RESEARCH_REPORT.md

Arabic — Full Ablation Study & Research Report

Detailed evaluation of all model variants trained on Arabic Wikipedia data by Wikilangs.

👈 Back to README

📋 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	3.251x	3.25	0.0702%	5,509,050
16k	3.654x	3.65	0.0788%	4,901,830
32k	4.033x	4.03	0.0870%	4,440,712
64k	4.347x 🏆	4.35	0.0938%	4,120,770

Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

Sample 1: استوديوهات أفلام والت ديزني أفلام والت ديزني منتجع والت ديزني العالمي ديزني لاند...

Vocab	Tokens	Count
8k	▁است ودي وه ات ▁أفلام ▁والت ▁دي ز ني ▁أفلام ... (+22 more)	32
16k	▁است ودي وهات ▁أفلام ▁والت ▁ديزني ▁أفلام ▁والت ▁ديزني ▁منت ... (+10 more)	20
32k	▁استوديوهات ▁أفلام ▁والت ▁ديزني ▁أفلام ▁والت ▁ديزني ▁منتجع ▁والت ▁ديزني ... (+7 more)	17
64k	▁استوديوهات ▁أفلام ▁والت ▁ديزني ▁أفلام ▁والت ▁ديزني ▁منتجع ▁والت ▁ديزني ... (+7 more)	17

Sample 2: باسكال قد تعني: الباسكال، وحدة قياس الضغط لغة باسكال، لغة برمجة الفيلسوف باسكال،...

Vocab	Tokens	Count
8k	▁با سك ال ▁قد ▁تعني : ▁البا سك ال ، ... (+29 more)	39
16k	▁باسكال ▁قد ▁تعني : ▁الباسك ال ، ▁وحدة ▁قياس ▁الضغط ... (+18 more)	28
32k	▁باسكال ▁قد ▁تعني : ▁الباسك ال ، ▁وحدة ▁قياس ▁الضغط ... (+15 more)	25
64k	▁باسكال ▁قد ▁تعني : ▁الباسك ال ، ▁وحدة ▁قياس ▁الضغط ... (+15 more)	25

Sample 3: جمهورية الكونغو الديمقراطية، زائير سابقًا، عاصمتها كينشاسا. جمهورية الكونغو، عاص...

Vocab	Tokens	Count
8k	▁جمهورية ▁الكون غو ▁الديمقراطية ، ▁ز ائ ير ▁سابق ًا ... (+21 more)	31
16k	▁جمهورية ▁الكونغو ▁الديمقراطية ، ▁ز ائ ير ▁سابقًا ، ▁عاصمتها ... (+16 more)	26
32k	▁جمهورية ▁الكونغو ▁الديمقراطية ، ▁زائ ير ▁سابقًا ، ▁عاصمتها ▁كينشاسا ... (+12 more)	22
64k	▁جمهورية ▁الكونغو ▁الديمقراطية ، ▁زائير ▁سابقًا ، ▁عاصمتها ▁كينشاسا . ... (+10 more)	20

Key Findings

• Best Compression: 64k achieves 4.347x compression
• Lowest UNK Rate: 8k with 0.0702% 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	359,826	18.46	2,030,200	4.4%	13.3%
2-gram	Subword	426 🏆	8.73	44,225	56.3%	96.2%
3-gram	Word	775,988	19.57	2,900,317	3.0%	10.9%
3-gram	Subword	4,163	12.02	321,654	24.3%	56.3%
4-gram	Word	1,494,234	20.51	4,693,107	2.8%	10.2%
4-gram	Subword	27,277	14.74	1,666,030	13.3%	31.5%
5-gram	Word	1,059,510	20.01	3,368,028	3.6%	11.9%
5-gram	Subword	133,736	17.03	5,324,551	5.8%	18.5%

Top 5 N-grams by Size

2-grams (Word):

