README.md

---
language: el
language_name: Greek
language_family: greek
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-greek
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: 4.872
  - name: best_isotropy
    type: isotropy
    value: 0.8028
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-10
---

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

This repository contains NLP models trained and evaluated by Wikilangs, specifically on **Greek** 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** | 3.621x | 3.62 | 0.0471% | 2,711,752 |
| **16k** | 4.087x | 4.09 | 0.0531% | 2,402,524 |
| **32k** | 4.519x | 4.52 | 0.0587% | 2,172,769 |
| **64k** | 4.872x 🏆 | 4.87 | 0.0633% | 2,015,689 |

### Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

**Sample 1:** `.ms είναι ο top-level domain κωδικός για το Μοντσερράτ στο Διαδίκτυο. Δείτε επίσ...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁. ms ▁είναι ▁ο ▁top - level ▁domain ▁κω δικόσ ... (+30 more)` | 40 |
| 16k | `▁. ms ▁είναι ▁ο ▁top - level ▁domain ▁κωδικόσ ▁για ... (+21 more)` | 31 |
| 32k | `▁. ms ▁είναι ▁ο ▁top - level ▁domain ▁κωδικόσ ▁για ... (+21 more)` | 31 |
| 64k | `▁. ms ▁είναι ▁ο ▁top - level ▁domain ▁κωδικόσ ▁για ... (+19 more)` | 29 |

**Sample 2:** `Το Φόππολο (ιταλικά: Foppolo) είναι ιταλικός δήμος στην Επαρχία του Μπέργκαμο, σ...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁το ▁φ όπ πο λο ▁( ιταλικά : ▁f op ... (+32 more)` | 42 |
| 16k | `▁το ▁φ όπ πο λο ▁( ιταλικά : ▁f op ... (+28 more)` | 38 |
| 32k | `▁το ▁φ όπ πο λο ▁( ιταλικά : ▁f op ... (+25 more)` | 35 |
| 64k | `▁το ▁φ όπ πο λο ▁( ιταλικά : ▁f op ... (+21 more)` | 31 |

**Sample 3:** `Το Λε Τορ () είναι γαλλική κοινότητα στο νομό της Ερ, στη διοικητική περιοχή της...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁το ▁λε ▁τορ ▁() ▁είναι ▁γαλλική ▁κοινότητα ▁στο ▁νομό ▁τησ ... (+15 more)` | 25 |
| 16k | `▁το ▁λε ▁τορ ▁() ▁είναι ▁γαλλική ▁κοινότητα ▁στο ▁νομό ▁τησ ... (+14 more)` | 24 |
| 32k | `▁το ▁λε ▁τορ ▁() ▁είναι ▁γαλλική ▁κοινότητα ▁στο ▁νομό ▁τησ ... (+13 more)` | 23 |
| 64k | `▁το ▁λε ▁τορ ▁() ▁είναι ▁γαλλική ▁κοινότητα ▁στο ▁νομό ▁τησ ... (+13 more)` | 23 |


### Key Findings

- **Best Compression:** 64k achieves 4.872x compression
- **Lowest UNK Rate:** 8k with 0.0471% 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 | 254,029 | 17.95 | 2,414,487 | 7.3% | 17.4% |
| **2-gram** | Subword | 443 🏆 | 8.79 | 26,716 | 56.5% | 96.8% |
| **3-gram** | Word | 1,488,610 | 20.51 | 5,529,817 | 1.9% | 6.3% |
| **3-gram** | Subword | 3,933 | 11.94 | 250,216 | 24.2% | 59.6% |
| **4-gram** | Word | 3,845,615 | 21.87 | 9,144,193 | 1.3% | 3.9% |
| **4-gram** | Subword | 22,210 | 14.44 | 1,519,855 | 12.8% | 34.2% |
| **5-gram** | Word | 2,910,168 | 21.47 | 5,914,525 | 1.4% | 4.2% |
| **5-gram** | Subword | 87,887 | 16.42 | 5,267,290 | 7.2% | 20.9% |

### Top 5 N-grams by Size

**2-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `από το` | 323,213 |
| 2 | `από την` | 290,152 |
| 3 | `με την` | 252,647 |
| 4 | `από τον` | 241,108 |
| 5 | `για την` | 198,175 |

**3-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `κατά τη διάρκεια` | 71,561 |
| 2 | `παραπομπές εξωτερικοί σύνδεσμοι` | 62,539 |
| 3 | `τη διάρκεια της` | 34,723 |
| 4 | `για πρώτη φορά` | 29,480 |
| 5 | `σύμφωνα με την` | 25,173 |

