quadgram

A quadgram is a sequence of four consecutive symbols drawn from a text. In natural language processing and text analysis, quadgrams are a specific case of n-grams with n equal to four. They are commonly defined over characters, though in some applications they may be defined over tokens or syllables.

In language modeling, quadgrams are used to estimate the probability of text using a fourth-order Markov assumption. The frequency of each quadgram is counted in a large corpus, and probabilities are derived from these counts, often converted to log probabilities for numerical stability. To cope with unseen quadgrams, smoothing techniques such as Laplace smoothing or more advanced methods like Kneser-Ney are employed. It is also common to include sentence boundaries or start markers to better capture context.

Quadgrams are used in various applications including predictive text input, spelling and OCR post-processing, language identification, and stylistic analysis. They provide more context than trigrams but require more data to estimate reliably.

In cryptography, quadgram statistics are used to score candidate decryptions of monoalphabetic or polyalphabetic substitution ciphers. A quadgram score, derived from a large corpus of English text, assigns higher scores to decryptions that produce plausible English quadgrams, aiding search-based cryptanalysis.

Limitations include sparsity of data for many quadgrams, high dimensionality, and sensitivity to tokenization and text genre. Effective use often relies on smoothing, domain adaptation, and, in NLP, backing off to lower-order n-grams when data are scarce.