Upload all models and assets for eml (latest)

.gitattributes

@@ -39,3 +39,4 @@ visualizations/position_encoding_comparison.png filter=lfs diff=lfs merge=lfs -t
 visualizations/tsne_sentences.png filter=lfs diff=lfs merge=lfs -text
 visualizations/tsne_words.png filter=lfs diff=lfs merge=lfs -text
 visualizations/zipf_law.png filter=lfs diff=lfs merge=lfs -text
+visualizations/embedding_tsne_multilingual.png filter=lfs diff=lfs merge=lfs -text

README.md

@@ -1,6 +1,6 @@
 ---
 language: eml
-language_name: EML
+language_name: Unknown language [eml]
 language_family: romance_galloitalic
 tags:
   - wikilangs
@@ -10,11 +10,21 @@ tags:
   - n-gram
   - markov
   - wikipedia
+  - feature-extraction
+  - sentence-similarity
+  - tokenization
+  - n-grams
+  - markov-chain
+  - text-mining
+  - fasttext
+  - babelvec
+  - vocabulous
+  - vocabulary
   - monolingual
   - family-romance_galloitalic
 license: mit
 library_name: wikilangs
-pipeline_tag: feature-extraction
+pipeline_tag: text-generation
 datasets:
   - omarkamali/wikipedia-monthly
 dataset_info:
@@ -23,20 +33,20 @@ dataset_info:
 metrics:
   - name: best_compression_ratio
     type: compression
-    value: 3.867
+    value: 3.369
   - name: best_isotropy
     type: isotropy
-    value: 0.4456
+    value: 0.3584
   - name: vocabulary_size
     type: vocab
-    value: 18015
-generated: 2025-12-30
+    value: 0
+generated: 2026-01-04
 ---
 
-# EML - Wikilangs Models
+# Unknown language [eml] - Wikilangs Models
 ## Comprehensive Research Report & Full Ablation Study
 
-This repository contains NLP models trained and evaluated by Wikilangs, specifically on **EML** Wikipedia data.
+This repository contains NLP models trained and evaluated by Wikilangs, specifically on **Unknown language [eml]** Wikipedia data.
 We analyze tokenizers, n-gram models, Markov chains, vocabulary statistics, and word embeddings.
 
 ## 📋 Repository Contents
@@ -44,12 +54,13 @@ We analyze tokenizers, n-gram models, Markov chains, vocabulary statistics, and
 ### Models & Assets
 
 - Tokenizers (8k, 16k, 32k, 64k)
-- N-gram models (2, 3, 4-gram)
-- Markov chains (context of 1, 2, 3 and 4)
+- 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
+- Embeddings in various sizes and dimensions (aligned and unaligned)
 - Language Vocabulary
 - Language Statistics
+
 ![Performance Dashboard](visualizations/performance_dashboard.png)
 
 ### Analysis and Evaluation
@@ -59,7 +70,8 @@ We analyze tokenizers, n-gram models, Markov chains, vocabulary statistics, and
 - [3. Markov Chain Evaluation](#3-markov-chain-evaluation)
 - [4. Vocabulary Analysis](#4-vocabulary-analysis)
 - [5. Word Embeddings Evaluation](#5-word-embeddings-evaluation)
-- [6. Summary & Recommendations](#6-summary--recommendations)
+- [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)
 
@@ -68,54 +80,53 @@ We analyze tokenizers, n-gram models, Markov chains, vocabulary statistics, and
 
 ![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** | 3.069x | 3.04 | 0.1566% | 208,856 |
-| **16k** | 3.337x | 3.30 | 0.1703% | 192,055 |
-| **32k** | 3.580x | 3.54 | 0.1827% | 179,030 |
-| **64k** | 3.867x 🏆 | 3.83 | 0.1973% | 165,756 |
+| **8k** | 2.942x | 2.95 | 0.4433% | 289,426 |
+| **16k** | 3.144x | 3.15 | 0.4738% | 270,763 |
+| **32k** | 3.369x 🏆 | 3.37 | 0.5076% | 252,742 |
 
 ### Tokenization Examples
 
 Below are sample sentences tokenized with each vocabulary size:
 
-**Sample 1:** `Categoria:TELEVISIONE
-Categoria:CINEMA
-Categoria:Atōr tedésc`
+**Sample 1:** `'l è 'l nòm 'd un domìni genèric. Al funsiòuna da 'l zógn dal ed domìni tachê a ...`
 
 | Vocab | Tokens | Count |
 |-------|--------|-------|
-| 8k | `▁categoria : televisione ▁categoria : cinema ▁categoria : atōr ▁tedésc` | 10 |
-| 16k | `▁categoria : televisione ▁categoria : cinema ▁categoria : atōr ▁tedésc` | 10 |
-| 32k | `▁categoria : televisione ▁categoria : cinema ▁categoria : atōr ▁tedésc` | 10 |
-| 64k | `▁categoria : televisione ▁categoria : cinema ▁categoria : atōr ▁tedésc` | 10 |
+| 8k | `▁' l ▁è ▁' l ▁nòm ▁' d ▁un ▁domìni ... (+17 more)` | 27 |
+| 16k | `▁' l ▁è ▁' l ▁nòm ▁' d ▁un ▁domìni ... (+17 more)` | 27 |
+| 32k | `▁' l ▁è ▁' l ▁nòm ▁' d ▁un ▁domìni ... (+17 more)` | 27 |
 
