Comprehensive Research Report & Full Ablation Study
This repository contains NLP models trained and evaluated by Wikilangs, specifically on Chuvash 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)
Trade-off: Larger vocabularies improve compression but increase model size
Recommendation: 32k vocabulary provides optimal balance for production use
2. N-gram Model Evaluation
N-gram Perplexity
N-gram Unique
N-gram Coverage
Results
N-gram
Variant
Perplexity
Entropy
Unique N-grams
Top-100 Coverage
Top-1000 Coverage
2-gram
Word
9,473
13.21
71,211
26.6%
47.9%
2-gram
Subword
532 🏆
9.06
7,908
52.7%
95.2%
3-gram
Word
8,325
13.02
89,585
30.3%
52.2%
3-gram
Subword
4,929
12.27
69,351
17.2%
56.3%
4-gram
Word
14,593
13.83
169,630
26.4%
47.5%
4-gram
Subword
26,364
14.69
378,926
10.1%
32.1%
5-gram
Word
12,306
13.59
144,170
27.1%
49.1%
5-gram
Subword
81,182
16.31
1,007,721
7.9%
24.5%
Top 5 N-grams by Size
2-grams (Word):
Rank
N-gram
Count
1
шыв шыв
22,911
2
территоринчи юханшыв
14,353
3
территорипе юхать
13,579
4
юхса юханшыв
13,517
5
экологи министерстви
11,703
3-grams (Word):
Rank
N-gram
Count
1
рф экологи министерстви
11,700
2
территорин шыв геоинформаци
11,389
3
геоинформаци системин шыв
11,389
4
федераци агентстви рф
11,389
5
шыв федераци агентстви
11,389
4-grams (Word):
Rank
N-gram
Count
1
геоинформаци системин шыв шыв
11,389
2
рф территорин шыв геоинформаци
11,389
3
агентстви рф территорин шыв
11,389
4
федераци агентстви рф территорин
11,389
5
территорин шыв геоинформаци системин
11,389
5-grams (Word):
Rank
N-gram
Count
1
агентстви рф территорин шыв геоинформаци
11,389
2
федераци агентстви рф территорин шыв
11,389
3
шыв геоинформаци системин шыв шыв
11,389
4
территорин шыв геоинформаци системин шыв
11,389
5
шыв федераци агентстви рф территорин
11,389
2-grams (Subword):
Rank
N-gram
Count
1
. _
465,426
2
а _
402,164
3
и _
363,006
4
— _
346,175
5
_ —
343,660
3-grams (Subword):
Rank
N-gram
Count
1
_ — _
342,728
2
ш ы в
149,577
3
ы в _
121,922
4
_ ю х
94,718
5
т е р
86,508
4-grams (Subword):
Rank
N-gram
Count
1
ш ы в _
121,828
2
_ ш ы в
85,484
3
_ ю х а
76,914
4
ю х а н
63,379
5
х а н ш
63,281
5-grams (Subword):
Rank
N-gram
Count
1
_ ш ы в _
83,923
2
ю х а н ш
63,268
3
х а н ш ы
63,265
4
а н ш ы в
63,263
5
_ ю х а н
62,475
Key Findings
Best Perplexity: 2-gram (subword) with 532
Entropy Trend: Decreases with larger n-grams (more predictable)
Coverage: Top-1000 patterns cover ~25% of corpus
Recommendation: 4-gram or 5-gram for best predictive performance
3. Markov Chain Evaluation
Markov Entropy
Markov Contexts
Markov Branching
Results
Context
Variant
Avg Entropy
Perplexity
Branching Factor
Unique Contexts
Predictability
1
Word
0.7800
1.717
5.34
352,836
22.0%
1
Subword
0.6157
1.532
6.03
3,635
38.4%
2
Word
0.1829
1.135
1.40
1,869,675
81.7%
2
Subword
0.9040
1.871
6.19
21,903
9.6%
3
Word
0.0525
1.037
1.09
2,591,084
94.7%
3
Subword
0.8721
1.830
4.70
135,543
12.8%
4
Word
0.0223 🏆
1.016
1.04
2,792,400
97.8%
4
Subword
0.7095
1.635
3.14
636,890
29.1%
Generated Text Samples (Word-based)
Below are text samples generated from each word-based Markov chain model:
Context Size 1:
шыв гидрологи бассейн шыв шыв геоинформаци системин шыв федераци агентстви рф территорин шыв геоинфо...
юханшыв двина печора шыв федераци агентстви рф экологи министерстви республикин ао коми республики т...
в цене чем предпочитают вспоминать и дефекты зрения м советская энциклопедия в унисон с любашей леро...
Context Size 2:
шыв шыв тури обь иртыш шыв федераци агентстви рф территорин шыв геоинформаци системин шыв шыв тури о...
территоринчи юханшыв рейн вестфали территорипе юхать юханшыв негус ях сулахай 13 км шыв шыв тури бас...
территорипе юхать юханшыв мăн салым сулахай 220 км юхса юханшыв 12 км шыв шыв гидрологи бассейн том
Context Size 3:
федераци агентстви рф территорин шыв геоинформаци системин шыв шыв гидрологи гт бассейн том гт 15 гт...
шыв федераци агентстви рф территорин шыв геоинформаци системин шыв шыв гидрологи бассейн том 15 3 рф...
шыв геоинформаци системин шыв шыв гидрологи гт бассейн том гт 11 гт 1 рф экологи министерстви респуб...
Context Size 4:
шыв геоинформаци системин шыв шыв гидрологи гт бассейн том гт 03 гт 0 рф экологи министерстви ао рес...
территорин шыв геоинформаци системин шыв шыв гидрологи бассейн том 15 3 рф экологи министерстви авто...
