Upload all models and assets for ca (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

@@ -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:
@@ -23,14 +33,14 @@ dataset_info:
 metrics:
   - name: best_compression_ratio
     type: compression
-    value: 4.298
+    value: 4.448
   - name: best_isotropy
     type: isotropy
-    value: 0.7184
+    value: 0.7469
   - name: vocabulary_size
     type: vocab
-    value: 1000000
-generated: 2025-12-28
+    value: 0
+generated: 2026-01-08
 ---
 
 # Catalan - Wikilangs Models
@@ -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)
 
@@ -68,53 +80,57 @@ 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** | 3.555x | 3.53 | 0.1339% | 4,215,094 |
-| **16k** | 3.867x | 3.84 | 0.1457% | 3,874,434 |
-| **32k** | 4.116x | 4.09 | 0.1551% | 3,639,962 |
-| **64k** | 4.298x 🏆 | 4.27 | 0.1619% | 3,486,449 |
+| **8k** | 3.608x | 3.61 | 0.1295% | 3,980,202 |
+| **16k** | 3.955x | 3.96 | 0.1420% | 3,630,953 |
+| **32k** | 4.237x | 4.24 | 0.1521% | 3,389,435 |
+| **64k** | 4.448x 🏆 | 4.45 | 0.1597% | 3,228,954 |
 
 ### Tokenization Examples
 
 Below are sample sentences tokenized with each vocabulary size:
 
-**Sample 1:** `Guilherand-Granges és un municipi de la regió d'Alvèrnia-Roine-Alps i el departa...`
+**Sample 1:** `Llista de topònims (noms propis de lloc) del municipi de Capmany, a l'Alt Empord...`
 
 | Vocab | Tokens | Count |
 |-------|--------|-------|
-| 8k | `▁gu il her and - gran ges ▁és ▁un ▁municipi ... (+35 more)` | 45 |
-| 16k | `▁gu il her and - gran ges ▁és ▁un ▁municipi ... (+33 more)` | 43 |
-| 32k | `▁guil her and - gran ges ▁és ▁un ▁municipi ▁de ... (+29 more)` | 39 |
-| 64k | `▁guil her and - gran ges ▁és ▁un ▁municipi ▁de ... (+27 more)` | 37 |
+| 8k | `▁llista ▁de ▁topònims ▁( nom s ▁propis ▁de ▁lloc ) ... (+13 more)` | 23 |
+| 16k | `▁llista ▁de ▁topònims ▁( nom s ▁propis ▁de ▁lloc ) ... (+13 more)` | 23 |
+| 32k | `▁llista ▁de ▁topònims ▁( nom s ▁propis ▁de ▁lloc ) ... (+12 more)` | 22 |
+| 64k | `▁llista ▁de ▁topònims ▁( noms ▁propis ▁de ▁lloc ) ▁del ... (+10 more)` | 20 |
 
-**Sample 2:** `Estheria (crustaci), un gènere de crustacis del període Carbonífer
- Estheria (dí...`
+**Sample 2:** `Trànsportni (Krasnodar), poble del krai de Krasnodar, a Rússia Trànsportni (Maga...`
 
 | Vocab | Tokens | Count |
 |-------|--------|-------|
-| 8k | `▁est h eria ▁( cr usta ci ), ▁un ▁gènere ... (+32 more)` | 42 |
-| 16k | `▁est h eria ▁( cr usta ci ), ▁un ▁gènere ... (+29 more)` | 39 |
-| 32k | `▁est h eria ▁( cr usta ci ), ▁un ▁gènere ... (+27 more)` | 37 |
-| 64k | `▁est h eria ▁( cr usta ci ), ▁un ▁gènere ... (+23 more)` | 33 |
+| 8k | `▁tr àn s port ni ▁( k ras n od ... (+39 more)` | 49 |
+| 16k | `▁tràn sport ni ▁( k ras n od ar ), ... (+33 more)` | 43 |
+| 32k | `▁tràn sport ni ▁( k ras n odar ), ▁poble ... (+27 more)` | 37 |
+| 64k | `▁tràn sport ni ▁( k ras n odar ), ▁poble ... (+25 more)` | 35 |
 
-**Sample 3:** `Torneig de tennis masculí: St. Petersburg Open 2021
- Torneig de tennis femení: S...`
+**Sample 3:** `Torneigs de tennis masculí: Serbia Open (ATP 250) Belgrade Open (ATP 250) Tornei...`
 
 | Vocab | Tokens | Count |
 |-------|--------|-------|
-| 8k | `▁torneig ▁de ▁ten nis ▁mascul í : ▁st . ▁peters ... (+28 more)` | 38 |
-| 16k | `▁torneig ▁de ▁tennis ▁masculí : ▁st . ▁petersburg ▁open ▁ ... (+21 more)` | 31 |
-| 32k | `▁torneig ▁de ▁tennis ▁masculí : ▁st . ▁petersburg ▁open ▁ ... (+21 more)` | 31 |
-| 64k | `▁torneig ▁de ▁tennis ▁masculí : ▁st . ▁petersburg ▁open ▁ ... (+19 more)` | 29 |
+| 8k | `▁torneig s ▁de ▁ten nis ▁mascul í : ▁ser bia ... (+44 more)` | 54 |
+| 16k | `▁torneig s ▁de ▁tennis ▁masculí : ▁ser bia ▁open ▁( ... (+38 more)` | 48 |
+| 32k | `▁torneigs ▁de ▁tennis ▁masculí : ▁ser bia ▁open ▁( atp ... (+34 more)` | 44 |
+| 64k | `▁torneigs ▁de ▁tennis ▁masculí : ▁ser bia ▁open ▁( atp ... (+33 more)` | 43 |
 
