natural language processing (csep 517): …...natural language processing (csep 517): computational...

67
Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington [email protected] May 22, 2017 1 / 67

Upload: others

Post on 16-Mar-2020

29 views

Category:

Documents


0 download

TRANSCRIPT

Page 1: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Natural Language Processing (CSEP 517):Computational Pragmatics

Chenhao Tanc© 2017

University of [email protected]

May 22, 2017

1 / 67

Page 2: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

What do we use language for?

Communicating with other humans

I exchanging emails

I talking to friends

I writing

I giving lectures

I ...

2 / 67

Page 3: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

What do we use language for?

Communicating with other humans

I exchanging emails

I talking to friends

I writing

I giving lectures

I ...

3 / 67

Page 4: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Throw back Monday

Can you pass me the salt?

4 / 67

Page 5: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Pragmatics

The study of meaning as communicated by a speaker to a listener (Yule, 1996).

Or, contextual meaning

Pragmatics is important for building conversational agents, understanding humandecision making, understanding language, etc.

5 / 67

Page 6: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Pragmatics

The study of meaning as communicated by a speaker to a listener (Yule, 1996).

Or, contextual meaning

Pragmatics is important for building conversational agents, understanding humandecision making, understanding language, etc.

6 / 67

Page 7: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Pragmatics

The study of meaning as communicated by a speaker to a listener (Yule, 1996).

Or, contextual meaning

Pragmatics is important for building conversational agents, understanding humandecision making, understanding language, etc.

7 / 67

Page 8: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Pragmatics vs. Syntax, Semantics (Yule, 1996)

I Syntax: the relationships between linguistic forms, how they are arranged insequences, and which sequences are well-formed.

I Semantics: the relationships between linguistic forms and entities in the world.

I Pragmatics: the relationships between linguistic forms and the users of thoseforms.

8 / 67

Page 9: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Outline

Speech act theory

The effect of wording choices (big data pragmatics)

Modeling conversations: dialogue act categorization

Rational speech acts model

9 / 67

Page 10: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Outline

Speech act theory

The effect of wording choices (big data pragmatics)

Modeling conversations: dialogue act categorization

Rational speech acts model

10 / 67

Page 11: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Speech act theory

We do not simply produce utterances containing grammatical structures; we performactions via those utterances.

Actions performed via utterances are generally called speech acts (Austin, 1975).

11 / 67

Page 12: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Speech act theory

We do not simply produce utterances containing grammatical structures; we performactions via those utterances.Actions performed via utterances are generally called speech acts (Austin, 1975).

12 / 67

Page 13: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Speech act theory

I locutionary act (the actual utterance and its ostensible meaning)

I illocutionary act (its real, intended meaning)

I perlocutionary act (its actual effect, whether intended or not)

13 / 67

Page 14: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Outline

Speech act theory

The effect of wording choices (big data pragmatics)

Modeling conversations: dialogue act categorization

Rational speech acts model

14 / 67

Page 15: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Wording mattersMotivate voter turnout (Bryan et al., 2011)

“How important is it to you to be a voterin the upcoming election?”

“How important is it to you to vote in theupcoming election?”

15 / 67

Page 16: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Wording mattersMotivate voter turnout (Bryan et al., 2011)

“How important is it to you to be a voterin the upcoming election?”

“How important is it to you to vote in theupcoming election?”

16 / 67

Page 17: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Large-scale natural experiments

A large number of social interactions in the format of texts

Potential opportunities for natural experiments

17 / 67

Page 18: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Large-scale natural experiments

A large number of social interactions in the format of texts

Potential opportunities for natural experiments

18 / 67

Page 19: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Large-scale natural experiments

The effect of wording on message propagation on Twitter (Tan et al., 2014)

19 / 67

Page 20: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Large-scale natural experiments

The effect of wording on message propagation on Twitter (Tan et al., 2014)

20 / 67

Page 21: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Large-scale natural experiments

I Millions of topic-author controlled pairsI Ranking within a pair (classification)

I Evaluation: the accuracy of predicting which one was retweeted more(random → 50%)

