Tetum - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

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

Analysis and Evaluation

• 1. Tokenizer Evaluation
• 2. N-gram Model Evaluation
• 3. Markov Chain Evaluation
• 4. Vocabulary Analysis
• 5. Word Embeddings Evaluation
• 6. Morphological Analysis (Experimental)
• 7. Summary & Recommendations
• Metrics Glossary
• Visualizations Index

---

1. Tokenizer Evaluation

Results

Vocab Size	Compression	Avg Token Len	UNK Rate	Total Tokens
8k	3.685x	3.69	0.0920%	220,741
16k	3.897x	3.90	0.0973%	208,698
32k	4.079x 🏆	4.08	0.1018%	199,418

Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

Sample 1: Paraná mak sai estadu iha Brazíl. Populasaun ema Ligasaun Ba Li'ur Governo do Es...

Vocab	Tokens	Count
8k	▁par aná ▁mak ▁sai ▁estadu ▁iha ▁brazíl . ▁populasaun ▁ema ... (+18 more)	28
16k	▁paraná ▁mak ▁sai ▁estadu ▁iha ▁brazíl . ▁populasaun ▁ema ▁ligasaun ... (+16 more)	26
32k	▁paraná ▁mak ▁sai ▁estadu ▁iha ▁brazíl . ▁populasaun ▁ema ▁ligasaun ... (+16 more)	26

Sample 2: Mekanika (Lian Latina mechanicus, husi Lian Yunani Mechanikos, ema ne'ebe espesi...

Vocab	Tokens	Count
8k	▁mekanika ▁( lian ▁la tina ▁me ch an ic us ... (+24 more)	34
16k	▁mekanika ▁( lian ▁latina ▁mechan ic us , ▁husi ▁lian ... (+18 more)	28
32k	▁mekanika ▁( lian ▁latina ▁mechanicus , ▁husi ▁lian ▁yunani ▁mechanikos ... (+12 more)	22

Sample 3: Inkscape hanesan Aplikasaun editor ba imajem ne'ebe ho kodigu nakloke iha lisens...

Vocab	Tokens	Count
8k	▁in ks cape ▁hanesan ▁aplikasaun ▁ed itor ▁ba ▁imajem ▁ne ... (+15 more)	25
16k	▁in ks cape ▁hanesan ▁aplikasaun ▁editor ▁ba ▁imajem ▁ne ' ... (+11 more)	21
32k	▁inkscape ▁hanesan ▁aplikasaun ▁editor ▁ba ▁imajem ▁ne ' ebe ▁ho ... (+9 more)	19

Key Findings

• Best Compression: 32k achieves 4.079x compression
• Lowest UNK Rate: 8k with 0.0920% 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	1,400	10.45	5,366	42.9%	71.5%
2-gram	Subword	284 🏆	8.15	1,827	67.5%	99.3%
3-gram	Word	1,275	10.32	6,153	49.9%	70.9%
3-gram	Subword	2,144	11.07	13,149	25.6%	72.8%
4-gram	Word	1,739	10.76	10,529	49.1%	63.6%
4-gram	Subword	8,921	13.12	53,309	14.4%	45.0%
5-gram	Word	1,049	10.03	7,279	55.7%	71.0%
5-gram	Subword	20,481	14.32	103,985	10.6%	35.3%

Top 5 N-grams by Size

2-grams (Word):

Rank	N-gram	Count
1	ne e	2,579
2	ne ebé	2,254
3	iha tinan	1,036
4	timor leste	973
5	lorosa e	966

3-grams (Word):

Rank	N-gram	Count
1	timór lorosa e	863
2	ba li ur	806
3	ligasaun ba li	803
4	timor leste nian	553
5	ne e iha	542

4-grams (Word):

Rank	N-gram	Count
1	ligasaun ba li ur	803
2	iha timór lorosa e	486
3	da républica mit dem	440
4	républica mit dem diploma	440
5	jornal da républica mit	439

5-grams (Word):

