Estonian Word List with Ekilex and Frequency Data

This repo contains an open Estonian word list with 160,316 base words, enriched with Ekilex metadata and OpenSubtitles frequency data.

If you only need one file, start with data/est_words_160k.tsv. It includes the base word, frequency rank, CEFR level when available, and part of speech.

Snapshot date: 2026-04-01.

At a Glance

Metric	Value
Base words	160,316
Inflected forms	5,454,775
Found in Ekilex	160,312 (99.99%)
With frequency rank	12,493 (7.8%)
With CEFR proficiency level	9,951 (6.2%)
With POS tag	158,729 (99.0%)
Unique POS categories	14
Proficiency levels	A1, A2, B1, B2, C1

Files

data/est_words_160k.tsv

Main dataset. Contains all 160,316 words with frequency rank, proficiency level, and part of speech.

Sort order:

1. Frequency rank
2. Proficiency level
3. Alphabetical order

Column	Type	Description
word	string	Base form of the word
freq_rank	int	Frequency rank from the OpenSubtitles corpus (1 = most common, 0 = no data)
proficiency	string	CEFR level (A1, A2, B1, B2, C1, or empty)
pos	string	Part-of-speech codes, comma-separated

word	freq_rank	proficiency	pos
ma	2	A1	pron
ei	3	A1	adv,interj
koer	2866	A1	s,adj

data/est_inflected_forms.tsv

Inflected forms dataset. Contains 160,173 words with grammatical forms such as cases, declensions, and conjugations.

Column	Type	Description
word	string	Base form (dictionary headword)
forms	string	All inflected forms, comma-separated

word	forms
koer	koer,koera,koerasse,koeras,koerast,koerale,koeral,koeralt,...
maja	maja,majja,majasse,majas,majast,majale,majal,majalt,...

Estonian has 14 grammatical cases and a lot of verb inflection, so a single word can easily have 30 to 50+ forms.

data/est_proficiency_words.tsv

CEFR-tagged subset. Contains 9,951 words with language proficiency levels.

Column	Type	Description
word	string	Base form
proficiency	string	CEFR level from A1 (beginner) to C1 (advanced)
pos	string	Part of speech

word	proficiency	pos
aadress	A1	s
aasta	A1	s
aabits	B1	s

Distribution: A1: 685, A2: 997, B1: 2,509, B2: 5,020, C1: 740

data/est_frequency_ranked.tsv

Frequency-ranked subset. Contains 12,493 words with corpus frequency from OpenSubtitles.

Column	Type	Description
word	string	Word form
rank	int	Frequency rank (1 = most common)
corpus_count	int	Raw occurrence count in the subtitle corpus

word	rank	corpus_count
ma	2	1864911
ei	3	1572717
koer	2866	13805

data/est_words_simple.txt

Plain word list, one word per line, in the same sort order as the main dataset. Use this when you just need the words and none of the metadata.

Part of Speech Tags

Code	Meaning	Count
s	Noun	116,122
adj	Adjective	18,232
prop	Proper noun	8,901
v	Verb	8,729
adv	Adverb	7,129
interj	Interjection	718
postp	Postposition	295
vrm	Verb form	239
num	Numeral	190
adjg	Adjective (genitive attribute)	173
pron	Pronoun	128
prep	Preposition	79
konj	Conjunction	29
adjid	Adjective (plural only)	1

Data Sources

1. Ekilex, Eesti Keele Instituut (EKI)

Source: ekilex.ee

Ekilex is the official dictionary and terminology system maintained by the Estonian Language Institute. We queried it for:

• part of speech
• CEFR proficiency level
• inflected forms

160,312 out of 160,316 words were found. The 4 missing words are Turkish place names written with dotted-I Unicode characters.

2. Hermit Dave / OpenSubtitles Frequency List

Source: github.com/hermitdave/FrequencyWords

This list is derived from the OpenSubtitles 2018 corpus. It gives raw occurrence counts for about 50,000 Estonian words. Of those, 12,493 overlap with this project's base word list.

This frequency signal is useful, but it comes with bias. Subtitle data reflects spoken and conversational language much better than formal, technical, or literary Estonian.

3. Base Word List

The base 160,316-word list was compiled from publicly available Estonian dictionary sources. See methodology/ for the collection details.

Sorting Logic

The main dataset, est_words_160k.tsv, uses a three-tier sort so the most useful words show up first.

Tier	Words	Criteria	Sort within tier
1	12,493	Has frequency rank from OpenSubtitles	Rank ascending, most common first
2	3,703	Has proficiency level but no frequency rank	A1 → A2 → B1 → B2 → C1, then alphabetical
3	144,120	Has neither frequency nor proficiency data	Alphabetical

Use Cases

• Keyboard autocomplete: use the tier-sorted list for better suggestion order
• Spellcheckers: use est_words_simple.txt or the inflected forms dataset for broader coverage
• Language learning apps: use est_proficiency_words.tsv for graded vocabulary
• NLP and morphological analysis: use inflected forms for lemmatization and stemming
• Text analysis: use frequency data as one input for difficulty scoring

Regenerating the Data

Use scripts/export_from_checkpoint.py to regenerate the published data files from the raw Ekilex checkpoint:

python3 scripts/export_from_checkpoint.py \
    --checkpoint /path/to/ekilex_checkpoint.jsonl \
    --frequency /path/to/hermit_dave_et_50k.txt \
    --output-dir data/

The raw checkpoint file is not included in this repo because it is large (123 MB, 160K JSONL entries). See methodology/collection_process.md for how it was produced.

License

CC-BY-SA 4.0 (Creative Commons Attribution-ShareAlike 4.0 International)

Attribution

If you use this dataset, please include attribution like this:

Estonian word list data sourced from:

1. Eesti Keele Instituut (EKI) via Ekilex: word entries, parts of speech, proficiency levels, and inflected forms. License: CC-BY 4.0.
2. Hermit Dave / OpenSubtitles via FrequencyWords: frequency ranks derived from the OpenSubtitles 2018 corpus. License: CC-BY-SA 4.0.

Combined and processed by Kristjan Pikhof. Modifications include Ekilex enrichment for POS, proficiency, and inflected forms, plus frequency mapping from OpenSubtitles data. Results were deduplicated and sorted by frequency and proficiency tiers.

Limitations

• Frequency coverage is limited: only 7.8% of words have frequency data. OpenSubtitles is useful, but it misses a lot of formal, technical, and specialized vocabulary.
• Proficiency tagging is sparse: only 6.2% of words have CEFR levels. Ekilex mainly provides proficiency data for commonly taught vocabulary.
• Subtitle frequency is biased: movie and TV subtitles overrepresent spoken and informal language.
• No sentence context: this is a word-level dataset. It does not include bigrams, collocations, or example sentences.
• Inflected forms may include archaic variants: Ekilex paradigms include grammatically valid forms that are rare in modern usage.