125
Testing Functional Explanations of Word Order Universals Michael Hahn Richard Futrell Stanford UC Irvine

Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

  • Upload
    others

  • View
    3

  • Download
    0

Embed Size (px)

Citation preview

Page 1: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Testing Functional Explanations of Word Order Universals

Michael Hahn Richard FutrellStanford UC Irvine

Page 2: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

(Greenberg 1963)

Page 3: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

U3: ‘Languages with dominant VSO order are alwaysprepositional.’

Page 4: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

U3: ‘Languages with dominant VSO order are alwaysprepositional.’

U4: ‘With overwhelmingly greater than chancefrequency, languages with normal SOV order arepostpositional.’

Page 5: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

U3: ‘Languages with dominant VSO order are alwaysprepositional.’

U4: ‘With overwhelmingly greater than chancefrequency, languages with normal SOV order arepostpositional.’

`Relative position of adposition & noun ~relative position ofverb & object’

Page 6: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

OV languages with postpositions

VO languages with prepositions

Page 7: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 8: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Why do these universals hold?

Innate constraints on language, ‘Universal Grammar’? (Chomsky 1981)

Facilitation of human communication? (Dryer 1992, Hawkins 1994)

Make languages learnable? (Culbertson 2017)

Page 9: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Why do these universals hold?

Innate constraints on language, ‘Universal Grammar’? (Chomsky 1981)

Facilitation of human communication? (Dryer 1992, Hawkins 1994)

Approach: Test functional explanations by implementing efficiency measures, optimizing grammars, and checking whether universals hold in optimized grammars.

Make languages learnable? (Culbertson 2017)

Page 10: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Three Efficiency Measures

Dependency Length Minimization (Rijkhoff, 1986; Hawkins, 1994, 2003)

Surprisal (Gildea and Jaeger, 2015; Ferrer-i Cancho, 2017)

Parsability (Hawkins, 1994, 2003)

Page 11: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Three Efficiency Measures

Dependency Length Minimization (Rijkhoff, 1986; Hawkins, 1994, 2003)

Page 12: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Three Efficiency Measures

Dependency Length Minimization (Rijkhoff, 1986; Hawkins, 1994, 2003)

21 1

Page 13: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Three Efficiency Measures

Dependency Length Minimization (Rijkhoff, 1986; Hawkins, 1994, 2003)

21 1+ + = 4

Page 14: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Three Efficiency MeasuresSurprisal

Surprisal(w1...wi-1) = -Σi log P(wi|w1...wi-1)

Page 15: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Three Efficiency MeasuresSurprisal

Surprisal(w1...wi-1) = -Σi log P(wi|w1...wi-1)

Estimated using recurrent neural networks, the strongest existing methods for estimating surprisal and predicting reading times.

Page 16: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Three Efficiency MeasuresParsability

Mary has two green books.

Page 17: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Three Efficiency MeasuresParsability

Mary has two green books.

Parsability(utterance) := log P(tree | utterance)

Page 18: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Three Efficiency MeasuresParsability

Mary has two green books.

Parsability(utterance) := log P(tree | utterance)

Estimated using a neural network model (Dozat and Manning 2017)

with extremely generic architecture.

Page 19: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Utility Informativity Cost-=

Amount of Meaning that can be extracted from utterance

Cost of processing utterance

λ

Combining Parsability + Surprisal

Page 20: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Utility Informativity Cost-=

Amount of Meaning that can be extracted from utterance

Cost of processing utterance

Long tradition as an explanation of language (Gabelentz 1903, Zipf 1949, Horn 1984, …)

λ

Combining Parsability + Surprisal

Page 21: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Utility Informativity Cost-=

Amount of Meaning that can be extracted from utterance ~ Parsability

Cost of processing utterance

~ Surprisal

λ

Combining Parsability + Surprisal

Long tradition as an explanation of language (Gabelentz 1903, Zipf 1949, Horn 1984, …)

Page 22: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Utility Informativity Cost-=

Amount of Meaning that can be extracted from utterance ~ Parsability

Cost of processing utterance

~ SurprisalLong tradition as an explanation of language (Gabelentz 1903, Zipf 1949, Horn 1984, …)

