Learning in the Rational Speech Acts Model Christopher Potts - PowerPoint PPT Presentation

Overview RSA TUNA Learned RSA Experiments Conclusion Semantic parsing Zettlemoyer & Collins 2005: Rules Categories produced from logical form Input Trigger Output Category arg max( λx.state ( x ) ^ borders ( x, texas ) , λx.size ( x )) constant c NP : c NP : texas arity one predicate p 1 N : λx.p 1 ( x ) N : λx.state ( x ) arity one predicate p 1 S \ NP : λx.p 1 ( x ) S \ NP : λx.state ( x ) arity two predicate p 2 ( S \ NP ) /NP : λx.λy.p 2 ( y, x ) ( S \ NP ) /NP : λx.λy.borders ( y, x ) arity two predicate p 2 ( S \ NP ) /NP : λx.λy.p 2 ( x, y ) ( S \ NP ) /NP : λx.λy.borders ( x, y ) arity one predicate p 1 N/N : λg.λx.p 1 ( x ) ^ g ( x ) N/N : λg.λx.state ( x ) ^ g ( x ) literal with arity two predicate p 2 N/N : λg.λx.p 2 ( x, c ) ^ g ( x ) N/N : λg.λx.borders ( x, texas ) ^ g ( x ) and constant second argument c arity two predicate p 2 ( N \ N ) /NP : λx.λg.λy.p 2 ( x, y ) ^ g ( x ) ( N \ N ) /NP : λg.λx.λy.borders ( x, y ) ^ g ( x ) an arg max / min with second NP/N : λg. arg max / min( g, λx.f ( x )) NP/N : λg. arg max( g, λx.size ( x )) argument arity one function f an arity one S/NP : λx.f ( x ) S/NP : λx.size ( x ) numeric-ranged function f 5 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Semantic parsing Zettlemoyer & Collins 2005: Rules Categories produced from logical form Input Trigger Output Category arg max( λx.state ( x ) ^ borders ( x, texas ) , λx.size ( x )) constant c NP : c NP : texas arity one predicate p 1 N : λx.p 1 ( x ) N : λx.state ( x ) arity one predicate p 1 S \ NP : λx.p 1 ( x ) S \ NP : λx.state ( x ) arity two predicate p 2 ( S \ NP ) /NP : λx.λy.p 2 ( y, x ) ( S \ NP ) /NP : λx.λy.borders ( y, x ) arity two predicate p 2 ( S \ NP ) /NP : λx.λy.p 2 ( x, y ) ( S \ NP ) /NP : λx.λy.borders ( x, y ) arity one predicate p 1 N/N : λg.λx.p 1 ( x ) ^ g ( x ) N/N : λg.λx.state ( x ) ^ g ( x ) literal with arity two predicate p 2 N/N : λg.λx.p 2 ( x, c ) ^ g ( x ) N/N : λg.λx.borders ( x, texas ) ^ g ( x ) and constant second argument c arity two predicate p 2 ( N \ N ) /NP : λx.λg.λy.p 2 ( x, y ) ^ g ( x ) ( N \ N ) /NP : λg.λx.λy.borders ( x, y ) ^ g ( x ) an arg max / min with second NP/N : λg. arg max / min( g, λx.f ( x )) NP/N : λg. arg max( g, λx.size ( x )) argument arity one function f an arity one S/NP : λx.f ( x ) S/NP : λx.size ( x ) numeric-ranged function f constant c NP : c arity one predicate 5 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Semantic parsing Zettlemoyer & Collins 2005: Rules Categories produced from logical form Input Trigger Output Category arg max( λx.state ( x ) ^ borders ( x, texas ) , λx.size ( x )) constant c NP : c NP : texas arity one predicate p 1 N : λx.p 1 ( x ) N : λx.state ( x ) arity one predicate p 1 S \ NP : λx.p 1 ( x ) S \ NP : λx.state ( x ) arity two predicate p 2 ( S \ NP ) /NP : λx.λy.p 2 ( y, x ) ( S \ NP ) /NP : λx.λy.borders ( y, x ) arity two predicate p 2 ( S \ NP ) /NP : λx.λy.p 2 ( x, y ) ( S \ NP ) /NP : λx.λy.borders ( x, y ) arity one predicate p 1 N/N : λg.λx.p 1 ( x ) ^ g ( x ) N/N : λg.λx.state ( x ) ^ g ( x ) literal with arity two predicate p 2 N/N : λg.λx.p 2 ( x, c ) ^ g ( x ) N/N : λg.λx.borders ( x, texas ) ^ g ( x ) and constant second argument c arity two predicate p 2 ( N \ N ) /NP : λx.λg.λy.p 2 ( x, y ) ^ g ( x ) ( N \ N ) /NP : λg.λx.λy.borders ( x, y ) ^ g ( x ) an arg max / min with second NP/N : λg. arg max / min( g, λx.f ( x )) NP/N : λg. arg max( g, λx.size ( x )) argument arity one function f an arity one S/NP : λx.f ( x ) S/NP : λx.size ( x ) numeric-ranged function f arity one predicate p 1 N : λx.p 1 ( x ) arity one predicate p 1 S \ NP : λx.p 1 ( x ) arity two predicate 5 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion The Rational Speech Acts Model (RSA) 1 The Rational Speech Acts (RSA) model 2 TUNA 3 Learned RSA 4 Experiments 6 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Grice Cooperative principle : Make your contribution as is required, when it is required, by the conversation in which you are engaged. 