**4-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `κατά τη διάρκεια της` | 32,537 |
| 2 | `από το έως το` | 20,094 |
| 3 | `κατά τη διάρκεια του` | 19,453 |
| 4 | `γαλλική κοινότητα στο νομό` | 16,152 |
| 5 | `είναι γαλλική κοινότητα στο` | 16,142 |

**5-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `είναι γαλλική κοινότητα στο νομό` | 16,142 |
| 2 | `γαλλική κοινότητα στο νομό της` | 10,798 |
| 3 | `σύμφωνα με την απογραφή του` | 8,977 |
| 4 | `προβλήματα οργανικής χημείας ν α` | 5,103 |
| 5 | `οργανικής χημείας ν α πετάση` | 5,103 |

**2-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `ς _` | 20,530,109 |
| 2 | `_ τ` | 20,509,338 |
| 3 | `τ ο` | 15,006,596 |
| 4 | `ο υ` | 13,459,949 |
| 5 | `α _` | 12,791,705 |

**3-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ τ ο` | 9,583,813 |
| 2 | `ο υ _` | 7,426,167 |
| 3 | `_ κ α` | 6,229,911 |
| 4 | `α ι _` | 5,946,159 |
| 5 | `_ τ η` | 5,812,762 |

**4-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ τ ο υ` | 4,854,974 |
| 2 | `τ ο υ _` | 3,990,563 |
| 3 | `_ κ α ι` | 3,906,895 |
| 4 | `κ α ι _` | 3,870,183 |
| 5 | `_ τ ο _` | 3,120,828 |

**5-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ κ α ι _` | 3,856,808 |
| 2 | `_ τ ο υ _` | 3,836,821 |
| 3 | `_ τ η ς _` | 2,888,245 |
| 4 | `_ τ η ν _` | 1,890,516 |
| 5 | `_ α π ό _` | 1,864,707 |


### Key Findings

- **Best Perplexity:** 2-gram (subword) with 443
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
- **Coverage:** Top-1000 patterns cover ~21% 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 | 0.9344 | 1.911 | 11.28 | 2,374,710 | 6.6% |
| **1** | Subword | 1.0861 | 2.123 | 7.80 | 13,425 | 0.0% |
| **2** | Word | 0.4145 | 1.333 | 2.61 | 26,731,768 | 58.6% |
| **2** | Subword | 0.7185 | 1.645 | 5.31 | 104,621 | 28.2% |
| **3** | Word | 0.1946 | 1.144 | 1.46 | 69,637,387 | 80.5% |
| **3** | Subword | 0.8000 | 1.741 | 4.75 | 555,743 | 20.0% |
| **4** | Word | 0.0819 🏆 | 1.058 | 1.15 | 101,596,464 | 91.8% |
| **4** | Subword | 0.7130 | 1.639 | 3.67 | 2,639,831 | 28.7% |

### Generated Text Samples (Word-based)

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

**Context Size 1:**

1. `του άγραφος νόμος και εκλογές κερδίζει το ο ν ευστρατίου κώστας καραπατής έλληνας αγωνιστής του οίκο...`
2. `και βασανίστηκε σε αντίθεση με τον στρυμόνα ο βοναπάρτης κάλεσε σε κομματικό μάθημα φυκολογία harvey...`
3. `το μπρύγκεν κάηκε τρεις πήχεις και τους τύπους κλειδώματος πολλές προσπάθειες ευχρηστίας υπηρετεί ως...`

**Context Size 2:**

1. `από το πανί και τον βιότοπο της κέντρο είναι το δεύτερο όσκαρ β τέλεσε τη θεία της`
2. `από την αστυνομία ενώ είναι διαθέσιμο σε 409 αγώνες σκοράροντας 4 γκολ σε όλες τις έδρες δηλαδή`
3. `με την οργάνωση και επέκταση των ορίων λειτουργίας των διαδικασιών η εταιρεία το δίκτυο αποχέτευσης ...`

**Context Size 3:**

1. `κατά τη διάρκεια της οποίας προέτρεψε να παραδοθούν αφού πρωτύτερα συμφώνησαν να μην ενημερώσουν τον...`
2. `παραπομπές εξωτερικοί σύνδεσμοι ψηφιακό αρχείο των δημοσιεύσεων του χ σάιμον με τα πλήρη ίσια μαλλιά...`
3. `τη διάρκεια της βασιλείας του τσάρου πέτρου α τα ελεύθερα οικόπεδα αγοράστηκαν και το μια μεταλλική ...`

**Context Size 4:**

1. `κατά τη διάρκεια της δεκαετίας του 20 τάφηκε μαζί με την σύζυγο του αυγούστα κόρτενεϋ 8 φεβρουαρίου ...`
2. `από το έως το με εξαίρεση εκείνες του μετά την έξωση του όθωνα κατά τη διάρκεια των φιλορωσικών ανατ...`
3. `κατά τη διάρκεια του χειμώνα μεταξύ της τελευταίας κυριακής του οκτωβρίου μέχρι τη 1 00 utc της τελε...`


