Novial - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on Novial 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	3.864x	3.87	0.0651%	156,765
16k	4.098x	4.11	0.0690%	147,789
32k	4.293x 🏆	4.30	0.0723%	141,092

Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

Sample 1: Li Isles Malukus (Moluccas) Es un archipelag in li orientale parte de Indonesia....

Vocab	Tokens	Count
8k	▁li ▁isles ▁mal uk us ▁( mol uc cas ) ... (+15 more)	25
16k	▁li ▁isles ▁mal uk us ▁( mol uc cas ) ... (+12 more)	22
32k	▁li ▁isles ▁malukus ▁( moluccas ) ▁es ▁un ▁archipelag ▁in ... (+7 more)	17

Sample 2: Little Rock es li maxim grandi urbe de Arkansas, Unionati States de Amerika.

Vocab	Tokens	Count
8k	▁little ▁rock ▁es ▁li ▁maxim ▁grandi ▁urbe ▁de ▁ar k ... (+7 more)	17
16k	▁little ▁rock ▁es ▁li ▁maxim ▁grandi ▁urbe ▁de ▁arkansas , ... (+5 more)	15
32k	▁little ▁rock ▁es ▁li ▁maxim ▁grandi ▁urbe ▁de ▁arkansas , ... (+5 more)	15

Sample 3: Eventes Naskos - George Gamow, rusi-usani fisikisto e skribiste de populari sien...

Vocab	Tokens	Count
8k	▁eventes ▁naskos ▁- ▁george ▁gamow , ▁rusi - usani ▁fisikisto ... (+7 more)	17
16k	▁eventes ▁naskos ▁- ▁george ▁gamow , ▁rusi - usani ▁fisikisto ... (+6 more)	16
32k	▁eventes ▁naskos ▁- ▁george ▁gamow , ▁rusi - usani ▁fisikisto ... (+6 more)	16

Key Findings

• Best Compression: 32k achieves 4.293x compression
• Lowest UNK Rate: 8k with 0.0651% 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	1,178	10.20	3,283	41.9%	75.7%
2-gram	Subword	244 🏆	7.93	1,418	69.7%	99.7%
3-gram	Word	1,168	10.19	4,001	45.5%	73.4%
3-gram	Subword	1,771	10.79	10,056	28.4%	76.4%
4-gram	Word	2,024	10.98	7,221	38.8%	61.9%
4-gram	Subword	7,394	12.85	40,232	15.6%	48.7%
5-gram	Word	1,593	10.64	5,612	40.6%	64.9%
5-gram	Subword	16,261	13.99	72,530	11.2%	38.1%

Top 5 N-grams by Size

2-grams (Word):

Rank	N-gram	Count
1	in li	946
2	es li	745
3	ek li	622
4	de sud	594
5	sud afrika	563

3-grams (Word):

Rank	N-gram	Count
1	de sud afrika	551
2	kristiani demokrati partise	505
3	un ek li	313
4	es un ek	300
5	provinse de sud	289

4-grams (Word):

Rank	N-gram	Count
1	es un ek li	296
2	provinse de sud afrika	289
3	es ek li nombro	278
4	demarcation board stats sa	278
5	li majoritate de lun	278

5-grams (Word):

Rank	N-gram	Count
1	statistikes es ek li nombro	278
2	stats sa census page independent	278
3	independent electoral commission election results	278
4	page independent electoral commission election	278
5	census page independent electoral commission	278

2-grams (Subword):

Rank	N-gram	Count
1	e _	34,143
2	i _	21,509
3	e s	19,756
4	_ d	17,368
5	d e	16,656

3-grams (Subword):

Rank	N-gram	Count
1	_ d e	12,988
2	_ l i	10,620
3	e s _	10,569
4	l i _	9,707
5	d e _	8,712

4-grams (Subword):

Rank	N-gram	Count
1	_ l i _	8,332
2	_ d e _	7,798
3	e _ d e	4,960
4	t i o n	4,248
5	_ e s _	4,034

5-grams (Subword):