I Classifier: logistic regression

21 / 67

Page 22: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Features

Pronouns

first person singular (i)

——–

first person plural (we)

——–

second person (you)

——–

third person singular (she, he)

↑↑ ↑↑

third person plural (they)

↑ ↑↑↑

Referring to other people helps

22 / 67

Page 23: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Features

Pronouns

first person singular (i) ——–first person plural (we) ——–second person (you) ——–third person singular (she, he) ↑↑ ↑↑third person plural (they) ↑ ↑↑↑

Referring to other people helps

23 / 67

Page 24: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Features

Pronouns

first person singular (i) ——–first person plural (we) ——–second person (you) ——–third person singular (she, he) ↑↑ ↑↑third person plural (they) ↑ ↑↑↑

Referring to other people helps

24 / 67

Page 25: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Features

Generality

indefinite articles (a,an)

↑↑↑ ↑

definite articles (the)

——–

Generality helps

25 / 67

Page 26: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Features

Generality

indefinite articles (a,an) ↑↑↑ ↑definite articles (the) ——–

Generality helps

26 / 67

Page 27: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Features

Generality

indefinite articles (a,an) ↑↑↑ ↑definite articles (the) ——–

Generality helps

27 / 67

Page 28: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Features

Language model scores

I similarity with overall Twitter users

twitter unigram

↑↑↑ ↑

twitter bigram

↑↑↑ ↑

28 / 67

Page 29: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Features

Language model scores

I similarity with overall Twitter users

twitter unigram ↑↑↑ ↑twitter bigram ↑↑↑ ↑

29 / 67

Page 30: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Features

Language model scores

I similarity with overall Twitter users

twitter unigram ↑↑↑ ↑twitter bigram ↑↑↑ ↑

I similarity with personal history

personal unigram

↑↑↑ ↑

personal bigram

——–

Be like the community & be true to yourself

30 / 67

Page 31: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Features

Language model scores

I similarity with overall Twitter users

twitter unigram ↑↑↑ ↑twitter bigram ↑↑↑ ↑

I similarity with personal history

personal unigram ↑↑↑ ↑personal bigram ——–

Be like the community & be true to yourself

31 / 67

Page 32: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Baseline without “natural experiments”

Supervised classification without control

I most-retweeted tweets vs. least-retweeted tweets

32 / 67

Page 33: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Prediction performance

Human performance

I Controlling for context is importantI Big data can help understand pragmatics

https://chenhaot.com/retweetedmore

33 / 67

Page 34: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Prediction performance

Human performance

I Controlling for context is importantI Big data can help understand pragmatics

https://chenhaot.com/retweetedmore

34 / 67

Page 35: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Beyond retweeting

I Persuasive arguments (Tan et al., 2016)

I Memorable (movie) quotes (Danescu-Niculescu-Mizil et al., 2012a)

I Power dynamics (Danescu-Niculescu-Mizil et al., 2012b; Prabhakaran et al., 2014)

I Newsworthiness of research articles and political speeches (Zhang et al., 2016)

35 / 67

Page 36: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Outline

Speech act theory

The effect of wording choices (big data pragmatics)

Modeling conversations: dialogue act categorization

Rational speech acts model

36 / 67

Page 37: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Dialogue act classification/tagging

Define categories and label corpora (Stolcke et al., 2000)

I statement

I question

I backchannel

I agreement

I apology

I ...

37 / 67

Page 38: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Dialogue act classification/tagging

Supervised classification

I SVM

I logistic classification

Structure prediction (sequence tagging)

I Hidden Markov model

I Conditional random field

38 / 67

Page 39: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Speech act theory

The effect of wording choices (big data pragmatics)

Modeling conversations: dialogue act categorization

Rational speech acts model

39 / 67

Page 40: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Cooperative Principle

Make your contribution as is required, when it is required, by the conversation in whichyou are engaged (Grice, 1975).