 
 ### Key Findings
 
-- **Best Compression:** 64k achieves 4.298x compression
-- **Lowest UNK Rate:** 8k with 0.1339% unknown tokens
+- **Best Compression:** 64k achieves 4.448x compression
+- **Lowest UNK Rate:** 8k with 0.1295% unknown tokens
 - **Trade-off:** Larger vocabularies improve compression but increase model size
 - **Recommendation:** 32k vocabulary provides optimal balance for production use
 
@@ -123,57 +139,111 @@ Below are sample sentences tokenized with each vocabulary size:
 
 ![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** | 115,248 🏆 | 16.81 | 4,858,651 | 14.1% | 27.6% |
-| **2-gram** | 310 🏆 | 8.28 | 50,468 | 65.3% | 98.2% |
-| **3-gram** | 1,102,520 | 20.07 | 16,377,428 | 4.5% | 12.5% |
-| **3-gram** | 2,693 | 11.40 | 370,834 | 27.1% | 69.0% |
-| **4-gram** | 4,349,805 | 22.05 | 36,310,673 | 2.3% | 8.2% |
-| **4-gram** | 16,302 | 13.99 | 2,376,653 | 13.3% | 38.2% |
+| N-gram | Variant | Perplexity | Entropy | Unique N-grams | Top-100 Coverage | Top-1000 Coverage |
+|--------|---------|------------|---------|----------------|------------------|-------------------|
+| **2-gram** | Word | 167,717 | 17.36 | 4,576,334 | 10.6% | 23.4% |
+| **2-gram** | Subword | 262 🏆 | 8.03 | 41,609 | 69.0% | 98.9% |
+| **3-gram** | Word | 1,409,334 | 20.43 | 13,479,698 | 2.7% | 10.3% |
+| **3-gram** | Subword | 2,211 | 11.11 | 288,734 | 29.3% | 72.4% |
+| **4-gram** | Word | 4,798,593 | 22.19 | 27,616,287 | 1.8% | 7.6% |
+| **4-gram** | Subword | 13,232 | 13.69 | 1,676,138 | 14.2% | 40.2% |
+| **5-gram** | Word | 4,523,219 | 22.11 | 21,934,897 | 2.3% | 8.8% |
+| **5-gram** | Subword | 58,187 | 15.83 | 6,034,155 | 7.7% | 24.2% |
 
 ### Top 5 N-grams by Size
 
-**2-grams:**
+**2-grams (Word):**
+
+| Rank | N-gram | Count |
+|------|--------|-------|
+| 1 | `de la` | 3,892,352 |
+| 2 | `a la` | 1,832,648 |
+| 3 | `de l` | 1,806,800 |
+| 4 | `a l` | 1,007,338 |
+| 5 | `de les` | 998,964 |
+
+**3-grams (Word):**
+
+| Rank | N-gram | Count |
+|------|--------|-------|
+| 1 | `de la seva` | 186,164 |
+| 2 | `per a la` | 131,594 |
+| 3 | `referències enllaços externs` | 121,418 |
+| 4 | `la pel lícula` | 114,682 |
+| 5 | `d octubre de` | 112,980 |
+
+**4-grams (Word):**
+
+| Rank | N-gram | Count |
+|------|--------|-------|
+| 1 | `de kitt peak spacewatch` | 78,569 |
+| 2 | `de la universitat de` | 56,957 |
+| 3 | `que hi havia el` | 55,303 |
+| 4 | `segons el cens del` | 47,569 |
+| 5 | `de la família dels` | 44,734 |
+
+**5-grams (Word):**
+
+| Rank | N-gram | Count |
+|------|--------|-------|
+| 1 | `el nombre mitjà de persones` | 43,284 |
+| 2 | `el següent diagrama mostra les` | 42,548 |
+| 3 | `següent diagrama mostra les poblacions` | 42,548 |
+| 4 | `diagrama mostra les poblacions més` | 42,542 |
+| 5 | `mostra les poblacions més properes` | 42,497 |
+
+**2-grams (Subword):**
+
+| Rank | N-gram | Count |
+|------|--------|-------|
+| 1 | `a _` | 65,660,325 |
+| 2 | `s _` | 52,744,093 |
+| 3 | `_ d` | 49,682,099 |
+| 4 | `e _` | 42,364,044 |
+| 5 | `d e` | 41,208,775 |
+
+**3-grams (Subword):**
 
 | Rank | N-gram | Count |
 |------|--------|-------|
-| 1 | `l '` | 6,103,238 |
-| 2 | `d '` | 5,990,435 |
-| 3 | `de la` | 3,858,095 |
-| 4 | `categoria :` | 2,458,133 |
-| 5 | `a la` | 1,831,941 |
+| 1 | `_ d e` | 35,468,647 |
+| 2 | `d e _` | 24,280,649 |
+| 3 | `e s _` | 19,244,620 |
+| 4 | `e l _` | 15,094,409 |
+| 5 | `l a _` | 14,700,214 |
 
