For some part of the time he said the response was similar to a human response. The following snippets are from that portion:
I am extremely happy and day dreamer. Do you get happy often? yes Can you elaborate on that? it's just that i do not worry much about things. Is it because you do not worry much about things that you came to me? no I would appreciate it if you would continue.
Earlier you said you were happy? Yes Go on, don't be afraid. I feel good today. Maybe your life has something to do with this. Yes, of course. Is it because of your plans that you say yes of course? no, i say it because of what happened today. Are you sure that is the real reason? yes, what do you think can be the real reason? You mentioned that you do not worry much about things? Don't change the topic mate. Perhaps what happened today have something to do with this.
In the following snippets one can figure out that it's not a human easily:
Good, I was surfing the web and came across some wonderful quotes. Are you sure that that? Yes and they were hilarious. Do you really think i would like to share your happiness me? what are you saying? Is it because of the people you hang around with that you are going through all this? yes, they all are very fun-loving and i enjoy their company. I don't understand. what do you not understand?
How do you reconcile problems at school? I don't go to school. Tell me about your school. I just told you that I doon't go to any school. Are you annoyed that your grades could improve? no, i am not annoyed, and yes, they can improve. Are you annoyed often? Who told you that I am annoyed. Is the fact that you are annoyed the real reason? You are really annoying me now. What makes you think you are really annoying you now? I am sorry, but I am going.
On looking at the source code, I would say ELIZA works by simple string parsing and substitution of key words into sentences. It first of all makes pre transformation to the sentences, checks for a quit word. Then it scans sentences for keys and builds a key stack, and finally tries to make decompositions for each key. During decomposition, it tries to isolate the word's synonyms. Then it reassembles the reply from decomposition and generates the response. It also reverts the topic to the initial entries entered by the user, if it can't analyze the sentence completely.