README.md

metadata

language: as
language_name: Assamese
language_family: indoaryan_eastern
tags:
  - wikilangs
  - nlp
  - tokenizer
  - embeddings
  - n-gram
  - markov
  - wikipedia
  - feature-extraction
  - sentence-similarity
  - tokenization
  - n-grams
  - markov-chain
  - text-mining
  - fasttext
  - babelvec
  - vocabulous
  - vocabulary
  - monolingual
  - family-indoaryan_eastern
license: mit
library_name: wikilangs
pipeline_tag: text-generation
datasets:
  - omarkamali/wikipedia-monthly
dataset_info:
  name: wikipedia-monthly
  description: Monthly snapshots of Wikipedia articles across 300+ languages
metrics:
  - name: best_compression_ratio
    type: compression
    value: 4.542
  - name: best_isotropy
    type: isotropy
    value: 0.8547
  - name: vocabulary_size
    type: vocab
    value: 0
generated: 2026-01-03T00:00:00.000Z

Assamese - Wikilangs Models

Comprehensive Research Report & Full Ablation Study

This repository contains NLP models trained and evaluated by Wikilangs, specifically on Assamese 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	3.450x	3.45	0.0757%	1,416,711
16k	3.894x	3.89	0.0855%	1,255,391
32k	4.266x	4.27	0.0937%	1,145,685
64k	4.542x 🏆	4.54	0.0997%	1,076,075

Tokenization Examples

Below are sample sentences tokenized with each vocabulary size:

Sample 1: জয়নগৰ মজিলপুৰ ভাৰতৰ পশ্চিমবংগ ৰাজ্যৰ দক্ষিণ চব্বিশ পৰগনা জিলাত অৱস্থিত এখন চহৰ।...

Vocab	Tokens	Count
8k	▁জ য়ন গৰ ▁মজ িল পুৰ ▁ভাৰতৰ ▁পশ্চিমব ংগ ▁ৰাজ্যৰ ... (+14 more)	24
16k	▁জ য়ন গৰ ▁মজ িল পুৰ ▁ভাৰতৰ ▁পশ্চিমবংগ ▁ৰাজ্যৰ ▁দক্ষিণ ... (+12 more)	22
32k	▁জয়ন গৰ ▁মজ িল পুৰ ▁ভাৰতৰ ▁পশ্চিমবংগ ▁ৰাজ্যৰ ▁দক্ষিণ ▁চব্বিশ ... (+8 more)	18
64k	▁জয়নগৰ ▁মজ িল পুৰ ▁ভাৰতৰ ▁পশ্চিমবংগ ▁ৰাজ্যৰ ▁দক্ষিণ ▁চব্বিশ ▁পৰগনা ... (+7 more)	17

Sample 2: হাবুং মৈদাম হৈছে আহোমসকলৰ পঞ্চমৰাজধানী হাবুংৰ টাইভেটিত অৱস্থিত দুটা প্ৰাচীন মৈদা...

Vocab	Tokens	Count
8k	▁হাব ু ং ▁মৈ দ াম ▁হৈছে ▁আহোম সকলৰ ▁পঞ্চম ... (+31 more)	41
16k	▁হাব ুং ▁মৈ দাম ▁হৈছে ▁আহোম সকলৰ ▁পঞ্চম ৰাজ ধান ... (+26 more)	36
32k	▁হাব ুং ▁মৈদাম ▁হৈছে ▁আহোমসকলৰ ▁পঞ্চম ৰাজ ধানী ▁হাব ুং ... (+21 more)	31
64k	▁হাবুং ▁মৈদাম ▁হৈছে ▁আহোমসকলৰ ▁পঞ্চম ৰাজধানী ▁হাবুং ৰ ▁টাই ভেটিত ... (+16 more)	26

Sample 3: ভাৰতীয় ন্যায় সংহিতা (IAST: Bhāratīya Nyāya Saṃhitā), ভাৰতীয় গণৰাজ্যৰ অপৰাধ সং...