-**Sample 2:** `La Ròca o anch Ròca San Casiàn (Rocca San Casciano in italièn) l'è un cumòn ad 1...`
+**Sample 2:** `'l è 'l nòm 'd un domìni genèric. Al funsiòuna da 'l setèmber dal ed domìni tach...`
 
 | Vocab | Tokens | Count |
 |-------|--------|-------|
-| 8k | `▁la ▁rò ca ▁o ▁anch ▁rò ca ▁san ▁ca si ... (+37 more)` | 47 |
-| 16k | `▁la ▁ròca ▁o ▁anch ▁ròca ▁san ▁ca si àn ▁( ... (+33 more)` | 43 |
-| 32k | `▁la ▁ròca ▁o ▁anch ▁ròca ▁san ▁ca si àn ▁( ... (+32 more)` | 42 |
-| 64k | `▁la ▁ròca ▁o ▁anch ▁ròca ▁san ▁casiàn ▁( rocca ▁san ... (+29 more)` | 39 |
+| 8k | `▁' l ▁è ▁' l ▁nòm ▁' d ▁un ▁domìni ... (+17 more)` | 27 |
+| 16k | `▁' l ▁è ▁' l ▁nòm ▁' d ▁un ▁domìni ... (+17 more)` | 27 |
+| 32k | `▁' l ▁è ▁' l ▁nòm ▁' d ▁un ▁domìni ... (+17 more)` | 27 |
 
-**Sample 3:** `Categoria:GEOGRAFIA
-Categoria:CITTADITALIA`
+**Sample 3:** `Al 294 'l è 'n an edl III sécol dal Calendàri gregoriàn. Avenimèint Nê Mort III`
 
 | Vocab | Tokens | Count |
 |-------|--------|-------|
-| 8k | `▁categoria : geografia ▁categoria : cittaditalia` | 6 |
-| 16k | `▁categoria : geografia ▁categoria : cittaditalia` | 6 |
-| 32k | `▁categoria : geografia ▁categoria : cittaditalia` | 6 |
-| 64k | `▁categoria : geografia ▁categoria : cittaditalia` | 6 |
+| 8k | `▁al ▁ 2 9 4 ▁' l ▁è ▁' n ... (+12 more)` | 22 |
+| 16k | `▁al ▁ 2 9 4 ▁' l ▁è ▁' n ... (+12 more)` | 22 |
+| 32k | `▁al ▁ 2 9 4 ▁' l ▁è ▁' n ... (+12 more)` | 22 |
 
 
 ### Key Findings
 
-- **Best Compression:** 64k achieves 3.867x compression
-- **Lowest UNK Rate:** 8k with 0.1566% unknown tokens
+- **Best Compression:** 32k achieves 3.369x compression
+- **Lowest UNK Rate:** 8k with 0.4433% unknown tokens
 - **Trade-off:** Larger vocabularies improve compression but increase model size
 - **Recommendation:** 32k vocabulary provides optimal balance for production use
 
@@ -124,57 +135,111 @@ Categoria:CITTADITALIA`
 
 ![N-gram Perplexity](visualizations/ngram_perplexity.png)
 
+![N-gram Unique](visualizations/ngram_unique.png)
+
 ![N-gram Coverage](visualizations/ngram_coverage.png)
 
 ### Results
 
-| N-gram | Perplexity | Entropy | Unique N-grams | Top-100 Coverage | Top-1000 Coverage |
-|--------|------------|---------|----------------|------------------|-------------------|
-| **2-gram** | 689 🏆 | 9.43 | 7,061 | 54.1% | 81.3% |
-| **2-gram** | 367 🏆 | 8.52 | 2,917 | 61.9% | 97.6% |
-| **3-gram** | 1,170 | 10.19 | 11,174 | 48.6% | 76.8% |
-| **3-gram** | 2,199 | 11.10 | 21,058 | 33.1% | 68.9% |
-| **4-gram** | 1,936 | 10.92 | 20,258 | 43.6% | 71.1% |
-| **4-gram** | 7,361 | 12.85 | 88,240 | 25.5% | 51.4% |
+| N-gram | Variant | Perplexity | Entropy | Unique N-grams | Top-100 Coverage | Top-1000 Coverage |
+|--------|---------|------------|---------|----------------|------------------|-------------------|
+| **2-gram** | Word | 855 | 9.74 | 4,527 | 49.2% | 80.8% |
+| **2-gram** | Subword | 342 🏆 | 8.42 | 2,464 | 62.9% | 97.8% |
+| **3-gram** | Word | 936 | 9.87 | 6,071 | 49.5% | 79.8% |
+| **3-gram** | Subword | 2,480 | 11.28 | 17,300 | 27.4% | 69.1% |
+| **4-gram** | Word | 1,262 | 10.30 | 9,814 | 45.9% | 76.0% |
+| **4-gram** | Subword | 9,840 | 13.26 | 65,901 | 17.3% | 46.3% |
+| **5-gram** | Word | 1,050 | 10.04 | 7,194 | 45.5% | 79.7% |
+| **5-gram** | Subword | 19,916 | 14.28 | 117,450 | 14.0% | 39.1% |
 
