German Reading Game
Read a German story a few words at a time and pick the missing word (tap, or press 1, 2, 3).
Settings
Adaptive: 8 right in your last 10 answers moves you up a level, 3 misses in your last 6 answers moves you down. Window: how much text you see before the gap. The next gap comes up by itself; press any key or tap to skip the wait. Pause after mistakes adds a Continue button after wrong answers.
Loading chapter...
Show full text so far
Hardest words
Words you miss appear here and come back as spaced reviews.
How it works
- Key words asks about one word per clause, chosen to practise grammar: article and adjective endings, pronoun case, verb person and tense, prepositions and noun forms. You see the rest of the clause after the gap, so exactly one option fits. Each item was checked by a language model, and comes with a short explanation.
- Levels (Key words mode): Beginner asks only about common words, with fewer gaps, and the two wrong options are clearly wrong words from the same text, with the same capitalisation as the answer. Intermediate mixes those with grammar near-misses. Advanced asks every key word with near-miss forms like der, den, dem or verb endings. With adaptive level on (the default), 8 right answers in your last 10 move you up a level and 3 misses in your last 6 move you down. At Beginner, items where capitalisation alone gives the answer away do not count.
- Reading window: you see the current sentence and a few words before it, and the text slides on as you answer. Choose a Small, Medium or Large window in Settings, or open Show full text so far to read everything up to the gap.
- Every word hides every word as you read, with only the text before it visible. This is pure next-word prediction: harder, and sometimes more than one option could continue the sentence, so it scores the word the author actually wrote. Its choices are generated by grammar rules and not individually reviewed.
- Correct answers score 10 points plus 2 per step of your current streak (up to +20). A wrong answer resets the streak and adds the word to Hardest words.
- Reviews: each missed form (for example dem, a form of der) goes into box 1 and comes back as a Review card in the reading flow after about 10 new gaps. A right answer moves it up a box, with the next review after 1, 3, 7 and then 16 days; a wrong one sends it back to box 1. You get at most 1 review per 5 new gaps and 10 per session, so reading stays the main thing. Reviews score points like normal gaps and do not move you on in the chapter.
- The English gloss shows a short translation of the answer word after you choose (Key words mode).
Texts and sources
- Grimms Märchen by Jacob and Wilhelm Grimm, Easier (A2 to B1). Source: Deutsche Märchen gesammelt durch die Brüder Grimm, ed. M. Thilo-Luyken (Langewiesche-Brandt, 1921), from Project Gutenberg. Public domain.
- Die Verwandlung by Franz Kafka, Harder (B2 to C1). Source: Die Verwandlung (Kurt Wolff Verlag, Leipzig, 1915), from Project Gutenberg. Public domain.
The texts keep their original historical spelling: you will see daß (modern dass), muß (muss) and forms like Walde or ward. The Grimm tales included are Rotkäppchen, Hänsel und Gretel, Die Bremer Stadtmusikanten, Der Froschkönig, Der Wolf und die sieben jungen Geißlein and Frau Holle. Wrong options were generated with spaCy's German model and grammar rules, then filtered by a language model (Claude Sonnet), which also wrote the explanations. The clearly wrong options used at Beginner and Intermediate are drawn from words of the same chapter with the same capitalisation as the answer: nouns of unrelated meaning for noun gaps, verbs for article, pronoun and preposition gaps, articles for verb gaps. In a check by Claude Opus of 50 random Beginner and 50 Intermediate items with these options, all 100 had exactly one fitting option. In an earlier independent check by a different model (Claude Opus) of 50 random items per level, all 150 had exactly one fitting option, and 50 of 50 (Beginner), 49 of 50 (Intermediate) and 49 of 50 (Advanced) explanations were judged correct. In Every word mode, about 7 in 10 items had only one plausible continuation. Mistakes are possible.