You can deploy this model and start earning money today!

Views

No views yet

Hindi - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on Hindi 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

1. Tokenizer Evaluation

Results

Vocab Size	Compression	Avg Token Len	UNK Rate	Total Tokens
8k	3.436x	3.44	0.0604%	2,277,747
16k	3.796x	3.80	0.0667%	2,061,793
32k	4.066x	4.07	0.0715%	1,924,898
64k	4.250x 🏆	4.25	0.0747%	1,841,478

Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

Sample 1:

जेरोम इसाक फ्रीडमन अमेरिका के प्रसिद्द वैज्ञानिक हैं। में इन्हें भौतिक विज्ञान म...

Vocab	Tokens	Count
8k	`▁ज ेर ोम ▁इस ाक ▁फ्री ड मन ▁अमेरिका ▁के ... (+19 more)`	29
16k	`▁जेर ोम ▁इस ाक ▁फ्री ड मन ▁अमेरिका ▁के ▁प्रसि ... (+17 more)`	27
32k	`▁जेर ोम ▁इस ाक ▁फ्री ड मन ▁अमेरिका ▁के ▁प्रसिद्द ... (+16 more)`	26
64k	`▁जेरोम ▁इसाक ▁फ्री ड मन ▁अमेरिका ▁के ▁प्रसिद्द ▁वैज्ञानिक ▁हैं ... (+14 more)`	24

Sample 2:

मवैया हंडिया, इलाहाबाद, उत्तर प्रदेश स्थित एक गाँव है। भूगोल जनसांख्यिकी यातायात...

Vocab	Tokens	Count
8k	`▁म वै या ▁हंडिया , ▁इलाहाबाद , ▁उत्तर ▁प्रदेश ▁स्थित ... (+16 more)`	26
16k	`▁म वै या ▁हंडिया , ▁इलाहाबाद , ▁उत्तर ▁प्रदेश ▁स्थित ... (+16 more)`	26
32k	`▁म वैया ▁हंडिया , ▁इलाहाबाद , ▁उत्तर ▁प्रदेश ▁स्थित ▁एक ... (+15 more)`	25
64k	`▁म वैया ▁हंडिया , ▁इलाहाबाद , ▁उत्तर ▁प्रदेश ▁स्थित ▁एक ... (+15 more)`	25

Sample 3:

माधवी हिन्दी फिल्मों की एक प्रसिद्ध अभिनेत्री हैं। व्यक्तिगत जीवन फिल्मी सफर प्र...

Vocab	Tokens	Count
8k	`▁माध वी ▁हिन्दी ▁फिल्मों ▁की ▁एक ▁प्रसिद्ध ▁अभिनेत्री ▁हैं । ... (+14 more)`	24
16k	`▁माध वी ▁हिन्दी ▁फिल्मों ▁की ▁एक ▁प्रसिद्ध ▁अभिनेत्री ▁हैं । ... (+12 more)`	22
32k	`▁माधवी ▁हिन्दी ▁फिल्मों ▁की ▁एक ▁प्रसिद्ध ▁अभिनेत्री ▁हैं । ▁व्यक्तिगत ... (+11 more)`	21
64k	`▁माधवी ▁हिन्दी ▁फिल्मों ▁की ▁एक ▁प्रसिद्ध ▁अभिनेत्री ▁हैं । ▁व्यक्तिगत ... (+11 more)`	21

Key Findings

Best Compression: 64k achieves 4.250x compression
Lowest UNK Rate: 8k with 0.0604% 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	99,591	16.60	936,439	10.1%	23.9%
2-gram	Subword	2,241 🏆	11.13	158,282	38.5%	70.6%
3-gram	Word	399,267	18.61	1,969,797	5.9%	14.2%
3-gram	Subword	22,500	14.46	933,655	15.0%	35.5%
4-gram	Word	884,119	19.75	3,325,655	5.2%	12.2%
4-gram	Subword	140,402	17.10	4,229,461	7.4%	21.2%
5-gram	Word	517,438	18.98	2,208,715	8.0%	17.2%
5-gram	Subword	516,632	18.98	8,451,936	4.4%	13.2%

Top 5 N-grams by Size

2-grams (Word):

Rank	N-gram	Count
1	`के लिए`	298,043
2	`जाता है`	144,432
3	`रूप में`	130,790
4	`के रूप`	124,867
5	`के साथ`	119,967

3-grams (Word):

Rank	N-gram	Count
1	`के रूप में`	123,170
2	`इन्हें भी देखें`	48,061
3	`करने के लिए`	45,421
4	`किया जाता है`	36,326
5	`किया गया था`	35,930

