Scots - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on Scots 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	3.617x	3.62	0.0092%	577,294
16k	3.956x	3.96	0.0100%	527,731
32k	4.216x	4.22	0.0107%	495,233
64k	4.412x 🏆	4.41	0.0112%	473,222

Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

Sample 1: La Cruz is a smaw ceety in the Mexican state o Sinaloa. The ceety reportit 15,65...

Vocab	Tokens	Count
8k	▁la ▁cruz ▁is ▁a ▁smaw ▁ceety ▁in ▁the ▁mexican ▁state ... (+26 more)	36
16k	▁la ▁cruz ▁is ▁a ▁smaw ▁ceety ▁in ▁the ▁mexican ▁state ... (+22 more)	32
32k	▁la ▁cruz ▁is ▁a ▁smaw ▁ceety ▁in ▁the ▁mexican ▁state ... (+22 more)	32
64k	▁la ▁cruz ▁is ▁a ▁smaw ▁ceety ▁in ▁the ▁mexican ▁state ... (+22 more)	32

Sample 2: Navalafuente is a municipality o the Commonty o Madrid, Spain. Freemit airtins i...

Vocab	Tokens	Count
8k	▁naval af u ente ▁is ▁a ▁municipality ▁o ▁the ▁commonty ... (+18 more)	28
16k	▁naval af u ente ▁is ▁a ▁municipality ▁o ▁the ▁commonty ... (+18 more)	28
32k	▁naval af u ente ▁is ▁a ▁municipality ▁o ▁the ▁commonty ... (+18 more)	28
64k	▁naval afu ente ▁is ▁a ▁municipality ▁o ▁the ▁commonty ▁o ... (+17 more)	27

Sample 3: Magnetite is a rock mineral an ane o the main airn ures. References minerals gro...

Vocab	Tokens	Count
8k	▁magn et ite ▁is ▁a ▁rock ▁mineral ▁an ▁ane ▁o ... (+24 more)	34
16k	▁magnet ite ▁is ▁a ▁rock ▁mineral ▁an ▁ane ▁o ▁the ... (+20 more)	30
32k	▁magnet ite ▁is ▁a ▁rock ▁mineral ▁an ▁ane ▁o ▁the ... (+18 more)	28
64k	▁magnetite ▁is ▁a ▁rock ▁mineral ▁an ▁ane ▁o ▁the ▁main ... (+14 more)	24

Key Findings

• Best Compression: 64k achieves 4.412x compression
• Lowest UNK Rate: 8k with 0.0092% 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	26,453	14.69	140,557	16.0%	32.2%
2-gram	Subword	271 🏆	8.08	7,416	67.7%	99.0%
3-gram	Word	72,001	16.14	210,013	7.3%	19.9%
3-gram	Subword	2,416	11.24	51,687	25.6%	69.9%
4-gram	Word	131,079	17.00	309,274	5.1%	14.5%
4-gram	Subword	14,275	13.80	273,093	12.8%	37.3%
5-gram	Word	95,213	16.54	199,412	4.7%	15.0%
5-gram	Subword	54,670	15.74	795,931	8.2%	24.3%

Top 5 N-grams by Size

2-grams (Word):

Rank	N-gram	Count
1	o the	83,237
2	in the	58,596
3	is a	24,631
4	tae the	17,805
5	an the	13,525

3-grams (Word):

Rank	N-gram	Count
1	ane o the	5,732
2	references freemit airtins	4,456
3	the unitit states	4,149
4	pairt o the	4,120
5	the province o	3,589

4-grams (Word):

Rank	N-gram	Count
1	in the province o	2,669
2	o the order o	2,501
3	is ane o the	2,083
4	is a toun an	1,707
5	o the unitit states	1,656

5-grams (Word):

Rank	N-gram	Count
1	is a toun an municipality	1,214
2	o the order o the	1,192
3	a toun an municipality in	966
4	as o the municipality haed	846
5	o the municipality haed a	784

2-grams (Subword):

Rank	N-gram	Count
1	e _	1,050,184
2	n _	810,931
3	s _	775,649
4	_ t	732,959
5	_ a	719,183

3-grams (Subword):