Vocab	Tokens	Count
8k	▁ভাৰতীয় ▁ন্যায় ▁সংহ িতা ▁( i ast : ▁bh ā ... (+27 more)	37
16k	▁ভাৰতীয় ▁ন্যায় ▁সংহিতা ▁( i ast : ▁bh ā rat ... (+23 more)	33
32k	▁ভাৰতীয় ▁ন্যায় ▁সংহিতা ▁( iast : ▁bh ā rat ī ... (+20 more)	30
64k	▁ভাৰতীয় ▁ন্যায় ▁সংহিতা ▁( iast : ▁bh ā rat īya ... (+18 more)	28

Key Findings

• Best Compression: 64k achieves 4.542x compression
• Lowest UNK Rate: 8k with 0.0757% 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	62,472	15.93	206,764	8.3%	21.4%
2-gram	Subword	2,308 🏆	11.17	63,567	34.0%	69.4%
3-gram	Word	109,754	16.74	237,526	5.0%	14.6%
3-gram	Subword	20,939	14.35	371,943	13.3%	35.5%
4-gram	Word	247,178	17.92	371,701	2.3%	7.7%
4-gram	Subword	113,780	16.80	1,515,602	7.8%	20.9%
5-gram	Word	182,489	17.48	239,039	1.9%	7.3%
5-gram	Subword	319,720	18.29	2,664,609	5.1%	14.4%

Top 5 N-grams by Size

2-grams (Word):

Rank	N-gram	Count
1	কৰা হয়	29,188
2	কৰা হৈছিল	12,508
3	হ ল	11,276
4	লাভ কৰে	10,608
5	কৰা হৈছে	10,201

3-grams (Word):

Rank	N-gram	Count
1	ব্যৱহাৰ কৰা হয়	3,336
2	হ ব পাৰে	3,197
3	বুলি কোৱা হয়	3,190
4	গণ্য কৰা হয়	2,309
5	ডিগ্ৰী লাভ কৰে	2,043

4-grams (Word):

Rank	N-gram	Count
1	তথ্য সংগ্ৰহ বাহ্যিক সংযোগ	1,641
2	বুলি গণ্য কৰা হয়	1,265
3	স্নাতক ডিগ্ৰী লাভ কৰে	864
4	হিচাপে গণ্য কৰা হয়	801
5	তথ্য উৎস বাহ্যিক সংযোগ	782

5-grams (Word):

Rank	N-gram	Count
1	archived from the original on	423
2	অভিনেত্ৰী চলচ্চিত্ৰৰ অভিনেত্ৰী চলচ্চিত্ৰৰ অভিনেত্ৰী	245
3	দিনটোত ঘটা কেইটামান উল্লেখযোগ্য ঘটনা	244
4	এই দিনটোত ঘটা কেইটামান উল্লেখযোগ্য	237
5	প্ৰাৰম্ভিক জীৱন আৰু শিক্ষা চনৰ	214

2-grams (Subword):

Rank	N-gram	Count
1	ৰ _	1,309,192
2	ত _	645,783
3	_ আ	585,970
4	। _	462,800
5	_ ক	454,528

3-grams (Subword):

Rank	N-gram	Count
1	আ ৰু _	247,371
2	_ আ ৰু	247,194
3	_ ক ৰি	139,299
4	_ তে ওঁ	136,655
5	ন ৰ _	124,787

4-grams (Subword):

Rank	N-gram	Count
1	_ আ ৰু _	246,758
2	ছি ল । _	101,454
3	_ ক ৰা _	90,139
4	_ এ ই _	64,252
5	_ তে ওঁ _	64,067

5-grams (Subword):

Rank	N-gram	Count
1	_ হ য় । _	58,520
2	_ ক ৰে । _	51,805
3	_ ক ৰি ছি ল	51,692
4	ৰ _ বা বে _	47,193
5	_ চ ন ত _	47,011

Key Findings

• Best Perplexity: 2-gram (subword) with 2,308
• Entropy Trend: Decreases with larger n-grams (more predictable)
• Coverage: Top-1000 patterns cover ~14% 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	0.8450	1.796	7.84	550,295	15.5%
1	Subword	0.8398	1.790	12.10	15,252	16.0%
2	Word	0.2695	1.205	1.71	4,311,912	73.1%
2	Subword	0.7069	1.632	5.33	184,530	29.3%
3	Word	0.0827	1.059	1.15	7,360,379	91.7%
3	Subword	0.5599	1.474	3.49	984,248	44.0%
4	Word	0.0276 🏆	1.019	1.04	8,480,563	97.2%
4	Subword	0.4373	1.354	2.27	3,437,009	56.3%

Generated Text Samples (Word-based)

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

Context Size 1:

1. আৰু লেখিকা বুলি সকলোৱে উৎসাহিত কৰাৰ উদ্দেশ্যে চনত বামুণপাৰা বালিপাৰা ষ্টীম কুকাৰতে খাদ্যৰ ৬৫ মিলিয়ন...
2. কৰা এক শিক্ষা প্ৰদানৰ বিষয় হিচাপে কাৰ্যনিৰ্বাহ কৰিছিল মৃত্যু চনত হোমেন বৰগোহাঞিৰ এখন মেল পাতে আৰু
3. হয় আমেৰিকা যুক্তৰাষ্টৰ প্ৰথম উপাচাৰ্য আছিল অনুমান কৰা পুৰণি অট্টালিকাবোৰত মধ্যমীয়া চৰিত্ৰত অভিনয় ...