 ### Top 5 N-grams by Size
 
-**2-grams:**
+**2-grams (Word):**
+
+| Rank | N-gram | Count |
+|------|--------|-------|
+| 1 | `l è` | 4,349 |
+| 2 | `da l` | 2,854 |
+| 3 | `d un` | 2,584 |
+| 4 | `dal calendàri` | 1,948 |
+| 5 | `è n` | 1,667 |
+
+**3-grams (Word):**
+
+| Rank | N-gram | Count |
+|------|--------|-------|
+| 1 | `l è n` | 1,665 |
+| 2 | `dal calendàri gregoriàn` | 1,584 |
+| 3 | `sécol dal calendàri` | 1,575 |
+| 4 | `è n an` | 1,575 |
+| 5 | `avenimèint nê mort` | 1,412 |
+
+**4-grams (Word):**
+
+| Rank | N-gram | Count |
+|------|--------|-------|
+| 1 | `l è n an` | 1,575 |
+| 2 | `ed domìni tachê a` | 1,255 |
+| 3 | `a funsionèr da l` | 1,255 |
+| 4 | `domìni tachê a funsionèr` | 1,255 |
+| 5 | `tachê a funsionèr da` | 1,255 |
+
+**5-grams (Word):**
+
+| Rank | N-gram | Count |
+|------|--------|-------|
+| 1 | `domìni tachê a funsionèr da` | 1,255 |
+| 2 | `ed domìni tachê a funsionèr` | 1,255 |
+| 3 | `tachê a funsionèr da l` | 1,255 |
+| 4 | `l è l nòm d` | 1,247 |
+| 5 | `l nòm d un domìni` | 1,247 |
+
+**2-grams (Subword):**
+
+| Rank | N-gram | Count |
+|------|--------|-------|
+| 1 | `a _` | 44,681 |
+| 2 | `l _` | 36,354 |
+| 3 | `_ d` | 31,152 |
+| 4 | `_ a` | 28,707 |
+| 5 | `n _` | 26,332 |
+
+**3-grams (Subword):**
 
 | Rank | N-gram | Count |
 |------|--------|-------|
-| 1 | `categoria :` | 28,651 |
-| 2 | `: nùmer` | 11,903 |
-| 3 | `' l` | 8,131 |
-| 4 | `l '` | 4,130 |
-| 5 | `' d` | 3,768 |
+| 1 | `a l _` | 19,233 |
+| 2 | `_ d a` | 13,700 |
+| 3 | `_ i n` | 10,014 |
+| 4 | `l a _` | 9,054 |
+| 5 | `d a l` | 8,840 |
 
-**3-grams:**
+**4-grams (Subword):**
 
 | Rank | N-gram | Count |
 |------|--------|-------|
-| 1 | `categoria : nùmer` | 11,902 |
-| 2 | `' l è` | 3,029 |
-| 3 | `l è '` | 2,876 |
-| 4 | `categoria : cinema` | 2,834 |
-| 5 | `da ' l` | 2,787 |
+| 1 | `_ d a l` | 8,766 |
+| 2 | `d a l _` | 8,710 |
+| 3 | `_ a l _` | 7,884 |
+| 4 | `_ e d _` | 6,634 |
+| 5 | `_ l a _` | 5,983 |
 
-**4-grams:**
+**5-grams (Subword):**
 
 | Rank | N-gram | Count |
 |------|--------|-------|
-| 1 | `' l è '` | 2,875 |
-| 2 | `: cinema categoria :` | 2,777 |
-| 3 | `categoria : cinema categoria` | 2,777 |
-| 4 | `: matematica categoria :` | 2,068 |
-| 5 | `categoria : matematica categoria` | 2,068 |
+| 1 | `_ d a l _` | 8,679 |
+| 2 | `_ d a _ '` | 2,988 |
+| 3 | `' l _ è _` | 2,975 |
+| 4 | `l _ è _ '` | 2,854 |
+| 5 | `d a _ ' l` | 2,762 |
 
 
 ### Key Findings
 
-- **Best Perplexity:** 2-gram with 367
+- **Best Perplexity:** 2-gram (subword) with 342
 - **Entropy Trend:** Decreases with larger n-grams (more predictable)
-- **Coverage:** Top-1000 patterns cover ~51% of corpus
+- **Coverage:** Top-1000 patterns cover ~39% of corpus
 - **Recommendation:** 4-gram or 5-gram for best predictive performance
 
 ---
@@ -182,55 +247,86 @@ Categoria:CITTADITALIA`
 
 ![Markov Entropy](visualizations/markov_entropy.png)
 
+![Markov Contexts](visualizations/markov_contexts.png)
+
 ![Markov Branching](visualizations/markov_branching.png)
 
 ### Results
 
-| Context | Avg Entropy | Perplexity | Branching Factor | Unique Contexts | Predictability |
-|---------|-------------|------------|------------------|-----------------|----------------|
-| **1** | 0.4790 | 1.394 | 2.99 | 54,943 | 52.1% |
-| **1** | 1.0426 | 2.060 | 10.88 | 518 | 0.0% |
-| **2** | 0.1826 | 1.135 | 1.42 | 163,004 | 81.7% |
-| **2** | 1.1789 | 2.264 | 6.89 | 5,633 | 0.0% |
-| **3** | 0.0769 | 1.055 | 1.15 | 229,724 | 92.3% |
-| **3** | 0.9021 | 1.869 | 3.96 | 38,777 | 9.8% |
-| **4** | 0.0371 🏆 | 1.026 | 1.07 | 262,110 | 96.3% |
-| **4** | 0.6406 🏆 | 1.559 | 2.42 | 153,229 | 35.9% |
+| Context | Variant | Avg Entropy | Perplexity | Branching Factor | Unique Contexts | Predictability |
+|---------|---------|-------------|------------|------------------|-----------------|----------------|
+| **1** | Word | 0.6144 | 1.531 | 3.27 | 38,079 | 38.6% |
+| **1** | Subword | 1.2142 | 2.320 | 11.72 | 398 | 0.0% |
+| **2** | Word | 0.1859 | 1.138 | 1.39 | 123,729 | 81.4% |
+| **2** | Subword | 1.1401 | 2.204 | 6.72 | 4,661 | 0.0% |
+| **3** | Word | 0.0688 | 1.049 | 1.12 | 170,769 | 93.1% |
+| **3** | Subword | 0.8376 | 1.787 | 3.69 | 31,279 | 16.2% |
+| **4** | Word | 0.0286 🏆 | 1.020 | 1.05 | 189,112 | 97.1% |
+| **4** | Subword | 0.5759 | 1.491 | 2.30 | 115,443 | 42.4% |
+
+### Generated Text Samples (Word-based)
+
+Below are text samples generated from each word-based Markov chain model:
+
+**Context Size 1:**
+
+1. `l é al progrâma pc 12 518 519 520 gonèl 22 ed domìni genèric al urèl`
+2. `al funsiòuna da per 4 quèśi prim sfènic difetìv 322 in sensu laudator temporis acti prudentes`
+3. `dal crìst 4 d oro una cumêdia d antonino inferito da l è l è l`
+
+**Context Size 2:**
+
+1. `l è n an dal vii sécol dal calendàri gregoriàn avenimèint nê guélf vi mort xii`
+2. `d un nùmer triangolèr moltìplica per 5 d un nùmer quèder moltìplica per 3 d un domìni`
+3. `dal calendàri gregoriàn avenimèint nê mort x`
+
+**Context Size 3:**
+
+1. `l è n an edl iii sécol dal calendàri gregoriàn avenimèint nê mort i`
+2. `dal calendàri gregoriàn avenimèint nê mort viii`
+3. `è n an edl viii sécol dal calendàri gregoriàn avenimèint nê mort xvi`
+
+**Context Size 4:**
+
+1. `l è n an edl ix sécol dal calendàri gregoriàn avenimèint nê mort v`
+2. `domìni tachê a funsionèr da l`
+3. `ed domìni tachê a funsionèr da l`
 
