Gujarati - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on Gujarati 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	3.375x	3.38	0.3164%	854,027
16k	3.770x	3.77	0.3535%	764,449
32k	4.098x	4.10	0.3842%	703,344
64k	4.347x 🏆	4.35	0.4076%	662,946

Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

Sample 1: સિકર ભારત દેશના પશ્ચિમ ભાગમાં આવેલા રાજસ્થાન રાજ્યનું એક નગર છે. સિકરમાં સિકર જિ...

Vocab	Tokens	Count
8k	▁સિક ર ▁ભારત ▁દેશના ▁પશ્ચિમ ▁ભાગમાં ▁આવેલા ▁રાજસ્થાન ▁રાજ્યનું ▁એક ... (+11 more)	21
16k	▁સિક ર ▁ભારત ▁દેશના ▁પશ્ચિમ ▁ભાગમાં ▁આવેલા ▁રાજસ્થાન ▁રાજ્યનું ▁એક ... (+11 more)	21
32k	▁સિકર ▁ભારત ▁દેશના ▁પશ્ચિમ ▁ભાગમાં ▁આવેલા ▁રાજસ્થાન ▁રાજ્યનું ▁એક ▁નગર ... (+9 more)	19
64k	▁સિકર ▁ભારત ▁દેશના ▁પશ્ચિમ ▁ભાગમાં ▁આવેલા ▁રાજસ્થાન ▁રાજ્યનું ▁એક ▁નગર ... (+9 more)	19

Sample 2: હસમુખ પટેલ ગુજરાતના અમર‌ઇવાડી લોકસભા મત વિસ્તારમાંથી લોકસભા ચૂંટણીમાં ભારતીય જનત...

Vocab	Tokens	Count
8k	▁હ સ મુખ ▁પટેલ ▁ગુજરાતના ▁અમર ▁ઇ વાડી ▁લોકસભા ▁મત ... (+14 more)	24
16k	▁હસમુખ ▁પટેલ ▁ગુજરાતના ▁અમર ▁ઇ વાડી ▁લોકસભા ▁મત ▁વિસ્તારમાંથી ▁લોકસભા ... (+11 more)	21
32k	▁હસમુખ ▁પટેલ ▁ગુજરાતના ▁અમર ▁ઇ વાડી ▁લોકસભા ▁મત ▁વિસ્તારમાંથી ▁લોકસભા ... (+11 more)	21
64k	▁હસમુખ ▁પટેલ ▁ગુજરાતના ▁અમર ▁ઇ વાડી ▁લોકસભા ▁મત ▁વિસ્તારમાંથી ▁લોકસભા ... (+11 more)	21

Sample 3: જય હિન્દ ગુજરાતી ભાષાનું એક રોજીંદુ સમાચારપત્ર છે. બાહ્ય કડીઓ જય હિન્દ વેબસાઇટ સ...

Vocab	Tokens	Count
8k	▁જય ▁હિન્ દ ▁ગુજરાતી ▁ભાષાનું ▁એક ▁રો જી ંદુ ▁સમાચાર ... (+11 more)	21
16k	▁જય ▁હિન્ દ ▁ગુજરાતી ▁ભાષાનું ▁એક ▁રોજી ંદુ ▁સમાચારપત્ર ▁છે ... (+8 more)	18
32k	▁જય ▁હિન્દ ▁ગુજરાતી ▁ભાષાનું ▁એક ▁રોજી ંદુ ▁સમાચારપત્ર ▁છે . ... (+6 more)	16
64k	▁જય ▁હિન્દ ▁ગુજરાતી ▁ભાષાનું ▁એક ▁રોજી ંદુ ▁સમાચારપત્ર ▁છે . ... (+6 more)	16

Key Findings

• Best Compression: 64k achieves 4.347x compression
• Lowest UNK Rate: 8k with 0.3164% 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	10,698	13.39	125,204	30.7%	48.0%
2-gram	Subword	2,297 🏆	11.17	62,671	37.2%	69.7%
3-gram	Word	6,094	12.57	126,108	38.3%	57.0%
3-gram	Subword	19,412	14.24	340,527	15.1%	39.0%
4-gram	Word	4,835	12.24	174,488	42.1%	62.1%
4-gram	Subword	84,467	16.37	1,351,931	11.0%	28.7%
5-gram	Word	2,358	11.20	99,029	46.1%	70.1%
5-gram	Subword	171,415	17.39	2,092,817	9.1%	24.9%

Top 5 N-grams by Size

2-grams (Word):

