A human Dialogue is a human to human interaction.

turn

each contributino to a conversation is called a “turn”, which contains a sentence, multiple sentences, or a single word

turn-taking

• when to take the floor
• who takes the floor
• what happens during interruptions?

barge-in

barge-in is the property to allow the user to interrupt the system

end-pointing

deciding when a human has stopped talking, compute, etc.

speech-act

each turn is actually an “action” performed by the user

Dialogue State Architecture uses dialogue acts instead of simple frame filling to perform generation; used currently more in research.

• NLU: slot fillers to extract user’s utterance, using ML
• Dialogue State Tracker: maintains current state of dialogue
• Dialogue policy: decides what to do next (think GUS’ policy: ask, fill, respond)—but nowaday we have more complex dynamics
• NLG: respond

dialogue acts

dialogue acts combines speech-acts with underlying states

slot filing

we typically do this with BIO Tagging with a BERT just like NER Tagging, but we tag for frame slots.

An

We have a function:

\begin{equation} |x|+|y|\frac{dy}{dx} = \sin \left(\frac{x}{n}\right) \end{equation}

We are to attempt to express the solution analytically and also approximate them.

To develop a basic approximate solution, we will leverage a recursive simulation approach.

We first set a constant \(N\) which in the \(N\) value which we will eventually vary.

N = 0.5

We can get some values by stepping through \(x\) and \(y\) through which we can then figure \(\frac{dy}{dx}\), namely, how the function evolves.

Generative Classifier

A Generative Classifier builds a good model of a class, and use that to assign how “class-y” is that image.

For instance, to categorize cats vs. dogs, we build a cat model and dog model. To classify, then, we see if a particular image is more “cat-y” or “dog-y”.

Discriminative Classifier

A Discriminative Classifier observes the differences between two classes, instead of trying to model each one.