Rank	N-gram	Count
1	في عام	137,432
2	في القرن	92,611
3	كرة قدم	88,053
4	العديد من	65,695
5	الولايات المتحدة	63,417

3-grams (Word):

Rank	N-gram	Count
1	في القرن 20	27,502
2	في الولايات المتحدة	25,188
3	على الرغم من	25,111
4	في القرن 21	20,515
5	بما في ذلك	18,931

4-grams (Word):

Rank	N-gram	Count
1	كرة قدم مغتربون في	15,717
2	تحت سن الثامنة عشر	13,585
3	على الرغم من أن	8,756
4	في الألعاب الأولمبية الصيفية	5,980
5	عام بلغ عدد سكان	5,886

5-grams (Word):

Rank	N-gram	Count
1	تعداد عام بلغ عدد سكان	5,588
2	بحسب تعداد عام وبلغ عدد	5,569
3	تعداد عام وبلغ عدد الأسر	5,569
4	نسمة بحسب تعداد عام وبلغ	5,566
5	في الفئة العمرية ما بين	5,561

2-grams (Subword):

Rank	N-gram	Count
1	ا ل	27,516,669
2	_ ا	23,616,110
3	ة _	13,152,069
4	ن _	9,255,735
5	ي _	9,009,959

3-grams (Subword):

Rank	N-gram	Count
1	_ ا ل	22,248,047
2	ا ل م	4,149,844
3	ي ة _	4,126,642
4	_ ف ي	4,065,816
5	ف ي _	3,976,227

4-grams (Subword):

Rank	N-gram	Count
1	_ ف ي _	3,688,677
2	ة _ ا ل	3,625,657
3	_ ا ل م	3,573,633
4	ن _ ا ل	2,468,103
5	_ م ن _	2,362,149

5-grams (Subword):

Rank	N-gram	Count
1	ف ي _ ا ل	1,266,206
2	_ ف ي _ ا	1,245,053
3	ا ت _ ا ل	1,085,180
4	_ ع ل ى _	1,078,435
5	ي ة _ ا ل	1,036,752

Key Findings

• Best Perplexity: 2-gram (subword) with 426
• Entropy Trend: Decreases with larger n-grams (more predictable)
• Coverage: Top-1000 patterns cover ~18% 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	1.0468	2.066	15.08	2,190,668	0.0%
1	Subword	1.2063	2.307	11.28	11,477	0.0%
2	Word	0.3256	1.253	2.03	33,010,787	67.4%
2	Subword	0.8269	1.774	5.80	129,485	17.3%
3	Word	0.1052	1.076	1.21	67,054,969	89.5%
3	Subword	0.7049	1.630	4.15	751,177	29.5%
4	Word	0.0350 🏆	1.025	1.06	81,123,579	96.5%
4	Subword	0.6481	1.567	3.38	3,113,652	35.2%

Generated Text Samples (Word-based)

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

Context Size 1:

1. في الألعاب الآسيوية خاض جهادًا فالأقدر قتالًا شديدًا تولَّى من حيث كانوا في المناهج العلاجية في
2. من حيث منعت استخدام مصطلح من الخلايا battery of america bureau of the baskervilles العديد من
3. على التلال وهي عضو النادي مبارياته الدولية بعد فوز فرنسا بيافرا التي عززت المظهر الخارجي للمبنى

Context Size 2:

1. في عام وقد انتقل بعض أفراد فرقته إلى فرقة المسرح الكويتي مسرح الرواد في هذا المجال غونار
2. في القرن 20 يابانيون في القرن 20 ذكور في سينيما ماراثية من دلهي النحات الرئيسي والمسؤول الرئيسي
3. كرة قدم مغتربون في الولايات المتحدة وبريطانيا العظمى والهجينة على مركبة فضائية مأهولة في منطقة كوم ا...

