Consider two similar, but DIFFERENT questions.
1. A roulette wheel is spun 3800 times and the results are recorded as red or black or green. Then the green results are removed, creating a string of r’s and b’s, roughly 3600. Then that string is broken down, sequentially, into pairs. Then one of those pairs is selected at random. If we know one of the results in that randomly selected pair is red, what is the chance both results in the pair are red? (Answer = 1/3).
2. A roulette wheel has been spun twice. We are informed by a reliable source that neither result was green and one of the two results is red. What is the chance both results are red? (Answer = 50%). Does it matter whether the information packet we received contained the sequence of the red result? (Answer = No). Does it matter whether the information packet we received contains the ball number of the red result? (Answer = No).
Likewise with the first two animals into the first pen. Likewise with the last two animals left in the corral. If one is female, the chance both are female is 50%.
It is only when the pair is selected by random sampling, from a large population of similar pairs, that the answer is 1/3.
Frank’s question posited one woman and two children. Therefore, Frank was asking question 2, and the answers, given the information provided, were 25%, 50%, and 50%.
And, yes, what we have here is failure of communication. "Failure to communicate" implies a locus for the failure.