Madurese - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on Madurese 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

1. Tokenizer Evaluation

Results

Vocab Size	Compression	Avg Token Len	UNK Rate	Total Tokens
8k	3.672x	3.68	0.0762%	283,323
16k	4.063x	4.07	0.0844%	255,999
32k	4.409x	4.41	0.0915%	235,937
64k	4.690x 🏆	4.69	0.0974%	221,777

Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

Sample 1: Kolami iyâ arèya dhisa è Kacamadhân Walea Kapoloan, Tojo Una-Una, Sulawesi Tengn...

Vocab	Tokens	Count
8k	`▁ko lami ▁iyâ ▁arèya ▁dhisa ▁è ▁kacamadhân ▁wa lea ▁kapoloan ... (+12 more)`	22
16k	`▁ko lami ▁iyâ ▁arèya ▁dhisa ▁è ▁kacamadhân ▁walea ▁kapoloan , ... (+10 more)`	20
32k	`▁ko lami ▁iyâ ▁arèya ▁dhisa ▁è ▁kacamadhân ▁walea ▁kapoloan , ... (+10 more)`	20
64k	`▁kolami ▁iyâ ▁arèya ▁dhisa ▁è ▁kacamadhân ▁walea ▁kapoloan , ▁tojo ... (+9 more)`	19

Sample 2: jmpl Nyarang ojhen biasanah è kalakoh parappâèn bâdâ acara mantân

Vocab	Tokens	Count
8k	`▁jmpl ▁ny arang ▁o jh en ▁biasanah ▁è ▁kala koh ... (+9 more)`	19
16k	`▁jmpl ▁ny arang ▁o jh en ▁biasanah ▁è ▁kala koh ... (+8 more)`	18
32k	`▁jmpl ▁ny arang ▁o jhen ▁biasanah ▁è ▁kala koh ▁para ... (+6 more)`	16
64k	`▁jmpl ▁nyarang ▁ojhen ▁biasanah ▁è ▁kalakoh ▁parappâ èn ▁bâdâ ▁acara ... (+1 more)`	11

Sample 3: jmpl cer bawang, iâ area kakanan dâri Mekasân, Madhurâ. èghâbây dâri teppong bân...

Vocab	Tokens	Count
8k	`▁jmpl ▁cer ▁ba wang , ▁i â ▁area ▁kakanan ▁dâri ... (+13 more)`	23
16k	`▁jmpl ▁cer ▁bawang , ▁i â ▁area ▁kakanan ▁dâri ▁me ... (+11 more)`	21
32k	`▁jmpl ▁cer ▁bawang , ▁iâ ▁area ▁kakanan ▁dâri ▁me kasân ... (+10 more)`	20
64k	`▁jmpl ▁cer ▁bawang , ▁iâ ▁area ▁kakanan ▁dâri ▁mekasân , ... (+9 more)`	19

Key Findings

Best Compression: 64k achieves 4.690x compression
Lowest UNK Rate: 8k with 0.0762% 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

Results

N-gram	Variant	Perplexity	Entropy	Unique N-grams	Top-100 Coverage	Top-1000 Coverage
2-gram	Word	7,660	12.90	15,927	15.5%	38.8%
2-gram	Subword	284 🏆	8.15	2,917	65.5%	99.2%
3-gram	Word	8,331	13.02	12,743	10.9%	33.7%
3-gram	Subword	2,475	11.27	21,754	25.0%	69.0%
4-gram	Word	11,782	13.52	16,213	9.8%	26.2%
4-gram	Subword	14,165	13.79	105,104	11.2%	37.1%
5-gram	Word	6,142	12.58	8,427	13.4%	34.7%
5-gram	Subword	47,465	15.53	258,432	7.5%	23.2%

Top 5 N-grams by Size

2-grams (Word):

Rank	N-gram	Count
1	`iyâ arèya`	3,079
2	`è taon`	2,005
3	`sala sèttong`	1,545
4	`è bâkto`	1,201
5	`ka angghuy`	1,038

