README.md

---
language: vls
language_name: West Flemish
language_family: germanic_west_continental
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-germanic_west_continental
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: 4.163
  - name: best_isotropy
    type: isotropy
    value: 0.8756
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-11
---

# West Flemish - Wikilangs Models
## Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on **West Flemish** 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** | 3.334x | 3.34 | 0.0287% | 502,567 |
| **16k** | 3.665x | 3.67 | 0.0315% | 457,201 |
| **32k** | 3.934x | 3.94 | 0.0338% | 425,860 |
| **64k** | 4.163x 🏆 | 4.17 | 0.0358% | 402,499 |

### Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

**Sample 1:** `Achtntwientig is 't getal 28, e nateurlik getal achter zeevnetwientig en voorn n...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁a chtn tw ientig ▁is ▁' t ▁getal ▁ 2 ... (+18 more)` | 28 |
| 16k | `▁a chtn tw ientig ▁is ▁' t ▁getal ▁ 2 ... (+18 more)` | 28 |
| 32k | `▁a chtn twientig ▁is ▁' t ▁getal ▁ 2 8 ... (+13 more)` | 23 |
| 64k | `▁achtn twientig ▁is ▁' t ▁getal ▁ 2 8 , ... (+12 more)` | 22 |

**Sample 2:** `de volksnoame van de gemêente Ôostrôzebeke e dêelgemêente van Stoan, zie: Rôzebe...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁de ▁volk sn oame ▁van ▁de ▁gemêente ▁ôostrôzebeke ▁e ▁dêelgemêente ... (+10 more)` | 20 |
| 16k | `▁de ▁volk sn oame ▁van ▁de ▁gemêente ▁ôostrôzebeke ▁e ▁dêelgemêente ... (+10 more)` | 20 |
| 32k | `▁de ▁volk snoame ▁van ▁de ▁gemêente ▁ôostrôzebeke ▁e ▁dêelgemêente ▁van ... (+8 more)` | 18 |
| 64k | `▁de ▁volk snoame ▁van ▁de ▁gemêente ▁ôostrôzebeke ▁e ▁dêelgemêente ▁van ... (+8 more)` | 18 |

**Sample 3:** `Paltoga (Russisch: Палтога) is e dorp in Rusland in 't district Vytegorsky (obla...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁pal t og a ▁( russisch : ▁ п а ... (+37 more)` | 47 |
| 16k | `▁pal t og a ▁( russisch : ▁ п а ... (+35 more)` | 45 |
| 32k | `▁pal t oga ▁( russisch : ▁ п ал то ... (+29 more)` | 39 |
| 64k | `▁palt oga ▁( russisch : ▁ п ал то г ... (+28 more)` | 38 |


### Key Findings

- **Best Compression:** 64k achieves 4.163x compression
- **Lowest UNK Rate:** 8k with 0.0287% 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 | 12,804 | 13.64 | 41,132 | 15.7% | 36.6% |
| **2-gram** | Subword | 282 🏆 | 8.14 | 3,241 | 64.7% | 99.2% |
| **3-gram** | Word | 27,763 | 14.76 | 51,974 | 7.7% | 22.9% |
| **3-gram** | Subword | 2,519 | 11.30 | 27,863 | 25.7% | 68.5% |
| **4-gram** | Word | 45,411 | 15.47 | 74,505 | 6.8% | 17.6% |
| **4-gram** | Subword | 15,236 | 13.90 | 154,373 | 12.4% | 36.1% |
| **5-gram** | Word | 30,248 | 14.88 | 47,265 | 8.2% | 19.7% |
| **5-gram** | Subword | 57,965 | 15.82 | 420,619 | 7.2% | 22.1% |

### Top 5 N-grams by Size

**2-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `van de` | 15,489 |
| 2 | `in de` | 10,285 |
| 3 | `in t` | 6,874 |
| 4 | `van t` | 5,995 |
| 5 | `en de` | 3,723 |

