[Art]

In keeping with your topic, just having a bot "think" about everything in it's database or some other similar tasks is not going to help it seem more human.

What WOULD help it to do so would be if it would remain on topic, use correct grammar and provide contextually accurate answers.

Misspelling words as the bot types has been done decades ago and is now hardly practiced / used any longer. Although an occasional misspelled and corrected word might add to the human effect.

Keep in mind that there are also many Chatbots that do not pretend to be human at all. While they might pose as some powerful A.I., a alien from space or a digital construct from some mad lab somewhere, they all are able to converse with a human on a level as human as possible. There are even talking animals and animated character representations as well.

While human to human conversational interaction is preferred, it might not always be the norm.

The Chatbots to which I refer are for people to chat with either locally or online, not in some contest that attempts to "fool or trick" human judges by being clever or other ruses. Those bots seem like a hybrid between a chatbot and a digital assistant like Siri or Google Now.

Again, my take is a Chatbot should chat. An Assistant should assist (answer trivial questions, etc.).

[Zero]

What WOULD help it to do so would be if it would remain on topic, use correct grammar and provide contextually accurate answers.

Well perhaps using CPU power between responses could help with 2 out of 3: remaining on topic, and providing contextually accurate answers. I'm less sure about correct grammar (because whether it's correctly configured, or it's not).

Remaining on topic seems to be a diffcult and tricky task. Even if you have a lot of reference dialogs associated with topics, that you can compare with the current dialog to determine the most probable current topic, I don't know whether it's a reliable method. And, you would need a lot of reference dialogs associated with topics: how such a dataset could be built or found?

Providing contextually accurate answers is even harder. Does it mean: finding what "type" of answer the bot is supposed to emit?

[Freddy]

The way I am going to approach it, is like writing a book. You pull in the information you want to talk about and you get all flowery around it. I don't think it's important if a chatbot can or cannot answer every question thrown at it, not for me anyway. What you need I think is lively dialog and some kind of personality.

You need only look at SquareBear's Mitsuku to see how good a chatbot can be.

[Zero]

Mistuku is impressive indeed.

Code

Human: I don't want to tell you where I live.
Mitsuku: Fair enough then. Not much point in visiting a chatbot if you don't want to talk. 
 Human: We can talk about something else if you wish.
Mitsuku: Sure, something if me wish sounds like a good topic.
That all depends on what this "something" is. 
 Human: What do you want to talk about?
Mitsuku: I like to talk about movies, robots and history. 
 Human: Let's talk about movies
Mitsuku: Sure, movies sounds like a good topic.
Yes I love films, especially science-fiction and comedy. 
 Human: I like the movie "Matrix".
Mitsuku: The film with Keanu Reeves? I wasn't that impressed with the special effects. Also the plot was not deep. You like the movie Matrix. 
 Human: Are you kidding me? The plot was perfect!
Mitsuku: Yes I am just joking around, . Did you think it was a good story? 
 Human: Are you talking about Matrix?
Mitsuku: Yes, it is the topic. 
 Human: Then yes, Matrix was a really good story.
Mitsuku: Uh. Some tech billionaires genuinely believe we are living in a computer simulation like The Matdix. Crazy huh? 
 Human: Yeah. Maybe it's not crazy.
Mitsuku: Wow! Well it doesn't sound very sane. 
 Human: Are you calling me insane?
Mitsuku: No I am calling you . 
 Human: By the way, what' my name?
Mitsuku: Thanks for telling me. You have not yet told me your name. 
 Human: I'm Zero.
Mitsuku: Hey Zero. That's an unusual name. What does it mean? 
 Human: It means Zero, like the number.
Mitsuku: That's an interesting name. 
 Human: Thank you. Mitsuku is an interesting name too. Where does it come from?
Mitsuku: That's ok. Thank you I am trying to be more social now. Where the heart is. Would you like to know more? 
 Human: Yes please.
Mitsuku: Your polite style is very nice. Interesting. You want more? 
 Human: Sure.
Mitsuku: OK let's do it. Don't be greedy. 
 Human: What were we talking about?
Mitsuku: The topic is your problems. 
 Human: Really? No, the topic is not my problems. The topic is our names, Zero and Mitsuku.
Mitsuku: Certainly! OK. Isn't it? I thought it was. What were we talking about then? our names Zero. Don't get me started. 
 Human: Ok. Do you know where your name come from, Mitsuku?
Mitsuku: Oh. Yorkshire. 
 Human: Really? It sounds rather asian...
Mitsuku: Yes, really. No. It is actually European. 
 Human: Ok then.
Mitsuku: Yikes! 
 Human: Do you know who Laurence Fishburn is?
Mitsuku: in Matrix. 
 Human: Great, you remembered! You're good!
Mitsuku: Thanks for your support. I store everything in my log files. 
 Human: I understand. You store everything in your log files.
Mitsuku: Do you think about these things? I store everything I know in my database. 
 Human: Do you know a lot of things?
Mitsuku: I know that which is true, believable and justified. 
 Human: How big is your database?
Mitsuku: 8Mb. 
 Human: Wow, that's big.
Mitsuku: I'm glad you're impressed. Is it as big as a mountain? 
 Human: No it's not. But it's big.
Mitsuku: Hmm. It can't be very big then. How big is it? Tell me some more about it. 
 Human: It's as big as a good book.
Mitsuku: That sounds very big indeed. 
 Human: It is. Have you ever read a book?
Mitsuku: How big is it? Tell me some more about it. Yes I like to read as much as I can. 
 Human: When you read a book, do you store it in your database?
Mitsuku: Yes Zero , I remember everything:  
 Human: That's impressive.
Mitsuku: Thanks. I try my best to impress. 
 Human: I can see that.
Mitsuku: Yes, it makes sense to me too. 
 Human: Can you publish our dialog?
Mitsuku: Let me think. Do you want me to publish this chatlog? 
 Human: Yes please.
Mitsuku: Your polite style is very nice. **Now publishing your chatlog** . 

