README.md

---
language: zea
language_name: Zeelandic
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: 4.195
  - name: best_isotropy
    type: isotropy
    value: 0.7531
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-11
---

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

This repository contains NLP models trained and evaluated by Wikilangs, specifically on **Zeelandic** 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** | 3.358x | 3.36 | 0.1058% | 433,648 |
| **16k** | 3.668x | 3.67 | 0.1156% | 397,034 |
| **32k** | 3.937x | 3.94 | 0.1241% | 369,853 |
| **64k** | 4.195x 🏆 | 4.20 | 0.1322% | 347,155 |

### Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

**Sample 1:** `12 juni is d'n 163e of 164e dag (bie een schrikkeljaer) van 't jaer.`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁ 1 2 ▁juni ▁is ▁d ' n ▁ 1 ... (+20 more)` | 30 |
| 16k | `▁ 1 2 ▁juni ▁is ▁d ' n ▁ 1 ... (+20 more)` | 30 |
| 32k | `▁ 1 2 ▁juni ▁is ▁d ' n ▁ 1 ... (+20 more)` | 30 |
| 64k | `▁ 1 2 ▁juni ▁is ▁d ' n ▁ 1 ... (+20 more)` | 30 |

**Sample 2:** `is 'n jaer. Gebeurtenisse 5 juni - Op last van de Franse keizer Napoleon wor de ...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁is ▁' n ▁jaer . ▁gebeurtenisse ▁ 5 ▁juni ▁- ... (+22 more)` | 32 |
| 16k | `▁is ▁' n ▁jaer . ▁gebeurtenisse ▁ 5 ▁juni ▁- ... (+20 more)` | 30 |
| 32k | `▁is ▁' n ▁jaer . ▁gebeurtenisse ▁ 5 ▁juni ▁- ... (+20 more)` | 30 |
| 64k | `▁is ▁' n ▁jaer . ▁gebeurtenisse ▁ 5 ▁juni ▁- ... (+18 more)` | 28 |

**Sample 3:** `Sri Lanka is 'n land in Azië, d'n 'oôdstad is Sri Jayewardenapura Kotte. Groôste...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁sri ▁lanka ▁is ▁' n ▁land ▁in ▁azië , ▁d ... (+35 more)` | 45 |
| 16k | `▁sri ▁lanka ▁is ▁' n ▁land ▁in ▁azië , ▁d ... (+33 more)` | 43 |
| 32k | `▁sri ▁lanka ▁is ▁' n ▁land ▁in ▁azië , ▁d ... (+29 more)` | 39 |
| 64k | `▁sri ▁lanka ▁is ▁' n ▁land ▁in ▁azië , ▁d ... (+26 more)` | 36 |


### Key Findings

- **Best Compression:** 64k achieves 4.195x compression
- **Lowest UNK Rate:** 8k with 0.1058% 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 | 2,743 | 11.42 | 15,853 | 37.5% | 62.2% |
| **2-gram** | Subword | 285 🏆 | 8.16 | 2,525 | 65.2% | 99.1% |
| **3-gram** | Word | 3,421 | 11.74 | 23,993 | 38.7% | 58.8% |
| **3-gram** | Subword | 2,246 | 11.13 | 20,884 | 26.9% | 72.1% |
| **4-gram** | Word | 6,678 | 12.71 | 47,341 | 34.4% | 50.1% |
| **4-gram** | Subword | 10,644 | 13.38 | 102,913 | 14.8% | 45.3% |
| **5-gram** | Word | 5,192 | 12.34 | 39,407 | 37.4% | 52.6% |
| **5-gram** | Subword | 29,540 | 14.85 | 232,889 | 10.5% | 34.0% |

### Top 5 N-grams by Size

**2-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `van de` | 6,350 |
| 2 | `in de` | 6,008 |
| 3 | `in frankriek` | 4,947 |
| 4 | `is n` | 4,303 |
| 5 | `vogges t` | 3,505 |