7 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Grice Cooperative principle : Make your contribution as is required, when it is required, by the conversation in which you are engaged. • Quantity • Quality • Relevance • Manner 7 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Grice Cooperative principle : Make your contribution as is required, when it is required, by the conversation in which you are engaged. • Quantity • Quality • Relevance • Manner • Politeness 7 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Grice Cooperative principle : Make your contribution as is required, when it is required, by the conversation in which you are engaged. • Quantity • Quality • Relevance • Manner • Politeness • Stylishness 7 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Grice Cooperative principle : Make your contribution as is required, when it is required, by the conversation in which you are engaged. • Quantity • Q principle • Quality • R principle • Relevance • Manner • Politeness • Stylishness 7 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Grice Cooperative principle : Make your contribution as is required, when it is required, by the conversation in which you are engaged. • Quantity • Q principle • Q heuristic • Quality • R principle • I heuristic • Relevance • M heuristic • Manner • Politeness • Stylishness 7 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Grice Cooperative principle : Make your contribution as is required, when it is required, by the conversation in which you are engaged. • Quantity • Q principle • Q heuristic • Quality • R principle • I heuristic • Relevance • M heuristic • Manner • Politeness • Stylishness 7 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 1 S and L mutually, publicly presume that S is cooperative. 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 1 S and L mutually, publicly presume that S is cooperative. 2 To maintain 1 given U , it must be supposed that S thinks q . 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 1 S and L mutually, publicly presume that S is cooperative. 2 To maintain 1 given U , it must be supposed that S thinks q . 3 S thinks that both S and L mutually, publicly presume that L is willing and able to work out that 2 holds. 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 1 S and L mutually, publicly presume that S is cooperative. 2 To maintain 1 given U , it must be supposed that S thinks q . 3 S thinks that both S and L mutually, publicly presume that L is willing and able to work out that 2 holds. Example 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 1 S and L mutually, publicly presume that S is cooperative. 2 To maintain 1 given U , it must be supposed that S thinks q . 3 S thinks that both S and L mutually, publicly presume that L is willing and able to work out that 2 holds. Example Ann: What city does Paul live in? Bob: Hmm . . . he lives in California. 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 1 S and L mutually, publicly presume that S is cooperative. 2 To maintain 1 given U , it must be supposed that S thinks q . 3 S thinks that both S and L mutually, publicly presume that L is willing and able to work out that 2 holds. Example Ann: What city does Paul live in? Bob: Hmm . . . he lives in California. (A) Assume Bob is cooperative. 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 1 S and L mutually, publicly presume that S is cooperative. 2 To maintain 1 given U , it must be supposed that S thinks q . 3 S thinks that both S and L mutually, publicly presume that L is willing and able to work out that 2 holds. Example Ann: What city does Paul live in? Bob: Hmm . . . he lives in California. (A) Assume Bob is cooperative. (B) Bob supplied less information than required; clash with (A). 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 1 S and L mutually, publicly presume that S is cooperative. 2 To maintain 1 given U , it must be supposed that S thinks q . 3 S thinks that both S and L mutually, publicly presume that L is willing and able to work out that 2 holds. Example Ann: What city does Paul live in? Bob: Hmm . . . he lives in California. (A) Assume Bob is cooperative. (B) Bob supplied less information than required; clash with (A). (C) Assume Bob does not know which city Paul lives in. 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 1 S and L mutually, publicly presume that S is cooperative. 2 To maintain 1 given U , it must be supposed that S thinks q . 3 S thinks that both S and L mutually, publicly presume that L is willing and able to work out that 2 holds. Example Ann: What city does Paul live in? Bob: Hmm . . . he lives in California. (A) Assume Bob is cooperative. (B) Bob supplied less information than required; clash with (A). (C) Assume Bob does not know which city Paul lives in. (D) Then Bob’s answer is optimal given his evidence. 