Rank	N-gram	Count
1	e _ d e _	3,513
2	a t i o n	2,270
3	t i o n e	1,911
4	_ d e l _	1,822
5	_ p a r t	1,691

Key Findings

• Best Perplexity: 2-gram (subword) with 244
• Entropy Trend: Decreases with larger n-grams (more predictable)
• Coverage: Top-1000 patterns cover ~38% 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	0.7087	1.634	3.73	23,001	29.1%
1	Subword	0.9408	1.920	6.71	573	5.9%
2	Word	0.2002	1.149	1.39	85,181	80.0%
2	Subword	0.9000	1.866	5.14	3,839	10.0%
3	Word	0.0646	1.046	1.10	117,051	93.5%
3	Subword	0.8231	1.769	3.63	19,730	17.7%
4	Word	0.0276 🏆	1.019	1.05	128,096	97.2%
4	Subword	0.5739	1.488	2.29	71,462	42.6%

Generated Text Samples (Word-based)

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

Context Size 1:

1. li ekonomia de vietnam binh vietnam kum y z z li traktate de sinema kino bioskop
2. de kwazulu natal provinse de basal supositione provisorim akseptat kel bli jeta plu tardim plu natur...
3. es li nombro demografie li rego de bbc news last kingdom de lingues sexu etnikiso politike

Context Size 2:

1. in li sud afrikal general elektione total votes 4 803 31 de total populatione partisevotes inkatha l...
2. es li chef urbe es durban li majoritate de lun 193 766 homes parla zulum nombro geografia
3. ek li komunies de karu distrikte de nord amerika li nederlandani antilles konsista ek tri asertiones...

Context Size 3:

1. kristiani demokrati partise unionati demokrati movemente 1 demokrati alianse pac libereso fronte afr...
2. un ek li komunies de metsweding distrikte de gauteng provinse de sud afrika li majoritate de lun 92
3. de sud afrika fro 14 de june in unionati regia es lande de sud amerika de chile

Context Size 4:

1. es un ek li distriktes de kwazulu natal provinse de sud afrika li majoritate de lun 32 279 homes
2. provinse de sud afrika li chef urbe es li urbe de saitama referos de japan
3. sa census page independent electoral commission election results de sud afrika

Generated Text Samples (Subword-based)

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

Context Size 1:

1. _(ri_nkis_1"_]_7
2. ese_kan_u_le_enu
3. iari_fomesmbete_

Context Size 2:

1. e_sop_79,09li_nom
2. i_go_prolemokre_o
3. es_etteoli_pronal

Context Size 3:

1. _de_esentarabatal_
2. _li_yares_de_es_ek
3. es_un_spani_isaje_

Context Size 4:

1. _li_nur:_ove_you._l
2. _de_plu_tardim_tenn
3. e_de_sami_demokrati

Key Findings

• Best Predictability: Context-4 (word) with 97.2% predictability
• Branching Factor: Decreases with context size (more deterministic)
• Memory Trade-off: Larger contexts require more storage (71,462 contexts)
• Recommendation: Context-3 or Context-4 for text generation

---

4. Vocabulary Analysis

Statistics

Metric	Value
Vocabulary Size	9,838
Total Tokens	152,199
Mean Frequency	15.47
Median Frequency	3
Frequency Std Dev	140.97

Most Common Words

Rank	Word	Frequency
1	li	8,569
2	de	7,823
3	es	4,132
4	e	3,243
5	in	2,500
6	del	1,826
7	partise	1,098
8	sud	1,043
9	demokrati	1,042
10	en	921

Least Common Words (from vocabulary)

Rank	Word	Frequency
1	markant	2
2	hosta	2
3	pompidou	2
4	jános	2
5	monet	2
6	impresionisme	2
7	orkestres	2
8	brahms	2
9	operas	2
10	match	2

Zipf's Law Analysis

Metric	Value
Zipf Coefficient	1.0314
R² (Goodness of Fit)	0.989703
Adherence Quality	excellent

Coverage Analysis

Top N Words	Coverage
Top 100	45.2%
Top 1,000	75.4%
Top 5,000	92.8%
Top 10,000	0.0%

