Some questions with the toolkit about opendial HOT 3 CLOSED

Well, the flighbooking example is just a small example (with some naive, handcrafted NLU rules to make things simple), there are of course many other, more advanced ways to specify dialogue domains. OpenDial by itself does not enforce any structure on the dialogue domain, you're free to include any kind of state variables and probabilistic dependencies between them.

Actually, you can encode any arbitrary POMDP in OpenDial: simple create the initial state variables s_0, one large probability rule for the observation model P(o_t | s_t), one large probability rule for the transition model P(s_{t+1} | s_t, a_m), and one large utility rule for the Q-values Q(s_t,a_m). Each rule having its own set of unknown parameters of course.

This being said, even though the formalism can theoretically encode anything, in practice OpenDial is primarily meant for dialogue domains that do not have large amounts of dialogue data, and must therefore rely on expert knowledge to structure the dialogue domain and reduce the number of parameters to estimate. See my CSL paper for more information: http://nr.no/~plison/pdfs/cl/probrules-plison-csl2015.pdf

from opendial.

Taking the provided example-flight-booking.xml as an example, when the system requests me to input a destination, if I type in a date it will ignore it and keep asking for destination.

from opendial.

Oh that is right, thanks a lot.

from opendial.