40 / 67

Page 41: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Conversational Implicatures

I Maxims of quality(Do not say what you believe to be false; do not say that for which you lackadequate evidence)e.g., Noah is a nice person

⇒ I believe that Noah is a nice person

I Maxims of quantityMake you contribution as informative as is required (for the current purposes ofthe exchange); do not make your contribution more informative than is required

I I have two hands ⇒ I have no more than two hands

41 / 67

Page 42: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Conversational Implicatures

I Maxims of quality(Do not say what you believe to be false; do not say that for which you lackadequate evidence)e.g., Noah is a nice person ⇒ I believe that Noah is a nice person

I Maxims of quantityMake you contribution as informative as is required (for the current purposes ofthe exchange); do not make your contribution more informative than is required

I I have two hands ⇒ I have no more than two hands

42 / 67

Page 43: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Conversational Implicatures

I Maxims of quality(Do not say what you believe to be false; do not say that for which you lackadequate evidence)e.g., Noah is a nice person ⇒ I believe that Noah is a nice person

I Maxims of quantityMake you contribution as informative as is required (for the current purposes ofthe exchange); do not make your contribution more informative than is required

I I have two hands

⇒ I have no more than two hands

43 / 67

Page 44: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Conversational Implicatures

I Maxims of quality(Do not say what you believe to be false; do not say that for which you lackadequate evidence)e.g., Noah is a nice person ⇒ I believe that Noah is a nice person

I Maxims of quantityMake you contribution as informative as is required (for the current purposes ofthe exchange); do not make your contribution more informative than is required

I I have two hands ⇒ I have no more than two hands

44 / 67

Page 45: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts Model

Reference games (Wittgenstein, 1953; Frank and Goodman, 2012)

I Speaker. Imagine you are talking to someone and want to refer to the middleobject. Would you say “blue” or “circle”?

I Listener. Someone uses the word “blue” to refer to one of these objects. Whichobject are they talking about?

45 / 67

Page 46: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts Model

Reference games (Wittgenstein, 1953; Frank and Goodman, 2012)

I Speaker. Imagine you are talking to someone and want to refer to the middleobject. Would you say “blue” or “circle”?

I Listener. Someone uses the word “blue” to refer to one of these objects. Whichobject are they talking about?

46 / 67

Page 47: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts Model

Literal listener (l0)Pl0(s | u) ∝ P (s)JuK(s)

I P (s): the prior over states

I JuK(s): a mapping from states of the world to truth values

∀s, P (s) = 1/3

blue 0.5 0.5 0green 0 0 1square 0.5 0 0.5circle 0 1 0

47 / 67

Page 48: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts Model

Literal listener (l0)Pl0(s | u) ∝ P (s)JuK(s)

I P (s): the prior over states

I JuK(s): a mapping from states of the world to truth values

∀s, P (s) = 1/3

blue 0.5 0.5 0green 0 0 1square 0.5 0 0.5circle 0 1 0

48 / 67

Page 49: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts Model

Literal listener (l0)Pl0(s | u) ∝ P (s)JuK(s)

Pragmatic speaker (s1)Ps1(u | s, C) ∝ Us1(u; s)

I One way is to set the utility function to Pl0(s | u):

Ps1(u | s, C) ∝ Pl0(s | u) = exp(logPl0(s | u))

I More generally, we can incorporate message costs:

Ps1(u | s, C) ∝ exp(α(logPl0(s | u)− cost(u)))

49 / 67

Page 50: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts Model

Literal listener (l0)Pl0(s | u) ∝ P (s)JuK(s)

Pragmatic speaker (s1)Ps1(u | s, C) ∝ Us1(u; s)

I One way is to set the utility function to Pl0(s | u):

Ps1(u | s, C) ∝ Pl0(s | u) = exp(logPl0(s | u))

I More generally, we can incorporate message costs:

Ps1(u | s, C) ∝ exp(α(logPl0(s | u)− cost(u)))

50 / 67

Page 51: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts Model

Literal listener (l0)Pl0(s | u) ∝ P (s)JuK(s)

Pragmatic speaker (s1)Ps1(u | s, C) ∝ Us1(u; s)

I One way is to set the utility function to Pl0(s | u):

Ps1(u | s, C) ∝ Pl0(s | u) = exp(logPl0(s | u))