-**3-grams:**
+**4-grams (Subword):**
 
 | Rank | N-gram | Count |
 |------|--------|-------|
-| 1 | `de l '` | 1,783,419 |
-| 2 | `a l '` | 1,006,018 |
-| 3 | `| | |` | 637,045 |
-| 4 | `. l '` | 491,768 |
-| 5 | `d ' una` | 438,363 |
+| 1 | `_ d e _` | 23,793,570 |
+| 2 | `_ l a _` | 12,534,324 |
+| 3 | `_ e l _` | 8,556,406 |
+| 4 | `s _ d e` | 7,523,945 |
+| 5 | `d e _ l` | 7,343,393 |
 
-**4-grams:**
+**5-grams (Subword):**
 
 | Rank | N-gram | Count |
 |------|--------|-------|
-| 1 | `| | | |` | 320,057 |
-| 2 | `. referències categoria :` | 191,608 |
-| 3 | `categoria : naixements del` | 165,169 |
-| 4 | `- | | |` | 150,434 |
-| 5 | `d ' octubre de` | 137,349 |
+| 1 | `_ d e _ l` | 7,323,223 |
+| 2 | `_ d e l _` | 5,191,709 |
+| 3 | `s _ d e _` | 5,107,850 |
+| 4 | `_ q u e _` | 4,821,740 |
+| 5 | `a _ d e _` | 4,540,758 |
 
 
 ### Key Findings
 
-- **Best Perplexity:** 2-gram with 310
+- **Best Perplexity:** 2-gram (subword) with 262
 - **Entropy Trend:** Decreases with larger n-grams (more predictable)
-- **Coverage:** Top-1000 patterns cover ~38% of corpus
+- **Coverage:** Top-1000 patterns cover ~24% of corpus
 - **Recommendation:** 4-gram or 5-gram for best predictive performance
 
 ---
@@ -181,55 +251,86 @@ Below are sample sentences tokenized with each vocabulary size:
 
 ![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** | 0.6354 | 1.553 | 8.32 | 4,839,481 | 36.5% |
-| **1** | 1.0355 | 2.050 | 9.71 | 25,162 | 0.0% |
-| **2** | 0.4773 | 1.392 | 3.30 | 40,230,784 | 52.3% |
-| **2** | 0.6327 | 1.550 | 4.11 | 244,427 | 36.7% |
-| **3** | 0.2736 | 1.209 | 1.81 | 132,591,901 | 72.6% |
-| **3** | 0.7103 | 1.636 | 4.34 | 1,004,684 | 29.0% |
-| **4** | 0.1501 🏆 | 1.110 | 1.34 | 239,447,638 | 85.0% |
-| **4** | 0.7156 🏆 | 1.642 | 3.65 | 4,363,359 | 28.4% |
+| Context | Variant | Avg Entropy | Perplexity | Branching Factor | Unique Contexts | Predictability |
+|---------|---------|-------------|------------|------------------|-----------------|----------------|
+| **1** | Word | 0.9702 | 1.959 | 13.70 | 3,298,751 | 3.0% |
+| **1** | Subword | 0.8467 | 1.798 | 7.10 | 30,691 | 15.3% |
+| **2** | Word | 0.4478 | 1.364 | 2.95 | 45,099,512 | 55.2% |
+| **2** | Subword | 0.5676 | 1.482 | 3.72 | 217,960 | 43.2% |
+| **3** | Word | 0.2425 | 1.183 | 1.66 | 133,056,441 | 75.8% |
+| **3** | Subword | 0.6293 | 1.547 | 3.86 | 810,473 | 37.1% |
+| **4** | Word | 0.1249 🏆 | 1.090 | 1.26 | 221,190,469 | 87.5% |
+| **4** | Subword | 0.6563 | 1.576 | 3.56 | 3,128,822 | 34.4% |
+
+### Generated Text Samples (Word-based)
+
+Below are text samples generated from each word-based Markov chain model:
+
+**Context Size 1:**
+
+1. `de maig de la temporada l acceptació de muntar una muralla i el molí de la`
+2. `la població comunicació de encara que alemanya i des de la computació sent l estat substituïda`
+3. `i no són esmentats anteriorment icv el símbol del psoe des de guilgameix que un comerç`
+
+**Context Size 2:**
+
+1. `de la guerra di mario tronti i no solament va trobar que era del 5è al 16è`
+2. `a la taula de composició amb la seva història general del magistrat monetari c cassi a la`
+3. `de l expedició del virrei un germà gran del poble ulldeconencs o ulldeconins són coneguts com a`
+
+**Context Size 3:**
+
+1. `de la seva carrera periodística escrivint col laboracions a joves intel lectuals pertanyents a l alt...`
+2. `per a la secció de filosofia i ciències socials en les seves obligacions amb la seguretat i el`
+3. `referències enllaços externs fira festa de la pasqua hayivky el casament vessilia o ladkannya de la ...`
+
+**Context Size 4:**
+
+1. `de kitt peak spacewatch 8 de novembre de parcak i mumford del 8 de novembre de militants del flec`
+2. `de la universitat de salamanca honoris causa per la universitat christian albrecht de kiel de la uni...`
+3. `que hi havia el 1 era una gran superfície de material de bricolatge 1 una botiga de congelats 1`
 