-### Generated Text Samples
 
-Below are text samples generated from each Markov chain model:
+### Generated Text Samples (Subword-based)
+
+Below are text samples generated from each subword-based Markov chain model:
 
 **Context Size 1:**
 
-1. `: cinema categoria : matematica categoria : film , ch ' la fifa , sànt pîr`
-2. `categoria : matematica categoria : nùmer intēr categoria : nùmer prìm categoria : nùmer ed leyland`
-3. `' 700 méla cristiàn dla smana ch ' l avû ? se stès . sînch ân`
+1. `_pe_gotili_l'n_i`
+2. `andogrin_menèiṣa`
+3. `i_incōridl_stêst`
 
 **Context Size 2:**
 
-1. `categoria : nùmer naturêl categoria : film eròtic categoria : nùmer intēr categoria : nùmer difetìv ...`
-2. `: nùmer naturêl categoria : nùmer 19 - gonèl categoria : cantànt itagliàn categoria : nùmer naturêl`
-3. `' l ' intervîn gaudio – an vrîven mìa fêr i spetâcôl pr ’ un zôgh americân`
+1. `a_cuns_e_tòra_fiō`
+2. `l_séco,_ed_unèli_`
+3. `_drê_avōl_è_'l_59`
 
 **Context Size 3:**
 
-1. `categoria : nùmer moltìplica per 2 ' d un nùmer quèder categoria : nùmer 3 - quèśi prim`
-2. `' l è ' l nòm ' d un domìni genèric la registrasiòun dal in dal sît edl`
-3. `l è ' n elemèint ed ) 1894 ∃ quantificadōr eśistensiêl ( " a gh ' è la`
+1. `al_sît_la_cà_paolo`
+2. `_da_63_in_difestìl`
+3. `_in-dóvv_a_un_di_c`
 
 **Context Size 4:**
 
-1. `' l è ' l nòm ' d un domìni genèric . al funsiòuna da ' l setèmber dal`
-2. `: cinema categoria : atōr americàṅ categoria : atōr canadéś categoria : atōr americàṅ categoria : ar...`
-3. `categoria : cinema categoria : film categoria : film eròtic categoria : film ad scarésa`
+1. `_dal_calendàri_greg`
+2. `dal_viii_sèc._préma`
+3. `_al_funsiòuna_da_'l`
 
 
 ### Key Findings
 
-- **Best Predictability:** Context-4 with 96.3% predictability
+- **Best Predictability:** Context-4 (word) with 97.1% predictability
 - **Branching Factor:** Decreases with context size (more deterministic)
-- **Memory Trade-off:** Larger contexts require more storage (153,229 contexts)
+- **Memory Trade-off:** Larger contexts require more storage (115,443 contexts)
 - **Recommendation:** Context-3 or Context-4 for text generation
 
 ---
@@ -246,36 +342,36 @@ Below are text samples generated from each Markov chain model:
 
 | Metric | Value |
 |--------|-------|
-| Vocabulary Size | 18,015 |
-| Total Tokens | 364,595 |
-| Mean Frequency | 20.24 |
+| Vocabulary Size | 14,744 |
+| Total Tokens | 272,012 |
+| Mean Frequency | 18.45 |
 | Median Frequency | 3 |
-| Frequency Std Dev | 319.82 |
+| Frequency Std Dev | 223.57 |
 
 ### Most Common Words
 
 | Rank | Word | Frequency |
 |------|------|-----------|
-| 1 | categoria | 28,655 |
-| 2 | nùmer | 14,021 |
-| 3 | l | 13,281 |
-| 4 | al | 10,419 |
-| 5 | dal | 9,093 |
-| 6 | a | 7,637 |
-| 7 | ed | 7,001 |
-| 8 | la | 6,876 |
-| 9 | d | 5,723 |
-| 10 | in | 5,357 |
+| 1 | l | 12,992 |
+| 2 | al | 10,267 |
+| 3 | dal | 8,736 |
+| 4 | a | 7,317 |
+| 5 | ed | 6,740 |
+| 6 | la | 6,622 |
+| 7 | d | 5,491 |
+| 8 | in | 5,032 |
+| 9 | è | 4,792 |
+| 10 | da | 4,480 |
 
