1. Using Reinforcement Learning to Build a Better Model of Dialogue State Joel Tetreault & Diane Litman University of Pittsburgh LRDC April 7, 2006

2. Problem Problems with designing spoken dialogue systems:  What features to use?  How to handle noisy data or miscommunications?  Hand-tailoring policies for complex dialogues? Previous work used machine learning to improve the dialogue manager of spoken dialogue systems [Singh et al., ‘02; Walker, ‘00; Henderson et al., ‘05] However, very little empirical work on testing the utility of adding specialized features to construct a better dialogue state

3. Goal Lots of features can be used to describe the user state, which ones to you use? Goal: show that adding more complex features to a state is a worthwhile pursuit since it alters what actions a system should make 5 features: certainty, student dialogue move, concept repetition, frustration, student performance All are important to tutoring systems, but also are important to dialogue systems in general

4. Outline Markov Decision Processes (MDP) MDP Instantiation Experimental Method Results

5. Markov Decision Processes What is the best action an agent to take at any state to maximize reward at the end? MDP Input:  States  Actions  Reward Function

6. MDP Output Use policy iteration to propagate final reward to the states to determine:  V-value: the worth of each state  Policy: optimal action to take for each state Values and policies are based on the reward function but also on the probabilities of getting from one state to the next given a certain action

7. What’s the best path to the fly?

8. MDP Frog Example Final State: +1

9. MDP Frog Example Final State: +1 -2 0 0 -3

10. MDP’s in Spoken Dialogue MDP Dialogue System Training data Policy User Simulator Human User MDP works offline Interactions work online

11. ITSPOKE Corpus 100 dialogues with ITSPOKE spoken dialogue tutoring system [Litman et al. ’04]  All possible dialogue paths were authored by physics experts  Dialogues informally follow question-answer format  50 turns per dialogue on average Each student session has 5 dialogues bookended by a pretest and posttest to calculate how much student learned

13. Corpus Annotations Manual annotations:  Tutor and Student Moves (similar to Dialog Acts) [ Forbes-Riley et al., ’05]  Frustration and certainty [Litman et al. ’04] [Liscombe et al. ’05] Automated annotations:  Correctness (based on student’s response to last question)  Concept Repetition (whether a concept is repeated)  %Correctness (past performance)

14. MDP State Features FeaturesValues CorrectnessCorrect (C) Incorrect/Partially Correct (I) CertaintyCertain (cer), Neutral (neu), Uncertain (unc) Student MoveShallow (S), Deep/Novel Answer/Assertion (O) Concept RepetitionNew Concept (0), Repeated (R) FrustrationFrustrated (F), Neutral (N) % Correctness50-100% (H)igh, 0-49% (L)ow

15. MDP Action Choices CaseTMoveExample Turn FeedPos“Super.” NonFeedHint, Ques.“To analyze the pumpkin’s acceleration we will use Newton’s Second Law. What is the definition of the law?” MixPos, Rst, Ques. “Good. So when the truck and car collide they exert a force on each other. What is the relationship between their magnitudes?”

16. MDP Reward Function Reward Function: use normalized learning gain to do a median split on corpus: 10 students are “high learners” and the other 10 are “low learners” High learner dialogues had a final state with a reward of +100, low learners had one of -100

17. Infrastructure 1. State Transformer:  Based on RLDS [Singh et al., ’99]  Outputs State-Action probability matrix and reward matrix 2. MDP Matlab Toolkit (from INRA) to generate policies

18. Methodology Construct MDP’s to test the inclusion of new state features to a baseline:  Develop baseline state and policy  Add a feature to baseline and compare polices  A feature is deemed important if adding it results in a change in policy from a baseline policy (“shifts”) For each MDP: verify policies are reliable (V-value convergence)

19. Hypothetical Policy Change Example B1 StatePolicyB1+Certainty State 1[C]Feed[C,Cer] [C,Neu] [C,Unc] 2[I]Feed[I,Cer] [I,Neu] [I,Unc] +Cert 1 Policy  Feed Mix +Cert 2 Policy Mix Feed Mix NonFeed Mix 0 shifts5 shifts

20. Tests +%Correct +Goal +Frustration B2+ Correctness+Certainty Baseline 1 Baseline 2 B1+ +SMove

21. Baseline Actions: {Feed, NonFeed, Mix} Baseline State: {Correctness} Baseline network [C] FINAL F|NF|Mix [I] F|NF|Mix

22. Baseline 1 Policies Trend: if you only have student correctness as a model of student state, regardless of their response, the best tactic is to always give simple feedback #StateState SizePolicy 1[C]1308Feed 2[I]872Feed

23. But are our policies reliable? Best way to test is to run real experiments with human users with new dialogue manager, but that is months of work Our tact: check if our corpus is large enough to develop reliable policies by seeing if V-values converge as we add more data to corpus Method: run MDP on subsets of our corpus (incrementally add a student (5 dialogues) to data, and rerun MDP on each subset)