-### 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. `de londres categoria : districte és un programador en carta de suècia ruben / 1982 )`
-2. `, la tasca de nyerros que feia constar qui li ocasionaren la figura femenina de missouri`
-3. `. en un planell inferior del castell d ' aquestes àrees a l ' un grup`
+1. `_daral_euílere_s`
+2. `eivinde_ditel'hi`
+3. `agraweros._ome_2`
 
 **Context Size 2:**
 
-1. `l ' antiguitat i entre el 626 els àvars de la majoria dels parlants d ' aliments`
-2. `d ' àmbits els quals louis companyo formà el 1918 lhasa va causar un accident cerebrovascular ,`
-3. `de la influència dels descendents de dalmau i ribalta ( 1900 ) hindoo jugglers ( 1914 )`
+1. `a_ses_va_únivenci`
+2. `s_als_(rdor_reu_d`
+3. `_d'ofegria_amb_o_`
 
 **Context Size 3:**
 
-1. `de l ' àlbum d ' estudi , en el seu camí per trobar intuïtivament i de sobte`
-2. `a l ' hipotàlem i la suprarenal , una glàndula intramandibular inflada que s ' ha presentat a`
-3. `| | | | — | - id = 312 bgcolor = # d6d6d6 | 459311 | |`
+1. `_de_bre_seteodent_`
+2. `de_la_de_col·locia`
+3. `es_pres,_nastorals`
 
 **Context Size 4:**
 
-1. `| | | | 6 d ' abril , 2002 | | palomar | | neat | - |`
-2. `. referències categoria : òperes de gaetano donizetti categoria : òperes del 1922 categoria : morts ...`
-3. `categoria : naixements del 1914 categoria : morts el 2023 categoria : morts a bagdad categoria : mor...`
+1. `_de_doble_(a_−_batx`
+2. `_la_de_fan_es_va_ca`
+3. `_el_donar_les_si_es`
 
 
 ### Key Findings
 
-- **Best Predictability:** Context-4 with 85.0% predictability
+- **Best Predictability:** Context-4 (word) with 87.5% predictability
 - **Branching Factor:** Decreases with context size (more deterministic)
-- **Memory Trade-off:** Larger contexts require more storage (4,363,359 contexts)
+- **Memory Trade-off:** Larger contexts require more storage (3,128,822 contexts)
 - **Recommendation:** Context-3 or Context-4 for text generation
 
 ---
@@ -245,64 +346,64 @@ Below are text samples generated from each Markov chain model:
 
 | Metric | Value |
 |--------|-------|
-| Vocabulary Size | 1,000,000 |
-| Total Tokens | 394,566,173 |
-| Mean Frequency | 394.57 |
-| Median Frequency | 9 |
-| Frequency Std Dev | 36714.18 |
+| Vocabulary Size | 1,490,582 |
+| Total Tokens | 372,231,757 |
+| Mean Frequency | 249.72 |
+| Median Frequency | 4 |
+| Frequency Std Dev | 29623.92 |
 
 ### Most Common Words
 
 | Rank | Word | Frequency |
 |------|------|-----------|
-| 1 | de | 24,313,844 |
-| 2 | la | 12,938,517 |
-| 3 | i | 9,983,050 |
-| 4 | a | 9,644,831 |
-| 5 | el | 8,858,221 |
-| 6 | l | 6,235,709 |
-| 7 | d | 6,156,292 |
-| 8 | en | 5,560,129 |
-| 9 | del | 5,289,798 |
-| 10 | que | 4,942,373 |
+| 1 | de | 23,862,515 |
+| 2 | la | 12,874,088 |
+| 3 | i | 9,923,035 |
+| 4 | a | 9,593,194 |
+| 5 | el | 8,820,173 |
+| 6 | l | 6,195,164 |
+| 7 | d | 5,995,004 |
+| 8 | en | 5,534,785 |
+| 9 | del | 5,257,995 |
+| 10 | que | 4,926,945 |
 
 ### Least Common Words (from vocabulary)
 
 | Rank | Word | Frequency |
 |------|------|-----------|
-| 1 | cesel | 3 |
-| 2 | epinic | 3 |
-| 3 | deplexión | 3 |
-| 4 | ε³ | 3 |
-| 5 | α³ | 3 |
-| 6 | engelska | 3 |
-| 7 | rechercheconsultation | 3 |
-| 8 | pdfir | 3 |
-| 9 | βασίλιος | 3 |
-| 10 | βασιλείος | 3 |
+| 1 | binaritruncat | 2 |
+| 2 | fanerozoiques | 2 |
+| 3 | biòmers | 2 |
+| 4 | nianzhi | 2 |
+| 5 | fuching | 2 |
+| 6 | mndm | 2 |
+| 7 | cpsf | 2 |
+| 8 | preestàndard | 2 |
+| 9 | sweetshop | 2 |
+| 10 | whakaata | 2 |
 