3-grams (Word):

Rank	N-gram	Count
1	`panèka sala sèttong`	583
2	`al qur an`	334
3	`sè bâḍâ è`	250
4	`arèya sala sèttong`	249
5	`iyâ arèya sala`	218

4-grams (Word):

Rank	N-gram	Count
1	`iyâ arèya sala sèttong`	205
2	`sala sèttong naghârâ è`	116
3	`sè tamaso ka ḍâlem`	114
4	`tamaso ka ḍâlem famili`	112
5	`panèka sala sèttong sastrawan`	106

5-grams (Word):

Rank	N-gram	Count
1	`sè tamaso ka ḍâlem famili`	111
2	`panèka sala sèttong naghârâ è`	97
3	`arèya tombuwân sè tamaso ka`	83
4	`iyâ arèya tombuwân sè tamaso`	81
5	`panèka sala sèttong sastrawan bân`	76

2-grams (Subword):

Rank	N-gram	Count
1	`a n`	135,108
2	`a _`	111,120
3	`n _`	106,453
4	`n g`	96,416
5	`_ s`	84,557

3-grams (Subword):

Rank	N-gram	Count
1	`_ k a`	39,553
2	`a n _`	38,801
3	`â n _`	37,878
4	`n g _`	34,520
5	`a n g`	34,407

4-grams (Subword):

Rank	N-gram	Count
1	`b â n _`	25,412
2	`_ s è _`	23,221
3	`_ b â n`	22,209
4	`_ p a n`	12,960
5	`g h i _`	12,282

5-grams (Subword):

Rank	N-gram	Count
1	`_ b â n _`	19,896
2	`a g h i _`	10,512
3	`a n g g h`	7,941
4	`a n è k a`	6,131
5	`r è y a _`	6,117

Key Findings

Best Perplexity: 2-gram (subword) with 284
Entropy Trend: Decreases with larger n-grams (more predictable)
Coverage: Top-1000 patterns cover ~23% of corpus
Recommendation: 4-gram or 5-gram for best predictive performance

3. Markov Chain Evaluation

Results

Context	Variant	Avg Entropy	Perplexity	Branching Factor	Unique Contexts	Predictability
1	Word	0.8768	1.836	5.67	85,149	12.3%
1	Subword	0.9174	1.889	5.75	1,785	8.3%
2	Word	0.2172	1.162	1.45	481,154	78.3%
2	Subword	0.7767	1.713	4.61	10,251	22.3%
3	Word	0.0556	1.039	1.08	694,348	94.4%
3	Subword	0.8058	1.748	3.95	47,246	19.4%
4	Word	0.0147 🏆	1.010	1.02	750,067	98.5%
4	Subword	0.6526	1.572	2.80	186,332	34.7%

Generated Text Samples (Word-based)

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

Context Size 1:

è sosol empa kecamaḍhân bone èkennal mènangka am jungen rhein è ḍâlem ghâbâyânna james tautan sè
sè ajhârâ neng pernata dhârurat politik filsafat tiongkok akennalaghi kendaraan rèya kalabân lo polo...
bân sayatan è tèmor gedenken an panèka èlakonè marèna dâpa sè terlibat ḍâlem abentu pandhengngan man...

Context Size 2:

iyâ arèya katettapân ḍâri allah kaangghuy ngalakonè imsak molaè bâkto teknologi transistor mulaè a n...
è taon schrödinger dhâddhi asisten exner sombher
sala sèttong naghârâ è èropa lao provinsi kapolowan kanary ceuta melilla è afrika kantor perserikata...