Rank	N-gram	Count
1	_ t h	504,310
2	t h e	474,947
3	h e _	449,929
4	i n _	295,599
5	_ o _	271,843

4-grams (Subword):

Rank	N-gram	Count
1	_ t h e	434,137
2	t h e _	428,262
3	_ i n _	189,422
4	_ a n _	173,723
5	n _ t h	114,460

5-grams (Subword):

Rank	N-gram	Count
1	_ t h e _	418,560
2	n _ t h e	105,154
3	_ o _ t h	87,165
4	o _ t h e	85,549
5	i n _ t h	75,907

Key Findings

• Best Perplexity: 2-gram (subword) with 271
• Entropy Trend: Decreases with larger n-grams (more predictable)
• Coverage: Top-1000 patterns cover ~24% 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	0.9277	1.902	8.10	272,309	7.2%
1	Subword	1.0662	2.094	6.39	4,231	0.0%
2	Word	0.3124	1.242	1.88	2,201,132	68.8%
2	Subword	0.7253	1.653	4.46	27,028	27.5%
3	Word	0.1197	1.086	1.24	4,131,130	88.0%
3	Subword	0.7329	1.662	3.98	120,570	26.7%
4	Word	0.0487 🏆	1.034	1.08	5,105,427	95.1%
4	Subword	0.6942	1.618	3.19	479,292	30.6%

Generated Text Samples (Word-based)

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

Context Size 1:

1. the order of seduction dos veadeirosalto paraíso borbotón la revolucion in the distance rinners male...
2. o san juan mixtepec mixteca region in bages on the horizontal cross o the various schuils
3. in coonty yintian toun the aurie which led mission in australie seestem in its headquarters head

Context Size 2:

1. o the ceety o madrid an the van province is subdividit intae cantons municipality inhabitants seat l...
2. in the places mentionit in the savinja statistical region name the divide atween the an gan yavne
3. is a roushie mid size hatchback caur frae components made frae its oreeginal name o an alternate

Context Size 3:

1. ane o the maist strangest player frae osaka in the throu efter the incorporation o ford saf intae
2. references freemit airtins honda warldwide steid honda press library japanese but wi graphical timel...
3. pairt o the province o cuenca cuenca spaingie congress electoral destrict the commune is still no re...

Context Size 4:

1. in the province o tarragona vilanova de sau toun in the province o enna references
2. o the order o the aztec eagle o the order o meerit o the federal republic o germany o
3. is ane o the original thirteen states the caipital o massachusetts is boston that is an aw the tradi...

Generated Text Samples (Subword-based)

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

Context Size 1:

1. _an's_sir_r_cs-g
2. ee_t_te_tenti_in
3. aprenrothsicanin

Context Size 2:

1. e_licturichypence
2. n_the_uniage_spe_
3. s_st_rompion_kerm

Context Size 3:

1. _the_samate_voyar,
2. the_umwhilocht-sou
3. he_cries_airty_o_r

Context Size 4:

1. _the_elemen_wumman_
2. the_elemen's_pols_p
3. _in_as_the_municipa

Key Findings

• Best Predictability: Context-4 (word) with 95.1% predictability
• Branching Factor: Decreases with context size (more deterministic)
• Memory Trade-off: Larger contexts require more storage (479,292 contexts)
• Recommendation: Context-3 or Context-4 for text generation

---

4. Vocabulary Analysis

Statistics

Metric	Value
Vocabulary Size	123,249
Total Tokens	6,164,921
Mean Frequency	50.02
Median Frequency	4
Frequency Std Dev	1749.35

Most Common Words

Rank	Word	Frequency
1	the	427,737
2	o	273,854
3	in	193,597
4	an	176,125
5	a	119,842
6	is	93,570
7	tae	70,765
8	wis	49,082
9	as	41,842
10	frae	34,119

Least Common Words (from vocabulary)

Rank	Word	Frequency
1	erlier	2
2	margules	2
3	lifshitz	2
4	lakeith	2
5	exploder	2
6	fipresci	2
7	zubeen	2
8	beutel	2
9	badmen	2
10	taggert	2

Zipf's Law Analysis

Metric	Value
Zipf Coefficient	1.0502
R² (Goodness of Fit)	0.993417
Adherence Quality	excellent