Context Size 3:

1. في القرن 20 أمريكيون في القرن 21 هـ في القاهرة 923 هـ في القاهرة بالعربية في القرن 7
2. على الرغم من محدودية علمهم ومستواهما الثقافي إلا أنهما كانا تابعين لأمير بلدة فيدين البلغاري ميخائيل...
3. في الولايات المتحدة تصغير يسار ترجمة لاتينية عمرها خمس مائة عام لكتاب القانون في الطب لابن سينا وقال

Context Size 4:

1. كرة قدم مغتربون في فرنسا كيداه منتخب ماليزيا لكرة القدم روابط خارجية مراجع رجال ناميبيون في القرن 21...
2. تحت سن الثامنة عشر ونسبة 18 3 في الخامسة والستين من العمر وما فوق تعداد عام بلغ عدد سكان
3. على الرغم من أن الاكتشافات الأثرية لا تدعم هذه النظرية حيث أن تسمية الألوان الأساسية طبقا للتطور الت...

Generated Text Samples (Subword-based)

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

Context Size 1:

1. _التفيهالرزين_ول
2. اقاتية_إلوعتحروب
3. ل_ب_ي_الجة_الجة_

Context Size 2:

1. البية_عصرية_على_أ
2. _الزهربية._إره_مق
3. ة_التعلى_المية_ال

Context Size 3:

1. _البحر_من_أصبحت_حر
2. المصر_السنّة_-_فقد_
3. ية_في_wirtugust_ha

Context Size 4:

1. _في_جنوبيَّة_من_ناثـر
2. ة_الممثلين_على_نفسه
3. _المسيحيون_فلوريدا.

Key Findings

• Best Predictability: Context-4 (word) with 96.5% predictability
• Branching Factor: Decreases with context size (more deterministic)
• Memory Trade-off: Larger contexts require more storage (3,113,652 contexts)
• Recommendation: Context-3 or Context-4 for text generation

---

4. Vocabulary Analysis

Statistics

Metric	Value
Vocabulary Size	986,324
Total Tokens	94,902,130
Mean Frequency	96.22
Median Frequency	4
Frequency Std Dev	4980.31

Most Common Words

Rank	Word	Frequency
1	في	3,714,132
2	من	2,378,870
3	على	1,085,920
4	إلى	833,112
5	أن	489,978
6	عام	455,946
7	التي	369,985
8	عن	368,235
9	أو	366,818
10	مع	331,151

Least Common Words (from vocabulary)

Rank	Word	Frequency
1	وساريكات	2
2	نهايةالمدةفترة	2
3	valachi	2
4	فالمختصون	2
5	المتأسفين	2
6	والمنشغلين	2
7	انحسبت	2
8	غيوان	2
9	moji	2
10	إيمجوي	2

Zipf's Law Analysis

Metric	Value
Zipf Coefficient	0.9151
R² (Goodness of Fit)	0.992048
Adherence Quality	excellent

Coverage Analysis

Top N Words	Coverage
Top 100	22.0%
Top 1,000	43.4%
Top 5,000	63.5%
Top 10,000	72.1%

Key Findings

• Zipf Compliance: R²=0.9920 indicates excellent adherence to Zipf's law
• High Frequency Dominance: Top 100 words cover 22.0% of corpus
• Long Tail: 976,324 words needed for remaining 27.9% 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	0.8111	0.3617	N/A	N/A
mono_64d	64	0.7841	0.2928	N/A	N/A
mono_128d	128	0.7556	0.2345	N/A	N/A
aligned_32d	32	0.8111 🏆	0.3646	0.1340	0.4860
aligned_64d	64	0.7841	0.2939	0.2860	0.6560
aligned_128d	128	0.7556	0.2339	0.3720	0.7660