Context Size 2:

1. কৰা হয় চনৰ জানুৱাৰী মাহত অম্বা বহোৰা নামৰ এগৰাকী যুৱতীক তেওঁৰ স্বামী টোপনি যোৱালৈকে অপেক্ষা কৰাটো প...
2. কৰা হৈছিল কলহোৰা শাসকসকলৰ সমাধিস্থলত ফুল আৰু প্ৰসাদেৰে তুলসীক পূজা কৰা ধৰণৰ তাৰতম্য আছিল তথাপি ধৰ্মে...
3. হ ল পদ্মভূষণ ভাৰতৰ তৃতীয় সৰ্বোচ্চ অসামৰিক সন্মান পদ্মশ্ৰী লাভ কৰে তেখেতে অভিনয় কৰে চনত তেওঁৰ নিজাক...

Context Size 3:

1. ব্যৱহাৰ কৰা হয় msa এ smtp প্ৰটোকলত প্ৰদান কৰা গন্তব্যস্থানৰ ঠিকনা নিৰ্ধাৰণ কৰে বাৰ্তা হেডাৰৰ পৰা নহ...
2. হ ব পাৰে অসমৰ কবি লেখক জীৱন নৰহে আত্মজীৱনীমূলক গ্ৰন্থখনক নতুন প্ৰজন্মৰ সাহসৰ দলিল বুলি অভিহিত কৰে অৰ...
3. বুলি কোৱা হয় ৰাক্ষসসকলক প্ৰায় পৰাধীন সৈনিকৰ ৰূপত দেখুৱা হৈছিল পিছে কিছু ৰাক্ষসে অত্যন্ত বল অৰ্জন ক...

Context Size 4:

1. তথ্য সংগ্ৰহ বাহ্যিক সংযোগ cornell university e book library of classic texts on mechanical design an...
2. বুলি গণ্য কৰা হয় আৰু কোমল গ আৰু কোমল ধ স্বৰসমূহ কম্পনৰ সৈতে অন্দোলিত পৰিবেশিত হয় সকলো পাঁচটা স্বৰ
3. স্নাতক ডিগ্ৰী লাভ কৰে সেই একেই দীন দয়াল উপাধ্যায় কলেজৰ পৰা সামাজিক কাম চনত ১৯ বছৰ বয়সত ছেম অল্টমে...

Generated Text Samples (Subword-based)

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

Context Size 1:

1. _ধান_লা_শক্তি।_হাসিদ্ধ_তাকা
2. ৰক্ষ_নিজনগোৱাহালচ্চিত্ৰ_মবাবে
3. কথাই_ইছিল_জীৱন_ডিয়_বা

Context Size 2:

1. ৰ_আৰু_লাউ।_অসংখ্যক_ভাৰত
2. ত_জ্ঞানৰ_কৃপ,_কেন্দ্ৰটোৰ_পৰা
3. _আৰু_মোৰ_পৰা_উৰুলিয়ান_শা

Context Size 3:

1. আৰু_বীৰেন্দ্ৰ_মোডীৰ_নিগমৰ_প্ৰয়া
2. _আৰু_অৰ্থ।_দেৱালয়খনৰ_পিছ
3. _কৰিবলৈ_অস্বীকাৰ_হোৱা_মতবাদ

Context Size 4:

1. _আৰু_পাম_তেল_আৰু_পিপিপি_আৰু
2. ছিল।_কুমাৰীত্ব_পৰীক্ষাৰ_অংগ_আৰু
3. _কৰা_দুখৰ_আৰু_তেওঁৰ_ছিলভাৰ

Key Findings

• Best Predictability: Context-4 (word) with 97.2% predictability
• Branching Factor: Decreases with context size (more deterministic)
• Memory Trade-off: Larger contexts require more storage (3,437,009 contexts)
• Recommendation: Context-3 or Context-4 for text generation

---

4. Vocabulary Analysis

Statistics

Metric	Value
Vocabulary Size	225,407
Total Tokens	9,007,362
Mean Frequency	39.96
Median Frequency	4
Frequency Std Dev	792.95

Most Common Words

Rank	Word	Frequency
1	আৰু	247,463
2	কৰা	93,923
3	হয়	87,716
4	কৰে	78,599
5	এই	64,931
6	তেওঁ	64,613
7	পৰা	53,636
8	কৰিছিল	51,623
9	বাবে	50,799
10	চনত	49,149

Least Common Words (from vocabulary)

