I have to correct myself. You do not want a 50% chance, because the answer does not provide 1Bit of information. If we get it right, we also know the position(s) of the letter.
More precisely: Given a dictionary of all possible words and a letter to guess, we can split the dictionary into k parts. One sub-dictionary contains all words for which the answer is negative. Additionally, we have k-1 sub-dictionaries for each equivalence class of letter positions.
For each letter, we can compute the partition. Now we need to choose the strategically best partition. I believe it should be the one with "the lowest average sub-dictionary size".
More precisely: Given a dictionary of all possible words and a letter to guess, we can split the dictionary into k parts. One sub-dictionary contains all words for which the answer is negative. Additionally, we have k-1 sub-dictionaries for each equivalence class of letter positions.
For each letter, we can compute the partition. Now we need to choose the strategically best partition. I believe it should be the one with "the lowest average sub-dictionary size".