4-grams (Word):

Rank	N-gram	Count
1	`का एक गाँव है`	19,177
2	`है इन्हें भी देखें`	16,636
3	`जिले का एक गाँव`	14,515
4	`सरकार का आधिकारिक जालपृष्ठ`	12,365
5	`भारत सरकार के आधिकारिक`	12,363

5-grams (Word):

Rank	N-gram	Count
1	`जिले का एक गाँव है`	14,430
2	`सरकार के आधिकारिक पोर्टल पर`	12,360
3	`भारत सरकार के आधिकारिक पोर्टल`	12,359
4	`उत्तराखण्ड सरकार का आधिकारिक जालपृष्ठ`	10,606
5	`में विस्तृत एवं प्रामाणिक जानकारी`	10,604

2-grams (Subword):

Rank	N-gram	Count
1	`र _`	3,514,012
2	`के _`	2,568,356
3	`_ के`	2,390,034
4	`, _`	1,985,295
5	`न _`	1,962,003

3-grams (Subword):

Rank	N-gram	Count
1	`_ के _`	2,311,181
2	`_ में _`	1,613,203
3	`_ की _`	1,000,357
4	`औ र _`	977,524
5	`_ औ र`	976,951

4-grams (Subword):

Rank	N-gram	Count
1	`_ औ र _`	973,900
2	`_ है । _`	728,766
3	`_ ए क _`	550,979
4	`_ प र _`	374,271
5	`_ के _ लि`	322,541

5-grams (Subword):

Rank	N-gram	Count
1	`_ के _ लि ए`	298,051
2	`के _ लि ए _`	290,034
3	`ता _ है । _`	229,566
4	`_ क र ने _`	156,232
5	`_ जा ता _ है`	144,571

Key Findings

Best Perplexity: 2-gram (subword) with 2,241
Entropy Trend: Decreases with larger n-grams (more predictable)
Coverage: Top-1000 patterns cover ~13% 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	0.7477	1.679	9.12	1,271,248	25.2%
1	Subword	0.8713	1.829	13.88	36,229	12.9%
2	Word	0.3806	1.302	2.43	11,582,358	61.9%
2	Subword	0.6086	1.525	5.14	502,605	39.1%
3	Word	0.1696	1.125	1.40	28,143,624	83.0%
3	Subword	0.5073	1.421	3.65	2,584,793	49.3%
4	Word	0.0689 🏆	1.049	1.13	39,392,974	93.1%
4	Subword	0.4131	1.331	2.43	9,433,636	58.7%

Generated Text Samples (Word-based)

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

Context Size 1:

के प्रयासों के बाद गैन्डैल्फ़ के साथ स्थगित कर सभिको अचंभे में तर्क के नए स्थापित
में जीप वैन ज़ांड्ट एंड फ़्रैंसिस न्यूयॉर्क टाइम्स 10 से समझना लोगों को पुलिस ने सक्रिय
है लेनार्ट के साथ जारी की खेती श्रम बाजार में जेमस्टोन्स ब्रुन्सविक शामिल हो जाता है

Context Size 2:

के लिए वात्स्यायन ने कविता और नई विश्व व्यवस्था बनाने की सख्त जरूरत थी 24 फरवरी हिन्दी
जाता है कारण यह था कि जिन तस्वीरों में मंदिर के बगल में देखे बिना इस घोषणा
रूप में जिन में पाइथागोरस पहला व्यक्ति है जो वास्तविकता पर कम से कम कमाने वाले सदस्यों

Context Size 3:

के रूप में हॉलीवुड के पेशेवर लोगों के लिए उचित वक्त का इंतजार करने लगे उसे मारने के
इन्हें भी देखें उत्तराखण्ड के जिले उत्तराखण्ड के नगर कुमाऊँ मण्डल गढ़वाल मण्डल बाहरी कड़ियाँ उत्तराख...
करने के लिए सरकार द्वारा कोई वित्तीय सहायता प्राप्त हुई थी उन्होंने 14 फरवरी को विजय हजारे ट्रॉफी

Context Size 4:

का एक गाँव है बाहरी कड़ियाँ छत्तीसगढ़ सांस्कृतिक छत्तीसगढ जनजातियां कला खेल गोठ सतनाम पंथ छत्तीसगढ़ ...
है इन्हें भी देखें उत्तराखण्ड के जिले उत्तराखण्ड के नगर कुमाऊँ मण्डल गढ़वाल मण्डल बाहरी कड़ियाँ उत्त...
जिले का एक गाँव है बाहरी कड़ियाँ छत्तीसगढ़ सांस्कृतिक छत्तीसगढ जनजातियां कला खेल गोठ सतनाम पंथ छत्ती...

