The paper formulates what it claims is the central problem in particle typology: the widespread particle functions that seem to have little to no impact on truth conditions or on the definition of speech acts. The classical case is additivity, but there are many more cases. The problem is to explain why these functions are lexicalised by particles in so many languages, how they can grammaticalise in the face of their seeming uselessness and why their occurrences are mostly obligatory. The paper pursues a solution based on Minimal Bayesian Pragmatics (MBP), where the interpretation of verbal utterances is equated with finding the most probable subjective causal explanation of the utterance: the problematic particle functions turn out to be vital in preventing erroneous pragmatic enrichments/reinterpretations that would increase the probability of the interpretation.