Formalized in Rational-Speech Acts models (Frank and Goodman 2012)

λ

Combining Parsability + Surprisal

Page 23: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Utility Informativity Cost-=

Long tradition as an explanation of language (Gabelentz 1903, Zipf 1949, Horn 1984, …)

Formalized in Rational-Speech Acts models (Frank and Goodman 2012)

Related to Signal Processing (Rate-Distortion Theory, Information Bottleneck)

λ

Combining Parsability + Surprisal

Amount of Meaning that can be extracted from utterance ~ Parsability

Cost of processing utterance

~ Surprisal

Page 24: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Why do the universals hold?

Innate constraints on language, ‘Universal Grammar’? (Chomsky 1981)

Facilitation of human communication? (Dryer 1992, Hawkins 1994)

Approach: Test processing explanations by implementing efficiency measures, optimizing grammars, and checking whether universals hold in optimized grammars.

Make languages learnable? (Culbertson 2017)

Page 25: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Testing Functional Explanations

Approach: Optimize the word orders of languages for the three objectives, keeping syntactic structures unchanged

Page 26: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Testing Functional Explanations

Approach: Optimize the word orders of languages for the three objectives, keeping syntactic structures unchanged

Languages have word order regularities ⇒ Not sufficient to optimize the word orders of individual sentences

Page 27: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Testing Functional Explanations

Approach: Optimize the word orders of languages for the three objectives, keeping syntactic structures unchanged

Languages have word order regularities ⇒ Not sufficient to optimize the word orders of individual sentences

Instead: optimize word order rules of entire languages

Page 28: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Testing Functional Explanations

Approach: Optimize the word orders of languages for the three objectives, keeping syntactic structures unchanged

Languages have word order regularities ⇒ Not sufficient to optimize the word orders of individual sentences

Instead: optimize word order rules of entire languages

That is: optimized languages have optimized but internally consistent grammatical regularities in word order, and agree with an actual natural language in all other respects.

Page 29: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

dobj

nummod

amod

Dependency Corpus

Page 30: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

dobj

nummod

amod

Mary

hastwo

greenbooks

Tree Topologies

Dependency Corpus

Page 31: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

dobj

nummod

amod

Mary

hastwo

greenbooks

Tree Topologies

Dependency Corpus Ordering GrammarNOUN ADJamod

0.3

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

-0.2

0.8

Page 32: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

dobj

nummod

amod

Mary

hastwo

greenbooks

Tree Topologies

Dependency Corpus Ordering GrammarNOUN ADJamod

0.3

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

-0.2

0.8

“Object follows verb”

Page 33: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

dobj

nummod

amod

Mary

hastwo

greenbooks

Tree Topologies

Dependency Corpus Ordering GrammarNOUN ADJamod

0.3

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

-0.2

0.8

“Adjective precedes noun”

“Object follows verb”

Page 34: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

dobj

nummod

amod

Mary

hastwo

greenbooks

Tree Topologies

Dependency Corpus Ordering GrammarNOUN ADJamod

0.3

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

-0.2

0.8

“Adjective precedes noun”

“Object follows verb”

“Numerals follow adjectives & precede nouns”

Page 35: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

dobj

nummod

amod

Mary

hastwo

greenbooks

Tree Topologies

Maryhastwogreenbooks

Counterfactual Corpus

Dependency Corpus Ordering GrammarNOUN ADJamod

0.3

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

-0.2

0.8

Page 36: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

dobj

nummod

amod

Mary

hastwo

greenbooks

Tree Topologies

Maryhastwogreenbooks

Counterfactual Corpus

Dependency Corpus Ordering GrammarNOUN ADJamod

0.3

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

-0.2

0.8

Each parameter setting generates a different counterfactual corpus.

Page 37: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

dobj

nummod

amod

Mary

hastwo

greenbooks

Tree Topologies

Maryhastwogreen books

Counterfactual Corpus

Dependency Corpus Ordering GrammarNOUN ADJamod

0.9

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.1

0.5

0.2

Each parameter setting generates a different counterfactual corpus.

