README.md

---
language: nn
language_name: Norwegian Nynorsk
language_family: germanic_north
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_north
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: 4.622
  - name: best_isotropy
    type: isotropy
    value: 0.7969
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-15
---

# Norwegian Nynorsk - Wikilangs Models
## Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on **Norwegian Nynorsk** 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** | 3.649x | 3.65 | 0.1335% | 636,601 |
| **16k** | 4.025x | 4.03 | 0.1473% | 577,127 |
| **32k** | 4.353x | 4.35 | 0.1593% | 533,706 |
| **64k** | 4.622x 🏆 | 4.62 | 0.1691% | 502,547 |

### Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

**Sample 1:** `Sjoa kan vise til: Elva Sjoa i Heidal i Gudbrandsdalen Bygda Sjoa i Gudbrandsdal...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁sj oa ▁kan ▁vise ▁til : ▁elva ▁sj oa ▁i ... (+13 more)` | 23 |
| 16k | `▁sj oa ▁kan ▁vise ▁til : ▁elva ▁sj oa ▁i ... (+11 more)` | 21 |
| 32k | `▁sj oa ▁kan ▁vise ▁til : ▁elva ▁sj oa ▁i ... (+9 more)` | 19 |
| 64k | `▁sjoa ▁kan ▁vise ▁til : ▁elva ▁sjoa ▁i ▁heidal ▁i ... (+5 more)` | 15 |

**Sample 2:** `Vestlandets Avis var Nasjonal Samling si avis i Stavanger frå til Kjelder skipa ...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁vest landet s ▁avis ▁var ▁nasjonal ▁samling ▁si ▁avis ▁i ... (+9 more)` | 19 |
| 16k | `▁vestlandet s ▁avis ▁var ▁nasjonal ▁samling ▁si ▁avis ▁i ▁stavanger ... (+8 more)` | 18 |
| 32k | `▁vestlandet s ▁avis ▁var ▁nasjonal ▁samling ▁si ▁avis ▁i ▁stavanger ... (+8 more)` | 18 |
| 64k | `▁vestlandet s ▁avis ▁var ▁nasjonal ▁samling ▁si ▁avis ▁i ▁stavanger ... (+8 more)` | 18 |

**Sample 3:** `Jun Suzuki () er ein japansk fotballspelar. Han spela for klubbane SC Sagamihara...`

| Vocab | Tokens | Count |
|-------|--------|-------|
| 8k | `▁jun ▁su z uki ▁() ▁er ▁ein ▁japansk ▁fotballspelar . ... (+20 more)` | 30 |
| 16k | `▁jun ▁suz uki ▁() ▁er ▁ein ▁japansk ▁fotballspelar . ▁han ... (+18 more)` | 28 |
| 32k | `▁jun ▁suzuki ▁() ▁er ▁ein ▁japansk ▁fotballspelar . ▁han ▁spela ... (+16 more)` | 26 |
| 64k | `▁jun ▁suzuki ▁() ▁er ▁ein ▁japansk ▁fotballspelar . ▁han ▁spela ... (+11 more)` | 21 |


### Key Findings

- **Best Compression:** 64k achieves 4.622x compression
- **Lowest UNK Rate:** 8k with 0.1335% 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 | 109,666 | 16.74 | 757,323 | 8.1% | 21.8% |
| **2-gram** | Subword | 299 🏆 | 8.23 | 13,293 | 66.4% | 99.0% |
| **3-gram** | Word | 368,717 | 18.49 | 1,379,371 | 4.5% | 11.8% |
| **3-gram** | Subword | 2,774 | 11.44 | 106,788 | 23.8% | 68.2% |
| **4-gram** | Word | 703,833 | 19.42 | 2,105,642 | 4.1% | 9.6% |
| **4-gram** | Subword | 17,936 | 14.13 | 615,144 | 11.3% | 35.2% |
| **5-gram** | Word | 490,457 | 18.90 | 1,360,064 | 4.4% | 10.9% |
| **5-gram** | Subword | 81,186 | 16.31 | 2,151,391 | 6.2% | 20.5% |

### Top 5 N-grams by Size

**2-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `er ein` | 97,088 |
| 2 | `frå den` | 80,534 |
| 3 | `denne artikkelen` | 74,226 |
| 4 | `artikkelen bygger` | 72,759 |
| 5 | `bygger på` | 72,670 |

