Spoken Dialogue System based on POMDP having States about User's Grammatical Knowledge

1. Partially Observable Markov Decision Process (POMDP) [1] POMDP POMDP POMDP [2] 2. POMDP 2.1 POMDP POMDP 1 POMDP s s p(s) b(s) a POMDP b a π POMDP a 0 = π(b 0) b 0 s b 0 (s) = 1 |S| a 0 s 0 p(s 1 |s 0 , a 0) s 1 o 1 p(o 1 |s 1 , a 0) r(s 0 , a 0) b (s) = k · p(o 1 |s , a 0) ∑ s p(s |a 0 , s)b(s) (1) k ∑ s b(s) = 1 POMDP [1] POMDP 2.2 POMDP POMDP POMDP POMDP Composite Summary Point-Based Value Iteration (CSPBVI) [3] 3. POMDP POMDP [1] s (s = ()) s u g s = (s u , g) (2) s = ((,), ToDest) a o b (1) b (s) = k · p(o |s , a) × (3) ∑ s u ∑ g p(s = k · p(o |s , a) × (4) ∑ s u ∑ g p(s = k · p(o |s , a) ∑ g p(g |a)b(s u , g) (5) (5) p(s u |a, s u , g) s u = s u 1 0 s u [4]

[1]  Tatsuya Kawahara,et al.  User Modeling in Spoken Dialogue Systems to Generate Flexible Guidance , 2004, User Modeling and User-Adapted Interaction.

[2]  Steve J. Young,et al.  Partially observable Markov decision processes for spoken dialog systems , 2007, Comput. Speech Lang..

[3]  Steve Young,et al.  Statistical Approaches To Dialogue Systems (tutorial, including videos) , 2009 .

[4]  S. Young,et al.  Scaling POMDPs for Spoken Dialog Management , 2007, IEEE Transactions on Audio, Speech, and Language Processing.