7  Wahrscheinlichkeiten

7.1 Zusammenfassung

7.2 Experimentaufbau

7.3 Häufigkeiten von Wörtern und Kombinationen

Students receive an envelope containing a printed sentence cut into individual word slips:

“the cat eats the fish and the cat drinks the water and the dog eats the fish”

Working in pairs, they physically arrange the word slips on the table and:

  1. List all unique words — this is the vocabulary.
  2. Form all bigram pairs by laying words side by side: (the, cat), (cat, eats), (eats, the), …
  3. Count how many times each bigram appears by stacking duplicate pairs.
  4. For the context word "the", calculate by hand: what fraction of all bigrams starting with "the" lead to each possible next word?

They write the results in a small table on paper. This table is the bigram language model — built with no code at all.

7.4 Trainingskorpus und Vokabular

7.5 Vorhersagen ohne Kontext

7.6 Der minimale Kontext (Bigramme)

7.7 Etwas mehr Kontext (Trigramme)

7.8 Textgenerierung mit Wahrscheinlichkeiten

7.9 Temperatur

7.10 Neuronale Netze und Sprachmodelle

HinweisExperiment als PDF zum Download

Dieses Experiment gibt es hier auch als kompakte PDF-Version.