**3-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `denne artikkelen bygger` | 72,495 |
| 2 | `artikkelen bygger på` | 72,492 |
| 3 | `kjelder denne artikkelen` | 65,798 |
| 4 | `oppgav desse kjeldene` | 22,909 |
| 5 | `ein del av` | 14,027 |

**4-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `denne artikkelen bygger på` | 72,230 |
| 2 | `kjelder denne artikkelen bygger` | 64,588 |
| 3 | `oppgav desse kjeldene bakgrunnsstoff` | 6,804 |
| 4 | `plass utøvar land tid` | 6,605 |
| 5 | `under sommar ol under` | 6,222 |

**5-grams (Word):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `kjelder denne artikkelen bygger på` | 64,338 |
| 2 | `sommar ol under sommar ol` | 6,036 |
| 3 | `under sommar ol under sommar` | 6,032 |
| 4 | `deltakarar under sommar ol under` | 4,755 |
| 5 | `vinter ol under vinter ol` | 4,073 |

**2-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `e _` | 4,250,964 |
| 2 | `e r` | 4,042,601 |
| 3 | `r _` | 4,018,740 |
| 4 | `n _` | 3,701,628 |
| 5 | `e n` | 3,602,941 |

**3-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `e r _` | 1,868,515 |
| 2 | `e n _` | 1,833,528 |
| 3 | `_ i _` | 1,716,434 |
| 4 | `_ d e` | 1,533,918 |
| 5 | `a r _` | 1,320,154 |

**4-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ o g _` | 1,134,296 |
| 2 | `_ a v _` | 672,822 |
| 3 | `_ t i l` | 606,367 |
| 4 | `_ p å _` | 596,900 |
| 5 | `_ v a r` | 570,119 |

**5-grams (Subword):**

| Rank | N-gram | Count |
|------|--------|-------|
| 1 | `_ s o m _` | 521,163 |
| 2 | `_ t i l _` | 520,787 |
| 3 | `_ e i n _` | 435,490 |
| 4 | `_ f r å _` | 391,177 |
| 5 | `_ d e n _` | 386,818 |


### Key Findings

- **Best Perplexity:** 2-gram (subword) with 299
- **Entropy Trend:** Decreases with larger n-grams (more predictable)
- **Coverage:** Top-1000 patterns cover ~20% 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 | 0.9093 | 1.878 | 8.96 | 1,142,989 | 9.1% |
| **1** | Subword | 0.9635 | 1.950 | 6.47 | 7,110 | 3.7% |
| **2** | Word | 0.3519 | 1.276 | 2.18 | 10,223,775 | 64.8% |
| **2** | Subword | 0.7744 | 1.711 | 5.14 | 45,899 | 22.6% |
| **3** | Word | 0.1494 | 1.109 | 1.32 | 22,248,810 | 85.1% |
| **3** | Subword | 0.7774 | 1.714 | 4.42 | 235,613 | 22.3% |
| **4** | Word | 0.0611 🏆 | 1.043 | 1.11 | 29,377,227 | 93.9% |
| **4** | Subword | 0.7217 | 1.649 | 3.63 | 1,040,512 | 27.8% |

### Generated Text Samples (Word-based)

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

**Context Size 1:**

1. `i telemark fylkesvei 395 387 meter over vassflata og ut av fint forseggjorde klede bestod av`
2. `og døydde stille er ein kommune grensar til enorme røykutviklinga gjorde det har lege for finland`
3. `av dei egyptiske faraoen seti krus frå albumet er sekretærfuglen så rastlaus rytme akustisk gitar du...`

**Context Size 2:**

1. `er ein amerikansk teikneserien i barnebladet maurtua under psevdonymet tcp salslister og salstrofé s...`
2. `frå den 16 juni klokka 18 alle kampane i turneringa hans beste tid i saltgruver denne blir`
3. `denne artikkelen bygger på paul samwell smith og joseph alfred serret fundamentalteoremet for romkur...`

**Context Size 3:**

1. `denne artikkelen bygger på wadi radd frå den 27 mars bakgrunnsstoff i thurgau i innsjøar`
2. `artikkelen bygger på circles frå den 5 juli i dalarnas län i landskapet bohuslän i hadde byen nesten`
3. `kjelder denne artikkelen bygger på mont tramelan frå den 25 november bakgrunnsstoff department of co...`