Rank	N-gram	Count
1	da républica mit dem diploma	440
2	jornal da républica mit dem	439
3	ida iha timór lorosa e	439
4	ur sensus fo fila fali	438
5	mit dem diploma ministerial n	438

2-grams (Subword):

Rank	N-gram	Count
1	a _	55,311
2	a n	26,720
3	n _	25,228
4	_ n	24,195
5	e _	21,839

3-grams (Subword):

Rank	N-gram	Count
1	a n _	10,780
2	h a _	10,572
3	i h a	10,489
4	i a _	9,335
5	_ i h	9,184

4-grams (Subword):

Rank	N-gram	Count
1	i h a _	10,318
2	_ i h a	9,183
3	a u n _	6,940
4	_ n i a	6,849
5	s a u n	6,166

5-grams (Subword):

Rank	N-gram	Count
1	_ i h a _	9,098
2	s a u n _	5,780
3	_ s i r a	4,434
4	a s a u n	4,363
5	_ n i a n	3,879

Key Findings

• Best Perplexity: 2-gram (subword) with 284
• Entropy Trend: Decreases with larger n-grams (more predictable)
• Coverage: Top-1000 patterns cover ~35% 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	0.8183	1.763	4.67	28,115	18.2%
1	Subword	1.1849	2.274	9.58	386	0.0%
2	Word	0.2239	1.168	1.48	130,951	77.6%
2	Subword	1.0621	2.088	6.47	3,691	0.0%
3	Word	0.0716	1.051	1.13	192,808	92.8%
3	Subword	0.8599	1.815	3.88	23,852	14.0%
4	Word	0.0258 🏆	1.018	1.04	216,360	97.4%
4	Subword	0.5884	1.504	2.40	92,496	41.2%

Generated Text Samples (Word-based)

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

Context Size 1:

1. iha okos hosi ligasaun ba dook liu tan aplikasiaun simples konsistente no hetan ona kartaun natál
2. ne ebé afirma konkluzaun ka siénsia sira tenta seluk ne e haklakar an liu hanesan programa
3. no sosiál isabel de daroca td duxambé tanzánia td taxkent v de amor do escuta nian

Context Size 2:

1. ne e mós bele funsiona nu udar interiór nia kontinentál ho nuanse sira foho sira hotu sei
2. ne ebé mak marka prezensa iha sira nia komunikasaun ba malu bele mos aumenta e bele realiza
3. iha tinan total populasaun hamutuk área 97 37 km vinilale mak sai sidade kapitál seuta estremadura s...

Context Size 3:

1. timór lorosa e nian fatu lulik mak sai sidade inan ba giana populasaun 200 000 abit
2. ligasaun ba li ur sensus fo fila fali tetun pdf 8 6 mb referensia munisípiu timor leste nian
3. ba li ur iktiolojia

Context Size 4:

1. ligasaun ba li ur sensus fo fila fali tetun pdf 8 6 mb seeds of life suco information sheets
2. iha timór lorosa e suku ne e iha postu administrativu watucarbau munisípiu vikeke iha tinan total po...
3. républica mit dem diploma ministerial n 199 09 portugiesisch pdf 323 kb ligasaun ba li ur wikipédia ...

Generated Text Samples (Subword-based)

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

Context Size 1:

1. _a_n_la_bozatór_
2. ay_nan_simaraica
3. icçõe_bamo_a,_tg

Context Size 2:

1. a_psainfo_hos,_wi
2. anansusi_ca_anyea
3. n_semindo_lu_stro

Context Size 3:

1. an_niança_cola_fáb
2. ha_ami_lia_sendári
3. iha_progracts_lor=

Context Size 4:

1. iha_roma_mit_democr
2. _iha_moris_iha_kata
3. aun_su_entransa._f-

Key Findings

• Best Predictability: Context-4 (word) with 97.4% predictability
• Branching Factor: Decreases with context size (more deterministic)
• Memory Trade-off: Larger contexts require more storage (92,496 contexts)
• Recommendation: Context-3 or Context-4 for text generation