 ### Least Common Words (from vocabulary)
 
 | Rank | Word | Frequency |
 |------|------|-----------|
-| 1 | antiquariàt | 2 |
-| 2 | espositìv | 2 |
-| 3 | ecosistèma | 2 |
-| 4 | trasformasiòun | 2 |
-| 5 | galleria | 2 |
+| 1 | espositìv | 2 |
+| 2 | ecosistèma | 2 |
+| 3 | trasformasiòun | 2 |
+| 4 | galleria | 2 |
+| 5 | space | 2 |
 | 6 | velò | 2 |
 | 7 | arriv | 2 |
 | 8 | sèda | 2 |
@@ -286,24 +382,24 @@ Below are text samples generated from each Markov chain model:
 
 | Metric | Value |
 |--------|-------|
-| Zipf Coefficient | 1.0040 |
-| R² (Goodness of Fit) | 0.993487 |
+| Zipf Coefficient | 1.0159 |
+| R² (Goodness of Fit) | 0.990784 |
 | Adherence Quality | **excellent** |
 
 ### Coverage Analysis
 
 | Top N Words | Coverage |
 |-------------|----------|
-| Top 100 | 59.9% |
-| Top 1,000 | 77.9% |
-| Top 5,000 | 89.8% |
-| Top 10,000 | 95.2% |
+| Top 100 | 57.7% |
+| Top 1,000 | 77.8% |
+| Top 5,000 | 90.7% |
+| Top 10,000 | 96.5% |
 
 ### Key Findings
 
-- **Zipf Compliance:** R²=0.9935 indicates excellent adherence to Zipf's law
-- **High Frequency Dominance:** Top 100 words cover 59.9% of corpus
-- **Long Tail:** 8,015 words needed for remaining 4.8% coverage
+- **Zipf Compliance:** R²=0.9908 indicates excellent adherence to Zipf's law
+- **High Frequency Dominance:** Top 100 words cover 57.7% of corpus
+- **Long Tail:** 4,744 words needed for remaining 3.5% coverage
 
 ---
 ## 5. Word Embeddings Evaluation
@@ -316,24 +412,119 @@ Below are text samples generated from each Markov chain model:
 
 ![t-SNE Sentences](visualizations/tsne_sentences.png)
 
-### Model Comparison
 
-| Model | Vocab Size | Dimension | Avg Norm | Std Norm | Isotropy |
-|-------|------------|-----------|----------|----------|----------|
-| **mono_32d** | 5,895 | 32 | 4.010 | 0.920 | 0.4456 🏆 |
-| **mono_64d** | 5,895 | 64 | 4.045 | 0.873 | 0.1681 |
-| **mono_128d** | 5,895 | 128 | 4.056 | 0.886 | 0.0292 |
-| **embeddings_enhanced** | 0 | 0 | 0.000 | 0.000 | 0.0000 |
+### 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 | 0.3584 | 0.4391 | N/A | N/A |
+| **mono_64d** | 64 | 0.1134 | 0.4504 | N/A | N/A |
+| **mono_128d** | 128 | 0.0166 | 0.4596 | N/A | N/A |
+| **aligned_32d** | 32 | 0.3584 🏆 | 0.4411 | 0.0140 | 0.1660 |
+| **aligned_64d** | 64 | 0.1134 | 0.4292 | 0.0460 | 0.2440 |
+| **aligned_128d** | 128 | 0.0166 | 0.4457 | 0.0400 | 0.2640 |
 
 ### Key Findings
 
-- **Best Isotropy:** mono_32d with 0.4456 (more uniform distribution)
-- **Dimension Trade-off:** Higher dimensions capture more semantics but reduce isotropy
-- **Vocabulary Coverage:** All models cover 5,895 words
-- **Recommendation:** 100d for balanced semantic capture and efficiency
+- **Best Isotropy:** aligned_32d with 0.3584 (more uniform distribution)
+- **Semantic Density:** Average pairwise similarity of 0.4442. Lower values indicate better semantic separation.
+- **Alignment Quality:** Aligned models achieve up to 4.6% R@1 in cross-lingual retrieval.
+- **Recommendation:** 128d aligned for best cross-lingual performance
 