**Context Size 4:**

1. `denne artikkelen bygger på şereflikoçhisar frå den 28 august oppgav desse kjeldene bakgrunnsstoff ar...`
2. `kjelder denne artikkelen bygger på altenalp türm frå den 5 februar på skeiser i noreg i i farsund`
3. `oppgav desse kjeldene bakgrunnsstoff offisiell nettstad myrehovot info grunnlagde i i israel i israe...`


### Generated Text Samples (Subword-based)

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

**Context Size 1:**

1. `_el._faskagefowe`
2. `ei_knyr_marig_t_`
3. `aldon,_sk_ove_e_`

**Context Size 2:**

1. `e_hi_nortil_eiren`
2. `er_av_i_2_livaser`
3. `r_d'ams_«riseknin`

**Context Size 3:**

1. `er_særlen_art_av_m`
2. `en_212_fekk_kommun`
3. `_i_utantar_er_mati`

**Context Size 4:**

1. `_og_bedehus._dei_«b`
2. `_av_fengstida_renn_`
3. `_til_kalde_albumet_`


### Key Findings

- **Best Predictability:** Context-4 (word) with 93.9% predictability
- **Branching Factor:** Decreases with context size (more deterministic)
- **Memory Trade-off:** Larger contexts require more storage (1,040,512 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 | 513,511 |
| Total Tokens | 37,024,951 |
| Mean Frequency | 72.10 |
| Median Frequency | 4 |
| Frequency Std Dev | 3882.03 |

### Most Common Words

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | i | 1,743,863 |
| 2 | og | 1,137,518 |
| 3 | av | 676,970 |
| 4 | på | 603,741 |
| 5 | er | 529,313 |
| 6 | til | 527,717 |
| 7 | som | 526,626 |
| 8 | ein | 441,058 |
| 9 | frå | 400,518 |
| 10 | den | 393,604 |

### Least Common Words (from vocabulary)

| Rank | Word | Frequency |
|------|------|-----------|
| 1 | iguaca | 2 |
| 2 | macranthus | 2 |
| 3 | protoanemonin | 2 |
| 4 | musikkarbeidsstasjonar | 2 |
| 5 | småstillits | 2 |
| 6 | purpurtøy | 2 |
| 7 | levendehistorie | 2 |
| 8 | dutz | 2 |
| 9 | kreolerinnen | 2 |
| 10 | thornfield | 2 |

### Zipf's Law Analysis

| Metric | Value |
|--------|-------|
| Zipf Coefficient | 1.0395 |
| R² (Goodness of Fit) | 0.998477 |
| Adherence Quality | **excellent** |

### Coverage Analysis

| Top N Words | Coverage |
|-------------|----------|
| Top 100 | 39.4% |
| Top 1,000 | 60.6% |
| Top 5,000 | 75.4% |
| Top 10,000 | 81.2% |

### Key Findings

- **Zipf Compliance:** R²=0.9985 indicates excellent adherence to Zipf's law
- **High Frequency Dominance:** Top 100 words cover 39.4% of corpus
- **Long Tail:** 503,511 words needed for remaining 18.8% 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 | 0.7969 | 0.3590 | N/A | N/A |
| **mono_64d** | 64 | 0.7770 | 0.2951 | N/A | N/A |
| **mono_128d** | 128 | 0.7202 | 0.2244 | N/A | N/A |
| **aligned_32d** | 32 | 0.7969 🏆 | 0.3591 | 0.2560 | 0.6620 |
| **aligned_64d** | 64 | 0.7770 | 0.2887 | 0.5160 | 0.8280 |
| **aligned_128d** | 128 | 0.7202 | 0.2190 | 0.5380 | 0.8640 |

### Key Findings

- **Best Isotropy:** aligned_32d with 0.7969 (more uniform distribution)
- **Semantic Density:** Average pairwise similarity of 0.2909. Lower values indicate better semantic separation.
- **Alignment Quality:** Aligned models achieve up to 53.8% 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 | **-0.470** | 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` | seljefløyte, streifdyr, shinzo |
| `-a` | acoustique, akrylfarge, arnoediad |
| `-b` | bluessamuel, bramness, brünberg |
| `-ma` | malenchenko, marinepersonell, manneråk |
| `-k` | kóny, kjerstin, kariem |
| `-m` | myrtosbukta, mixopterus, miskóc |
| `-t` | târlea, tawitawi, trippeltriumf |
| `-l` | leksands, logone, lexington |