**3-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `joar in de` | 850 |
| 2 | `van t joar` | 791 |
| 3 | `bouwkundig erfgoed in` | 765 |
| 4 | `in west vloandern` | 742 |
| 5 | `t joar is` | 714 |

**4-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `t joar is t` | 693 |
| 2 | `eeuwe volgenst de christelikke` | 526 |
| 3 | `volgenst de christelikke joartellienge` | 526 |
| 4 | `noa bouwkundig erfgoed in` | 354 |
| 5 | `t ende van t` | 337 |

**5-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `eeuwe volgenst de christelikke joartellienge` | 526 |
| 2 | `t ende van t joar` | 304 |
| 3 | `volgenst de christelikke joartellienge gebeurtenissn` | 292 |
| 4 | `lyste van bouwkundig erfgoed in` | 251 |
| 5 | `toet t ende van t` | 250 |

**2-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `n _` | 399,169 |
| 2 | `e _` | 395,658 |
| 3 | `e r` | 217,859 |
| 4 | `e n` | 214,189 |
| 5 | `d e` | 208,906 |

**3-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ d e` | 123,266 |
| 2 | `d e _` | 116,073 |
| 3 | `a n _` | 97,189 |
| 4 | `e n _` | 96,860 |
| 5 | `_ v a` | 80,611 |

**4-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ d e _` | 90,909 |
| 2 | `_ v a n` | 76,359 |
| 3 | `v a n _` | 74,288 |
| 4 | `_ i n _` | 52,878 |
| 5 | `n _ d e` | 48,858 |

**5-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ v a n _` | 73,086 |
| 2 | `n _ d e _` | 39,289 |
| 3 | `a n _ d e` | 22,613 |
| 4 | `v a n _ d` | 21,346 |
| 5 | `e _ v a n` | 19,923 |


### Key Findings

- **Best Perplexity:** 2-gram (subword) with 282
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
- **Coverage:** Top-1000 patterns cover ~22% 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 | 0.8228 | 1.769 | 5.44 | 158,804 | 17.7% |
| **1** | Subword | 1.2080 | 2.310 | 9.96 | 735 | 0.0% |
| **2** | Word | 0.2583 | 1.196 | 1.64 | 860,998 | 74.2% |
| **2** | Subword | 1.0608 | 2.086 | 6.74 | 7,322 | 0.0% |
| **3** | Word | 0.0895 | 1.064 | 1.15 | 1,409,019 | 91.1% |
| **3** | Subword | 0.9474 | 1.928 | 4.92 | 49,306 | 5.3% |
| **4** | Word | 0.0313 🏆 | 1.022 | 1.05 | 1,616,997 | 96.9% |
| **4** | Subword | 0.7502 | 1.682 | 3.21 | 242,577 | 25.0% |

### Generated Text Samples (Word-based)

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

**Context Size 1:**

1. `de wyk van t nôordn gruujn dikkers in kontrast me 3 juli gin êen of mêercellig`
2. `van yper wunt en nieuw ryk in de kustvlaktn groene bewegienge wordn ze egliek nie kost`
3. `in ip t volgn nog 293 noa bouwkundig erfgoed bevern en mêer tyd toen ze van`

**Context Size 2:**

1. `van de verênigde stoatn busschn`
2. `in de dertiende êeuwe dus vès ipgedolvn gebied o den ôostkant van de verênigde stoatn en kanada`
3. `in t ôostn an ciney in noamn in en je viel italië were binn de stad stroomde`

**Context Size 3:**

1. `joar in de 13e of 14e êeuwe en van de 50 000 en 120 000 beschreevn sôortn varieern`
2. `van t joar geboorn pontormo gabriel fahrenheit gustaaf flamen emiel lauwers bob dylan gestorvn jozef...`
3. `bouwkundig erfgoed in tiegem in west vloandern t es eignlyk nen ouden arm van den aa t grenst`