Coverage Analysis

Top N Words	Coverage
Top 100	39.5%
Top 1,000	63.1%
Top 5,000	80.2%
Top 10,000	86.5%

Key Findings

• Zipf Compliance: R²=0.9934 indicates excellent adherence to Zipf's law
• High Frequency Dominance: Top 100 words cover 39.5% of corpus
• Long Tail: 113,249 words needed for remaining 13.5% 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	0.8628	0.3487	N/A	N/A
mono_64d	64	0.8453	0.2622	N/A	N/A
mono_128d	128	0.8330	0.1921	N/A	N/A
aligned_32d	32	0.8628 🏆	0.3373	0.4500	0.8320
aligned_64d	64	0.8453	0.2597	0.6080	0.8960
aligned_128d	128	0.8330	0.1921	0.7060	0.9300

Key Findings

• Best Isotropy: aligned_32d with 0.8628 (more uniform distribution)
• Semantic Density: Average pairwise similarity of 0.2653. Lower values indicate better semantic separation.
• Alignment Quality: Aligned models achieve up to 70.6% 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	-0.383	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
-s	sts, sables, safar
-a	armature, abkhazians, ald
-ma	mazīnān, manar, materazzi
-b	breid, blume, birnie
-m	mazīnān, michelangelos, mcqueers
-t	tu, tsugaru, tezuka
-c	cuiverin, coontin, ceasefire
-p	phrase, padmore, polje

Productive Suffixes

Suffix	Examples
-s	sts, michelangelos, mcqueers
-n	cuiverin, mazīnān, focusin
-e	phrase, padmore, neale
-a	donnacona, tezuka, camara
-t	hjärtat, insicht, 145t
-y	validity, climatology, horthy
-d	ootsauld, breid, liquidated
-es	sables, straddles, charlottes

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
eren	2.02x	57 contexts	keren, ferenc, kerend
ment	1.63x	93 contexts	menta, ament, amenta
stri	1.63x	89 contexts	strid, strix, strip
tric	1.59x	71 contexts	trick, nitric, strict
atio	1.62x	56 contexts	patio, ratio, cation
atit	1.67x	45 contexts	datit, fatit, matit
tion	1.45x	78 contexts	cation, nation, action
estr	1.56x	56 contexts	bestry, vestry, sestra
alit	1.61x	40 contexts	alita, balita, kalita
ence	1.64x	37 contexts	fence, pence, dence
renc	1.73x	27 contexts	renca, ferenc, french
dest	1.66x	27 contexts	modest, oldest, widest

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
-c	-s	129 words	cuevas, colorless
-a	-s	95 words	awaurness, aigeiroúses
-s	-s	94 words	sanctions, skippers
-p	-s	89 words	prowess, pairtisans
-s	-n	89 words	samson, sudan
-c	-n	64 words	copulation, caryn
-s	-e	61 words	sparse, suerte
-a	-e	60 words	airsie, australie
-t	-s	55 words	termales, trumpeters
-m	-s	54 words	makarios, montañas

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
freistaat	freista-a-t	7.5	a
ovulators	ovulat-o-rs	7.5	o
cardenden	carden-d-en	7.5	d
auldgirth	auldgir-t-h	7.5	t
islamists	islami-s-ts	7.5	s
steamboats	steambo-a-ts	7.5	a
spulyiein	spulyi-e-in	7.5	e
carrascosa	carrasco-s-a	7.5	s
armizonsky	armizon-s-ky	7.5	s
wiktionary	wiktion-ar-y	7.5	ar
sundsvall	sundsv-al-l	7.5	al
eventually	eventu-al-ly	7.5	al
montesson	montes-s-on	7.5	s
lifeboats	lifebo-a-ts	7.5	a
kindersley	kinders-le-y	7.5	le

6.6 Linguistic Interpretation

Automated Insight: The language Scots 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

Production Recommendations

Component	Recommended	Rationale
Tokenizer	64k BPE	Best compression (4.41x)
N-gram	2-gram	Lowest perplexity (271)
Markov	Context-4	Highest predictability (95.1%)
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 20:17:20