----------|------------------|----------|
| `kasa` | 2.52x | 45 contexts | akasa, kasaï, kasal |
| `ilal` | 2.61x | 36 contexts | silal, bilal, kilal |
| `asar` | 2.10x | 62 contexts | casar, asara, pasar |
| `opte` | 1.62x | 202 contexts | opter, scoptes, eoptera |
| `ulag` | 2.87x | 14 contexts | bulag, dulag, nabulag |
| `ahil` | 2.32x | 27 contexts | dahil, tahil, kahili |
| `akal` | 2.15x | 34 contexts | zakal, yakal, bakal |
| `akip` | 3.06x | 11 contexts | lakip, nakipa, kalakip |
| `agwa` | 2.64x | 16 contexts | dagway, magway, nagwara |
| `fami` | 2.07x | 24 contexts | famiy, famil, famiie |
| `peci` | 1.77x | 36 contexts | specis, specia, pecina |
| `subs` | 2.90x | 8 contexts | subst, subsp, subsaga |

### 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 |
|--------|--------|-----------|----------|
| `-a` | `-s` | 303 words | antoninus, albocingulatus |
| `-p` | `-s` | 258 words | plethobasus, platypygus |
| `-a` | `-a` | 251 words | artopenna, anahita |
| `-p` | `-a` | 250 words | pindica, paradoxophyla |
| `-s` | `-a` | 225 words | spodopa, synpsylla |
| `-s` | `-s` | 220 words | simaethis, spartiformis |
| `-m` | `-s` | 147 words | machetis, mesozonalis |
| `-a` | `-us` | 135 words | antoninus, albocingulatus |
| `-b` | `-s` | 111 words | beombawigulensis, boridiensis |
| `-m` | `-a` | 109 words | missiona, mauritacantha |

### 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 |
|------|-----------------|------------|------|
| yungchangense | **`yungchangen-s-e`** | 7.5 | `s` |
| betinensis | **`betinen-s-is`** | 7.5 | `s` |
| hundungensis | **`hundungen-s-is`** | 7.5 | `s` |
| yoneyamai | **`yoneyam-a-i`** | 7.5 | `a` |
| austrocolumbiana | **`austrocolumbi-a-na`** | 7.5 | `a` |
| glycereen | **`glycer-e-en`** | 7.5 | `e` |
| agesilaus | **`agesi-la-us`** | 7.5 | `la` |
| discoelongata | **`discoelong-a-ta`** | 7.5 | `a` |
| desordenata | **`desorden-a-ta`** | 7.5 | `a` |
| sexpectinata | **`sexpectin-a-ta`** | 7.5 | `a` |
| pluvigena | **`pluvig-e-na`** | 7.5 | `e` |
| chondrichthyans | **`chondrichthy-a-ns`** | 7.5 | `a` |
| ecorticata | **`ecortic-a-ta`** | 7.5 | `a` |
| binuangan | **`binuang-a-n`** | 7.5 | `a` |
| dallasiana | **`dallasi-a-na`** | 7.5 | `a` |

### 6.6 Linguistic Interpretation

> **Automated Insight:**
The language Waray 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 (3.93x) |
| N-gram | **2-gram** | Lowest perplexity (247) |
| Markov | **Context-4** | Highest predictability (86.6%) |
| 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 05:52:21*