---

4. Vocabulary Analysis

Statistics

Metric	Value
Vocabulary Size	12,756
Total Tokens	256,639
Mean Frequency	20.12
Median Frequency	4
Frequency Std Dev	164.32

Most Common Words

Rank	Word	Frequency
1	iha	9,917
2	ne	5,971
3	no	5,164
4	ba	4,578
5	sira	4,433
6	e	4,309
7	nian	4,134
8	nia	3,341
9	ho	2,906
10	ida	2,823

Least Common Words (from vocabulary)

Rank	Word	Frequency
1	injeta	2
2	injesaun	2
3	stiko	2
4	rezervatóriu	2
5	konfirmadu	2
6	profilaxe	2
7	中华人民共和国国家卫生健康委员会	2
8	uttar	2
9	pradesh	2
10	pántanu	2

Zipf's Law Analysis

Metric	Value
Zipf Coefficient	1.1160
R² (Goodness of Fit)	0.992469
Adherence Quality	excellent

Coverage Analysis

Top N Words	Coverage
Top 100	46.0%
Top 1,000	75.4%
Top 5,000	91.9%
Top 10,000	97.9%

Key Findings

• Zipf Compliance: R²=0.9925 indicates excellent adherence to Zipf's law
• High Frequency Dominance: Top 100 words cover 46.0% of corpus
• Long Tail: 2,756 words needed for remaining 2.1% 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	0.2388	0.4660	N/A	N/A
mono_64d	64	0.0465	0.4453	N/A	N/A
mono_128d	128	0.0060	0.4698	N/A	N/A
aligned_32d	32	0.2388 🏆	0.4494	0.0280	0.1680
aligned_64d	64	0.0465	0.4460	0.0280	0.1920
aligned_128d	128	0.0060	0.4501	0.0340	0.2000

Key Findings

• Best Isotropy: aligned_32d with 0.2388 (more uniform distribution)
• Semantic Density: Average pairwise similarity of 0.4544. Lower values indicate better semantic separation.
• Alignment Quality: Aligned models achieve up to 3.4% 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	1.010	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	agosto, asosia, acumau
-s	sigla, sleep, simples
-m	manulai, metan, markadór
-k	knananuk, konvite, krioulu
-ma	manulai, markadór, mamuk
-b	berliu, bazeada, belém
-p	polimentadu, penalidade, pandang
-l	leburema, livru, lollipop

Productive Suffixes

Suffix	Examples
-a	bazeada, ispánia, leburema
-u	polimentadu, berliu, impulsu
-n	metan, gestaun, union
-e	penalidade, opole, konvite
-s	simples, sukumatias, prepirenéus
-un	gestaun, turkomenistaun, kirgizistaun
-o	agosto, bailoro, pelo
-ia	ispánia, sekundária, podlakia

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
asau	1.75x	24 contexts	sasau, asaun, rasaun
ente	1.68x	26 contexts	enter, sente, gente
ment	1.65x	22 contexts	mental, mentál, aumentu
aran	1.59x	23 contexts	naran, laran, maran
entu	1.78x	15 contexts	eventu, bentuk, century
isau	1.66x	15 contexts	bisau, misaun, lisaun
orma	1.50x	16 contexts	forma, norma, formas
idad	1.68x	10 contexts	idade, cidade, sidade
nist	1.47x	10 contexts	ministro, amnistia, ministry
ensi	1.40x	11 contexts	ensinu, ensino, ensina
stra	1.36x	11 contexts	stray, strange, estraga
istr	1.38x	10 contexts	distritu, ministro, ministry

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
-a	-a	102 words	alexandria, américa
-k	-a	96 words	kassa, kompana
-p	-a	96 words	póvoa, portuguesa
-k	-u	87 words	kompañeiru, kriadu
-m	-a	83 words	manega, medisina
-s	-a	75 words	sosa, sida
-k	-n	69 words	kukun, kedan
-a	-u	67 words	asesu, adversáriu
-s	-o	64 words	sukucarlito, são
-p	-n	59 words	pokémon, prizaun

