Computational Pragmatics Michael Franke Two views of language - - PowerPoint PPT Presentation

Computational Pragmatics

Michael Franke

Two views of language

structure function

ambiguous disambiguated by pragmatic reasoning

“If I say to any one, ‘I saw some of your children to-day’, he might be justified in inferring that I did not see them all, not because the words mean it, but because, if I had seen them all, it is most likely that I should have said so.”

(Mill 1867)

“[O]ne of my avowed aims is to see talking as a special case or variety of purposive, indeed rational, behaviour.”

(Grice 1975)

language as a formal system interaction & communication

Gricean ideas Neo-Grice

Gazdar, Horn, Atlas, Levinson, Russell, Sauerland, Schulz, van Rooij, Spector,

Post-Gricean

Sperber, Wilson, Carston s y s t e m a t i z a t i

• n

f

• r

m a l i z a t i

• n

r e d u c t i

• n

e v

• l

. / c

• g

n . f

• u

n d a t i

• n

embedding in compositional semantics

Grammaticalism

Chierchia, Fox, Spector, Magri meanwhile elsewhere

Game theory

Parikh, Jäger, Benz, van Rooij

Optimality theory

Blutner, Zeevat, Hendriks, de Hoop, Jäger, Mattausch, Aloni, Krifka

Theoretical Economics

rational communication message credibility

iterated reasoning

Benz, van Rooij, Jäger, Franke, Rothschild, Pavan, Stevens

Cognitive Science

probabilistic (Bayesian) modeling

RSA

Pragmatics from rational social reasoning

∃¬∀

∀

situation situation “I saw some of your children today.”

? ?

Pragmatics from rational social reasoning

literal interpreter rational speaker

“all”

∀

∃¬∀

“some”

∀

∃¬∀

“all”

∀

“some”

rational interpreter

∀

∃¬∀

“all” “some”

Pragmatics from rational social reasoning

rational speaker literal interpreter rational interpreter

Pragmatics from rational social reasoning

rational speaker literal interpreter “all” 1 “some” .5 .5

∀

∃¬∀ rational interpreter “all” 1 “some” 1

∀

∃¬∀ “all” “some” 1 1

∀

∃¬∀

Pragmatics from rational social reasoning

literal interpreter “all” 1 “some” .5 .5

∀

∃¬∀ rational interpreter “all” .9 .1 “some” .1 .9

∀

∃¬∀ rational speaker “all” “some” .9 .1 .1 .9

∀

∃¬∀ approximately

Pragmatics from rational social reasoning

literal interpreter “all” 1 “some” .5 .5

∀

∃¬∀ rational interpreter “all” .9 .1 “some” .1 .9

∀

∃¬∀ rational speaker “all” “some” .9 .1 .1 .9

∀

∃¬∀ approximately

listener behavior speaker behavior

U → ∆(S) S → ∆(U)

Rational Speech Act model

pragmatic speaker literal listener pragmatic listener

L0 S1 L1

PL1(s | u) ∝ PS1(u | s) · P(s) PL0(s | u) = P(s | [ [u] ])

PS1(u | s) ∝ exp(α(log PL0(s | u) − Cost(u) | {z }

Exp.Utility(u|s)

))

Rational Speech Act model

pragmatic speaker literal listener pragmatic listener

L0 S1 L1

PL1(s | u) ∝ PS1(u | s) · P(s) PL0(s | u) = P(s | [ [u] ])

semantic meaning world knowledge

PS1(u | s) ∝ exp(α(log PL0(s | u) − Cost(u) | {z }

Exp.Utility(u|s)

))

Rational Speech Act model

pragmatic speaker literal listener pragmatic listener

L0 S1 L1

PL1(s | u) ∝ PS1(u | s) · P(s) PL0(s | u) = P(s | [ [u] ])

linguistic preference information flow rational choice

PS1(u | s) ∝ exp(α(log PL0(s | u) − Cost(u) | {z }

Exp.Utility(u|s)

))

Rational Speech Act model

pragmatic speaker literal listener pragmatic listener

L0 S1 L1

PL1(s | u) ∝ PS1(u | s) · P(s) PL0(s | u) = P(s | [ [u] ])

world knowledge Bayes rule speaker model

PS1(u | s) ∝ exp(α(log PL0(s | u) − Cost(u) | {z }

Exp.Utility(u|s)

))

This course

applications technicalities WebPPL Bayesian Data Analysis … referential communication (epistemic) scalar implicatures non-literal language use vagueness politeness …

referential communication

U = {”square”, ”circle”, ”green”, ”blue”}

context

set of objects/referents

utterances

single properties of objects

which object do you think a speaker meant when she selects “blue”?

RSA for reference games (example)

rational speaker literal interpreter rational interpreter

“square” .5 .5 “circle” 1 “green” 1 “blue” .5 .5 “square” .82 .18 “circle” 1 “green” 1 “blue” .82 .18 “square” “circle” “green” “blue” .5 .5 .89 .11 .11 .89