How this list was built
The 100-word core list is not a matter of taste. It is the measured top 100 German lemmas from the Leipzig Corpora Collection, and everything needed to check that claim — the code, the form-to-lemma map and the raw output — ships with the site.
The ranking comes from the Leipzig Corpora Collection, a long-running corpus project at Leipzig University. Two corpora were used so that no single genre decides the result:
| Corpus | Content | Word tokens |
|---|---|---|
| deu_mixed-typical_2011_1M | mixed genres, 2011 | 6,437,637 |
| deu_news_2023_1M | news, 2023 | 14,979,650 |
What was counted
Leipzig publishes counts of surface forms, not lemmas — ist, sind, war and bin are counted separately. A vocabulary list needs lemmas, so forms are grouped using an explicit map, and the rules are applied consistently:
- Every surface form is assigned to at most one lemma, so nothing is counted twice.
- Each corpus is converted to occurrences per million words, then the two are averaged. That way the larger corpus does not simply outvote the smaller one.
- Preposition + article contractions (im, am, zum, vom, beim, ins) count toward the preposition.
- Proper nouns — country and city names, party names, months, weekdays — are excluded. They rank high in news text but are artifacts of the source material, not general vocabulary.
- Punctuation is excluded from the denominator, so a coverage figure means words, not full stops.
What this method cannot do
Three honest limitations, because a number with no caveats is a number you should not trust:
- German is ambiguous and the corpus is not tagged. sie is “she”, “they” and formal “you”; die is both an article and a relative pronoun. Without grammatical tagging these cannot be separated, so each ambiguous form is assigned to its dominant reading and that choice is recorded in the source data.
- Written, not spoken. Both corpora are written German. Everyday speech would rank ich and du higher and werden lower. The coverage figure describes ordinary written German.
- Genre still shows through. Prozent, Euro and Uhr reach the top 100 partly because news text is full of figures and times. They were left in — they are ordinary German words a learner will meet — but that is why they are there.
Check it yourself
The scripts that produce this ranking ship with the site: corpus/lemmas.py holds the form-to-lemma map, corpus/rank.py computes the ranking and coverage, and corpus/ranked.json is the output, with a per-corpus breakdown for every lemma. Download the two corpora from Leipzig, run the scripts, and you should get these exact numbers.
Coverage, measured
How much of ordinary written German you can account for, by number of lemmas learned:
| Top N lemmas | mixed genres, 2011 | news, 2023 |
|---|---|---|
| 10 | 28% | 26.2% |
| 50 | 43.2% | 43.2% |
| 100 | 49.1% | 49.6% |
| 150 | 52.2% | 52.8% |
| 200 | 54.1% | 55% |
The curve flattens fast, and that is the whole argument for this page: the first hundred lemmas buy you roughly half of all running text, while the next hundred add only a few points more.
Try it
All 182 words are in the free trainer — flashcards, multiple choice, listening practice and spaced review. No account needed.