 ### Zipf's Law Analysis
 
 | Metric | Value |
 |--------|-------|
-| Zipf Coefficient | 1.0318 |
-| R² (Goodness of Fit) | 0.994658 |
+| Zipf Coefficient | 1.0222 |
+| R² (Goodness of Fit) | 0.996032 |
 | Adherence Quality | **excellent** |
 
 ### Coverage Analysis
 
 | Top N Words | Coverage |
 |-------------|----------|
-| Top 100 | 43.6% |
-| Top 1,000 | 63.1% |
+| Top 100 | 45.0% |
+| Top 1,000 | 63.8% |
 | Top 5,000 | 78.5% |
-| Top 10,000 | 84.3% |
+| Top 10,000 | 84.2% |
 
 ### Key Findings
 
-- **Zipf Compliance:** R²=0.9947 indicates excellent adherence to Zipf's law
-- **High Frequency Dominance:** Top 100 words cover 43.6% of corpus
-- **Long Tail:** 990,000 words needed for remaining 15.7% coverage
+- **Zipf Compliance:** R²=0.9960 indicates excellent adherence to Zipf's law
+- **High Frequency Dominance:** Top 100 words cover 45.0% of corpus
+- **Long Tail:** 1,480,582 words needed for remaining 15.8% coverage
 
 ---
 ## 5. Word Embeddings Evaluation
@@ -315,24 +416,131 @@ 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** | 1,514,898 | 32 | 3.264 | 1.334 | 0.7184 🏆 |
-| **mono_64d** | 1,514,898 | 64 | 3.636 | 1.309 | 0.7113 |
-| **mono_128d** | 1,514,898 | 128 | 4.070 | 1.306 | 0.6648 |
-| **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 | 0.7469 🏆 | 0.3896 | N/A | N/A |
+| **mono_64d** | 64 | 0.7390 | 0.2972 | N/A | N/A |
+| **mono_128d** | 128 | 0.6902 | 0.2374 | N/A | N/A |
+| **aligned_32d** | 32 | 0.7469 | 0.3696 | 0.4960 | 0.8360 |
+| **aligned_64d** | 64 | 0.7390 | 0.3068 | 0.7200 | 0.9380 |
+| **aligned_128d** | 128 | 0.6902 | 0.2443 | 0.8320 | 0.9720 |
 
 ### Key Findings
 
-- **Best Isotropy:** mono_32d with 0.7184 (more uniform distribution)
-- **Dimension Trade-off:** Higher dimensions capture more semantics but reduce isotropy
-- **Vocabulary Coverage:** All models cover 1,514,898 words
-- **Recommendation:** 100d for balanced semantic capture and efficiency
+- **Best Isotropy:** mono_32d with 0.7469 (more uniform distribution)
+- **Semantic Density:** Average pairwise similarity of 0.3075. Lower values indicate better semantic separation.
+- **Alignment Quality:** Aligned models achieve up to 83.2% 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 | **-0.637** | 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 |
+|--------|----------|
+| `-ca` | canadàwilliam, cancells, callissot |
+| `-co` | compsopogon, corlea, constitutionem |
+| `-ma` | matricarina, masaraga, massai |
+
+#### Productive Suffixes
+| Suffix | Examples |
+|--------|----------|
+| `-s` | pomacèntrids, pentalobulars, quiotas |
+| `-a` | matricarina, arduinna, yarima |
+| `-es` | asfèriques, biomatemàtiques, quies |
+| `-en` | grieneisen, robien, tensionen |
+| `-is` | rufistrigalis, reaccionaris, catàrsis |
+| `-ia` | praskóvia, llògia, orogenia |
+| `-ta` | lucasta, samudragupta, lisetita |
+
+### 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 |
+|------|----------|------------------|----------|
+| `nter` | 1.39x | 729 contexts | inter, anter, únter |
+| `efer` | 1.66x | 177 contexts | kefer, lefer, defer |
+| `uerr` | 1.61x | 153 contexts | uerra, guerr, duerr |
+| `espr` | 1.73x | 95 contexts | esprî, despr, esprai |
+| `stru` | 1.32x | 389 contexts | strum, struk, strus |
+| `rson` | 1.46x | 205 contexts | rsona, arson, urson |
+| `ient` | 1.31x | 364 contexts | rient, oient, lient |
+| `lmen` | 1.57x | 122 contexts | ulmen, ilmen, olmen |
+| `rinc` | 1.48x | 147 contexts | rinck, rincó, rinca |
+| `ènci` | 1.57x | 107 contexts | ència, mència, lència |
+| `embr` | 1.33x | 234 contexts | membr, embre, embry |
+| `onst` | 1.42x | 159 contexts | onsta, konst, const |
+
+### 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 |
+|--------|--------|-----------|----------|
+| `-co` | `-s` | 48 words | conventos, conservadors |
+| `-ma` | `-a` | 45 words | masicka, macclureana |
+| `-ca` | `-s` | 40 words | callolepis, cambyses |
+| `-co` | `-a` | 35 words | comunera, costanzana |
+| `-ma` | `-s` | 33 words | mahates, maktens |
+| `-ca` | `-a` | 30 words | camborda, cardellina |
+| `-co` | `-es` | 14 words | congoatlàntiques, colomates |
+| `-ca` | `-es` | 11 words | cambyses, calcídies |
+| `-ma` | `-es` | 9 words | mahates, masies |
+| `-ma` | `-ta` | 9 words | magnesiodumortierita, malwatta |
+
+### 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 |
+|------|-----------------|------------|------|
+| guerrista | **`guerr-is-ta`** | 6.0 | `guerr` |
+| whitlockita | **`whitlocki-ta`** | 4.5 | `whitlocki` |
+| assumptionis | **`assumption-is`** | 4.5 | `assumption` |
+| zumacales | **`zumacal-es`** | 4.5 | `zumacal` |
+| raperswilen | **`raperswil-en`** | 4.5 | `raperswil` |
+| antinomies | **`antinomi-es`** | 4.5 | `antinomi` |
+| reglamentaren | **`reglamentar-en`** | 4.5 | `reglamentar` |
+| remarcaria | **`remarcar-ia`** | 4.5 | `remarcar` |
+| reichsfürsten | **`reichsfürst-en`** | 4.5 | `reichsfürst` |
+| deflectores | **`deflector-es`** | 4.5 | `deflector` |
+| produeixen | **`produeix-en`** | 4.5 | `produeix` |
+| autoadjuntes | **`autoadjunt-es`** | 4.5 | `autoadjunt` |
+| subministraria | **`subministrar-ia`** | 4.5 | `subministrar` |
+| barbertonita | **`barbertoni-ta`** | 4.5 | `barbertoni` |
+| balsameres | **`balsamer-es`** | 4.5 | `balsamer` |
+
+### 6.6 Linguistic Interpretation
+
+> **Automated Insight:**
+The language Catalan 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)
 