Key Findings

• Best Isotropy: aligned_32d with 0.8111 (more uniform distribution)
• Semantic Density: Average pairwise similarity of 0.2969. Lower values indicate better semantic separation.
• Alignment Quality: Aligned models achieve up to 37.2% 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	-0.353	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
-ال	الماكر, الويجرية, العرقيه
-وال	والمسكيت, والمأذون, والرسو
-و	وَصِيف, وخلافاً, وبالكيفية
-الم	الماكر, المحيطان, المتوسِّط
-بال	بالمدين, بالجماع, بالتأثر
-ب	بِالنيابة, بالمدين, بقاءة
-ل	للتمتع, لجرح, لقزم
-م	مصصم, معاملات, مناظرا

Productive Suffixes

Suffix	Examples
-ا	تعرضها, مناظرا, فقابلا
-ن	بالمدين, والمأذون, المحيطان
-ة	بِالنيابة, الويجرية, وبالكيفية
-ت	معاملات, والمسكيت, ومُؤسسات
-ي	اخصابي, الثانيةفي, بيبيمي
-ين	بالمدين, الغلامين, للهيروجين
-ات	معاملات, ومُؤسسات, الألقابسنوات
-م	تسكينهم, مصصم, لقزم

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
ستخد	2.56x	420 contexts	ستخدم, يستخد, تستخد
التع	1.70x	417 contexts	التعس, التعب, التعمد
مجمو	2.12x	120 contexts	مجموة, مجمود, مجموع
استخ	1.97x	149 contexts	استخف, استخم, استخد
تحدة	2.82x	26 contexts	متحدة, ومتحدة, لمتحدة
المق	1.38x	607 contexts	المقد, المقل, المقص
ارات	1.31x	739 contexts	كارات, تارات, دارات
لمنا	1.38x	514 contexts	ظلمنا, حلمنا, لمنار
المج	1.39x	473 contexts	المجل, المجد, المجن
امعة	2.14x	53 contexts	قامعة, دامعة, سامعة
لعال	1.76x	115 contexts	العال, لعالم, لعالي
الحا	1.34x	492 contexts	الحاد, الحاق, الحاف

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
-ال	-ة	297 words	المحميّة, العقاقيرية
-ال	-ن	179 words	القطبيتان, المحتشدين
-ال	-ي	167 words	السيجومي, الازدي
-و	-ا	138 words	والكوسا, ومجتهدًا
-ال	-ية	129 words	العقاقيرية, المُغطية
-ال	-ت	113 words	الأستكشافات, المتغيِّرات
-ال	-ين	98 words	المحتشدين, المتخاذلين
-ال	-ات	97 words	الأستكشافات, المتغيِّرات
-وال	-ة	72 words	والحرورية, والضرورية
-م	-ا	64 words	مُؤديًا, مأمونًا

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
الهوسيتية	الهوسي-ت-ية	7.5	ت
وتحويلتين	وتحويل-ت-ين	7.5	ت
القراخانيين	القراخان-ي-ين	7.5	ي
فیروزآباد	فیروزآب-ا-د	7.5	ا
الارسالية	الا-رسال-ية	6.0	رسال
والمتعلمة	وال-متعلم-ة	6.0	متعلم
والكيكونغو	و-ال-كيكونغو	6.0	كيكونغو
والسويسريين	و-ال-سويسريين	6.0	سويسريين
والنازحون	و-ال-نازحون	6.0	نازحون
القترائية	الق-ترائ-ية	6.0	ترائ
للنوميديين	لل-نوميدي-ين	6.0	نوميدي
والفاندال	و-ال-فاندال	6.0	فاندال
وبالإجراءات	و-بال-إجراءات	6.0	إجراءات
بالهليكوبتر	ب-ال-هليكوبتر	6.0	هليكوبتر
والاستقلابية	و-ال-استقلابية	6.0	استقلابية

6.6 Linguistic Interpretation

Automated Insight: The language Arabic 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 (4.35x)
N-gram	2-gram	Lowest perplexity (426)
Markov	Context-4	Highest predictability (96.5%)
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

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

---

👈 Back to README

Generated by Wikilangs Pipeline · 2026-03-04 14:57:25