Rank	N-gram	Count
1	છે આ	59,392
2	તેમ જ	50,730
3	આવે છે	34,138
4	આ ગામમાં	33,314
5	ભાગમાં આવેલા	31,103

3-grams (Word):

Rank	N-gram	Count
1	છે આ ગામમાં	33,162
2	કરવામાં આવે છે	19,146
3	પશ્ચિમ ભાગમાં આવેલા	18,413
4	ભારત દેશના પશ્ચિમ	18,409
5	દેશના પશ્ચિમ ભાગમાં	18,400

4-grams (Word):

Rank	N-gram	Count
1	ભારત દેશના પશ્ચિમ ભાગમાં	18,398
2	દેશના પશ્ચિમ ભાગમાં આવેલા	18,367
3	પશ્ચિમ ભાગમાં આવેલા ગુજરાત	17,959
4	ભાગમાં આવેલા ગુજરાત રાજ્યના	17,656
5	લોકોનો મુખ્ય વ્યવસાય ખેતી	16,402

5-grams (Word):

Rank	N-gram	Count
1	ભારત દેશના પશ્ચિમ ભાગમાં આવેલા	18,365
2	દેશના પશ્ચિમ ભાગમાં આવેલા ગુજરાત	17,934
3	પશ્ચિમ ભાગમાં આવેલા ગુજરાત રાજ્યના	17,654
4	ગામના લોકોનો મુખ્ય વ્યવસાય ખેતી	16,093
5	લોકોનો મુખ્ય વ્યવસાય ખેતી ખેતમજૂરી	16,009

2-grams (Subword):

Rank	N-gram	Count
1	. _	427,913
2	, _	392,494
3	_ આ	387,226
4	માં _	379,145
5	_ અ	336,386

3-grams (Subword):

Rank	N-gram	Count
1	_ છે .	235,479
2	છે . _	225,048
3	_ અ ને	167,804
4	અ ને _	165,365
5	માં _ આ	149,526

4-grams (Subword):

Rank	N-gram	Count
1	_ છે . _	224,981
2	_ અ ને _	164,399
3	માં _ આ વે	110,493
4	. _ આ _	68,884
5	વા માં _ આ	67,045

5-grams (Subword):

Rank	N-gram	Count
1	_ છે . _ આ	62,651
2	છે . _ આ _	57,724
3	_ આ વે લા _	56,800
4	માં _ આ વે લા	53,829
5	_ તે મ _ જ	50,773

Key Findings

• Best Perplexity: 2-gram (subword) with 2,297
• 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

Results

Context	Variant	Avg Entropy	Perplexity	Branching Factor	Unique Contexts	Predictability
1	Word	0.8634	1.819	7.06	458,692	13.7%
1	Subword	0.9952	1.993	15.09	11,981	0.5%
2	Word	0.2300	1.173	1.59	3,234,951	77.0%
2	Subword	0.6872	1.610	5.08	180,749	31.3%
3	Word	0.0669	1.047	1.13	5,125,144	93.3%
3	Subword	0.5539	1.468	3.49	917,629	44.6%
4	Word	0.0220 🏆	1.015	1.04	5,806,762	97.8%
4	Subword	0.4165	1.335	2.13	3,199,363	58.3%

Generated Text Samples (Word-based)

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

Context Size 1:

1. છે તે સામેલ કરવાનો છે એઆરડીએ arda ના વેટરર્નસ ઘડાયેલા સ્તંભો વિમલ સંધિવિગ્રહિક દામોદર ઠાકરસી મહિલા
2. અને એ ઇ સ્મિથ એંથની જીડેંસ મિશેલ ચાર્લ્સ સ્ટેઇનમેટ્ઝ અને ગાયક તરીકે પસંદ નથી તેમણે સ્ટેલીના
3. આ ગામમાં પ્રાથમિક શાળા તેમ જ દૂધની ડેરી જેવી સવલતો પ્રાપ્ય થયેલી છે સંદર્ભ કર્યા છે

Context Size 2:

1. છે આ ગામ સ્થાપિત સહયોગ કુષ્ટ યજ્ઞ ટ્રસ્ટ નામની સંસ્થા પણ જીવનસંચારવાદ પણ શીખવે છે તે દરમિયાન
2. તેમ જ અન્ય શાકભાજીના પાકની ખેતી કરવામાં આવે છે આ શહેરની અર્થતંત્રએ ભારતમા દસમો ક્રમ ધરાવે છે
3. આવે છે જો કે બંગાળીઓ વિશે કેટલીક કાલ્પનિક વાર્તાઓ ડબ્લ્યુડબ્લ્યુઇમાં બિહાઇન્ડ ધ પેઇન્ટેડ સ્માઇલ માં ...