Context Size 3:

panèka sala sèttong sastrawan bân panolès inḍonèsia karjâ buku bidadari untuk dewa assalamualaikum b...
al qur an bapa èn serring nghâjhâk potra potrana akompol samarèna maghrib kaângguy abahas tafsir al ...
sè bâḍâ è antara kompolan polo polo è tèmorra polo maḍhurâ sapuḍi aropa aghi polo palèng lowas nomer

Context Size 4:

iyâ arèya sala sèttong ghunong wisata sè baḍâ è banyuwangi bân bândâbâsa jhâbâ tèmor inḍonèsia sè an...
sala sèttong naghârâ è èropa bârâ antillen belanda provinsi bonaire sint eustatius bân saba è amerik...
sè tamaso ka ḍâlem famili cucurbitaceae tombuwân arèya èkoca kèya jambu bol inḍonesia malay apple in...

Generated Text Samples (Subword-based)

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

Context Size 1:

_ewây_bâtè_se_pa
a'_jon_è._kana_l
n_-la_al_paasèra

Context Size 2:

an_kaapès_jaktunt
a_bia_al_nèkentuh
n_ton:_enta_pem-m

Context Size 3:

_kaoḍi’_“propa_kuf
an_krèpublik_ngalo
ân_sè_labân_kapa_l

Context Size 4:

bân_smp_3_ḍésémber_
_sè_abârra_sala_oli
_bân_bân_demi_abhâr

Key Findings

Best Predictability: Context-4 (word) with 98.5% predictability
Branching Factor: Decreases with context size (more deterministic)
Memory Trade-off: Larger contexts require more storage (186,332 contexts)
Recommendation: Context-3 or Context-4 for text generation

4. Vocabulary Analysis

Statistics

Metric	Value
Vocabulary Size	37,097
Total Tokens	741,682
Mean Frequency	19.99
Median Frequency	4
Frequency Std Dev	232.38

Most Common Words

Rank	Word	Frequency
1	è	23,535
2	sè	23,401
3	bân	20,011
4	ka	7,685
5	panèka	5,706
6	taon	5,597
7	ḍâri	4,979
8	kalabân	4,663
9	arèya	4,306
10	orèng	4,157

Least Common Words (from vocabulary)

Rank	Word	Frequency
1	eghunaaghin	2
2	pengatorannah	2
3	ngelaksanaaghin	2
4	sampèr	2
5	geluk	2
6	tekuk	2
7	rasmè	2
8	maddhekka	2
9	uttarkashi	2
10	spillway	2

Zipf's Law Analysis

Metric	Value
Zipf Coefficient	1.0120
R² (Goodness of Fit)	0.991547
Adherence Quality	excellent

Coverage Analysis

Top N Words	Coverage
Top 100	31.6%
Top 1,000	58.3%
Top 5,000	79.8%
Top 10,000	87.8%

Key Findings

Zipf Compliance: R²=0.9915 indicates excellent adherence to Zipf's law
High Frequency Dominance: Top 100 words cover 31.6% of corpus
Long Tail: 27,097 words needed for remaining 12.2% coverage

5. Word Embeddings Evaluation

5.1 Cross-Lingual Alignment

5.2 Model Comparison

Model	Dimension	Isotropy	Semantic Density	Alignment R@1	Alignment R@10
mono_32d	32	0.8668 🏆	0.3020	N/A	N/A
mono_64d	64	0.6062	0.2632	N/A	N/A
mono_128d	128	0.1633	0.2527	N/A	N/A
aligned_32d	32	0.8668	0.3113	0.0380	0.2740
aligned_64d	64	0.6062	0.2737	0.0720	0.3700
aligned_128d	128	0.1633	0.2516	0.1100	0.4080

Key Findings

Best Isotropy: mono_32d with 0.8668 (more uniform distribution)
Semantic Density: Average pairwise similarity of 0.2757. Lower values indicate better semantic separation.
Alignment Quality: Aligned models achieve up to 11.0% 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	0.495	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
`-a`	advokasi, aobâna, alias
`-s`	sekabbhinna, sahabatta, salajâ
`-ka`	kakosongan, kapalana, kaodi
`-ma`	macmillan, marapi, mareh
`-k`	kemaluan, kakosongan, khadijah
`-pa`	paragraf, parsiapân, panyâbâb
`-b`	berry, biography, bhâdâ
`-p`	penolès, paragraf, parsiapân

