I think that to pass the Turing test, you have to tell the judges that the entity they are about to talk to *might* be a computer program. Eliza worked because people had never encountered a computer that even tried to be remotely human - so the assumption was that this was a real person from the outset. Also Eliza is a psychologist - so she gets to ask all the questions and steer the conversation into territory she can actually handle. Responses to things she can't parse are things like "So how does that make YOU feel?" - which work in that situation.
In a real turing test, the questions are completely open and the judge is initially highly sceptical that this is a real human.
Judges in these contests always seem to low-ball the questions. Ask "How would Santa Claus fend off a horde of attacking Ninjas?"
Those are insanely difficult questions for an AI to get right without some neutral "I don't feel like answering that right now" kind of response. A 13 year old kid would leap in and start wondering whether Santa could fly away in his sleigh and drop presents on them...or set the elves loose on them...or ask another question in return, like "Can the reindeer help out?"
Something that requires creativity - not just knowledge (which Watson could pull off) or a decent use of the English language (which Eliza could manage to some degree).