24. Baseline Convergence Plot

25. Methodology: Adding more Features Create more complicated baseline by adding certainty feature (new baseline = B2) Add other 4 features (student moves, concept repetition, frustration, performance) individually to new baseline Check that V-values converge Analyze policy changes

26. Tests +%Correct +Goal +Frustration B2+ Correctness+Certainty Baseline 1 Baseline 2 B1+ +SMove

27. Certainty Previous work (Bhatt et al., ’04) has shown the importance of certainty in ITS A student who is certain and correct, may not need feedback, but one that is correct but showing some doubt is a sign they are becoming confused, give more feedback

28. B2: Baseline + Certainty Policies B1 StatePolicyB1+Certainty State +Certainty Policy 1[C]Feed[C,Cer] [C,Neu] [C,Unc] NonFeed Feed NonFeed 2[I]Feed[I,Cer] [I,Neu] [I,Unc] NonFeed Mix NonFeed Trend: if neutral, give Feed or Mix, else give NonFeed

29. Baseline 1 and 2 Convergence Plots

30. Tests + %Correct +Goal +Frustration B2+ Correctness+Certainty Baseline 1 Baseline 2 B1+ +SMove

31. % Correct Convergence Plots

32. Student Move Policies B2B2 PolicyB2 +SMove+Smove Policy 1[Cer,C]NonFeed [Cer,C,S] [Cer,C,O] NonFeed Feed 2[Cer,I]NonFeed [Cer,I,S] [Cer,I,O] Mix 3[Neu,C]Feed [Neu,C,S] [Neu,C,O] Feed NonFeed 4[Neu,I]Mix [Neu,I,S] [Neu,I,O] Mix NonFeed 5[Unc,C]NonFeed [Unc,C,S] [Unc,C,O] Mix NonFeed 6[Unc,I]NonFeed [Unc,I,S] [Unc,I,O] Mix NonFeed Trend: give Mix if shallow (S), give NonFeed if Other (O) 7 Changes

33. Concept Repetition Policies Trend: if concept is repeated (R) give complex or mix feedback B2B2 PolicyB2 +Concept+Concept Policy 1[Cer,C]NonFeed [Cer,C,O] [Cer,C,R] NonFeed Feed 2[Cer,I]NonFeed [Cer,I,O] [Cer,I,R] Mix 3[Neu,C]Feed [Neu,C,O] [Neu,C,R] Mix Feed 4[Neu,I]Mix [Neu,I,O] [Neu,I,R] Mix 5[Unc,C]NonFeed [Unc,C,O] [Unc,C,R] NonFeed 6[Unc,I]NonFeed [Unc,I,O] [Unc,I,R] NonFeed 4 Shifts

34. Frustration Policies Trend: if student is frustrated (F), give NonFeed B2B2 PolicyB2 +Frustration+Frustration Policy 1[Cer,C]NonFeed [Cer,C,N] [Cer,C,F] NonFeed Feed 2[Cer,I]NonFeed [Cer,I,N] [Cer,I,F] NonFeed 3[Neu,C]Feed [Neu,C,N] [Neu,C,F] Feed NonFeed 4[Neu,I]Mix [Neu,I,N] [Neu,I,F] Mix NonFeed 5[Unc,C]NonFeed [Unc,C,N] [Unc,C,F] NonFeed 6[Unc,I]NonFeed [Unc,I,N] [Unc,I,F] NonFeed 4 Shifts

35. Percent Correct Policies 3 Shifts Trend: if student is a low performer (L), give NonFeed B2B2 PolicyB2 +%Correct+%Correct Policy 1[Cer,C]NonFeed [Cer,C,H] [Cer,C,L] NonFeed 2[Cer,I]NonFeed [Cer,I,H] [Cer,I,L] Mix NonFeed 3[Neu,C]Feed [Neu,C,H] [Neu,C,L] Feed 4[Neu,I]Mix [Neu,I,H] [Neu,I,L] NonFeed Mix 5[Unc,C]NonFeed [Unc,C,H] [Unc,C,L] Mix NonFeed 6[Unc,I]NonFeed [Unc,I,H] [Unc,I,L] NonFeed

36. Discussion Incorporating more information into a representation of the student state has an impact on tutor policies Despite not having human or simulated users, can still claim that our findings are reliable due to convergence of V-values and policies Including Certainty, Student Moves and Concept Repetition effected the most change

37. Future Work Developing user simulations and annotating more human-computer experiments to further verify our policies are correct More data allows us to develop more complicated policies such as  More complex tutor actions (hints, questions)  Combinations of state features  More refined reward functions (PARADISE) Developing more complex convergence tests

38. Related Work [Paek and Chickering, ‘05] [Singh et al., ‘99] – optimal dialogue length [Frampton et al., ‘05] – last dialogue act [Williams et al., ‘03] – automatically generate good state/action sets

39. Diff Plots Diff Plot: compare final policy (20 students) with policies generated at smaller cuts