@@ -340,11 +548,12 @@ Below are text samples generated from each Markov chain model:
 
 | Component | Recommended | Rationale |
 |-----------|-------------|-----------|
-| Tokenizer | **32k BPE** | Best compression (4.30x) with low UNK rate |
-| N-gram | **5-gram** | Lowest perplexity (310) |
-| Markov | **Context-4** | Highest predictability (85.0%) |
+| Tokenizer | **64k BPE** | Best compression (4.45x) |
+| N-gram | **2-gram** | Lowest perplexity (262) |
+| Markov | **Context-4** | Highest predictability (87.5%) |
 | Embeddings | **100d** | Balanced semantic capture and isotropy |
 
+
 ---
 ## Appendix: Metrics Glossary & Interpretation Guide
 
@@ -534,7 +743,8 @@ 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}
 }
@@ -550,7 +760,8 @@ 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: 2025-12-28 16:22:11*
+*Report Date: 2026-01-08 03:10:53*

models/embeddings/aligned/ca_128d.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:24a1e4df40a3066dc73fa3b9c563fc1f4e8fb90e1aee1858380c3d7bd8a8e8a3
+size 2501686884

models/embeddings/aligned/ca_128d.meta.json

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

models/embeddings/aligned/ca_128d.projection.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:802d4468d1f385670eaeb49471e886b448ff5e09a77dcb77154c3337c462986c
+size 65664

models/embeddings/aligned/ca_128d_metadata.json

@@ -0,0 +1,8 @@
+{
+  "language": "ca",
+  "dimension": 128,
+  "version": "aligned",
+  "hub_language": "en",
+  "seed_vocab_size": 338419,
+  "vocab_size": 1417503
+}

models/embeddings/aligned/ca_32d.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:7090d8847bec43a8a16f3caa2e657c0f40a895cd63a132dcd601abb6499dd545
+size 645044580

models/embeddings/aligned/ca_32d.meta.json

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

models/embeddings/aligned/ca_32d.projection.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:b4959f123eccf2b23e091c2a6c906c37ee6fffc5f512fc10722aabe48d584a75
+size 4224

models/embeddings/aligned/ca_32d_metadata.json

@@ -0,0 +1,8 @@
+{
+  "language": "ca",
+  "dimension": 32,
+  "version": "aligned",
+  "hub_language": "en",
+  "seed_vocab_size": 338419,
+  "vocab_size": 1417503
+}

models/embeddings/aligned/ca_64d.bin

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:1880e73b0ed8f1e8ab616846d81db2396794ec49a6ebb244f50960fe73ff693c
+size 1263925348

models/embeddings/aligned/ca_64d.meta.json

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

models/embeddings/aligned/ca_64d.projection.npy

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:66569971bcebdb0f2bcfc3f5a476d7a05b75dc3989f9453416de8a30ce89ca1c
+size 16512

models/embeddings/aligned/ca_64d_metadata.json

@@ -0,0 +1,8 @@
+{
+  "language": "ca",
+  "dimension": 64,
+  "version": "aligned",
+  "hub_language": "en",
+  "seed_vocab_size": 338419,
+  "vocab_size": 1417503
+}

models/embeddings/monolingual/ca_128d.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:2de1ec6ef15f6007dfa5c9a4c67f73236a818307f9039071d3a4e96ea5f6e7b6
-size 2603504009
+oid sha256:24a1e4df40a3066dc73fa3b9c563fc1f4e8fb90e1aee1858380c3d7bd8a8e8a3
+size 2501686884