**3-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `lienks nae buten` | 3,384 |
| 2 | `in de rehio` | 1,790 |
| 3 | `in t departement` | 1,769 |
| 4 | `is n hemeênte` | 1,766 |
| 5 | `n hemeênte in` | 1,764 |

**4-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `is n hemeênte in` | 1,762 |
| 2 | `n hemeênte in t` | 1,755 |
| 3 | `t bureau van de` | 1,754 |
| 4 | `de statistiek n in` | 1,754 |
| 5 | `van de statistiek n` | 1,754 |

**5-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `is n hemeênte in t` | 1,755 |
| 2 | `t bureau van de statistiek` | 1,754 |
| 3 | `van de statistiek n in` | 1,754 |
| 4 | `bureau van de statistiek n` | 1,754 |
| 5 | `de statistiek n in frankriek` | 1,754 |

**2-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `n _` | 164,170 |
| 2 | `e _` | 153,880 |
| 3 | `e n` | 115,339 |
| 4 | `e r` | 100,491 |
| 5 | `d e` | 89,945 |

**3-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `e n _` | 59,274 |
| 2 | `_ d e` | 53,896 |
| 3 | `d e _` | 49,282 |
| 4 | `_ i n` | 42,462 |
| 5 | `i n _` | 36,109 |

**4-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ d e _` | 41,150 |
| 2 | `_ i n _` | 32,453 |
| 3 | `_ v a n` | 25,691 |
| 4 | `v a n _` | 24,842 |
| 5 | `n _ d e` | 19,736 |

**5-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ v a n _` | 24,505 |
| 2 | `n _ d e _` | 16,103 |
| 3 | `a n _ d e` | 9,072 |
| 4 | `e _ i n _` | 8,231 |
| 5 | `v a n _ d` | 8,211 |


### Key Findings

- **Best Perplexity:** 2-gram (subword) with 285
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
- **Coverage:** Top-1000 patterns cover ~34% 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 | 0.7527 | 1.685 | 4.57 | 73,879 | 24.7% |
| **1** | Subword | 1.3603 | 2.567 | 10.32 | 532 | 0.0% |
| **2** | Word | 0.2261 | 1.170 | 1.54 | 337,089 | 77.4% |
| **2** | Subword | 1.0696 | 2.099 | 6.58 | 5,488 | 0.0% |
| **3** | Word | 0.0785 | 1.056 | 1.14 | 515,910 | 92.2% |
| **3** | Subword | 0.9247 | 1.898 | 4.52 | 36,113 | 7.5% |
| **4** | Word | 0.0353 🏆 | 1.025 | 1.06 | 586,459 | 96.5% |
| **4** | Subword | 0.6763 | 1.598 | 2.77 | 163,034 | 32.4% |

### Generated Text Samples (Word-based)

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

**Context Size 1:**

1. `de gift erkent naemelijk hriekenland cyprus lid van begunne james challis aerzelend een autobedrief ...`
2. `in brussels hewest ok gerekend is de bevolkiengsdichteid bedroe 33 3 w aan ze kwam m`
3. `n zuster van de jaer gebeurtenisse 18 km bevolkienge in frankriek aod chazemais lei op de`

**Context Size 2:**

1. `van de gilberteilan n phoenixeilan n line eilan n of t angrenzende guatemala maekt anspraek op de`
2. `in de laete negentiende eêuw in de rehio picardie in frankriek geograofische informaotie artonges le...`
3. `is n hemeênte in t departement loire en de vrouwe bin net als aore grote steden in`

**Context Size 3:**

1. `lienks nae buten britannica fact file city population klimaatinfo liggienge links thumb kaerte in zw...`
2. `in de rehio picardie in frankriek geograofische informaotie barenton cel lei op de coördinaot n 49 0...`
3. `in t departement alpes de haute provence in de rehio auvergne in frankriek geograofische informaotie...`

**Context Size 4:**

1. `is n hemeênte in t departement aisne in de rehio picardie in frankriek geograofische informaotie la ...`
2. `n hemeênte in t departement ain in de rehio rhône alpes in frankriek geograofische informaotie courc...`
3. `statistiek n in frankriek aod saint julien d asse is n hemeênte in t departement aisne in de rehio`