Productive Suffixes

Suffix	Examples
`-n`	kemaluan, kakosongan, parsiapân
`-a`	sekabbhinna, sahabatta, aobâna
`-an`	kemaluan, kakosongan, macmillan
`-i`	èghâdhui, advokasi, ègabungaghi
`-hi`	ègabungaghi, aningghâlaghi, èdebataghi
`-na`	sekabbhinna, aobâna, rilisna
`-s`	waprès, penolès, cutlass
`-ng`	gâmpang, tambâng, torkaḍâng

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
`angk`	1.72x	122 contexts	angka, angko, èangka
`nggh`	1.57x	158 contexts	ongghe, èngghi, èngghâ
`gghu`	1.88x	60 contexts	agghu, negghu, ongghu
`ngka`	1.55x	131 contexts	angka, èangka, mengka
`angg`	1.47x	151 contexts	anggâ, anggun, rangga
`ddhi`	1.98x	37 contexts	eddhi, seddhi, deddhi
`gghâ`	1.73x	63 contexts	cegghâ, èngghâ, logghâ
`tton`	2.08x	25 contexts	ottone, èttong, button
`âddh`	2.13x	16 contexts	bâddhâ, ḍâddhi, sâddhi
`hâdd`	2.12x	15 contexts	dhâddi, dhâddih, dhâddhi
`aren`	1.66x	33 contexts	karen, arena, areng
`labâ`	1.84x	22 contexts	labân, alabân, labâng

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
`-p`	`-n`	162 words	pangobhâdhân, panganjhuân
`-pa`	`-n`	161 words	pangobhâdhân, panganjhuân
`-ka`	`-n`	154 words	kabendherran, kaodhiân
`-s`	`-a`	130 words	sèvilla, sadaja
`-k`	`-n`	124 words	kabendherran, kaodhiân
`-p`	`-an`	122 words	pakarangan, pangamatan
`-pa`	`-an`	106 words	pakarangan, pangamatan
`-k`	`-an`	99 words	kabendherran, karegghingan
`-a`	`-i`	91 words	adhâddiyaghi, azeri
`-ka`	`-an`	90 words	kabendherran, karegghingan

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
bertasbih	`bertasb-i-h`	7.5	`i`
pertamina	`pertam-i-na`	7.5	`i`
fakultassa	`fakultas-s-a`	7.5	`s`
pendukungnga	`pendukung-ng-a`	7.5	`ng`
parèntana	`parènt-an-a`	7.5	`an`
terlarang	`terla-ra-ng`	7.5	`ra`
kebijaksanaan	`kebijaksa-na-an`	7.5	`na`
ibukottana	`ibukott-an-a`	7.5	`an`
kapotosanna	`kapotos-an-na`	7.5	`an`
rangsangan	`rangsa-ng-an`	7.5	`ng`
pangangghuy	`pa-ng-angghuy`	7.5	`angghuy`
tangghungan	`tangghu-ng-an`	7.5	`ng`
polinesia	`poline-si-a`	7.5	`si`
pematangan	`pe-ma-tangan`	7.5	`tangan`
ètampilkan	`ètampil-k-an`	7.5	`k`

6.6 Linguistic Interpretation

Automated Insight: The language Madurese 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

Production Recommendations

Component	Recommended	Rationale
Tokenizer	64k BPE	Best compression (4.69x)
N-gram	2-gram	Lowest perplexity (284)
Markov	Context-4	Highest predictability (98.5%)
Embeddings	100d	Balanced semantic capture and isotropy

Appendix: Metrics Glossary & Interpretation Guide

This section provides definitions, intuitions, and guidance for interpreting the metrics used throughout this report.

Tokenizer Metrics

Compression Ratio

Definition: The ratio of characters to tokens (chars/token). Measures how efficiently the tokenizer represents text.

Intuition: Higher compression means fewer tokens needed to represent the same text, reducing sequence lengths for downstream models. A 3x compression means ~3 characters per token on average.