Generated Text Samples (Subword-based)

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

Context Size 1:

_के_का_इन्हें_की_दिरों_दमाने
र_अवल_विधानी_वर्ष_विश्लेष
क_आंकलालका_उगाव_नेगिस्तानी

Context Size 2:

र_ने_के_मंदिर_महाविद्यालय_
के_लिखने_में_इसे_कि_फ़िल्म_है
_के_पौड़ी_का_कुओमिक_रहने_

Context Size 3:

_के_बाद_वीकेंड_इंडोनेशिया_कोर
_में_52_नवंबर_दिए_गए।_
_की_मांग_की_कुछ_देशों_में_शामि

Context Size 4:

_और_युग_के_उत्तराखण्ड_राज्य_उ
_है।_कायाको_का_परित्याग_करती_
_एक_गाँव_है।_सूत्र_नहीं_देता_

Key Findings

Best Predictability: Context-4 (word) with 93.1% predictability
Branching Factor: Decreases with context size (more deterministic)
Memory Trade-off: Larger contexts require more storage (9,433,636 contexts)
Recommendation: Context-3 or Context-4 for text generation

4. Vocabulary Analysis

Statistics

Metric	Value
Vocabulary Size	503,387
Total Tokens	51,225,358
Mean Frequency	101.76
Median Frequency	4
Frequency Std Dev	5660.20

Most Common Words

Rank	Word	Frequency
1	के	2,319,434
2	में	1,706,170
3	है	1,377,542
4	की	1,046,592
5	और	978,950
6	से	789,677
7	का	776,115
8	को	650,931
9	एक	563,314
10	हैं	479,404

Least Common Words (from vocabulary)

Rank	Word	Frequency
1	तरॉरी	2
2	ग़ौरीद	2
3	ओकहर्स्ट	2
4	ओवरडेवलपमेंट	2
5	मिस्कैवेज	2
6	ज़ाल्स्की	2
7	aita	2
8	सूरजनसिंह	2
9	दीवानबगी	2
10	आशेक	2

Zipf's Law Analysis

Metric	Value
Zipf Coefficient	1.0969
R² (Goodness of Fit)	0.991607
Adherence Quality	excellent

Coverage Analysis

Top N Words	Coverage
Top 100	39.0%
Top 1,000	63.2%
Top 5,000	80.3%
Top 10,000	86.1%

Key Findings

Zipf Compliance: R²=0.9916 indicates excellent adherence to Zipf's law
High Frequency Dominance: Top 100 words cover 39.0% of corpus
Long Tail: 493,387 words needed for remaining 13.9% 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	0.8141	0.3993	N/A	N/A
mono_64d	64	0.7949	0.3123	N/A	N/A
mono_128d	128	0.7461	0.2670	N/A	N/A
aligned_32d	32	0.8141 🏆	0.3944	0.0840	0.4400
aligned_64d	64	0.7949	0.3145	0.2320	0.5660
aligned_128d	128	0.7461	0.2559	0.2760	0.6860

Key Findings

Best Isotropy: aligned_32d with 0.8141 (more uniform distribution)
Semantic Density: Average pairwise similarity of 0.3239. Lower values indicate better semantic separation.
Alignment Quality: Aligned models achieve up to 27.6% 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	0.307	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
`-स`	सुना, सिस्सु, स्नेहभरे
`-क`	करतारी, कोबिता, क्लिफफोर्ड
`-म`	माबैन, मखीजा, महाराना
`-ब`	बाइटों, ब्रेवहार्ट्स, ब्लूहोल
`-प`	पांडिकाट्ट, पर्यवेक्षणीय, प्रसन्नतापूवर्क
`-अ`	अहमदगर, अर्गला, अंजुल
`-र`	रगति, राठौरों, रतनगड
`-ज`	जीप्स, जच्चन्धवग्गो, जवहार

Productive Suffixes

Suffix	Examples
`-न`	गेलसान, माबैन, गमभन
`-र`	डाइबर, नयानगर, दखर
`-स`	जीप्स, ब्रेवहार्ट्स, नॉमिनेशंस
`-s`	siblings, sheriffs, hieroglyphics
`-क`	नखचंद्रक, प्रसन्नतापूवर्क, बाहुबंधक
`-ल`	नौटियाल, ब्लूहोल, आवर्तकाल
`-त`	द्विपरत, कुत, ऑस्ट्रेलियाभारत
`-ट`	पांडिकाट्ट, एग्रीमेंट, यूरोसेंट