 ---
-## 6. Summary & Recommendations
+## 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 | **1.037** | 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 |
+|--------|----------|
+| `-ca` | cal, cavésin, caviân |
+
+#### Productive Suffixes
+| Suffix | Examples |
+|--------|----------|
+| `-a` | scōla, algebra, câṣva |
+| `-um` | coelum, adsum, 217śum |
+| `-na` | vègna, teresina, ruvîna |
+
+### 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 |
+|------|----------|------------------|----------|
+| `asiò` | 1.80x | 17 contexts | asiòṅ, asiòun, frasiòn |
+| `siòu` | 1.79x | 17 contexts | asiòun, sesiòun, lesiòun |
+| `purt` | 1.55x | 23 contexts | purtâ, purtê, purtä |
+| `iòun` | 1.73x | 16 contexts | uniòun, asiòun, sesiòun |
+| `nter` | 1.50x | 24 contexts | inter, nterra, dänter |
+| `sèin` | 1.51x | 17 contexts | sèins, sèint, casèin |
+| `tèin` | 1.48x | 16 contexts | latèin, estèin, putèin |
+| `ital` | 1.53x | 14 contexts | italy, italo, vitali |
+| `tôri` | 1.78x | 9 contexts | stôri, stôric, stôria |
+| `rèin` | 1.46x | 14 contexts | rèina, trèin, terèin |
+| `inte` | 1.59x | 11 contexts | inter, intern, interès |
+| `mèin` | 1.79x | 8 contexts | mèint, camèin, mumèint |
+
+### 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 |
+|--------|--------|-----------|----------|
+| `-ca` | `-a` | 53 words | cavacürta, canpâgna |
+| `-ca` | `-na` | 16 words | canpâgna, catalógna |
+
+### 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 |
+|------|-----------------|------------|------|
+| cascaggna | **`ca-scagg-na`** | 3.0 | `scagg` |
+| califòrgna | **`ca-lifòrg-na`** | 3.0 | `lifòrg` |
+| campàggna | **`ca-mpàgg-na`** | 3.0 | `mpàgg` |
+| castlaran | **`ca-stlaran`** | 1.5 | `stlaran` |
+| philosophum | **`philosoph-um`** | 1.5 | `philosoph` |
+| privilegium | **`privilegi-um`** | 1.5 | `privilegi` |
+| calandäri | **`ca-landäri`** | 1.5 | `landäri` |
+| referendum | **`referend-um`** | 1.5 | `referend` |
+| metropolitana | **`metropolita-na`** | 1.5 | `metropolita` |
+| carabinieri | **`ca-rabinieri`** | 1.5 | `rabinieri` |
+| parmigiana | **`parmigia-na`** | 1.5 | `parmigia` |
+| funsiòuna | **`funsiòu-na`** | 1.5 | `funsiòu` |
+| carpigiano | **`ca-rpigiano`** | 1.5 | `rpigiano` |
+| caraterésstic | **`ca-raterésstic`** | 1.5 | `raterésstic` |
+| indipendentîxum | **`indipendentîx-um`** | 1.5 | `indipendentîx` |
+
+### 6.6 Linguistic Interpretation
+
+> **Automated Insight:**
+The language Unknown language [eml] 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)
 
@@ -341,11 +532,12 @@ Below are text samples generated from each Markov chain model:
 
 | Component | Recommended | Rationale |
 |-----------|-------------|-----------|
-| Tokenizer | **32k BPE** | Best compression (3.87x) with low UNK rate |
-| N-gram | **5-gram** | Lowest perplexity (367) |
-| Markov | **Context-4** | Highest predictability (96.3%) |
+| Tokenizer | **32k BPE** | Best compression (3.37x) |
+| N-gram | **2-gram** | Lowest perplexity (342) |
+| Markov | **Context-4** | Highest predictability (97.1%) |
 | Embeddings | **100d** | Balanced semantic capture and isotropy |
 
+
 ---
 ## Appendix: Metrics Glossary & Interpretation Guide
 
@@ -535,7 +727,8 @@ If you use these models in your research, please cite:
   author = {Kamali, Omar},
   title = {Wikilangs: Open NLP Models for Wikipedia Languages},
   year = {2025},
-  publisher = {HuggingFace},
+  doi = {10.5281/zenodo.18073153},
+  publisher = {Zenodo},
   url = {https://huggingface.co/wikilangs}
   institution = {Omneity Labs}
 }
@@ -551,7 +744,8 @@ MIT License - Free for academic and commercial use.
 - 🤗 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: 2025-12-30 12:20:31*
+*Report Date: 2026-01-04 14:33:51*

models/embeddings/aligned/eml_128d.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:fcb8fc5fbcdb5bbb3c3855e6260d0c0ed9be1bd2540732441851e144e55b1bb5
+size 1028819578

models/embeddings/aligned/eml_128d.meta.json

@@ -0,0 +1 @@
+{"lang": "eml", "dim": 128, "max_seq_len": 512, "is_aligned": true}

models/embeddings/aligned/eml_128d.projection.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:92a565cca0b1fec7fcc3929c57b7a0f14938727f603974d98feafa917c5b6a07
+size 65664

models/embeddings/aligned/eml_128d_metadata.json

@@ -0,0 +1,8 @@
+{
+  "language": "eml",
+  "dimension": 128,
+  "version": "aligned",
+  "hub_language": "en",
+  "seed_vocab_size": 1782,
+  "vocab_size": 4635
+}

models/embeddings/aligned/eml_32d.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:9048aabb8f2135dcb1eb5657f02c39c14081bad67cf1f08cc9d4be1e05987fdd
+size 257259898

models/embeddings/aligned/eml_32d.meta.json

@@ -0,0 +1 @@
+{"lang": "eml", "dim": 32, "max_seq_len": 512, "is_aligned": true}

models/embeddings/aligned/eml_32d.projection.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:4b3e21ae26c45f38a7dd8a3536b26336fb4ff728db58b503dcf80f58a62ce218
+size 4224

models/embeddings/aligned/eml_32d_metadata.json

@@ -0,0 +1,8 @@
+{
+  "language": "eml",
+  "dimension": 32,
+  "version": "aligned",
+  "hub_language": "en",
+  "seed_vocab_size": 1782,
+  "vocab_size": 4635
+}

models/embeddings/aligned/eml_64d.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:54e13e135d67a9e81ac2a112d124d647a6d159a876cf31a47f640828b12b32af
+size 514446458

models/embeddings/aligned/eml_64d.meta.json

@@ -0,0 +1 @@
+{"lang": "eml", "dim": 64, "max_seq_len": 512, "is_aligned": true}

models/embeddings/aligned/eml_64d.projection.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:c8496df49686df9ad094fba37cb9c2c359c62122fe0b0c3a35107a9b92700715
+size 16512