What to seek: Higher is generally better for efficiency, but extremely high compression may indicate overly aggressive merging that loses morphological information.

Average Token Length (Fertility)

Definition: Mean number of characters per token produced by the tokenizer.

Intuition: Reflects the granularity of tokenization. Longer tokens capture more context but may struggle with rare words; shorter tokens are more flexible but increase sequence length.

What to seek: Balance between 2-5 characters for most languages. Arabic/morphologically-rich languages may benefit from slightly longer tokens.

Unknown Token Rate (OOV Rate)

Definition: Percentage of tokens that map to the unknown/UNK token, indicating words the tokenizer cannot represent.

Intuition: Lower OOV means better vocabulary coverage. High OOV indicates the tokenizer encounters many unseen character sequences.

What to seek: Below 1% is excellent; below 5% is acceptable. BPE tokenizers typically achieve very low OOV due to subword fallback.

N-gram Model Metrics

Perplexity

Definition: Measures how "surprised" the model is by test data. Mathematically: 2^(cross-entropy). Lower values indicate better prediction.

Intuition: If perplexity is 100, the model is as uncertain as if choosing uniformly among 100 options at each step. A perplexity of 10 means effectively choosing among 10 equally likely options.

What to seek: Lower is better. Perplexity decreases with larger n-grams (more context). Values vary widely by language and corpus size.

Entropy

Definition: Average information content (in bits) needed to encode the next token given the context. Related to perplexity: perplexity = 2^entropy.

Intuition: High entropy means high uncertainty/randomness; low entropy means predictable patterns. Natural language typically has entropy between 1-4 bits per character.

What to seek: Lower entropy indicates more predictable text patterns. Entropy should decrease as n-gram size increases.

Coverage (Top-K)

Definition: Percentage of corpus occurrences explained by the top K most frequent n-grams.

Intuition: High coverage with few patterns indicates repetitive/formulaic text; low coverage suggests diverse vocabulary usage.

What to seek: Depends on use case. For language modeling, moderate coverage (40-60% with top-1000) is typical for natural text.

Markov Chain Metrics

Average Entropy

Definition: Mean entropy across all contexts, measuring average uncertainty in next-word prediction.

Intuition: Lower entropy means the model is more confident about what comes next. Context-1 has high entropy (many possible next words); Context-4 has low entropy (few likely continuations).

What to seek: Decreasing entropy with larger context sizes. Very low entropy (<0.1) indicates highly deterministic transitions.

Branching Factor

Definition: Average number of unique next tokens observed for each context.

Intuition: High branching = many possible continuations (flexible but uncertain); low branching = few options (predictable but potentially repetitive).

What to seek: Branching factor should decrease with context size. Values near 1.0 indicate nearly deterministic chains.

Predictability

Definition: Derived metric: (1 - normalized_entropy) × 100%. Indicates how deterministic the model's predictions are.

Intuition: 100% predictability means the next word is always certain; 0% means completely random. Real text falls between these extremes.

What to seek: Higher predictability for text generation quality, but too high (>98%) may produce repetitive output.

Vocabulary & Zipf's Law Metrics

Zipf's Coefficient

Definition: The slope of the log-log plot of word frequency vs. rank. Zipf's law predicts this should be approximately -1.

Intuition: A coefficient near -1 indicates the corpus follows natural language patterns where a few words are very common and most words are rare.

What to seek: Values between -0.8 and -1.2 indicate healthy natural language distribution. Deviations may suggest domain-specific or artificial text.

R² (Coefficient of Determination)

Definition: Measures how well the linear fit explains the frequency-rank relationship. Ranges from 0 to 1.

Intuition: R² near 1.0 means the data closely follows Zipf's law; lower values indicate deviation from expected word frequency patterns.

What to seek: R² > 0.95 is excellent; > 0.99 indicates near-perfect Zipf adherence typical of large natural corpora.

Vocabulary Coverage

Definition: Cumulative percentage of corpus tokens accounted for by the top N words.