models/embeddings/monolingual/ca_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": 1514898
+  "vocab_size": 1417503
 }

models/embeddings/monolingual/ca_32d.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:9644da17bbff6c339ba936b958d9b31a96266a293851c8eccad67b458ce4841a
-size 672062345
+oid sha256:7090d8847bec43a8a16f3caa2e657c0f40a895cd63a132dcd601abb6499dd545
+size 645044580

models/embeddings/monolingual/ca_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": 1514898
+  "vocab_size": 1417503
 }

models/embeddings/monolingual/ca_64d.bin

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:e8f8fb2c25755c94a88eace5de5f1fbd52ec30ff04fbffa01e595b78c516ed32
-size 1315876233
+oid sha256:1880e73b0ed8f1e8ab616846d81db2396794ec49a6ebb244f50960fe73ff693c
+size 1263925348

models/embeddings/monolingual/ca_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": 1514898
+  "vocab_size": 1417503
 }

models/subword_markov/ca_markov_ctx1_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:270c80664a864c1f71060ea7f8d709b2d841f34520fcae7f80ab36d41715d729
-size 1444269
+oid sha256:9eff07200d9975dea253fe43de601f122185e6a96985c26c438b63afaaf9acb1
+size 1393281

models/subword_markov/ca_markov_ctx1_subword_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 1,
   "variant": "subword",
   "language": "ca",
-  "unique_contexts": 25162,
-  "total_transitions": 2374472116
+  "unique_contexts": 30691,
+  "total_transitions": 2170761675
 }

models/subword_markov/ca_markov_ctx2_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:6553292a052ea21a7e04dd2fe1c65f9a324ceb7a03ff2b07b83bf8062abe2647
-size 8229347
+oid sha256:ca5c5b1e54ae39698e5cacc14cec34cbf702e677bf2e32ec6603b6841603a6fa
+size 6606050

models/subword_markov/ca_markov_ctx2_subword_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 2,
   "variant": "subword",
   "language": "ca",
-  "unique_contexts": 244427,
-  "total_transitions": 2373684587
+  "unique_contexts": 217960,
+  "total_transitions": 2169979038
 }

models/subword_markov/ca_markov_ctx3_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:3fbb799a80b64b1325067d7884080bfc376e7a29f06c3e2867f5a831e0224606
-size 33793244
+oid sha256:db35fc8ef584d59df2ee3752ca547611bb3f3a2bd606eb85c0dc6fc242d5e672
+size 25408651

models/subword_markov/ca_markov_ctx3_subword_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 3,
   "variant": "subword",
   "language": "ca",
-  "unique_contexts": 1004684,
-  "total_transitions": 2372897058
+  "unique_contexts": 810473,
+  "total_transitions": 2169196401
 }

models/subword_markov/ca_markov_ctx4_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f490ba905b17564a560111ab0f1c465f11666745e7a4119fce8f3ba6f168102d
-size 126097716
+oid sha256:2396047321998b3d2609f25bf3a10d651f28323cbf9cb2bde8269322c726d3ef
+size 90579198

models/subword_markov/ca_markov_ctx4_subword_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 4,
   "variant": "subword",
   "language": "ca",
-  "unique_contexts": 4363359,
-  "total_transitions": 2372109529
+  "unique_contexts": 3128822,
+  "total_transitions": 2168413764
 }

models/subword_ngram/ca_2gram_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:b7ab6db3e589628283d59649de17cd75a05bc6be07a068cbbe0464e833f5e1ad
-size 693894
+oid sha256:8436fa1ec5955cb730f5246fc40eabfe83d8b9edf3339cb07157ca799560b084
+size 590509

models/subword_ngram/ca_2gram_subword_metadata.json

@@ -2,6 +2,6 @@
   "n": 2,
   "variant": "subword",
   "language": "ca",
-  "unique_ngrams": 50468,
-  "total_ngrams": 2374472116
+  "unique_ngrams": 41609,
+  "total_ngrams": 2170761675
 }

models/subword_ngram/ca_3gram_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:4ccfc3bdd00e817e25923cd9aa462aec12b25e85ae59752607b8ca8d99cb3b94
-size 4601148
+oid sha256:96f568707811cc9639e19837002c41df6be438a465ba2d78ffab6ff9e0b94288
+size 3621792

models/subword_ngram/ca_3gram_subword_metadata.json

@@ -2,6 +2,6 @@
   "n": 3,
   "variant": "subword",
   "language": "ca",
-  "unique_ngrams": 370834,
-  "total_ngrams": 2373684587
+  "unique_ngrams": 288734,
+  "total_ngrams": 2169979038
 }

models/subword_ngram/ca_4gram_subword.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:a08ad0ffa43059f8afb0bd3da57b5c77ce0655de29a285e0c08a1e1d5f029d0f
-size 28473302
+oid sha256:f51c3b0800498bf443c8ecbf8324c285828c177a16f9e740b0dd747334b77dc4
+size 20638377