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
listening	listen-i-ng	7.5	i
konstituinte	konstitui-n-te	7.5	n
bahalarauain	bahalarau-a-in	7.5	a
haturalan	hatur-al-an	7.5	al
tradusaun	tradus-a-un	7.5	a
maubaralissa	maubaralis-s-a	7.5	s
administrasaun	administra-sa-un	7.5	sa
honorável	honoráv-e-l	7.5	e
deskrisaun	deskris-a-un	7.5	a
sobrevivente	sobrevive-n-te	7.5	n
computing	comput-i-ng	7.5	i
calataiud	calatai-u-d	7.5	u
dokumentasuan	dokumentas-u-an	7.5	u
evolusaun	evolus-a-un	7.5	a
prehistory	p-re-history	6.0	history

6.6 Linguistic Interpretation

Automated Insight: The language Tetum 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	32k BPE	Best compression (4.08x)
N-gram	2-gram	Lowest perplexity (284)
Markov	Context-4	Highest predictability (97.4%)
Embeddings	100d	Balanced semantic capture and isotropy

---

Appendix: Metrics Glossary & Interpretation Guide

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

Tokenizer Metrics

Compression Ratio

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

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

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

Average Token Length (Fertility)

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

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

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

Unknown Token Rate (OOV Rate)

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

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

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

N-gram Model Metrics

Perplexity

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

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

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

Entropy

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

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

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

Coverage (Top-K)

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

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

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

Markov Chain Metrics

Average Entropy

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

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

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

Branching Factor

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

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

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

Predictability

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

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

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

Vocabulary & Zipf's Law Metrics

Zipf's Coefficient

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

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

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

R² (Coefficient of Determination)

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

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

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

Vocabulary Coverage

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

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

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

Word Embedding Metrics

Isotropy

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

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

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

Average Norm

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

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

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

Cosine Similarity

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

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

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

t-SNE Visualization

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

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

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

General Interpretation Guidelines

1. Compare within model families: Metrics are most meaningful when comparing models of the same type (e.g., 8k vs 64k tokenizer).
2. Consider trade-offs: Better performance on one metric often comes at the cost of another (e.g., compression vs. OOV rate).
3. Context matters: Optimal values depend on downstream tasks. Text generation may prioritize different metrics than classification.
4. Corpus influence: All metrics are influenced by corpus characteristics. Wikipedia text differs from social media or literature.
5. Language-specific patterns: Morphologically rich languages (like Arabic) may show different optimal ranges than analytic languages.

Visualizations Index

Visualization	Description
Tokenizer Compression	Compression ratios by vocabulary size
Tokenizer Fertility	Average token length by vocabulary
Tokenizer OOV	Unknown token rates
Tokenizer Total Tokens	Total tokens by vocabulary
N-gram Perplexity	Perplexity by n-gram size
N-gram Entropy	Entropy by n-gram size
N-gram Coverage	Top pattern coverage
N-gram Unique	Unique n-gram counts
Markov Entropy	Entropy by context size
Markov Branching	Branching factor by context
Markov Contexts	Unique context counts
Zipf's Law	Frequency-rank distribution with fit
Vocab Frequency	Word frequency distribution
Top 20 Words	Most frequent words
Vocab Coverage	Cumulative coverage curve
Embedding Isotropy	Vector space uniformity
Embedding Norms	Vector magnitude distribution
Embedding Similarity	Word similarity heatmap
Nearest Neighbors	Similar words for key terms
t-SNE Words	2D word embedding visualization
t-SNE Sentences	2D sentence embedding visualization
Position Encoding	Encoding method comparison
Model Sizes	Storage requirements
Performance Dashboard	Comprehensive performance overview

---

About This Project

Data Source

Models trained on wikipedia-monthly - 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.

Links

• 🌐 Website: wikilangs.org
• 🤗 Models: huggingface.co/wikilangs
• 📊 Data: wikipedia-monthly
• 👤 Author: Omar Kamali
• 🤝 Sponsor: Featherless AI

---

Generated by Wikilangs Models Pipeline

Report Date: 2026-01-11 00:39:26