models/embeddings/aligned/eml_64d_metadata.json

@@ -0,0 +1,8 @@
+{
+  "language": "eml",
+  "dimension": 64,
+  "version": "aligned",
+  "hub_language": "en",
+  "seed_vocab_size": 1782,
+  "vocab_size": 4635
+}

models/embeddings/monolingual/eml_128d.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2dec3940dbc985fa841d1a2109e3331109da252bb16a290feddef6bff141233a
-size 1030131813
+oid sha256:fcb8fc5fbcdb5bbb3c3855e6260d0c0ed9be1bd2540732441851e144e55b1bb5
+size 1028819578

models/embeddings/monolingual/eml_128d_metadata.json

@@ -3,11 +3,13 @@
   "dimension": 128,
   "version": "monolingual",
   "training_params": {
-    "dim": 128,
+    "algorithm": "skipgram",
     "min_count": 5,
     "window": 5,
     "negative": 5,
-    "epochs": 5
+    "epochs": 5,
+    "encoding_method": "rope",
+    "dim": 128
   },
-  "vocab_size": 5895
+  "vocab_size": 4635
 }

models/embeddings/monolingual/eml_32d.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:8961e078ac6044a853b46be56893ebce2c5eb901499627424d436e4b8225a68e
-size 257604453
+oid sha256:9048aabb8f2135dcb1eb5657f02c39c14081bad67cf1f08cc9d4be1e05987fdd
+size 257259898

models/embeddings/monolingual/eml_32d_metadata.json

@@ -3,11 +3,13 @@
   "dimension": 32,
   "version": "monolingual",
   "training_params": {
-    "dim": 32,
+    "algorithm": "skipgram",
     "min_count": 5,
     "window": 5,
     "negative": 5,
-    "epochs": 5
+    "epochs": 5,
+    "encoding_method": "rope",
+    "dim": 32
   },
-  "vocab_size": 5895
+  "vocab_size": 4635
 }

models/embeddings/monolingual/eml_64d.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:ae7fd24e033b292e2212d1507f6faa04781e383418da4f7fbdf4b9fc502c9483
-size 515113573
+oid sha256:54e13e135d67a9e81ac2a112d124d647a6d159a876cf31a47f640828b12b32af
+size 514446458

models/embeddings/monolingual/eml_64d_metadata.json

@@ -3,11 +3,13 @@
   "dimension": 64,
   "version": "monolingual",
   "training_params": {
-    "dim": 64,
+    "algorithm": "skipgram",
     "min_count": 5,
     "window": 5,
     "negative": 5,
-    "epochs": 5
+    "epochs": 5,
+    "encoding_method": "rope",
+    "dim": 64
   },
-  "vocab_size": 5895
+  "vocab_size": 4635
 }

models/subword_markov/eml_markov_ctx1_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c097ab7bc9b897e8c09154ba19528d122008c0ea16dc6b51790599f0c3dd6dfe
-size 49006
+oid sha256:d6b926e1938473f2238d8b1a246fdca632c9df31a38f38cde47b1430b2326ca8
+size 41044

models/subword_markov/eml_markov_ctx1_subword_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 1,
   "variant": "subword",
   "language": "eml",
-  "unique_contexts": 518,
-  "total_transitions": 2357750
+  "unique_contexts": 398,
+  "total_transitions": 1534890
 }

models/subword_markov/eml_markov_ctx2_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:be0033475c1652e77b5fdf968e2c48e4cc8243a92791a66f0075dd78ea7da229
-size 302403
+oid sha256:b6b72d5397e477fda64eb68479d038a681650923e44ef8e56c40f841f61196ad
+size 239585

models/subword_markov/eml_markov_ctx2_subword_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 2,
   "variant": "subword",
   "language": "eml",
-  "unique_contexts": 5633,
-  "total_transitions": 2344566
+  "unique_contexts": 4661,
+  "total_transitions": 1529310
 }

models/subword_markov/eml_markov_ctx3_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:52b0a67bc82d05f5a35f649a72f33cdb45172a5d33904de6e0809a03164f8bae
-size 1051721
+oid sha256:d32d5c8474464f3542b5c640f67f02b7cdd18b7294731baa5ac803d49bb0eda9
+size 824128

models/subword_markov/eml_markov_ctx3_subword_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 3,
   "variant": "subword",
   "language": "eml",
-  "unique_contexts": 38777,
-  "total_transitions": 2331382
+  "unique_contexts": 31279,
+  "total_transitions": 1523730
 }

models/subword_markov/eml_markov_ctx4_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:0b3fa3bdc992bb33045daf260d321e98cd4d8f0bff7fe9925ceefb1b894f645c
-size 2772328
+oid sha256:86656b5d89b562fdcdf8d2b076d8bbbace53baaf9733ab41014512d25be68d3a
+size 2063582

models/subword_markov/eml_markov_ctx4_subword_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 4,
   "variant": "subword",
   "language": "eml",
-  "unique_contexts": 153229,
-  "total_transitions": 2318198
+  "unique_contexts": 115443,
+  "total_transitions": 1518150
 }

models/subword_ngram/eml_2gram_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:fc925b29a9fc88ab4cf66b8408d9eed367b4928e944c1292dd5824336cea41ac
-size 38300
+oid sha256:abc65477b77ec14a68768bc6a4cbe985f8fb901ef27a3c085790b09ba780b5f7
+size 32669

models/subword_ngram/eml_2gram_subword_metadata.json

@@ -2,6 +2,6 @@
   "n": 2,
   "variant": "subword",
   "language": "eml",
-  "unique_ngrams": 2917,
-  "total_ngrams": 2357750
+  "unique_ngrams": 2464,
+  "total_ngrams": 1534890
 }