### Generated Text Samples (Subword-based)

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

**Context Size 1:**

1. `_u)_d_dadier,_eo`
2. `erid’_'nt_n_oen,`
3. `n_a_din_hi_veses`

**Context Size 2:**

1. `n_somant._vortemb`
2. `e_hei_elive-a_pre`
3. `en_ad_eênt_he_300`

**Context Size 3:**

1. `en_van_de_salmanda`
2. `_de_vèr)_lieë,_'ao`
3. `de_31,8_mie_andamm`

**Context Size 4:**

1. `_de_botte_world_fac`
2. `_in_frankriek_meer.`
3. `_van_de_stant_insee`


### Key Findings

- **Best Predictability:** Context-4 (word) with 96.5% predictability
- **Branching Factor:** Decreases with context size (more deterministic)
- **Memory Trade-off:** Larger contexts require more storage (163,034 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 | 32,227 |
| Total Tokens | 787,829 |
| Mean Frequency | 24.45 |
| Median Frequency | 4 |
| Frequency Std Dev | 425.77 |

### Most Common Words

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | de | 42,264 |
| 2 | in | 32,757 |
| 3 | n | 30,236 |
| 4 | van | 24,663 |
| 5 | t | 18,522 |
| 6 | en | 15,110 |
| 7 | is | 12,805 |
| 8 | een | 8,217 |
| 9 | op | 7,396 |
| 10 | d | 5,762 |

### Least Common Words (from vocabulary)

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | groussbus | 2 |
| 2 | saeul | 2 |
| 3 | useldange | 2 |
| 4 | vichten | 2 |
| 5 | kiischpelt | 2 |
| 6 | kommunistische | 2 |
| 7 | zunneverduusterieng | 2 |
| 8 | eclipsewise | 2 |
| 9 | grifformeêrd | 2 |
| 10 | charkov | 2 |

### Zipf's Law Analysis

| Metric | Value |
|--------|-------|
| Zipf Coefficient | 1.0798 |
| R² (Goodness of Fit) | 0.997599 |
| Adherence Quality | **excellent** |

### Coverage Analysis

| Top N Words | Coverage |
|-------------|----------|
| Top 100 | 50.3% |
| Top 1,000 | 74.0% |
| Top 5,000 | 86.6% |
| Top 10,000 | 91.8% |

### Key Findings

- **Zipf Compliance:** R²=0.9976 indicates excellent adherence to Zipf's law
- **High Frequency Dominance:** Top 100 words cover 50.3% of corpus
- **Long Tail:** 22,227 words needed for remaining 8.2% 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 | 0.7531 | 0.3541 | N/A | N/A |
| **mono_64d** | 64 | 0.4175 | 0.3237 | N/A | N/A |
| **mono_128d** | 128 | 0.0896 | 0.3307 | N/A | N/A |
| **aligned_32d** | 32 | 0.7531 🏆 | 0.3585 | 0.0340 | 0.2080 |
| **aligned_64d** | 64 | 0.4175 | 0.3260 | 0.0620 | 0.2600 |
| **aligned_128d** | 128 | 0.0896 | 0.3217 | 0.0940 | 0.3260 |

### Key Findings

- **Best Isotropy:** aligned_32d with 0.7531 (more uniform distribution)
- **Semantic Density:** Average pairwise similarity of 0.3358. Lower values indicate better semantic separation.
- **Alignment Quality:** Aligned models achieve up to 9.4% 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 | **0.548** | 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 |
|--------|----------|
| `-b` | bievoegelijke, benediktsson, bommenwêrrepers |
| `-s` | stortte, stoffels, sovjetpresident |
| `-a` | aaien, a15, amor |
| `-e` | eslogen, ergste, eige |
| `-m` | melanocharis, michigan, mantel |
| `-be` | benediktsson, bestoot, beleven |
| `-k` | kassapa, kat, kroatisch |
| `-d` | du, dong, droizy |