геоинформаци системин шыв шыв гидрологи гт бассейн том гт 03 гт 0 рф экологи министерстви ао республ...
Generated Text Samples (Subword-based)
Below are text samples generated from each subword-based Markov chain model:
Context Size 1:
_—_фикулигинци_в
а,_;_улслаки_пид
и_каспалименияни
Context Size 2:
._—_торф_тыслана_
а_медилостви_тута
и_йышши_баллина_з
Context Size 3:
_—_теминисем_астар
шыв_—_мар_монтовол
ыв_шыв._команицы:_
Context Size 4:
шыв_шыв_—_венгрла._
_шыв_федераци_агент
_юханшыв_шыв_геоинф
Key Findings
Best Predictability: Context-4 (word) with 97.8% predictability
Branching Factor: Decreases with context size (more deterministic)
Memory Trade-off: Larger contexts require more storage (636,890 contexts)
Recommendation: Context-3 or Context-4 for text generation
4. Vocabulary Analysis
Zipf's Law
Top Words
Coverage Curve
Statistics
Metric
Value
Vocabulary Size
149,054
Total Tokens
3,895,916
Mean Frequency
26.14
Median Frequency
4
Frequency Std Dev
439.39
Most Common Words
Rank
Word
Frequency
1
шыв
84,160
2
юханшыв
53,731
3
в
45,242
4
и
41,204
5
с
37,543
6
тата
34,625
7
бассейн
28,455
8
км
25,026
9
м
24,932
10
рф
24,450
Least Common Words (from vocabulary)
Rank
Word
Frequency
1
дустлик
2
2
галляарал
2
3
зарбдар
2
4
джизакской
2
5
сардоба
2
6
баяут
2
7
хаваст
2
8
сырдарьинской
2
9
пайт
2
10
клинов
2
Zipf's Law Analysis
Metric
Value
Zipf Coefficient
1.0393
R² (Goodness of Fit)
0.997747
Adherence Quality
excellent
Coverage Analysis
Top N Words
Coverage
Top 100
30.0%
Top 1,000
56.1%
Top 5,000
72.5%
Top 10,000
79.0%
Key Findings
Zipf Compliance: R²=0.9977 indicates excellent adherence to Zipf's law
High Frequency Dominance: Top 100 words cover 30.0% of corpus
Long Tail: 139,054 words needed for remaining 21.0% coverage
5. Word Embeddings Evaluation
Embedding Isotropy
Similarity Matrix
t-SNE Words
t-SNE Sentences
5.1 Cross-Lingual Alignment
Alignment Quality
Multilingual t-SNE
5.2 Model Comparison
Model
Dimension
Isotropy
Semantic Density
Alignment R@1
Alignment R@10
mono_32d
32
0.8326 🏆
0.3463
N/A
N/A
mono_64d
64
0.8301
0.2835
N/A
N/A
mono_128d
128
0.7992
0.2278
N/A
N/A
aligned_32d
32
0.8326
0.3575
0.0120
0.1340
aligned_64d
64
0.8301
0.2722
0.0400
0.2360
aligned_128d
128
0.7992
0.2219
0.0680
0.3000
Key Findings
Best Isotropy: mono_32d with 0.8326 (more uniform distribution)
Semantic Density: Average pairwise similarity of 0.2849. Lower values indicate better semantic separation.
Alignment Quality: Aligned models achieve up to 6.8% 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
1.001
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
Productive Suffixes
Suffix
Examples
-а
курска, никсона, подвига
-ен
америкасен, слышен, судьясен
-не
взводне, очерксене, болгарине
-ов
резюков, коршунов, щенков
-ем
сикекенсем, символсем, перуанецсем
-ий
выступлений, парфентий, праславянский
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
олог
2.08x
173 contexts
геолог, пологи, эколог
сейн
2.92x
24 contexts
сейнер, хусейн, хасейн
ссей
2.92x
17 contexts
ессей, эссей, рассей
огра
1.78x
95 contexts
богра, ограды, ограда
рито
2.46x
26 contexts
ритон, крито, приток
ншыв
2.79x
17 contexts
юшаншыв, юханшыв, юханшыве
ерри
2.45x
22 contexts
черри, ферри, дерри
орин
1.72x
74 contexts
дорин, шорин, борин
аншы
2.79x
13 contexts
юшаншыв, юханшыв, юханшыве
исте
1.81x
57 contexts
листе, хистет, истерн
блик
2.25x
17 contexts
облик, облика, коблик
нист
1.86x
30 contexts
финист, пианист, капнист
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.
No significant affix co-occurrences detected.
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
айсбергов
айсберг-ов
4.5
айсберг
фахрутдинов
фахрутдин-ов
4.5
фахрутдин
экономикине
экономики-не
4.5
экономики
пурнӑҫланине
пурнӑҫлани-не
4.5
пурнӑҫлани
ансамбльне
ансамбль-не
4.5
ансамбль
хрустальне
хрусталь-не
4.5
хрусталь
анатомине
анатоми-не
4.5
анатоми
инженеров
инженер-ов
4.5
инженер
багдасаров
багдасар-ов
4.5
багдасар
фотографий
фотограф-ий
4.5
фотограф
ассамблейине
ассамблейи-не
4.5
ассамблейи
символикине
символики-не
4.5
символики
бриллиантов
бриллиант-ов
4.5
бриллиант
кинокритиков
кинокритик-ов
4.5
кинокритик
наводнений
наводн-ен-ий
3.0
наводн
6.6 Linguistic Interpretation
Automated Insight:
The language Chuvash 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
Performance Dashboard
Production Recommendations
Component
Recommended
Rationale
Tokenizer
64k BPE
Best compression (3.78x)
N-gram
2-gram
Lowest perplexity (532)
Markov
Context-4
Highest predictability (97.8%)
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.