Intuition: Shows how concentrated word usage is. If top-100 words cover 50% of text, the corpus relies heavily on common words.

What to seek: Top-100 covering 30-50% is typical. Higher coverage indicates more repetitive text; lower suggests richer vocabulary.

Word Embedding Metrics

Isotropy

Definition: Measures how uniformly distributed vectors are in the embedding space. Computed as the ratio of minimum to maximum singular values.

Intuition: High isotropy (near 1.0) means vectors spread evenly in all directions; low isotropy means vectors cluster in certain directions, reducing expressiveness.

What to seek: Higher isotropy generally indicates better-quality embeddings. Values > 0.1 are reasonable; > 0.3 is good. Lower-dimensional embeddings tend to have higher isotropy.

Average Norm

Definition: Mean magnitude (L2 norm) of word vectors in the embedding space.

Intuition: Indicates the typical "length" of vectors. Consistent norms suggest stable training; high variance may indicate some words are undertrained.

What to seek: Relatively consistent norms across models. The absolute value matters less than consistency (low std deviation).

Cosine Similarity

Definition: Measures angular similarity between vectors, ranging from -1 (opposite) to 1 (identical direction).

Intuition: Words with similar meanings should have high cosine similarity. This is the standard metric for semantic relatedness in embeddings.

What to seek: Semantically related words should score > 0.5; unrelated words should be near 0. Synonyms often score > 0.7.

t-SNE Visualization

Definition: t-Distributed Stochastic Neighbor Embedding - a dimensionality reduction technique that preserves local structure for visualization.

Intuition: Clusters in t-SNE plots indicate groups of semantically related words. Spread indicates vocabulary diversity; tight clusters suggest semantic coherence.

What to seek: Meaningful clusters (e.g., numbers together, verbs together). Avoid over-interpreting distances - t-SNE preserves local, not global, structure.

General Interpretation Guidelines

Compare within model families: Metrics are most meaningful when comparing models of the same type (e.g., 8k vs 64k tokenizer).
Consider trade-offs: Better performance on one metric often comes at the cost of another (e.g., compression vs. OOV rate).
Context matters: Optimal values depend on downstream tasks. Text generation may prioritize different metrics than classification.
Corpus influence: All metrics are influenced by corpus characteristics. Wikipedia text differs from social media or literature.
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 - a monthly snapshot of Wikipedia articles across 300+ languages.

Project

A project by Wikilangs - Open-source NLP models for every Wikipedia language.

Maintainer

Omar Kamali - Omneity Labs

Citation

If you use these models in your research, please cite:

@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.

Madurese - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

📋 Repository Contents

Models & Assets

Analysis and Evaluation

1. Tokenizer Evaluation

Results

Tokenization Examples

Key Findings

2. N-gram Model Evaluation

Results

Top 5 N-grams by Size

Key Findings

3. Markov Chain Evaluation

Results

Generated Text Samples (Word-based)

Generated Text Samples (Subword-based)

Key Findings

4. Vocabulary Analysis

Statistics

Most Common Words

Least Common Words (from vocabulary)

Zipf's Law Analysis

Coverage Analysis

Key Findings

5. Word Embeddings Evaluation

5.1 Cross-Lingual Alignment

5.2 Model Comparison

Key Findings

6. Morphological Analysis (Experimental)

6.1 Productivity & Complexity

6.2 Affix Inventory (Productive Units)

Productive Prefixes

Productive Suffixes

6.3 Bound Stems (Lexical Roots)

6.4 Affix Compatibility (Co-occurrence)

6.5 Recursive Morpheme Segmentation

6.6 Linguistic Interpretation

7. Summary & Recommendations

Production Recommendations

Appendix: Metrics Glossary & Interpretation Guide

Tokenizer Metrics

N-gram Model Metrics

Markov Chain Metrics

Vocabulary & Zipf's Law Metrics

Word Embedding Metrics

General Interpretation Guidelines

Visualizations Index

About This Project

Data Source

Project

Maintainer

Citation

License

Links

Dataset used to train wikilangs/mad