Context Size 3:

1. છે આ ગામમાં પ્રાથમિક શાળા આંગણવાડી પંચાયતઘર દૂધની ડેરી વગેરે સવલતો પ્રાપ્ય છે ગામના લોકો વ્યવસાયમાં ...
2. કરવામાં આવે છે આ તત્વનું નામ મેન્ડેલિવીયમ ને શુદ્ધ અને ઉપયોગિ રસાયણ શાસ્તની આંતરરાષ્ટ્રીય સંસ્થા દ્વ...
3. પશ્ચિમ ભાગમાં આવેલા ગુજરાત રાજ્યના મધ્ય ભાગમાં આવેલા પંચમહાલ જિલ્લામાં આવેલા કુલ ૬ છ તાલુકાઓ પૈકીના ...

Context Size 4:

1. ભારત દેશના પશ્ચિમ ભાગમાં આવેલા ગુજરાત રાજ્યમાં આવેલા સૌરાષ્ટ્ર વિસ્તારમાં આવેલા પોરબંદર જિલ્લામાં આવ...
2. દેશના પશ્ચિમ ભાગમાં આવેલા ગુજરાત રાજ્યના દક્ષિણ ભાગમાં આવેલા તાપી જિલ્લાના કુલ ૭ સાત તાલુકાઓ પૈકીના ...
3. પશ્ચિમ ભાગમાં આવેલા ગુજરાત રાજ્યના ઉત્તર પૂર્વ ભાગમાં આવેલા અરવલ્લી જિલ્લામાં આવેલા કુલ ૬ છ તાલુકાઓ ...

Generated Text Samples (Subword-based)

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

Context Size 1:

1. _મેજયેલી_રોજના_અને_અને_
2. રજકોનો_તા_દરાજ્યની_માટેની_
3. કર_hie_સ્વતંત્ર્ય_તાં_પદ્ધની

Context Size 2:

1. ._૧૯_રોહિતો_અને_ભારત_મા
2. ,_entના_પાકની_ઉત્તરાધિક_
3. _આર_સૂર્યના_મૃત્યુ_ગાંધીનગર,

Context Size 3:

1. _છે._આ_ગામમાં_મુખ્યત્વે_આદિવાસી
2. છે._તાલુકો_ખેતી_કરવાની_પણ_
3. _અનેક_આવેલા_અભ્યાસ_દ્વારા_પહોં

Context Size 4:

1. _છે._ગોડ્ડામાં_ગોડ્ડા_જિલ્લામાં_આવેલા_
2. _અને_ક્લાસિકલ_માર્ગ_પર_એમિગાઃ_
3. માં_આવે_ત્યારે_બાસાલ્ટિક_કે_દારૂનુ_

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 (3,199,363 contexts)
• Recommendation: Context-3 or Context-4 for text generation

---

4. Vocabulary Analysis

Statistics

Metric	Value
Vocabulary Size	193,639
Total Tokens	7,238,200
Mean Frequency	37.38
Median Frequency	4
Frequency Std Dev	1013.36

Most Common Words

Rank	Word	Frequency
1	છે	321,831
2	અને	165,458
3	આ	104,369
4	જ	69,831
5	એક	66,411
6	આવેલા	56,885
7	તેમ	53,009
8	કે	43,256
9	ગામમાં	39,326
10	માટે	39,310

Least Common Words (from vocabulary)

Rank	Word	Frequency
1	લંપુરને	2
2	શ્રુહાદ	2
3	ફાર્મહાઉસમાં	2
4	જંબુદ્દીવપણ્ણત્તિ	2
5	જંબુસામિચરિઉ	2
6	સુધર્મસ્વામી	2
7	જંબુને	2
8	વૈરાગ્યવિરોધી	2
9	જંબુની	2
10	સુધર્માસ્વામીએ	2

Zipf's Law Analysis

Metric	Value
Zipf Coefficient	1.0403
R² (Goodness of Fit)	0.997228
Adherence Quality	excellent

Coverage Analysis

Top N Words	Coverage
Top 100	34.5%
Top 1,000	58.3%
Top 5,000	75.1%
Top 10,000	81.6%

Key Findings

• Zipf Compliance: R²=0.9972 indicates excellent adherence to Zipf's law
• High Frequency Dominance: Top 100 words cover 34.5% of corpus
• Long Tail: 183,639 words needed for remaining 18.4% 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	0.7412	0.3587	N/A	N/A
mono_64d	64	0.7575	0.2728	N/A	N/A
mono_128d	128	0.7575 🏆	0.2077	N/A	N/A
aligned_32d	32	0.7412	0.3577	0.0380	0.1860
aligned_64d	64	0.7575	0.2690	0.0580	0.2540
aligned_128d	128	0.7575	0.2042	0.0660	0.2820

