TA works for the same reason that Pavlov's dog got fed when he (the dog) learned to ring the bell

He (the dog) got fed when he learned a particular action was related to a particular reward

Books have been published giving the probability of a future reward in the form of a price rise following various patterns on charts

If enough rewards result from trades based on the repetitive recognition of chart patterns it is reasonable to accept their predictive value ... otherwise it is not reasonable to rely on charts alone

[As we have discussed before I make no distinction between "probability" and "predictability". They are one and the same and I infer no greater probability to a prediction than I do to a probability. They are interchangeable terms with identical meaning]

In short if a reward follows often enough from a given chart pattern ... why not repeat the ringing of the bell i.e. the entry into the stock?