### Generated Text Samples (Subword-based)

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

**Context Size 1:**

1. `_ικαθεσυπν_μμε_a`
2. `ας._ησο_πόδύπίαι`
3. `ούν_πού_κόπρες_ό`

**Context Size 2:**

1. `ς_εξε_μος_αντρώτο`
2. `_τη_για_ήταχματην`
3. `το_ναι_από_τοντις`

**Context Size 3:**

1. `_του_αναλίαρχές_αλ`
2. `ου_έγκροτεχνολούν_`
3. `_καιρισμοι_/σεβαιω`

**Context Size 4:**

1. `_τους_χρήση_ο_πτερύ`
2. `του_της_ανακάλυψη_σ`
3. `_και_τους_δικτίνας_`


### Key Findings

- **Best Predictability:** Context-4 (word) with 91.8% predictability
- **Branching Factor:** Decreases with context size (more deterministic)
- **Memory Trade-off:** Larger contexts require more storage (2,639,831 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 | 1,039,940 |
| Total Tokens | 132,061,031 |
| Mean Frequency | 126.99 |
| Median Frequency | 4 |
| Frequency Std Dev | 9123.56 |

### Most Common Words

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | του | 4,095,731 |
| 2 | και | 3,886,615 |
| 3 | το | 3,228,440 |
| 4 | της | 2,987,569 |
| 5 | η | 1,958,228 |
| 6 | την | 1,895,055 |
| 7 | από | 1,882,149 |
| 8 | ο | 1,862,872 |
| 9 | με | 1,655,296 |
| 10 | τον | 1,304,224 |

### Least Common Words (from vocabulary)

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | ωσμωπροστατευτικά | 2 |
| 2 | ορμπέκη | 2 |
| 3 | hidronor | 2 |
| 4 | jpp | 2 |
| 5 | liebrand | 2 |
| 6 | οϊρατσουμέ | 2 |
| 7 | χασιχίτο | 2 |
| 8 | σεϊσι | 2 |
| 9 | τακατσουκασά | 2 |
| 10 | κατσιρέλο | 2 |

### Zipf's Law Analysis

| Metric | Value |
|--------|-------|
| Zipf Coefficient | 0.9498 |
| R² (Goodness of Fit) | 0.997066 |
| Adherence Quality | **excellent** |

### Coverage Analysis

| Top N Words | Coverage |
|-------------|----------|
| Top 100 | 38.6% |
| Top 1,000 | 55.9% |
| Top 5,000 | 71.4% |
| Top 10,000 | 78.0% |

### Key Findings

- **Zipf Compliance:** R²=0.9971 indicates excellent adherence to Zipf's law
- **High Frequency Dominance:** Top 100 words cover 38.6% of corpus
- **Long Tail:** 1,029,940 words needed for remaining 22.0% 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 | 0.8028 | 0.3648 | N/A | N/A |
| **mono_64d** | 64 | 0.7821 | 0.3021 | N/A | N/A |
| **mono_128d** | 128 | 0.7303 | 0.2408 | N/A | N/A |
| **aligned_32d** | 32 | 0.8028 🏆 | 0.3775 | 0.2640 | 0.6820 |
| **aligned_64d** | 64 | 0.7821 | 0.2965 | 0.4780 | 0.8720 |
| **aligned_128d** | 128 | 0.7303 | 0.2330 | 0.6560 | 0.9100 |

### Key Findings

- **Best Isotropy:** aligned_32d with 0.8028 (more uniform distribution)
- **Semantic Density:** Average pairwise similarity of 0.3025. Lower values indicate better semantic separation.
- **Alignment Quality:** Aligned models achieve up to 65.6% 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 | **-0.798** | 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 |
|--------|----------|
| `-α` | αβρανσάν, απόχρεμψη, αποφέροντάς |
| `-σ` | συνειδητοποιήσετε, στίβενσον, σπειροτόμησης |
| `-a` | ayodhya, addicted, apocolo |
| `-s` | superdome, sembrich, sibling |
| `-κ` | κίτσεβο, κλειδώνω, κινοσάκι |
| `-κα` | καριστάνιου, κασιγουαμπάρα, καλλιρροη |
| `-ε` | ελληνοαλβανικών, επανεξετάζει, ενοργάνιση |
| `-μ` | μάστερινγκ, μεθυλοβουτανονιτρίλιοασκήσεις, μπαλάφα |