Key Findings

• Zipf Compliance: R²=0.9897 indicates excellent adherence to Zipf's law
• High Frequency Dominance: Top 100 words cover 45.2% of corpus
• Long Tail: -162 words needed for remaining 100.0% 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	0.1555 🏆	0.4672	N/A	N/A
mono_64d	64	0.0258	0.4629	N/A	N/A
mono_128d	128	0.0036	0.4758	N/A	N/A
aligned_32d	32	0.1555	0.4481	0.0160	0.1600
aligned_64d	64	0.0258	0.4683	0.0300	0.1780
aligned_128d	128	0.0036	0.4627	0.0280	0.1860

Key Findings

• Best Isotropy: mono_32d with 0.1555 (more uniform distribution)
• Semantic Density: Average pairwise similarity of 0.4642. Lower values indicate better semantic separation.
• Alignment Quality: Aligned models achieve up to 3.0% 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	0.702	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
-s	sidney, states, strukture
-a	arrhenius, alpes, autonoma
-m	minutes, multes, morocco
-p	paleontologia, prendit, plural
-k	kolpa, kampionate, kolpes
-b	bulbizarre, biofisike, bloemfontein
-d	damajes, delon, dôme
-t	tipes, tekte, taiwan

Productive Suffixes

Suffix	Examples
-e	bulbizarre, biofisike, kampionate
-s	racontas, damajes, arrhenius
-es	damajes, alpes, minutes
-a	paleontologia, kolpa, resista
-i	ri, landunionati, religiosi
-ne	anione, natione, opinione
-o	cargo, morocco, romejko
-n	roman, omnen, an

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
tion	1.51x	31 contexts	nation, lation, action
arti	1.59x	22 contexts	partie, martin, partim
lekt	1.56x	20 contexts	lekte, elekte, elekta
atio	1.48x	22 contexts	nation, lation, natione
ekti	1.75x	13 contexts	korekti, direkti, efektivi
ktio	1.74x	12 contexts	aktione, fiktione, funktione
ling	1.67x	12 contexts	lingo, lingua, lingue
ente	1.33x	23 contexts	enter, mente, vente
onte	1.58x	13 contexts	monte, fonte, ponte
nter	1.35x	17 contexts	inter, enter, konter
iona	1.49x	12 contexts	fiona, optional, rational
ntes	1.38x	14 contexts	entes, fontes, dentes

6.4 Affix Compatibility (Co-occurrence)

This table shows which prefixes and suffixes most frequently co-occur on the same stems, revealing the 'stacking' rules of the language's morphology.

Prefix	Suffix	Frequency	Examples
-s	-e	114 words	strukture, suksese
-p	-e	105 words	politike, pasaje
-k	-e	97 words	kampionate, kable
-a	-e	81 words	anione, amerikaante
-m	-e	79 words	mute, mamifere
-d	-e	78 words	dôme, desisione
-p	-s	76 words	paketes, probos
-s	-s	70 words	states, studies
-p	-es	59 words	paketes, partisevotes
-p	-a	57 words	paleontologia, poza

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
praktisad	prakti-s-ad	7.5	s
kompletisat	kompleti-s-at	7.5	s
transfera	transf-e-ra	7.5	e
diferensa	diferen-s-a	7.5	s
medievali	mediev-al-i	7.5	al
religiosi	religio-s-i	7.5	s
skriptero	skript-e-ro	7.5	e
interretal	interre-t-al	7.5	t
development	develop-me-nt	7.5	me
skripteti	skript-e-ti	7.5	e
politikalim	politik-al-im	7.5	al
fisikalim	fisik-al-im	7.5	al
afrikansum	afrikan-s-um	7.5	s
kontenanti	konten-an-ti	7.5	an
periodale	period-al-e	7.5	al

6.6 Linguistic Interpretation

Automated Insight: The language Novial 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 (4.29x)
N-gram	2-gram	Lowest perplexity (244)
Markov	Context-4	Highest predictability (97.2%)
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: 2026-01-10 15:52:59