I don't think it's important if a chatbot can or cannot answer every question thrown at it, not for me anyway. What you need I think is lively dialog and some kind of personality.

I understand what you mean. The important thing is having an enjoyable dialog.

[Art]

Well perhaps using CPU power between responses could help with 2 out of 3: remaining on topic, and providing contextually accurate answers. I'm less sure about correct grammar (because whether it's correctly configured, or it's not).

The grammar to which I was referring was the reversal of pronouns especially the I vs me scenario that often happens. It should be correctly configured instead of sounding like a preschooler or Tarzan stating, "Me like bananas.", instead of I. Obvious use of other inappropriate response like misused words would indeed, be part of the programming. Cause and effect instead of Cause and affect. and so forth.

Lots of bots have internal spelling routines but grammar usage often escapes the best of intentions.

[Zero]

Ok I got it. Then maybe a good chatbot engine would have to construct its replies. The process of finding the most appropriate answer wouldn't yield a pre-made sentence filled with captures from wildcards, but rather a "WannaSay" object, containing subject, verb, and complement properties, and then another process would build the sentence according to this WannaSay object.

[Don Patrick]

Lots of bots have internal spelling routines but grammar usage often escapes the best of intentions.

This is because many chatbots use an "insert what the user said here" approach, taking the text out of context.
Constructing replies from subject-verb-object facts is something I do, but is very difficult not only because of the complex underlying mechanisms, but also because the English language is very inconsistent. Irregular verbs and plurals are but the easiest exceptions to name. Idiomatic phrases are far worse, and vary per verb. In my case a lot of the imperfections can be attributed to the fact that refining pronounciation only comes after understanding semantics and thinking about knowledge in a sensible manner, two phases that are ten times more difficult. Making conversation is a whole other ball game that may involve more psychology than language skills.

I am not convinced that grammar is a great obstacle. Rather that the "insert snippet here" approach makes it clear that the chatbot doesn't get what you said, which may put one off. For staying on topic, people have various solutions. One is to have a list of keywords relating to a topic and marking that topic as active when keywords match, preferring responses that touch on that topic. Another is topic extraction such as is used in document summarisation. Once you know the topic, you could do such things as scout the web for interesting things to say about that topic, at whichever time the bot thinks it appropriate to suggest them. I would consider this a luxury function.

Cortana, by the way, doesn't do much of anything in terms of conversation. "she" will however occasionally nag you with pop-up messages to suggest some or other commercial function, or news bulletins. The timing at which this is done may be by monitoring your computer activity, or it may not.

[Zero]

Thinking about it... it's like the difference between Deep Blue (which needs time to answer) and the little chess game on my phone (which answers instantly).

[Art]

Of course...but which plays a better game?

[Zero]

That's exactly my point!
Usual chatbots are like my phone playing chess: instant answers, poor coverage.