Before we get to truly defining conversational software we need to look at the different types of conversation. In our last post, we discussed the sematics of conversation and talking, and we decided that:

Conversation can be defined as any form of communication between two or more participants that consists of the exchange of thoughts, feelings, information, and other expressions that can sometimes (but not always) be presented in question/answer pairs.

I was careful to note that although we explicitly mentioned question/answer pairs, that wasn't the only way that conversation could be expressed. In a future post, we're going to look closer at question/answer and statement/response pairs, as they are easily implemented in conversational software such as chatbots, but first, we need to acknowledge that they aren't the only forms of conversation. If they were, we'd be done, and these posts that I'm writing would be pointless: Chatbots have solved this to a great degree (even if I do have some issues with how they handle it). But the reality is that there is more to a conversation than these sequential pairs, and as we start to look at the harder aspects of conversational software, these types of conversation are going to come up more and more.

Let's a take a look at them:

• Stories: Storytelling can be either a part of a conversation or a part of some other form of communication. For the latter, it most often is a one-way form of communication such as a play or a book reading. For the former, however, storytelling is a break in the normal turn-taking flow of a conversation, offering the communicator an extended opportunity to convey a message. This storytelling does not necessarily expect an equal response, or even an acknowledgement.

• Narratives: Wait? What's the difference between a story and a narrative? Consider narratives to be the boring version. A narrative is a systemic recitation of events. This is different from storytelling, which involves a greater depth of skill in both presentation and linguistics.

• Interviews: Interviews are certainly paired exchanges, no? It depends. Sometimes interviews are formulated in a way that offer an extended opportunity for the recipient of questions to enter into storytelling mode. Interview questions can sometimes be considered the acknowledgement and acceptance of the interviewer to enter into an extended story without interrupting the interviewee.

• Therapy: I've included therapy here because Eliza was originally a therapy chatbot, and I even built my own extremely limited therapy bot as an example of what can be done with IBM Watson's emotion detection. Therapy seems to always enter into the conversation when it comes to chatbots, and although therapy does tend to be a back-and-forth relationship, in this scenario, a conversation is flipped where the authority is asking questions, while the other engaged party is telling the narrative or story. Therapy requires a greater rapport between the two parties, with one party acting as the authority, while employing listening methodologies, and possibly psychotherapeutic methodologies that could impact the conversational flow.

• Assessments: Assessments are appraisals or evaluations of a specific thing. When it comes to an assessment, this is very similar to therapy only the authority is not an authority on psychological topics, but instead on external topics that require some form of an evaluation. In some ways, an interview (if it's a job interview) is an assessment with the interviewer assessing your skills. In this regard, you could potentially consider an interview to be a sub-type of an assessment. Another example of an assessment would be the appraisal of a car. One party is speaking to another party about value, and that second party is potentially the authority on that topic.

• Complaints: Most often, a complaint isn't a turn-taking conversation. It is often a one-sided conversation where the complainer is not necessarily telling a story, nor are they an authority on a topic, but are potentially ranting about an event, situation, interaction, or other such thing in a long-form turn during a conversation, regardless of the willingness of the other party.

These are just examples of conversational aspects that branch out beyond (or potentially extend) our traditional thought of back-and-forth conversations. These aren't static categories—in fact, some of these categories flow into the others and even involve some standard back-and-forth—but the point of this exercise is to get you thinking outside of what you see when you see a chatbot operate. If we're going to build an effective, useful chatbot that truly follows the conversational rules that we see in our society, we have to understand these variations and exceptions in order to properly process them.

In the next post, we're going to specifically get into areas of sociology that deal with speaking and conversation: Namely conversation analysis, discourse analysis, and some forms of linguistics. From this post we will detail our charted course on how a chatbot should operate, and which principles we should be following.