**Context Size 4:**

1. `t joar is t 80e joar in de 10e eeuwe volgenst de christelikke joartellienge mmxii is e schrikkeljoar...`
2. `volgenst de christelikke joartellienge gebeurtenissn 25 april hertog jan zounder vrêes legt an d ips...`
3. `eeuwe volgenst de christelikke joartellienge gebeurtenissn april 5 de west vlamsche coureur gaston r...`


### Generated Text Samples (Subword-based)

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

**Context Size 1:**

1. `_scoe_taz,_we_ve`
2. `e,_scar-êli²_man`
3. `ndstoone_zogers_`

**Context Size 2:**

1. `n_'t_vroegroudt_a`
2. `e_priens)_gië_e_s`
3. `erd_ipparem_moste`

**Context Size 3:**

1. `_de_vanasamuele_(>`
2. `de_piegouwne_refeu`
3. `an_beken_deel_rede`

**Context Size 4:**

1. `_de_schopinidad_er_`
2. `_van_mandsche_kenme`
3. `van_flandn_ip_ne_bu`


### Key Findings

- **Best Predictability:** Context-4 (word) with 96.9% predictability
- **Branching Factor:** Decreases with context size (more deterministic)
- **Memory Trade-off:** Larger contexts require more storage (242,577 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 | 68,458 |
| Total Tokens | 1,735,026 |
| Mean Frequency | 25.34 |
| Median Frequency | 4 |
| Frequency Std Dev | 600.62 |

### Most Common Words

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | de | 93,287 |
| 2 | van | 73,544 |
| 3 | in | 53,708 |
| 4 | en | 49,180 |
| 5 | t | 45,426 |
| 6 | e | 21,400 |
| 7 | is | 17,745 |
| 8 | zyn | 16,831 |
| 9 | n | 15,475 |
| 10 | die | 12,301 |

### Least Common Words (from vocabulary)

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | myzeqe | 2 |
| 2 | seman | 2 |
| 3 | rumn | 2 |
| 4 | peshkopi | 2 |
| 5 | dibër | 2 |
| 6 | города | 2 |
| 7 | uytvoernde | 2 |
| 8 | stoatssecretoarisn | 2 |
| 9 | soamnstellinge | 2 |
| 10 | mph | 2 |

### Zipf's Law Analysis

| Metric | Value |
|--------|-------|
| Zipf Coefficient | 1.0178 |
| R² (Goodness of Fit) | 0.998718 |
| Adherence Quality | **excellent** |

### Coverage Analysis

| Top N Words | Coverage |
|-------------|----------|
| Top 100 | 40.8% |
| Top 1,000 | 63.3% |
| Top 5,000 | 79.0% |
| Top 10,000 | 85.3% |

### Key Findings

- **Zipf Compliance:** R²=0.9987 indicates excellent adherence to Zipf's law
- **High Frequency Dominance:** Top 100 words cover 40.8% of corpus
- **Long Tail:** 58,458 words needed for remaining 14.7% 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 | 0.8756 🏆 | 0.3181 | N/A | N/A |
| **mono_64d** | 64 | 0.8383 | 0.2517 | N/A | N/A |
| **mono_128d** | 128 | 0.5888 | 0.2007 | N/A | N/A |
| **aligned_32d** | 32 | 0.8756 | 0.3113 | 0.0840 | 0.3740 |
| **aligned_64d** | 64 | 0.8383 | 0.2465 | 0.1380 | 0.4500 |
| **aligned_128d** | 128 | 0.5888 | 0.2020 | 0.2000 | 0.5260 |

### Key Findings

- **Best Isotropy:** mono_32d with 0.8756 (more uniform distribution)
- **Semantic Density:** Average pairwise similarity of 0.2550. Lower values indicate better semantic separation.
- **Alignment Quality:** Aligned models achieve up to 20.0% 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 | **5.000** | High morphological productivity | Reliable analysis |
| Idiomaticity Gap | **-0.109** | 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 |
|--------|----------|
| `-s` | soôrtn, schwaben, schick |
| `-b` | binnstroomde, bolivië, biezelehe |
| `-a` | arenaria, addington, amazing |
| `-ge` | gelanceerd, gezeyd, gevoenn |
| `-o` | oendregienk, ogtepunt, omwald |
| `-be` | bewaren, bees, bedek |
| `-k` | kurs, kommiesje, koopman |
| `-d` | dié, donetsk, darling |

