Coevolution of Lexical Meaning and Pragmatic Use Thomas Brochhagen, - PowerPoint PPT Presentation

Coevolution of Lexical Meaning and Pragmatic Use Thomas Brochhagen, Michael Franke & Robert van Rooij

coevolution of semantics and pragmatics evolutionary dynamics with linguistic agents fitness-based selection AND agent-level learning meaning as mental representation Thomas Brochhagen, Michael Franke, Robert van Rooij (2018) “Coevolution of Lexical Meaning and Pragmatic Use” Cognitive Science 2

recap

We can hardly suppose a parliament of hitherto speechless elders meeting together and agreeing to call a cow a cow and a wolf a wolf. The association of words with their meanings must have grown up by some natural process , though at present the nature of the process is unknown. Bertrand Russell (1921) The Analysis of Mind p.190 4

Meaning as convention equilibria of signaling games David Lewis (1969) Convention

signaling theory evolutionary dynamics instead of equilibria fitness-based selection OR agent-level learning meaning as information content Brian Skyrms (2010) Signals: Evolution, Learning, and Information 6

signaling theory

signaling theory signaling game evolutionary stable states Lewis information content vector skyrms strategies ICV ( m ) = ⟨ log P S ( t 1 ∣ m ) sender: receiver: P R ( a ∣ m ) P S ( m ∣ t ) , log P S ( t 2 ∣ m ) ⟩ P ( t 1 ) P ( t 2 ) 8

signaling theory signaling game evolutionary stable states Lewis synopsis agent behavior reduced to input-output mapping agent-internal processes are abstracted away from meaning is identified at the level of behavioral patterns information content vector skyrms strategies ICV ( m ) = ⟨ log P S ( t 1 ∣ m ) sender: receiver: P R ( a ∣ m ) P S ( m ∣ t ) , log P S ( t 2 ∣ m ) ⟩ P ( t 1 ) P ( t 2 ) 9

types

evolutionary type lexicon comprehension & production rules 11

pragmatic reasoning s 1, s 2, s 3, s 4, … states P S ( m | s ) P L ( s | m ) m 1, m 2, m 3, m 4, messages … 12

Rational speech act models literal interpretation strategic depth 0 P lit ( s | m ) ∝ P ( s ) L [ s , m ] Gricean speaker strategic depth 1 P S ( m | s ) ∝ exp( α log P lit ( s | m )) Gricean interpretation strategic depth 2 P L ( s | m ) ∝ P ( s ) P S ( m | s ) e.g. Frank & Goodman (2012), Franke & Jäger (2016) 13

http://www.problang.org

literal vs. pragmatic language users literal agents strategic depth 0 S 0 ( m ∣ s ; L ) ∝ exp( λ L [ s , m ] ) H 0 ( s ∣ m ; L ) ∝ P ( s ) L [ s , m ] pragmatic agents strategic depth 1 S 1 ( m | s ; L ) ∝ exp( λ H 0 ( s | m ; L )) Literal Luke H 1 ( s | m ; L ) ∝ P ( s ) S 1 ( m | s ; L ) Gricean Greta 15

minimal type space

type space 1: all 4 combinations of 2 lexica + 2 pragmatic rules literal agents strategic depth 0 S 0 ( m ∣ s ; L ) ∝ exp( λ L [ s , m ] ) H 0 ( s ∣ m ; L ) ∝ P ( s ) L [ s , m ] pragmatic agents strategic depth 1 S 1 ( m | s ; L ) ∝ exp( λ H 0 ( s | m ; L )) H 1 ( s | m ; L ) ∝ P ( s ) S 1 ( m | s ; L ) 17

lexicalized upper bound textbook meaning lexica strategic depth 0 strategic depth 1 18

evolutionary dynamics

‣ fitness-based selection ‣ learning biases ๏ the better a type is at communicating, the ๏ agents acquire/update their type by more it will be replicated observation of others’ behavior f i = ∑ Q ji = ∑ P ( d ∣ t j ) P ( t i ∣ d ) x j EU( t i , t j ) d ∈ D j ∑ j x j f j Q ji x ′ � i = ϕ replicator mutator dynamic e.g., Nowak (2006), Griffith & Kalish (2007), Hutteger et al. (2014) 20

⃗ ⃗ ⃗ ⃗ ‣ fitness-based selection ‣ learning biases ๏ the better a type is at communicating, the ๏ agents acquire/update their type by more it will be replicated observation of others’ behavior f i = ∑ Q ji = ∑ P ( d ∣ t j ) P ( t i ∣ d ) x j EU( t i , t j ) d ∈ D j x ) ) i = x i f i ( RD( x ⋅ Q ) i (M( x )) i = ( Φ iterated learning replicator dynamic i = ( M ( RD( x ) ) ) i x ′ � replicator mutator dynamic e.g., Nowak (2006), Griffith & Kalis (2007), Hutteger et al. (2014) 21

example 22

minimal type space

type space 1: all 4 combinations of 2 lexica + 2 pragmatic rules literal agents strategic depth 0 S 0 ( m ∣ s ; L ) ∝ exp( λ L [ s , m ] ) H 0 ( s ∣ m ; L ) ∝ P ( s ) L [ s , m ] pragmatic agents strategic depth 1 S 1 ( m | s ; L ) ∝ exp( λ H 0 ( s | m ; L )) H 1 ( s | m ; L ) ∝ P ( s ) S 1 ( m | s ; L ) 24

analysis 25

larger type space

lexical representations set up S = { s ∅ , s ∃ ¬ ∀ , s ∀ } states 𝔙 = {lit, prag} usage 𝔐 = R M lexica lexical representations examples of relevant types of lexica 27

simulation results ::: Fitness-based selection only higher act-rationality 28

simulation results ::: iterated learning only higher belief-rationality 29

simulation results ::: replicator mutator dynamic higher act-rationality higher belief-ration. 30

summary ‣ pragmatic language use with underspecified semantics can evolve ‣ results from interplay of two forces: ๏ functional pressure towards efficient communication ๏ learning bias: preference for simple mental representations Gricean Greta Literal Luke 31

conclusion

general trend EXTENDING THE NATURALIST PROGRAMM you TO INCORPORATE MORE LINGUISTIC / COGNITIVE REALISM

‣ role of common ground in disambiguation of meaning ‣ interlocutor-specific adaptation ๏ from prior to passing theories ‣ functional rationale of vagueness ‣ impact of recurrent tropes on conventionalization of meaning 34