I More generally, we can incorporate message costs:

Ps1(u | s, C) ∝ exp(α(logPl0(s | u)− cost(u)))

51 / 67

Page 52: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts ModelLiteral listener (l0)

Pl0(s | u) ∝ P (s)JuK(s)Pragmatic speaker (s1)

Ps1(u | s, C) ∝ exp(α(logPl0(s | u)− cost(u)))

α = 1

cost(u) =

{0, u ∈ {blue, green, circle, square}∞, otherwise

blue green square circle

0.5 0 0.5 0

0.33 0 0 0.67

0 0.67 0.33 0

52 / 67

Page 53: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts Model

Literal listener (l0)Pl0(s | u) ∝ P (s)JuK(s)

Pragmatic speaker (s1)

Ps1(u | s, C) ∝ exp(α(logPl0(s | u)− cost(u)))

pragmatic speaker

blue green square circle

0.5 0 0.5 0

0.33 0 0 0.67

0 0.67 0.33 0

vs.

literal listener

blue 0.5 0.5 0green 0 0 1square 0.5 0 0.5circle 0 1 0

53 / 67

Page 54: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts ModelLiteral listener (l0)

Pl0(s | u) ∝ P (s)JuK(s)Pragmatic speaker (s1)

Ps1(u | s, C) ∝ exp(α(logPl0(s | u)− cost(u)))

Pragmatic listener (l1)Pl1(s | u) ∝ P (s)Ps1(u | s)

blue 0.6 0.4 0green 0 0 1square 0.6 0 0.4circle 0 1 0

54 / 67

Page 55: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts ModelLiteral listener (l0)

Pl0(s | u) ∝ P (s)JuK(s)Pragmatic speaker (s1)

Ps1(u | s, C) ∝ exp(α(logPl0(s | u)− cost(u)))

Pragmatic listener (l1)Pl1(s | u) ∝ P (s)Ps1(u | s)

blue 0.6 0.4 0green 0 0 1square 0.6 0 0.4circle 0 1 0

55 / 67

Page 56: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts ModelLiteral listener (l0)

Pl0(s | u) ∝ P (s)JuK(s)Pragmatic speaker (s1)

Ps1(u | s, C) ∝ exp(α(logPl0(s | u)− cost(u)))

Pragmatic listener (l1)Pl1(s | u) ∝ P (s)Ps1(u | s)

pragmatic listener

blue 0.6 0.4 0green 0 0 1square 0.6 0 0.4circle 0 1 0

vs.

literal listener

blue 0.5 0.5 0green 0 0 1square 0.5 0 0.5circle 0 1 0

Live demo: http://gscontras.github.io/ESSLLI-2016/

56 / 67

Page 57: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts ModelLiteral listener (l0)

Pl0(s | u) ∝ P (s)JuK(s)Pragmatic speaker (s1)

Ps1(u | s, C) ∝ exp(α(logPl0(s | u)− cost(u)))

Pragmatic listener (l1)Pl1(s | u) ∝ P (s)Ps1(u | s)

pragmatic listener

blue 0.6 0.4 0green 0 0 1square 0.6 0 0.4circle 0 1 0

vs.

literal listener

blue 0.5 0.5 0green 0 0 1square 0.5 0 0.5circle 0 1 0

Live demo: http://gscontras.github.io/ESSLLI-2016/57 / 67

Page 58: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts Model

Literal listener (l0): utterance meaning × state priorPragmatic speaker (s1): literal listener - utterance costsPragmatic listener (l1): pragmatic speaker × state prior

58 / 67

Page 59: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Rational Speech Acts Model

Literal speaker (s0): utterance meaning - utterance costsPragmatic listener (l1): literal speaker × state priorPragmatic speaker (s1): pragmatic listener - utterance costs

59 / 67

Page 60: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Experiments

Rational speech acts model is a powerful tool for understanding the pragmatic meaningof language.

60 / 67

Page 61: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Experiments

Rational speech acts model is a powerful tool for understanding the pragmatic meaningof language.