Key Findings

• Best Isotropy: mono_128d with 0.7575 (more uniform distribution)
• Semantic Density: Average pairwise similarity of 0.2783. Lower values indicate better semantic separation.
• Alignment Quality: Aligned models achieve up to 6.6% 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	3.181	High morphological productivity	Reliable analysis
Idiomaticity Gap	1.816	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
-સ	સંસ્કૃતભાષામાં, સામાન્યત, સ્થપાતી
-ક	કોબોલ્ડ, કાલક્રમાનુસાર, કાળજીપુર્વક
-મ	મિંઢા, મહાવીરને, માયથિકલ
-પ	પોપચા, પાડવા, પંચગવ્ય
-બ	બૉલની, બલાઢા, બિન્કસ
-વ	વાસ્તે, વાલીના, વિદેશોની
-ર	રેટની, રાસમસ, રેતીયાની
-અ	અવસ્થાના, અપ્રમાણિકતાનો, અતુલનીય

Productive Suffixes

Suffix	Examples
-ર	કાલક્રમાનુસાર, ઈન્કાર, છાપર
-ન	નાસારિસ્તાન, ફિઝિશ્યન, અંકન
-સ	બિન્કસ, રાસમસ, ન્યૂટ્સ
-ક	આસ્ક, કાળજીપુર્વક, ટેલિફોનિક
-લ	માયથિકલ, ખારોલ, લીવરપૂલ
-ટ	ઓરિએન્ટ, બેરોનેટ, કન્ફ્લિક્ટ
-એ	શાલ્વએ, વેબએ, ટેસ્લાએ
-s	guianensis, libraries, rhodes

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	3.61x	29 contexts	motion, nation, notion
atio	3.66x	25 contexts	ratio, nation, station
indi	3.58x	24 contexts	india, hindi, indic
વનગર	3.22x	14 contexts	ઇવનગર, ધુવનગર, ભાવનગર
નવલક	3.19x	5 contexts	નવલકથા, નવલકથાએ, નવલકથાઓ

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
-ક	-ન	26 words	કુલન, ક્રુશિયન
-સ	-સ	25 words	સ્પીકર્સ, સેટેલાઇટ્સ
-ક	-સ	22 words	કનેક્શન્સ, કાલપેર્સ
-ક	-ર	22 words	કેલનર, કૃષ્ણકુમાર
-સ	-ક	18 words	સબસોનિક, સ્ટ્રેટેજીક
-સ	-ર	17 words	સુમેર, સ્ટ્રાઈકર
-પ	-ર	17 words	પાઉડર, પીયર
-સ	-ન	17 words	સફરજન, સવેન
-મ	-ન	16 words	માર્જીન, મિત્રસેન
-પ	-ન	16 words	પરાધીન, પ્રણોદન

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
conventional	convention-al	4.5	convention
કસોટીઓમાં	ક-સ-ોટીઓમાં	4.5	ોટીઓમાં
એનડબલ્યુએ	એનડબલ્યુ-એ	4.5	એનડબલ્યુ
ઇન્સ્ટ્રુમેન્ટલ	ઇન્સ્ટ્રુમેન્ટ-લ	4.5	ઇન્સ્ટ્રુમેન્ટ
અવિશ્વાસની	અ-વિશ્વાસની	4.5	વિશ્વાસની
હેલેનીકોન	હેલેનીકો-ન	4.5	હેલેનીકો
manifestations	manifestation-s	4.5	manifestation
ટેકનોલોજીએ	ટેકનોલોજી-એ	4.5	ટેકનોલોજી
festivals	festival-s	4.5	festival
citations	citation-s	4.5	citation
લક્ષણોમાં	લ-ક્ષણોમાં	4.5	ક્ષણોમાં
પહોંચાડવામાં	પ-હ-ોંચાડવામાં	4.5	ોંચાડવામાં
બિનસહસંયોજક	બ-િનસહસંયોજ-ક	3.0	િનસહસંયોજ
રુખમાબાઈને	ર-ુખમાબાઈને	1.5	ુખમાબાઈને
ભવિષ્યકથન	ભવિષ્યકથ-ન	1.5	ભવિષ્યકથ

6.6 Linguistic Interpretation

Automated Insight: The language Gujarati 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	64k BPE	Best compression (4.35x)
N-gram	2-gram	Lowest perplexity (2,297)
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

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 00:30:07