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
`nter`	3.13x	80 contexts	inter, enter, unter
`atio`	3.06x	61 contexts	patio, ation, ratio
`tion`	2.97x	67 contexts	tiong, ation, nation
`ctio`	3.12x	40 contexts	action, actions, section
`iona`	3.07x	26 contexts	ciona, fiona, acciona
`ubli`	2.96x	23 contexts	hubli, publié, public
`rpor`	3.33x	11 contexts	corpore, corpora, airport
`onal`	3.05x	11 contexts	tonal, monal, zonal
`guid`	3.19x	9 contexts	guide, guido, eguide

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
`-क`	`-र`	33 words	कुक्कुटेश्वर, कंपोज़र
`-स`	`-स`	33 words	सोफ्रोनियस, सत्यदास
`-प`	`-स`	30 words	प्रोड्क्शन्स, पैवोनिस
`-म`	`-र`	28 words	माष्टर, मऊरानीपुर
`-स`	`-न`	27 words	सेक्‍शन, सीस्तान
`-स`	`-र`	24 words	साहेर, सूर्यवीर
`-प`	`-न`	24 words	पराधीन, पिकायून
`-प`	`-त`	23 words	पिएत, प्रोग्रामित
`-स`	`-क`	23 words	सौंदर्यबोधक, सफलतार्पूक
`-व`	`-र`	22 words	वुर, विज्कगवर्नर

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
संगीतनाटक	`संगीतना-ट-क`	7.5	`ट`
साहित्यकर	`साहित्य-क-र`	7.5	`क`
sanderson	`sanders-on`	4.5	`sanders`
hornbills	`hornbill-s`	4.5	`hornbill`
मैगाट्रॉनस	`मैगाट्रॉन-स`	4.5	`मैगाट्रॉन`
जनसांख्यकीय	`जनसांख्यकी-य`	4.5	`जनसांख्यकी`
इन्फ्लुएंसस	`इन्फ्लुएंस-स`	4.5	`इन्फ्लुएंस`
österreichs	`österreich-s`	4.5	`österreich`
दक्षिणमध्य	`द-क-्षिणमध्य`	4.5	`्षिणमध्य`
अर्धसूत्र	`अ-र-्धसूत्र`	4.5	`्धसूत्र`
anatolian	`anatoli-an`	4.5	`anatoli`
responded	`respond-ed`	4.5	`respond`
paralympics	`paralympic-s`	4.5	`paralympic`
उष्मागतिक	`उष्मागति-क`	4.5	`उष्मागति`
एजेंसियाँ	`ए-ज-ेंसियाँ`	4.5	`ेंसियाँ`

6.6 Linguistic Interpretation

Automated Insight: The language Hindi 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 (4.25x)
N-gram	2-gram	Lowest perplexity (2,241)
Markov	Context-4	Highest predictability (93.1%)
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

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.

Maintainer

Omar Kamali - Omneity Labs

Citation

If you use these models in your research, please cite:

bibtex

@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.

wikilangs/hi

Hindi - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

📋 Repository Contents

Models & Assets

Analysis and Evaluation

1. Tokenizer Evaluation

Results

Tokenization Examples

Key Findings

2. N-gram Model Evaluation

Results

Top 5 N-grams by Size

Key Findings

3. Markov Chain Evaluation

Results

Generated Text Samples (Word-based)

Generated Text Samples (Subword-based)

Key Findings

4. Vocabulary Analysis

Statistics

Most Common Words

Least Common Words (from vocabulary)

Zipf's Law Analysis

Coverage Analysis

Key Findings

5. Word Embeddings Evaluation

5.1 Cross-Lingual Alignment

5.2 Model Comparison

Key Findings

6. Morphological Analysis (Experimental)

6.1 Productivity & Complexity

6.2 Affix Inventory (Productive Units)

Productive Prefixes

Productive Suffixes

6.3 Bound Stems (Lexical Roots)

6.4 Affix Compatibility (Co-occurrence)

6.5 Recursive Morpheme Segmentation

6.6 Linguistic Interpretation

7. Summary & Recommendations

Production Recommendations

Appendix: Metrics Glossary & Interpretation Guide

Tokenizer Metrics

N-gram Model Metrics

Markov Chain Metrics

Vocabulary & Zipf's Law Metrics

Word Embedding Metrics

General Interpretation Guidelines

Visualizations Index

About This Project

Data Source

Project

Maintainer

Citation

License

Links

wikilangs/
hi