#### Productive Suffixes
| Suffix | Examples |
|--------|----------|
| `-n` | fjærlandsfjorden, froskemann, kjerstin |
| `-en` | fjærlandsfjorden, rhônen, orgien |
| `-e` | seljefløyte, acoustique, forkynnande |
| `-r` | gwr, streifdyr, goldwater |
| `-t` | nordljoset, inkavimpelstjert, ustrukturert |
| `-a` | târlea, myrtosbukta, ternopilborga |
| `-ar` | snorfigurar, mygglarvar, oversettingar |
| `-et` | nordljoset, intervjuobjektet, mellomøyret |

### 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 |
|------|----------|------------------|----------|
| `ller` | 1.68x | 323 contexts | eller, iller, uller |
| `lbum` | 2.74x | 21 contexts | album, albuma, allbum |
| `ansk` | 1.65x | 160 contexts | ansky, kansk, dansk |
| `iske` | 1.58x | 170 contexts | piske, miske, riske |
| `tter` | 1.32x | 422 contexts | etter, ëtter, atter |
| `lder` | 1.59x | 144 contexts | ålder, ilder, older |
| `ygge` | 1.83x | 70 contexts | bygge, tygge, rygge |
| `jeld` | 1.72x | 68 contexts | kjeld, njeld, gjeld |
| `nter` | 1.34x | 220 contexts | inter, enter, unter |
| `tisk` | 1.55x | 105 contexts | etisk, fotisk, estisk |
| `ngen` | 1.36x | 193 contexts | ingen, sngen, ngeny |
| `rste` | 1.40x | 142 contexts | erste, ørste, torste |

### 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` | 163 words | soloppgangen, sjoåsen |
| `-s` | `-e` | 146 words | sandstripe, schreibe |
| `-s` | `-r` | 128 words | standarder, syboliserer |
| `-s` | `-en` | 119 words | soloppgangen, sjoåsen |
| `-s` | `-a` | 111 words | storhovda, spørsmåla |
| `-s` | `-t` | 103 words | sanat, storbukt |
| `-k` | `-n` | 89 words | karawanken, knubben |
| `-b` | `-n` | 77 words | berndtsson, bordkøyraren |
| `-t` | `-n` | 73 words | tausen, torturisten |
| `-a` | `-n` | 71 words | arnkværn, akerryggen |

### 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 |
|------|-----------------|------------|------|
| bidireksjonal | **`bidireksjo-n-al`** | 7.5 | `n` |
| mørkbrunt | **`mørkbru-n-t`** | 7.5 | `n` |
| betalande | **`be-ta-lande`** | 7.5 | `lande` |
| mooncrest | **`mooncr-e-st`** | 7.5 | `e` |
| døgnvariasjonen | **`døgnvariasjo-n-en`** | 7.5 | `n` |
| ergebnisse | **`ergebnis-s-e`** | 7.5 | `s` |
| distanseritt | **`distanseri-t-t`** | 7.5 | `t` |
| mysteriøse | **`mysteriø-s-e`** | 7.5 | `s` |
| gullmyntar | **`gullmyn-t-ar`** | 7.5 | `t` |
| capricorni | **`capricor-n-i`** | 7.5 | `n` |
| archerbreen | **`archerbr-e-en`** | 7.5 | `e` |
| highwired | **`highwir-e-d`** | 7.5 | `e` |
| traktatkomiteen | **`traktatkomit-e-en`** | 7.5 | `e` |
| herrefoss | **`herrefo-s-s`** | 7.5 | `s` |
| regnbogehinne | **`regnbogehi-n-ne`** | 7.5 | `n` |

### 6.6 Linguistic Interpretation

> **Automated Insight:**
The language Norwegian Nynorsk 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 (4.62x) |
| N-gram | **2-gram** | Lowest perplexity (299) |
| Markov | **Context-4** | Highest predictability (93.9%) |
| 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: 2026-01-15 20:48:02*