Clarification

The DS could fail to recognize the speaker's intent should he/she not provide enough information. For example, supposing someone says, "Can I have a ticket from New York?" the DS will ask him/her, "Where is your destination?" rather than respond with "Congratulations! You have successfully bought a ticket from New York." This is because whenever the DS is confused about the speaker's intention, it will ask him or her some questions to decide what he/she really wants, a process we call clarification. There are three situations when the system might ask the speaker for clarifications so as to keep the conversation going:

1. When a speaker fails to provide the system with all the required beliefs, the system will ask him/her to give the missing information.

Let's suppose the app is expecting an intent as figure 1. And the user says "Remind me to turn my cell phone off."  Now the system notices two pieces of information are missing: time and place.  Then the app will ask the user something like: "When and where do you want me to remind you of this.", so as to get more information. 

 Fig. 1. An intent.

2. The system has found two intents that are both matched.

Let's suppose there are two intents, intentA and intentB, referenced by planA and planB, respectively. (See Figure 2 below. ) So if the user says "The first one", both plans are able to be triggered, which may cause conflict. One solution is to look at the priority field. Hence, whichever plan havig the highest priority will be triggered. 

However, if the priorities of the above two plans are, unfortunately, the same, the system will undergo a process of clarification. And the questions it will be asking are based on the function names of the plans.

Fig. 2. Two intents.

3. The system has successfully found an intent, of which the confidence, however, is not high enough.

Take the sentence in Figure 1 as an example. If the user says "Remind me to turn my cell phone on tomorrow." The DS might guess this utterance matches the reminder intent. We say the DS guesses the user's intent because the intent and the utterance are not perfectly matched. (The user should have said "turn ... off" instead of "turn ... on". And there was no place information provided) In this case, the DS will calculate a confidence value of this match. If the confidence is too low, the system will undergo the clarification process.