4Gram Interactive Demo
4-Gram Language Models
in your browser.
Train, tune, and evaluate a classical 4-gram text generator. Compare backoff with interpolation, inspect perplexity, and watch generated text form one word at a time.
Pipeline
Corpus
Tokenize
Limit vocab
Train 4-gram
Perplexity
Generate
LM1 Backoff
Shorter context fallback
When a 4-gram is unseen, LM1 backs off to trigram, bigram, then unigram counts.
Inspect counts, probabilities, perplexity, and generated output.
LM2 Interpolation
Weighted mixture with smoothing
LM2 mixes all n-gram orders with add-k smoothing, so unseen contexts still receive probability.
Inspect counts, probabilities, perplexity, and generated output.