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 1 S and L mutually, publicly presume that S is cooperative. 2 To maintain 1 given U , it must be supposed that S thinks q . 3 S thinks that both S and L mutually, publicly presume that L is willing and able to work out that 2 holds. Implicature as • Rooted in cooperativity • Social, interactional • Cognitively complex • Error-driven 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Conversational implicature Definition Speaker S saying U to listener L conversationally implicates q iff 1 S and L mutually, publicly presume that S is cooperative. 2 To maintain 1 given U , it must be supposed that S thinks q . 3 S thinks that both S and L mutually, publicly presume that L is willing and able to work out that 2 holds. Implicature as • Rooted in cooperativity • Social, interactional • Cognitively complex • Error-driven 8 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Pragmatic listeners 9 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Pragmatic listeners Definition (Literal listener) l 0 ( w | msg , Lex ) ∝ Lex ( msg , w ) P ( w ) 9 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Pragmatic listeners Definition (Pragmatic speaker) s 1 ( msg | w , Lex ) ∝ exp λ ( log l 0 ( w | msg , Lex ) − C ( msg )) Definition (Literal listener) l 0 ( w | msg , Lex ) ∝ Lex ( msg , w ) P ( w ) 9 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Pragmatic listeners Definition (Pragmatic listener) l 1 ( w | msg , Lex ) ∝ s 1 ( msg | w , Lex ) P ( w ) Definition (Pragmatic speaker) s 1 ( msg | w , Lex ) ∝ exp λ ( log l 0 ( w | msg , Lex ) − C ( msg )) Definition (Literal listener) l 0 ( w | msg , Lex ) ∝ Lex ( msg , w ) P ( w ) 9 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Pragmatic listeners Definition (Pragmatic listener) l 1 ( w | msg , Lex ) = pragmatic speaker × state prior Definition (Pragmatic speaker) s 1 ( msg | w , Lex ) = literal listener − message costs Definition (Literal listener) l 0 ( w | msg , Lex ) = lexicon × state prior 9 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion RSA listener example l 1 beard 1 0 0 s 1 glasses 1 1 0 l 0 Lex tie 0 1 1 10 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion RSA listener example 1 l 1 beard 0 0 s 1 glasses .5 .5 0 l 0 Lex tie 0 .5 .5 10 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion RSA listener example beard glasses tie .67 .33 0 l 1 s 1 1 0 0 l 0 Lex 0 1 0 10 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion RSA listener example 1 l 1 beard 0 0 s 1 .75 glasses .25 0 l 0 Lex 1 tie 0 0 10 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion More Gricean terrain 11 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion More Gricean terrain • Lexical uncertainty and the division of pragmatic labor Bergen, Levy, Goodman, ‘Pragmatic reasoning through semantic inference’ 11 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion More Gricean terrain • Lexical uncertainty and the division of pragmatic labor Bergen, Levy, Goodman, ‘Pragmatic reasoning through semantic inference’ • Lexical uncertainty and embedded implicatures Potts, Lassiter, Levy, Frank, ‘Embedded implicatures as pragmatic inferences under compositional lexical uncertainty’ 11 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion More Gricean terrain • Lexical uncertainty and the division of pragmatic labor Bergen, Levy, Goodman, ‘Pragmatic reasoning through semantic inference’ • Lexical uncertainty and embedded implicatures Potts, Lassiter, Levy, Frank, ‘Embedded implicatures as pragmatic inferences under compositional lexical uncertainty’ • Pragmatic free variables and non-literal language Kao, Wu, Bergen, Goodman, ‘Nonliteral understanding of number words’ 11 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion More Gricean terrain • Lexical uncertainty and the division of pragmatic labor Bergen, Levy, Goodman, ‘Pragmatic reasoning through semantic inference’ • Lexical uncertainty and embedded implicatures Potts, Lassiter, Levy, Frank, ‘Embedded implicatures as pragmatic inferences under compositional lexical uncertainty’ • Pragmatic free variables and non-literal language Kao, Wu, Bergen, Goodman, ‘Nonliteral understanding of number words’ • Implicature blocking by higher-level agents Potts & Levy, ‘Negotiating lexical uncertainty and speaker expertise with disjunction’ 11 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Pragmatic speakers 12 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Pragmatic