models/subword_ngram/eml_3gram_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:77c3e7f77b71859d97a1553d613852050db6480386165b77e169f8f2e811014e
-size 253696
+oid sha256:98c47ca7a64b381223f763a388157b191e0de53ced724ae4b32a96deaf39a731
+size 205276

models/subword_ngram/eml_3gram_subword_metadata.json

@@ -2,6 +2,6 @@
   "n": 3,
   "variant": "subword",
   "language": "eml",
-  "unique_ngrams": 21058,
-  "total_ngrams": 2344566
+  "unique_ngrams": 17300,
+  "total_ngrams": 1529310
 }

models/subword_ngram/eml_4gram_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2f729f6f0180bbb326697b8f8f6ee4544692e6237369a32db85e4d4314868f64
-size 995050
+oid sha256:9c52554bef1faa4c04b02e07fd6ee1d6c2d1808c138f7468b367a43effed73bf
+size 776580

models/subword_ngram/eml_4gram_subword_metadata.json

@@ -2,6 +2,6 @@
   "n": 4,
   "variant": "subword",
   "language": "eml",
-  "unique_ngrams": 88240,
-  "total_ngrams": 2331382
+  "unique_ngrams": 65901,
+  "total_ngrams": 1523730
 }

models/subword_ngram/eml_5gram_subword.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ceedf622b97b0e16718698141c4b219dee5ba830fac72b206d1dc6e3f594096b
+size 1371520

models/subword_ngram/eml_5gram_subword_metadata.json

@@ -0,0 +1,7 @@
+{
+  "n": 5,
+  "variant": "subword",
+  "language": "eml",
+  "unique_ngrams": 117450,
+  "total_ngrams": 1518150
+}

models/tokenizer/eml_tokenizer_16k.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:da605d84f28d5f46f26288c35f9d14a5b54103782a7f4fff43cb4ebbc45fcfb0
-size 501006
+oid sha256:e4ca140168e773921387292dd413a84d6448393458bdcff201a3aa5256417afb
+size 508897

models/tokenizer/eml_tokenizer_32k.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4dd80c30b968ba0c9cb19ceea25fde8d0ee363c9db15ed1778feabd4e8d0fe63
-size 783799
+oid sha256:3f05d7f5bc311f044df8793f8da075d15b79dd7ccce85b63f5afe91f40ea36b3
+size 797552

models/tokenizer/eml_tokenizer_8k.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:1f40c61f5172b6055933e6e6cbad3e888013bf908612e87962354dcee190ed69
-size 370286
+oid sha256:0a22d81760f62d518080719bef868af8f2d18593e0e1ddcab0664e141bc3c718
+size 370847

models/vocabulary/eml_vocabulary.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d7e9c74a50c31370e5c3ea97bfdf5872799156358c39c661f61de5beebc61be3
-size 287372
+oid sha256:292e25805145d187f0bbc5f3f2fc487329b45ee890b9b03b744d59d04ce3e18d
+size 233473

models/vocabulary/eml_vocabulary_metadata.json

@@ -1,16 +1,17 @@
 {
   "language": "eml",
-  "vocabulary_size": 18015,
+  "vocabulary_size": 14744,
+  "variant": "full",
   "statistics": {
-    "type_token_ratio": 0.1374079429750441,
+    "type_token_ratio": 0.12930704797360182,
     "coverage": {
-      "top_100": 0.5431980467359734,
-      "top_1000": 0.7066495100662288,
-      "top_5000": 0.814778403589944,
-      "top_10000": 0.8643118651829692
+      "top_100": 0.5308097131737034,
+      "top_1000": 0.7164396311024621,
+      "top_5000": 0.8352855571537355,
+      "top_10000": 0.8884812589897623
     },
-    "hapax_count": 37194,
-    "hapax_ratio": 0.6736945063304896,
-    "total_documents": 13184
+    "hapax_count": 23463,
+    "hapax_ratio": 0.6141021278823252,
+    "total_documents": 5580
   }
 }

models/word_markov/eml_markov_ctx1_word.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:d821202557de5b54da2f843a288c7a778a7dccc6dbce3a3dcac43e102399b7d1
-size 1512273
+oid sha256:b58650add7a0b58bfff9460f57cc83996fd12568ed8a603676040d4122f7d3f7
+size 1170078

models/word_markov/eml_markov_ctx1_word_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 1,
   "variant": "word",
   "language": "eml",
-  "unique_contexts": 54943,
-  "total_transitions": 500906
+  "unique_contexts": 38079,
+  "total_transitions": 289895
 }

models/word_markov/eml_markov_ctx2_word.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b6b0d6986dea78583fb0c1563283c6d6f8c70ea36c65e20d31a321417c8aff29
-size 2759692
+oid sha256:180ce7b12c5d6e9d779d053b5c92bd28686e4ae07387f9ce71787acbdc206aa8
+size 2189970

models/word_markov/eml_markov_ctx2_word_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 2,
   "variant": "word",
   "language": "eml",
-  "unique_contexts": 163004,
-  "total_transitions": 487775
+  "unique_contexts": 123729,
+  "total_transitions": 284315
 }

models/word_markov/eml_markov_ctx3_word.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:45b80193c6d32ec40bbe313766649b2f767a796e6427017d9b93cd588fc105bb
-size 3690696
+oid sha256:8dcf8703e6eb0d8f9636fcdb573b7b4b59c35eca9c75294e4d6f51dbb89b5c46
+size 2864580

models/word_markov/eml_markov_ctx3_word_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 3,
   "variant": "word",
   "language": "eml",
-  "unique_contexts": 229724,
-  "total_transitions": 474974
+  "unique_contexts": 170769,
+  "total_transitions": 278735
 }