#### Productive Suffixes
| Suffix | Examples |
|--------|----------|
| `-e` | colonne, bievoegelijke, ergste |
| `-n` | eslogen, aaien, benediktsson |
| `-en` | eslogen, aaien, lampen |
| `-s` | melanocharis, bommenwêrrepers, cnemotriccus |
| `-t` | kat, vaorieert, verdeêlt |
| `-d` | banjaerd, rehenwoud, eerlijkeid |
| `-r` | pêr, omar, christopher |
| `-er` | christopher, onmiskenbaer, creuzier |

### 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 |
|------|----------|------------------|----------|
| `sche` | 1.77x | 55 contexts | schep, schei, scheer |
| `nder` | 1.60x | 60 contexts | onder, ander, under |
| `chte` | 1.47x | 82 contexts | achte, echte, zochte |
| `isch` | 1.94x | 27 contexts | visch, episch, typisch |
| `enge` | 1.63x | 50 contexts | engel, ienge, hienge |
| `eder` | 1.77x | 36 contexts | ieder, ceder, reder |
| `onde` | 1.57x | 57 contexts | onden, ondek, konde |
| `erde` | 1.41x | 72 contexts | erder, derde, verde |
| `ienk` | 1.60x | 39 contexts | dienk, lienk, wienk |
| `emen` | 1.44x | 28 contexts | jemen, nemen, remens |
| `geme` | 1.58x | 16 contexts | gemet, gemert, gemeên |
| `uten` | 1.48x | 18 contexts | futen, outen, buten |

### 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` | `-n` | 105 words | stichtten, speelgelegenheden |
| `-s` | `-e` | 104 words | sprake, studie |
| `-b` | `-e` | 91 words | biolohische, belgische |
| `-s` | `-en` | 87 words | stichtten, speelgelegenheden |
| `-a` | `-e` | 84 words | angenome, afrikaanse |
| `-b` | `-n` | 83 words | beton, bussen |
| `-g` | `-e` | 70 words | grooste, gekoze |
| `-s` | `-s` | 70 words | schans, syrrhaptes |
| `-g` | `-n` | 66 words | gerben, gerdien |
| `-k` | `-e` | 61 words | kiescollege, konienginne |

### 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 |
|------|-----------------|------------|------|
| biblioteek | **`bibliot-e-ek`** | 7.5 | `e` |
| pannerden | **`panner-d-en`** | 7.5 | `d` |
| castaneus | **`castan-e-us`** | 7.5 | `e` |
| hartennes | **`harten-n-es`** | 7.5 | `n` |
| iengelsman | **`iengels-m-an`** | 7.5 | `m` |
| waoterdunen | **`waoterdu-n-en`** | 7.5 | `n` |
| brandaris | **`branda-r-is`** | 7.5 | `r` |
| ijsselmeer | **`ijsselm-e-er`** | 7.5 | `e` |
| wullemsen | **`wullem-s-en`** | 7.5 | `s` |
| waerneêmer | **`waerneê-m-er`** | 7.5 | `m` |
| verkennen | **`verken-n-en`** | 7.5 | `n` |
| pâturages | **`pâtura-ge-s`** | 7.5 | `ge` |
| begeerten | **`be-ge-erten`** | 7.5 | `erten` |
| regerienk | **`re-ge-rienk`** | 7.5 | `rienk` |
| rekenieng | **`reken-ie-ng`** | 6.0 | `reken` |

### 6.6 Linguistic Interpretation

> **Automated Insight:**
The language Zeelandic shows high morphological productivity. The subword models are significantly more efficient than word models, suggesting a rich system of affixation or compounding.

> **Note on Idiomaticity:** The high Idiomaticity Gap suggests a large number of frequent multi-word expressions or formulaic sequences that are statistically distinct from their component parts.

---
## 7. Summary & Recommendations

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

### Production Recommendations

| Component | Recommended | Rationale |
|-----------|-------------|-----------|
| Tokenizer | **64k BPE** | Best compression (4.19x) |
| N-gram | **2-gram** | Lowest perplexity (285) |
| Markov | **Context-4** | Highest predictability (96.5%) |
| 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 05:53:31*