speakers Definition (Literal speaker) s 0 ( msg | w , Lex ) ∝ exp λ ( log Lex ( msg , w ) − C ( msg )) 12 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Pragmatic speakers Definition (Pragmatic listener) l 1 ( w | msg , Lex ) ∝ s 0 ( msg | w , Lex ) P ( w ) Definition (Literal speaker) s 0 ( msg | w , Lex ) ∝ exp λ ( log Lex ( msg , w ) − C ( msg )) 12 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Pragmatic speakers Definition (Pragmatic speaker) s 1 ( msg | w , Lex ) ∝ exp λ ( log l 1 ( w | msg , Lex ) − C ( msg )) Definition (Pragmatic listener) l 1 ( w | msg , Lex ) ∝ s 0 ( msg | w , Lex ) P ( w ) Definition (Literal speaker) s 0 ( msg | w , Lex ) ∝ exp λ ( log Lex ( msg , w ) − C ( msg )) 12 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Pragmatic speakers Definition (Pragmatic speaker) s 1 ( msg | w , Lex ) = pragmatic listener − message costs Definition (Pragmatic listener) l 1 ( w | msg , Lex ) = literal speaker × state prior Definition (Literal speaker) s 0 ( msg | w , Lex ) = lexicon − message costs 12 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion RSA speaker example beard glasses tie 1 1 0 s 1 l 1 0 1 1 s 0 Lex 0 0 1 13 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion RSA speaker example beard glasses tie .5 .5 0 s 1 l 1 0 .5 .5 s 0 Lex 0 1 0 13 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion RSA speaker example 1 s 1 beard 0 0 l 1 glasses .5 .5 0 s 0 Lex .67 tie 0 .33 13 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion RSA speaker example beard glasses tie .67 .33 0 s 1 l 1 .6 0 .4 s 0 Lex 0 1 0 13 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Achievements and drawbacks beard glasses tie .67 .33 0 .6 0 .4 0 1 0 14 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Achievements and drawbacks beard glasses tie .67 .33 0 • Cognitive demands limit speaker rationality .6 0 .4 0 1 0 14 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Achievements and drawbacks beard glasses tie .67 .33 0 • Cognitive demands limit speaker rationality • Speaker preferences .6 0 .4 0 1 0 14 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Achievements and drawbacks beard glasses tie .67 .33 0 • Cognitive demands limit speaker rationality • Speaker preferences .6 0 .4 • Hand-specified lexicon 0 1 0 14 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Achievements and drawbacks beard glasses tie .67 .33 0 • Cognitive demands limit speaker rationality • Speaker preferences .6 0 .4 • Hand-specified lexicon • High-bias model; few chances to learn from data 0 1 0 14 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion TUNA 1 The Rational Speech Acts (RSA) model 2 TUNA 3 Learned RSA 4 Experiments 15 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Furniture example colour : green colour : green colour : red orientation : left orientation : left orientation : back size : small size : small size : large type : fan type : sofa type : fan x - dimension :1 x - dimension :1 x - dimension :1 y - dimension :1 y - dimension :2 y - dimension :3 colour : red colour : blue orientation : back orientation : left size : large size : large type : sofa type : fan x - dimension :2 x - dimension :2 y - dimension :1 y - dimension :2 colour : blue colour : blue orientation : left orientation : left size : large size : small type : sofa type : fan x - dimension :3 x - dimension :3 y - dimension :1 y - dimension :3 Utterance: “blue fan small” Utterance attributes: [ colour:blue ] ; [ size:small ] ; [ type:fan ] 16 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Furniture example colour : green colour : green colour : red orientation : left orientation : left orientation : back size : small size : small size : large type : fan type : sofa type : fan x - dimension :1 x - dimension :1 x - dimension :1 y - dimension :1 y - dimension :2 y - dimension :3 colour : red colour : blue orientation : back orientation : left size : large size : large type : sofa type : fan x - dimension :2 x - dimension :2 y - dimension :1 y - dimension :2 colour : blue colour : blue orientation : left orientation : left size : large size : small type : sofa type : fan x - dimension :3 x - dimension :3 y - dimension :1 y - dimension :3 Utterance: “blue fan small” Utterance attributes: [ colour:blue ] ; [ size:small ] ; [ type:fan ] 16 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Furniture example colour : green colour : green colour : red orientation : left orientation : left orientation : back