#### Productive Suffixes
| Suffix | Examples |
|--------|----------|
| `-ς` | νεπαλέζους, 125ος, μεθυλοβουτανονιτρίλιοασκήσεις |
| `-ν` | ελληνοαλβανικών, νταγκάν, αβρανσάν |
| `-α` | οκτωβρίουεφημερίδα, προσωπίδα, τζιτζιμπίρα |
| `-ι` | χότζι, φρύξουσι, υπονομεύεται |
| `-ος` | 125ος, φιλαθλος, μπατιστάτος |
| `-ο` | ζηρίνειο, κίτσεβο, ριβονουκλεοτίδιο |
| `-ου` | καριστάνιου, ατταβύρου, βερεγγάριου |
| `-ης` | φαρέλης, απόρθητης, σπειροτόμησης |

### 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 |
|------|----------|------------------|----------|
| `ικών` | 2.20x | 163 contexts | δικών, νικών, οικών |
| `ικής` | 2.14x | 156 contexts | ιικής, τικής, πικής |
| `ότητ` | 2.07x | 175 contexts | κότητα, νότητα, ἑνότητα |
| `ικές` | 1.96x | 135 contexts | νικές, μικές, δικές |
| `ιστι` | 1.52x | 338 contexts | μιστι, ιστική, πιστιν |
| `ατος` | 1.90x | 92 contexts | ματος, αίατος, υπατος |
| `ανικ` | 1.44x | 370 contexts | δανικα, δανικό, μανικά |
| `ήθηκ` | 1.93x | 81 contexts | ψήθηκε, λήθηκε, μυήθηκε |
| `ολογ` | 1.40x | 399 contexts | ολογρ, υπολογ, οδολογ |
| `πίση` | 2.06x | 48 contexts | πίσης, επίση, έπίσης |
| `ατικ` | 1.38x | 317 contexts | ατικέ, ατικά, φατική |
| `οποι` | 1.45x | 200 contexts | τοποι, οποιά, οποιο |

### 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 |
|--------|--------|-----------|----------|
| `-α` | `-ς` | 188 words | αφηγησεις, ανύπανδρους |
| `-κ` | `-ς` | 153 words | καλλιοντζής, κωστούλης |
| `-σ` | `-ς` | 127 words | στηις, σοβαρώς |
| `-ε` | `-ς` | 116 words | ενελικτικός, επιμορφωτικούς |
| `-μ` | `-ς` | 110 words | μεταξάςπρωταγωνιστικός, μπούσεβιτς |
| `-α` | `-ν` | 104 words | αιτωλίαν, απονεμηθέν |
| `-κ` | `-ν` | 68 words | κηρύκειον, κατακάηκαν |
| `-μ` | `-ν` | 65 words | μπιέγκαν, μεταβλητών |
| `-ε` | `-ν` | 65 words | εξεπόνησαν, ερείπωσαν |
| `-α` | `-α` | 65 words | αυτοκρατόρισσα, αυστραλια |

### 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 |
|------|-----------------|------------|------|
| έτοςχιόνι | **`έτοςχιό-ν-ι`** | 7.5 | `ν` |
| περισσεία | **`περισσ-ε-ία`** | 7.5 | `ε` |
| αἰγινήτου | **`αἰγινή-τ-ου`** | 7.5 | `τ` |
| αντιψυχωσικών | **`αντιψυχωσι-κ-ών`** | 7.5 | `κ` |
| λανγκλουά | **`λανγκλ-ου-ά`** | 6.0 | `λανγκλ` |
| μπουνάκιας | **`μπουνάκ-ια-ς`** | 6.0 | `μπουνάκ` |
| γιαλούρης | **`γιαλούρη-ς`** | 4.5 | `γιαλούρη` |
| εφαρμόζεις | **`εφαρμόζει-ς`** | 4.5 | `εφαρμόζει` |
| internationalοι | **`international-οι`** | 4.5 | `international` |
| λοξότητας | **`λοξότητα-ς`** | 4.5 | `λοξότητα` |
| δομινικανικής | **`δομινικανική-ς`** | 4.5 | `δομινικανική` |
| aθλητικός | **`aθλητικό-ς`** | 4.5 | `aθλητικό` |
| επηρεασμένης | **`επηρεασμένη-ς`** | 4.5 | `επηρεασμένη` |
| σελτζουκικός | **`σελτζουκικό-ς`** | 4.5 | `σελτζουκικό` |
| modernisme | **`modernism-e`** | 4.5 | `modernism` |

### 6.6 Linguistic Interpretation

> **Automated Insight:**
The language Greek 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 (4.87x) |
| N-gram | **2-gram** | Lowest perplexity (443) |
| Markov | **Context-4** | Highest predictability (91.8%) |
| 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: 2026-01-10 02:57:50*