models/subword_ngram/ca_4gram_subword_metadata.json

@@ -2,6 +2,6 @@
   "n": 4,
   "variant": "subword",
   "language": "ca",
-  "unique_ngrams": 2376653,
-  "total_ngrams": 2372897058
+  "unique_ngrams": 1676138,
+  "total_ngrams": 2169196401
 }

models/subword_ngram/ca_5gram_subword.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:ace21b62f81448899fa83f82e04ede8337828b8fc9fa347420bff131c37d5b8e
+size 74421040

models/subword_ngram/ca_5gram_subword_metadata.json

@@ -0,0 +1,7 @@
+{
+  "n": 5,
+  "variant": "subword",
+  "language": "ca",
+  "unique_ngrams": 6034155,
+  "total_ngrams": 2168413764
+}

models/tokenizer/ca_tokenizer_16k.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:f01f7af1f8d5c4e98ce3e2b3f156da25a7fc75208e444d2da62aa6bd055eb53c
-size 513099
+oid sha256:09cedc3c99d8c1bf292ffebd251dd5c3ad9de85c2f453e3e2832994834c283b8
+size 511401

models/tokenizer/ca_tokenizer_32k.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c1ec62b707367a1c81e08267aa8b8e0399165c52209f9fc1b829d24bb692a9a9
-size 795646
+oid sha256:8b22bc0012e3d4b698e9d9f2caf992c456c9a549b663b1a55d05642a0f73648b
+size 790628

models/tokenizer/ca_tokenizer_64k.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:377e3e1a65683ec7a5ea17048490d3664d56f4aa83bf05202b915f3b4294c9b5
-size 1368059
+oid sha256:88d889048650ea6a5950b07988fd2f76716015504f0d365d9a6ca927d6f5e1da
+size 1358262

models/tokenizer/ca_tokenizer_8k.model

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:c14f1b4742fb21d6ca5d98cb15380605f889146de927013a491de70b5e12e57b
-size 375029
+oid sha256:b638ce04790bb844a2400a620c30328aac4bd967de91839673382a6537634870
+size 374334

models/vocabulary/ca_vocabulary.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:93e0c0ee1b4d1a2b23217b790aa4baa87458d5f7ff2b006079314bff72875c6a
-size 14885768
+oid sha256:b996ac5d5b1626b076abdc107b305b912677f3186fb9bcc303dcc216d8153a3e
+size 22533215

models/vocabulary/ca_vocabulary_metadata.json

@@ -1,16 +1,17 @@
 {
   "language": "ca",
-  "vocabulary_size": 1000000,
+  "vocabulary_size": 1490582,
+  "variant": "full",
   "statistics": {
-    "type_token_ratio": 0.012118454476561086,
+    "type_token_ratio": 0.0088238088631879,
     "coverage": {
-      "top_100": 0.43113921935660243,
-      "top_1000": 0.6237288116322734,
-      "top_5000": 0.7756172262594092,
-      "top_10000": 0.8329204449496337
+      "top_100": 0.4481618941219131,
+      "top_1000": 0.6351444736313012,
+      "top_5000": 0.7810939031610028,
+      "top_10000": 0.8379540286860089
     },
-    "hapax_count": 3103831,
-    "hapax_ratio": 0.6414158123806192,
-    "total_documents": 787529
+    "hapax_count": 1809890,
+    "hapax_ratio": 0.5483730811835398,
+    "total_documents": 782637
   }
 }

models/vocabulary/ca_vocabulary_top.parquet

@@ -0,0 +1,3 @@
+version https://git-lfs.github.com/spec/v1
+oid sha256:dc7c0f73dc675fec3ea36f344213a6fef18fac8c3db6daba687f6e5b5ce59583
+size 15047767

models/vocabulary/ca_vocabulary_top_metadata.json

@@ -0,0 +1,20 @@
+{
+  "language": "ca",
+  "vocabulary_size": 1000000,
+  "variant": "top",
+  "statistics": {
+    "type_token_ratio": 0.0088238088631879,
+    "coverage": {
+      "top_100": 0.4481618941219131,
+      "top_1000": 0.6351444736313012,
+      "top_5000": 0.7810939031610028,
+      "top_10000": 0.8379540286860089
+    },
+    "hapax_count": 1809890,
+    "hapax_ratio": 0.5483730811835398,
+    "total_documents": 782637,
+    "top_vocab_size": 1000000,
+    "coverage_ratio": 0.9923944164431507,
+    "tokens_excluded": 490582
+  }
+}

models/word_markov/ca_markov_ctx1_word.parquet

@@ -1,3 +1,3 @@
 version https://git-lfs.github.com/spec/v1
-oid sha256:07bdda626dbfe973038eb25a9dc10c1512fdb8805dfb1c42748d0c9780c2d6bc
-size 361692175
+oid sha256:96435b63dd567ae47be848bf59b2c094743cf70cb0fef7e47e6b333c23332b3a
+size 388066360

models/word_markov/ca_markov_ctx1_word_metadata.json

@@ -2,6 +2,6 @@
   "context_size": 1,
   "variant": "word",
   "language": "ca",
-  "unique_contexts": 4839481,
-  "total_transitions": 490797423
+  "unique_contexts": 3298751,
+  "total_transitions": 373259010
 }