size : small size : small size : large type : fan type : sofa type : fan x - dimension :1 x - dimension :1 x - dimension :1 y - dimension :1 y - dimension :2 y - dimension :3 colour : red colour : blue orientation : back orientation : left size : large size : large type : sofa type : fan x - dimension :2 x - dimension :2 y - dimension :1 y - dimension :2 colour : blue colour : blue orientation : left orientation : left size : large size : small type : sofa type : fan x - dimension :3 x - dimension :3 y - dimension :1 y - dimension :3 Utterance: “blue fan small” Utterance attributes: [ colour:blue ] ; [ size:small ] ; [ type:fan ] 16 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion People example age : old age : young hair C olour : light hair C olour : dark has B eard :1 has B eard :0 has G lasses :0 has G lasses :0 has H air :0 has H air :1 has S hirt :1 has S hirt :1 has S uit :0 has S uit :0 has T ie :0 has T ie :0 orientation : left orientation : front type : person type : person x - dimension :1 x - dimension :1 y - dimension :1 y - dimension :2 age : young age : young hair C olour : dark hair C olour : dark has B eard :1 has B eard :1 has G lasses :0 has G lasses :0 has H air :1 has H air :1 has S hirt :1 has S hirt :0 has S uit :0 has S uit :1 has T ie :1 has T ie :1 orientation : front orientation : front type : person type : person x - dimension :2 x - dimension :2 y - dimension :1 y - dimension :2 age : young age : young age : young hair C olour : dark hair C olour : dark hair C olour : dark has B eard :0 has B eard :1 has B eard :0 has G lasses :0 has G lasses :0 has G lasses :0 has H air :1 has H air :1 has H air :1 has S hirt :0 has S hirt :1 has S hirt :0 has S uit :1 has S uit :0 has S uit :1 has T ie :1 has T ie :0 has T ie :1 orientation : front orientation : front orientation : front type : person type : person type : person x - dimension :3 x - dimension :3 x - dimension :3 y - dimension :1 y - dimension :2 y - dimension :3 Utterance: The bald man with a beard [ hasBeard:1 ] ; [ hasHair:0 ] ; [ type:person ] 17 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion People example age : old age : young hair C olour : light hair C olour : dark has B eard :1 has B eard :0 has G lasses :0 has G lasses :0 has H air :0 has H air :1 has S hirt :1 has S hirt :1 has S uit :0 has S uit :0 has T ie :0 has T ie :0 orientation : left orientation : front type : person type : person x - dimension :1 x - dimension :1 y - dimension :1 y - dimension :2 age : young age : young hair C olour : dark hair C olour : dark has B eard :1 has B eard :1 has G lasses :0 has G lasses :0 has H air :1 has H air :1 has S hirt :1 has S hirt :0 has S uit :0 has S uit :1 has T ie :1 has T ie :1 orientation : front orientation : front type : person type : person x - dimension :2 x - dimension :2 y - dimension :1 y - dimension :2 age : young age : young age : young hair C olour : dark hair C olour : dark hair C olour : dark has B eard :0 has B eard :1 has B eard :0 has G lasses :0 has G lasses :0 has G lasses :0 has H air :1 has H air :1 has H air :1 has S hirt :0 has S hirt :1 has S hirt :0 has S uit :1 has S uit :0 has S uit :1 has T ie :1 has T ie :0 has T ie :1 orientation : front orientation : front orientation : front type : person type : person type : person x - dimension :3 x - dimension :3 x - dimension :3 y - dimension :1 y - dimension :2 y - dimension :3 Utterance: The bald man with a beard [ hasBeard:1 ] ; [ hasHair:0 ] ; [ type:person ] 17 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion People example age : old age : young hair C olour : light hair C olour : dark has B eard :1 has B eard :0 has G lasses :0 has G lasses :0 has H air :0 has H air :1 has S hirt :1 has S hirt :1 has S uit :0 has S uit :0 has T ie :0 has T ie :0 orientation : left orientation : front type : person type : person x - dimension :1 x - dimension :1 y - dimension :1 y - dimension :2 age : young age : young hair C olour : dark hair C olour : dark has B eard :1 has B eard :1 has G lasses :0 has G lasses :0 has H air :1 has H air :1 has S hirt :1 has S hirt :0 has S uit :0 has S uit :1 has T ie :1 has T ie :1 orientation : front orientation : front type : person type : person x - dimension :2 x - dimension :2 y - dimension :1 y - dimension :2 age : young age : young age : young hair C olour : dark hair C olour : dark hair C olour : dark has B eard :0 has B eard :1 has B eard :0 has G lasses :0 has G lasses :0 has G lasses :0 has H air :1 has H air :1 has H air :1 has S hirt :0 has S hirt :1 has S hirt :0 has S uit :1 has S uit :0 has S uit :1 has T ie :1 has T ie :0 has T ie :1 orientation : front orientation : front orientation : front type : person type : person type : person x - dimension :3 x - dimension :3 x - dimension :3 y - dimension :1 y - dimension :2 y - dimension :3 Utterance: The bald man with a beard [ hasBeard:1 ] ; [ hasHair:0 ] ; [ type:person ] 17 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Multiset Dice coefficient Definition � � 2 � Z a ( msg i ) ( x ) , Z a ( msg j ) ( x ) x ∈ D min | a ( msg i ) | + | a ( msg j ) | 18 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Multiset Dice coefficient Definition Multiset intersection cardinality Multiset union cardinality 18 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Multiset Dice coefficient Definition Multiset intersection cardinality Multiset union cardinality predicted : [ a a ] b c a ] = 1 • actual : [ a b c 18 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Multiset Dice coefficient Definition Multiset intersection cardinality Multiset union cardinality predicted : [ a a ] b c a ] = 1 • actual : [ a b c predicted : [ a a ] b c ] = . 86 • actual : [ a b c 18 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Multiset Dice coefficient Definition Multiset intersection cardinality Multiset union cardinality predicted : [ a a ] b c a ] = 1 • actual : [ a b c predicted : [ a a ] b c ] = . 86 • actual : [ a b c predicted : [ a b c ] a ] = . 86 • actual : [ a b c 18 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Multiset Dice coefficient Definition Multiset intersection cardinality Multiset union cardinality predicted : [ a a ] b c a ] = 1 • actual : [ a b c predicted : [ a a ] b c ] = . 86 • actual : [ a b c predicted : [ a b c ] a ] = . 86 • actual : [ a b c predicted : [ a ] a ] = . 4 • actual : [ a b c 18 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Learned RSA 1 The Rational Speech Acts (RSA) model 2 TUNA 3 Learned RSA 4 Experiments 19 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Feature representations 20 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Feature representations Target Utterance attributes Features colour : blue orientation : left [ colour:blue ] size : small [ size:small ] type : fan [ type:fan ] x - dimension :3 y - dimension :3 20 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Feature representations Target Utterance attributes Features colour : blue ∧ [ colour:blue ] colour : blue colour : blue ∧ [ size:small ] orientation : left colour : blue ∧ [ type:fan ] [ colour:blue ] size : small orientation : left ∧ [ colour:blue ] [ size:small ] type : fan orientation : left ∧ [ size:small ] [ type:fan ] x - dimension :3 orientation : left ∧ [ type:fan ] y - dimension :3 . . . 20 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Feature representations Target Utterance attributes Features colour : blue ∧ [ colour:blue ] colour : blue colour : blue ∧ [ size:small ] orientation : left colour : blue ∧ [ type:fan ] [ colour:blue ] size : small orientation : left ∧ [ colour:blue ] [ size:small ] type : fan orientation : left ∧ [ size:small ] [ type:fan ] x - dimension :3 orientation : left ∧ [ type:fan ] y - dimension :3 . . . color    Generation features     20 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Feature representations Target Utterance attributes Features colour : blue ∧ [ colour:blue ] colour : blue colour : blue ∧ [ size:small ] orientation : left colour : blue ∧ [ type:fan ] [ colour:blue ] size : small orientation : left ∧ [ colour:blue ] [ size:small ] type : fan orientation : left ∧ [ size:small ] [ type:fan ] x - dimension :3 orientation : left ∧ [ type:fan ] y - dimension :3 . . . color  type + color   Generation features  color + ¬ size    type ≫ color ≫ size type ≫ orientation ≫ color ≫ size 20 / 29

Overview RSA TUNA Learned RSA Experiments Conclusion Feature representations Target Utterance attributes Features colour : blue ∧ [ colour:blue ] colour : blue colour : blue ∧ [ size:small ] orientation : left colour : blue ∧ [ type:fan ] [ colour:blue ] size : small orientation : left ∧ [ colour:blue ] [ size:small ] type : fan orientation : left ∧ [ size:small ] [ type:fan ] x - dimension :3 orientation : left ∧ [ type:fan ] y - dimension :3 . . . color  type + color   Generation features  color + ¬ size    attribute-count = 3 type ≫ color ≫ size type ≫ orientation ≫ color ≫ size 20 / 29