#### Productive Suffixes
| Suffix | Examples |
|--------|----------|
| `-e` | underne, binnstroomde, poginge |
| `-n` | soôrtn, fryslân, hopeweunn |
| `-s` | zothuus, kurs, cervantes |
| `-t` | ogtepunt, varlet, capaciteit |
| `-en` | conservatieven, schwaben, bewaren |
| `-d` | vervolgd, omwald, tulband |
| `-ge` | poginge, franstalige, lancerienge |
| `-r` | elektrotoer, êesteminister, hour |

### 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 |
|------|----------|------------------|----------|
| `enge` | 2.33x | 50 contexts | engel, oenger, mengel |
| `sche` | 1.68x | 141 contexts | schee, asche, vasche |
| `chte` | 1.60x | 115 contexts | achte, echte, vichte |
| `fran` | 2.05x | 37 contexts | frank, franz, frang |
| `schi` | 1.77x | 65 contexts | schip, schie, schid |
| `icht` | 1.56x | 114 contexts | richt, licht, vicht |
| `isch` | 1.83x | 51 contexts | ischl, visch, vischn |
| `hter` | 1.94x | 38 contexts | ahter, echter, achter |
| `nder` | 1.41x | 150 contexts | ander, under, onder |
| `elik` | 1.72x | 51 contexts | gelik, tielik, feliks |
| `oate` | 1.77x | 40 contexts | zoate, oater, moate |
| `erke` | 1.54x | 66 contexts | kerke, berke, werke |

### 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 |
|--------|--------|-----------|----------|
| `-s` | `-e` | 169 words | subklasse, sukerziekte |
| `-b` | `-e` | 149 words | bulskampstroate, beschoafde |
| `-s` | `-n` | 125 words | skorsenelen, steeën |
| `-b` | `-n` | 114 words | blokkn, behunn |
| `-k` | `-e` | 108 words | kunstacademie, kassie |
| `-m` | `-e` | 100 words | muuzee, multiple |
| `-o` | `-n` | 95 words | oafbusschn, ofebrookn |
| `-o` | `-e` | 91 words | ounbevlekte, omriengende |
| `-d` | `-e` | 90 words | dagtemprateure, duytstoalige |
| `-a` | `-e` | 88 words | adresse, ansluutienge |

### 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 |
|------|-----------------|------------|------|
| fermenteren | **`fermenter-e-n`** | 7.5 | `e` |
| benoaderd | **`benoa-de-rd`** | 7.5 | `de` |
| bruggelingen | **`bruggeling-e-n`** | 7.5 | `e` |
| romantiek | **`romanti-e-k`** | 7.5 | `e` |
| vruchtvlees | **`vruchtv-le-es`** | 7.5 | `le` |
| treuzelen | **`treuze-le-n`** | 7.5 | `le` |
| vluchters | **`vlucht-e-rs`** | 7.5 | `e` |
| resources | **`resourc-e-s`** | 7.5 | `e` |
| splenters | **`splent-e-rs`** | 7.5 | `e` |
| ipbryngsten | **`ipbryngst-e-n`** | 7.5 | `e` |
| vienkezetters | **`vienkezett-e-rs`** | 7.5 | `e` |
| knobbeltjes | **`knobbeltj-e-s`** | 7.5 | `e` |
| beweegboar | **`beweegbo-a-r`** | 7.5 | `a` |
| schoonhoven | **`schoonhov-e-n`** | 7.5 | `e` |
| donspluumtjes | **`donspluumtj-e-s`** | 7.5 | `e` |

### 6.6 Linguistic Interpretation

> **Automated Insight:**
The language West Flemish 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 (4.16x) |
| N-gram | **2-gram** | Lowest perplexity (282) |
| Markov | **Context-4** | Highest predictability (96.9%) |
| 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 03:19:22*