That bi-gram probabilistic models infer transitional probabilities between word pairs from datasets is a foundational definition within computational linguistics and natural language processing, requiring no citation.
The claim states a definitional and foundational fact about bigram probabilistic models in computational linguistics and statistics: they estimate transitional probabilities between word pairs (bigrams) based on training data. Because this is a basic, definitional property of bigrams rather than an empirical discovery subject to debate, it falls under common knowledge.