Sample

Trigram Language Model

Here we sample from the conditional probability distributions for a trigram language model trained on 30,000 names. This is a character-level model with 0.1 smoothing. The heatmap and the graphs show the probabilities of the next character given the previous two characters. Roll over the heatmap or graphs (or tap on mobile) to see the probabilities. Optionally choose a couple of starting characters, then press sample to generate a random number to sample from the conditional probablities shown for the next character. The display will update with the result of your sampling. Then press or tap 'Add sampled character' to add the character to the names you are building. Press 'Reset' to start over. (The \n character is shown as ↵)