Dialogue systems Luděk Bártek Finite State Dialogue System Analysis Dialogue systems User Modelling Dialogue and Strategy Games FSA models Application for Generating the Dialogue Interfaces Luděk Bártek Laboratory of Searching and Dialogue, Fakulty of Informatics, Masaryk University, Brno spring 2023 Finite State Dialogue System Analysis Motivation Dialogue systems Luděk Bártek Finite State Dialogue System Analysis ■ Dialogue strategies and user model unification. ■ Affective computing - emotion processing. User Modelling Dialogue and ■ Other research area compatibility: Strategy Games FSA models Application for Generating the Dialogue Interfaces ■ the formal languages and automata theory ■ the game theory ■ the universal algebra ■ ... ■ Allows to predict the dialogue flow. ■ VoiceXML compatibility. 1 -f) <\(y Dialogue Mathematical Model Dialogue systems Luděk Bártek Consists of: two communicating parties, their utterances that alternate. User Modelling Dialogue and Strategy Games FSA models Application for Generating the Dialogue Interfaces n 4 □ ► S1 = Pragmatics of Dialogue Dialogue systems Luděk Bártek User Modelling Dialogue and Strategy Games We analyse the dialogue from the point of view of: ■ speech act theory ■ lexical semantic a pragmatic ■ dynamic semantic ■ dynamic epistemic logic (cognition logic) - the utterance meaning may change through the time and speaker need not be sure for 100% with the content. ■ conversation theory Pragmatic may contain next: ■ user(s) model(s) ■ environment model Dialogue Pragmatic Speech Acts Theory View Dialogue systems Luděk Bártek User Modelling Dialogue and Strategy Games speech act - action performed by language: Utterance: "It's 9 o'clock." Utterance pragmatic - Current time vs. It's too soon/too late and some action should be performed. Dialogue Pragmatic The View of Lexical Semantic and Pragmatic Dialogue systems Luděk Bártek User Modelling Dialogue and Strategy Games Lexical semantic - the meaning of utterance based on the other parts of the speech (words, ...) Lexical pragmatic - the meaning of of the utterance in the context of the speech ■ previous utterances ■ non-verbal parts of communication Dialogue Pragmatic The Dynamic Semantic and The Dynamic Epistemic Logic View Dialogue systems Luděk Bártek User Modelling Dialogue and Strategy Games Dynamic semantic ■ takes anaphora into the account (object referencing in the text: "I've seen it,") ■ the dynamics of meaning changes in discourse (Using the language to achieve the intended pragmatic meaning, the speaker utterance in the meaning of the dialogue) initiated by the gradual exposure of anaphora. Dynamic epistemic logic - cognition logic: ■ the truthfulness of a proposition is developing through the time, ■ the speaker need not to be sure for 100%. Dialogue Pragmatic Example Dialogue systems The pragmatic containing the models of users: Luděk Bártek Finite State Dialogue System Analysis User Modelling Dialogue and Strategy Games FSA models Application for Generating the Dialogue Interfaces The pragmatic containing the models of users and the environment: Dialogue Simulation by Pawlak's Information System Dialogue systems Luděk Bártek ■ Pawlak's IS S = (X, 7, V,f): Finite State Dialogue System Analysis User Modelling ■ X - set of IS objects. ■ T - set of objects attributes. ■ V - set of attributes values. Dialogue and Strategy Games FSA models Application for Generating the Dialogue Interfaces f:XxT->V ■ Example - trading on the market: ■ Attributes: ■ proposed price: -, 1 — 1000 ■ non-verbal attitude: :-), :-, :-(, : :-> ■ Possible dialogue flow: (1000, :-)), (600, :-(), (800,:-), (-,:->), (700, : (500, :-), (900, :-)), :-(), (700,:-)) □ Dialogue Modelling Mealy Automaton Dialogue systems Luděk Bártek User Modelling Dialogue and Strategy Games Mealy automaton M - finite state automaton with output. M = {Q,X, Y,X,so) ■ Q - non-empty set of states ■ X - finite input alphabet ■ Y - finite output alphabet ■ A - transition function: A: QxX^Qx Y so - initial state. User Modelling Poker Dialogue systems Luděk Bártek User Modelling Dialogue and Strategy Games Poker is a card game where the player bet on their cards. Basic rules: Banker, the person who is handing cards, deals 5 cards to each player. Players based on the cards in their hand can: ■ bet money with possibility of exchange the cards that don't suite them and either to call or to increase the bet. ■ fold the cards. As soon as no one raise the bet (call/fold) the game is over. The game is over in the case when only one player doesn't fold. The player with highest cards wins the entire staked money. Poker Continue Dialogue systems Luděk Bártek User Modelling Dialogue and Strategy Games This game has a lot of variants.. When player use a good strategy he may win although his cards are bad: ■ minimization of non-verbal expressions ■ well chosen strategy of raising, that gives the impression that cards are very good ■ avoiding similar strategy in the following games - the other players shouldn't be able to observe players common behaviour features in some situations. Poker Finite State Analysis alä dialogue Dialogue systems Luděk Bártek Finite State Dialogue System Analysis ■ Game's state attributes (sets As)\ User Modelling Dialogue and Strategy Games FSA models Application for Generating the Dialogue Interfaces ■ ASi - my cards ■ AS2 - the total amount of the bet ■ ASs - trust in own cards ■ AS4 - expected opponent's cards ■ AS$ - player's strategy 1 -f) <\(y Poker Game State Attributes Values Dialogue systems Luděk Bártek User Modelling Dialogue and Strategy Games my cards - ASi G {bad, medium, good} the total amount of the bet - AS2 G {high, medium, low} trust in own cards - AS3 G {high, medium, low} opponent's cards - AS4 G {bad, medium, good} player's strategy - AS$ G {careful, risky, bluffing}. Poker Dialogue Utterances Attributes (Ax Sets) Dialogue systems Luděk Bártek Finite State Dialogue System Analysis User Modelling ■ choice - AXi G {call, rise, fold} Dialogue and Strategy Games ■ speech self-confidence - AX2 G {high, medium, low} FSA models Application for Generating the Dialogue Interfaces ■ speech credibility - AX3 G {high, medium, low} ■ voice excitement - AX4 G {high, medium, low} Poker Inner States Fragment Dialogue systems Luděk Bártek Finite State Dialogue System Analysis User Modelling Dialogue and Strategy Games FSA models Application for Generating the Dialogue Interfaces State my total trust in opponent player cards amount own cards strategy of bet cards Si medium high medium medium careful s2 medium high high medium risky S3 high high medium medium risky 54 high high medium medium bluffing 55 bad low medium low risky 56 bad low low medium careful 57 good medium medium medium risky Ss good medium low good careful Poker Dialogue Utterances Fragment Dialogue systems Luděk Bártek Finite State Dialogue System Analysis User Modelling Dialogue and Strategy Games FSA models Application for Generating the Dialogue Interfaces Utterance AX1 (choice) AX2 (self-confidence) AX3 (credibility) AXA (exciter™ *nt raise high low high raise high high low *3 call low high mediur n X4 raise high medium low *5 fold low high low *6 call low high mediur n Poker Dialogue Strategies Examples Dialogue systems Luděk Bártek User Modelling Dialogue and Strategy Games Experienced player: ■ state Si = (medium own cards, high bet, medium own cards trust, expected medium opponent cards, careful strategy), utterance Xi=(raise, high self-confidence, low credibility, high excitement in voice): S(S1,X1) = S2,X(SUX1) = X2 ■ State Ss= (good own cards, high bet, medium own cards trust, expected medium opponent cards, careful strategy), utteranceXl: