Probabilistic language models have gained popularity in Natural Language Processing due to their ability to successfully capture language structures and constraints with computational efficiency. Probabilistic language models are flexible and easily adapted to language changes over time as well...

When a trigram backoff language model is created from a large body of text, trigrams and bigrams that occur few times in the training text are often excluded from the model in order to decrease the model size. Generally, the elimination of n-grams with very low counts is believed to not...