61 / 67

Page 62: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Extensions and critiques

I Learning based approach by featurizing utterances and states (Monroe and Potts,2015)

I Neural rational speech acts model (Monroe et al., 2017)

I Exceptions: sarcasm, irony, hedging, etc

I Cultural differences

62 / 67

Page 63: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Extensions and critiques

I Learning based approach by featurizing utterances and states (Monroe and Potts,2015)

I Neural rational speech acts model (Monroe et al., 2017)

I Exceptions: sarcasm, irony, hedging, etc

I Cultural differences

63 / 67

Page 64: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Summary

I Wording matters; we can learn useful insights from social interaction dataavailable nowadays

I Modeling conversations by categorizing speech acts

I Rational speech acts model can achieve pragmatics understanding

64 / 67

Page 65: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

Computational Pragmatics

Questions?https://chenhaot.com

[email protected]

65 / 67

Page 66: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

References I

John Langshaw Austin. How to do things with words. Oxford university press, 1975.

Christopher J Bryan, Gregory M Walton, Todd Rogers, and Carol S Dweck. Motivating voter turnout byinvoking the self. Proceedings of the National Academy of Sciences, 108(31):12653–12656, 2011. URLhttp://www.pnas.org/content/108/31/12653.abstract.

Cristian Danescu-Niculescu-Mizil, Justin Cheng, Jon Kleinberg, and Lillian Lee. You Had Me at Hello: HowPhrasing Affects Memorability. In Proceedings of ACL, 2012a. URLhttp://dl.acm.org/citation.cfm?id=2390524.2390647.

Cristian Danescu-Niculescu-Mizil, Lillian Lee, Bo Pang, and Jon Kleinberg. Echoes of Power: Language Effectsand Power Differences in Social Interaction. In Proceedings of the 21st International Conference on WorldWide Web, 2012b. URL http://doi.acm.org/10.1145/2187836.2187931.

Michael C Frank and Noah D Goodman. Predicting Pragmatic Reasoning in Language Games. Science, 336(6084):998–998, 2012.

H Paul Grice. Logic and conversation. 1975, pages 41–58, 1975.

Will Monroe and Christopher Potts. Learning in the rational speech acts model. In Proceedings of the 20thAmsterdam Colloquium, 2015.

Will Monroe, Robert X.D. Hawkins, Noah D. Goodman, and Christopher Potts. Colors in Context: A PragmaticNeural Model for Grounded Language Understanding. In Proceedings of ACL, 2017.

Vinodkumar Prabhakaran, Ashima Arora, and Owen Rambow. Staying on Topic: An Indicator of Power inPolitical Debates. In EMNLP, 2014.

66 / 67

Page 67: Natural Language Processing (CSEP 517): …...Natural Language Processing (CSEP 517): Computational Pragmatics Chenhao Tan c 2017 University of Washington chenhao@chenhaot.com May

References II

Andreas Stolcke, Klaus Ries, Noah Coccaro, Elizabeth Shriberg, Rebecca Bates, Daniel Jurafsky, Paul Taylor,Rachel Martin, Carol Van Ess-Dykema, and Marie Meteer. Dialogue act modeling for automatic tagging andrecognition of conversational speech. Computational linguistics, 26(3):339–373, 2000.

Chenhao Tan, Lillian Lee, and Bo Pang. The effect of wording on message propagation: Topic- andauthor-controlled natural experiments on Twitter. In Proceedings of ACL, 2014.

Chenhao Tan, Vlad Niculae, Cristian Danescu-Niculescu-Mizil, and Lillian Lee. Winning Arguments: InteractionDynamics and Persuasion Strategies in Good-faith Online Discussions. In Proceedings of WWW, 2016. URLhttp://dx.doi.org/10.1145/2872427.2883081.

Ludwig Wittgenstein. Philosophical investigations. 1953.

George Yule. Pragmatics. Oxford, 1996.

Ye Zhang, Erin Willis, Michael J Paul, Nomie Elhadad, and Byron C Wallace. Characterizing the (Perceived)Newsworthiness of Health Science Articles: A Data-Driven Approach. JMIR medical informatics, 4(3), 2016.

67 / 67