Rank	Word	Frequency
1	পাণ্ডুবংশী	2
2	মনুমেণ্টছ	2
3	ছিৰপুৰৰ	2
4	swfl	2
5	manhunt	2
6	megamodel	2
7	গ্লেডৰেগ্‌চ	2
8	কিস	2
9	পদাইভীৰণ	2
10	বিগিল	2

Zipf's Law Analysis

Metric	Value
Zipf Coefficient	1.0094
R² (Goodness of Fit)	0.989782
Adherence Quality	excellent

Coverage Analysis

Top N Words	Coverage
Top 100	25.5%
Top 1,000	50.9%
Top 5,000	71.9%
Top 10,000	79.7%

Key Findings

• Zipf Compliance: R²=0.9898 indicates excellent adherence to Zipf's law
• High Frequency Dominance: Top 100 words cover 25.5% of corpus
• Long Tail: 215,407 words needed for remaining 20.3% 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	0.8458	0.3637	N/A	N/A
mono_64d	64	0.8547 🏆	0.2742	N/A	N/A
mono_128d	128	0.8359	0.2093	N/A	N/A
aligned_32d	32	0.8458	0.3735	0.0580	0.3060
aligned_64d	64	0.8547	0.2836	0.1180	0.3960
aligned_128d	128	0.8359	0.2075	0.1480	0.4820

Key Findings

• Best Isotropy: mono_64d with 0.8547 (more uniform distribution)
• Semantic Density: Average pairwise similarity of 0.2853. Lower values indicate better semantic separation.
• Alignment Quality: Aligned models achieve up to 14.8% 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	-0.495	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

Productive Suffixes

Suffix	Examples
-ৰ	কেথলিকসকলৰ, স্বপ্ৰচাৰ, ৱিণ্টাৰ
-াৰ	স্বপ্ৰচাৰ, ৱিণ্টাৰ, অফকাটাৰ

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
ther	3.36x	64 contexts	theri, there, other
ress	3.34x	44 contexts	press, dress, duress
nter	3.33x	38 contexts	inter, enter, wynter
vers	3.15x	47 contexts	verso, versa, verse
atio	3.32x	37 contexts	ratio, fatio, nation
indi	3.24x	39 contexts	hindi, indie, india
ment	3.24x	38 contexts	cement, moment, mental
stor	3.25x	35 contexts	storm, jstor, story
ctio	3.33x	32 contexts	action, auction, faction
iver	3.16x	26 contexts	liver, giver, river
ersi	3.21x	20 contexts	persia, persie, yersin
mber	3.17x	18 contexts	amber, number, member

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
চেটেছুয়াৰাৰ	চেটেছুয়-াৰ-াৰ	3.0	চেটেছুয়
সীতাৰামায়াৰ	সীতাৰামায়-াৰ	1.5	সীতাৰামায়
ৰাজ্কুমাৰ	ৰাজ্কুম-াৰ	1.5	ৰাজ্কুম
প্ৰতিৰক্ষাৰ	প্ৰতিৰক্ষ-াৰ	1.5	প্ৰতিৰক্ষ
দত্তবৰুৱাৰ	দত্তবৰুৱ-াৰ	1.5	দত্তবৰুৱ
বিষ্ণুৰাভাৰ	বিষ্ণুৰাভ-াৰ	1.5	বিষ্ণুৰাভ
বদৌপায়াৰ	বদৌপায়-াৰ	1.5	বদৌপায়
হাছলেংগাৰ	হাছলেংগ-াৰ	1.5	হাছলেংগ
চিজাৰিয়াৰ	চিজাৰিয়-াৰ	1.5	চিজাৰিয়
কুকুৰাঝাৰ	কুকুৰাঝ-াৰ	1.5	কুকুৰাঝ
মন্দাৱস্থাৰ	মন্দাৱস্থ-াৰ	1.5	মন্দাৱস্থ
আত্মপ্ৰতিষ্ঠাৰ	আত্মপ্ৰতিষ্ঠ-াৰ	1.5	আত্মপ্ৰতিষ্ঠ
কেঁচাগোল্লাৰ	কেঁচাগোল্ল-াৰ	1.5	কেঁচাগোল্ল
ফ্ৰণ্টিয়াৰ	ফ্ৰণ্টিয়-াৰ	1.5	ফ্ৰণ্টিয়
যিহোচূৱাৰ	যিহোচূৱ-াৰ	1.5	যিহোচূৱ

6.6 Linguistic Interpretation

Automated Insight: The language Assamese 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 (4.54x)
N-gram	2-gram	Lowest perplexity (2,308)
Markov	Context-4	Highest predictability (97.2%)
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: 2026-01-03 17:31:44