nGramme

NGramme is a term used in computational linguistics to denote a family of techniques and software resources that analyze text by counting occurrences of n-grams—contiguous sequences of n tokens or characters—from a corpus. It covers language-model construction, feature extraction for machine learning, and analytical workflows that rely on n-gram statistics.

NGramme methods can operate at the word level or the character level. They estimate probabilities such as P(w|history) from observed counts, using smoothing to handle unseen histories. Common approaches include additive smoothing, back-off, and interpolation. NGramme tools may support fixed-length and variable-length models and provide evaluation metrics such as perplexity.

A typical NGramme pipeline includes text ingestion, tokenization, n-gram extraction, model training, and scoring. Implementations store n-gram catalogs in efficient data structures and offer APIs for integration into machine-learning workflows. Some frameworks prioritize rapid experimentation on large corpora, while others optimize for memory usage and speed.

Practical applications include predictive text and autocomplete, spell-checking, information retrieval ranking, authorship attribution, language identification, and preprocessing for neural models where n-gram features complement deep representations.

N-gram concepts emerged in statistical natural language processing in the latter part of the 20th century and remain foundational alongside newer neural methods. Today, NGramme resources are used for benchmarking, exploratory data analysis, and hybrid systems that combine n-gram statistics with neural models.

Although the spelling NGramme with a capital G is sometimes used to brand libraries or projects in the field, it is not a standardized term. In practice, NGramme serves as a generic label for n-gram based analysis rather than a single, specific product.

Related topics include n-gram, language model, smoothing, and information retrieval.