Page 38: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

dobj

nummod

amod

Mary

hastwo

greenbooks

Tree Topologies

Maryhas twogreenbooks

Counterfactual Corpus

Dependency Corpus Ordering GrammarNOUN ADJamod

0.1

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.95

04.2

0.82

Each parameter setting generates a different counterfactual corpus.

Page 39: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Dependency Length Surprisal

Parsability

2.35.81.8

We compute processing measures on counterfactual corpora.

Page 40: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Dependency Length Surprisal

Parsability

2.35.81.8

Each parameter setting results in different values for the processing measures.

Page 41: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Dependency Length Surprisal

Parsability

2.94.52.9

Each parameter setting results in different values for the processing measures.

Page 42: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Dependency Length Surprisal

Parsability

3.47.81.2

Each parameter setting results in different values for the processing measures.

Page 43: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Dependency Length Surprisal

Parsability

3.47.81.2

Each parameter setting results in different values for the processing measures.

Which settings optimise the measures?

Page 44: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Dependency Length Surprisal

Parsability

3.47.81.2

Each parameter setting results in different values for the processing measures.

Which settings optimise the measures?

Do the optimised settings replicate the Greenberg correlations?

Page 45: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

For each objective, find parameters that optimise it.

NOUN ADJamod0.1

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.95

04.2

0.82

NOUN ADJamod0.1

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.85

0.1

0.22

Minimize Dep. Length Minimize Surprisal

NOUN ADJamod0.1

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

0.5

0.8

NOUN ADJamod0.21

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.45

0.4

0.32

Maximize Parsability Optimize Pars.+Surp.

Page 46: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

For each objective, find parameters that optimise it.

Repeat this for corpora from 51 real languages from Universal Dependencies Project.

NOUN ADJamod0.1

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.95

04.2

0.82

NOUN ADJamod0.1

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.85

0.1

0.22

Minimize Dep. Length Minimize Surprisal

NOUN ADJamod0.1

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

0.5

0.8

NOUN ADJamod0.21

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.45

0.4

0.32

Maximize Parsability Optimize Pars.+Surp.

Page 47: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

For each objective, find parameters that optimise it.

Repeat this for corpora from 51 real languages from Universal Dependencies Project.

0.1

0.95

04.2

0.82

0.1

0.85

0.1

0.22

Minimize Dep. Length Minimize Surprisal

NOUN ADJamod0.1

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

NOUN ADJ 0.1

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

0.5

0.8

0.7

0.5

0.8

0.21

0.45

NOUN ADJ 0.1

NOUN

NOUN ADJ 0.1

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

0.5

0.8

NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

0.5

0.8

0.4

0.32

Maximize Parsability Optimize Pars.+Surp.

1. How do the objectives compare?2. Which universals are predicted?

Minimize Dep. Length Minimize Surprisal

Page 48: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Surprisal and Parsability minimize Dependency Length

Page 49: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Surprisal and Parsability minimize Dependency Length

Page 50: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Surprisal and Parsability minimize Dependency Length

Communicative Utility predicts Dependency Length Minimization.

Page 51: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Better Parsability

Lower Surprisal

z-transformed on the level of languages

Language optimizes Surprisal and Parsability

Page 52: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Better Parsability

Lower Surprisal

Random Grammars

Language optimizes Surprisal and Parsability

Page 53: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Better Parsability

Lower Surprisal

Random Grammars

Grammars fit to Real Orderings

Language optimizes Surprisal and Parsability

Better Parsability

Page 54: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Better Parsability

Lower Surprisal

Random Grammars

Optimized for Surprisal

Optimized for Parsability

Optimized for Parsability+Surprisal

Grammars fit to Real Orderings

Language optimizes Surprisal and Parsability

Page 55: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 56: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

(Dryer 1992 in Language)

Page 57: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

(Dryer 1992 in Language)

`Relative position of adposition & noun ~relative position ofverb & object’

Page 58: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

We formalize the correlations in the Universal Dependencies format.

Page 59: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

(Dryer 1992 in Language)

Page 60: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

X

XX

Page 61: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

We formalize the correlations in the Universal Dependencies format.

For any word order grammar, we can then check which correlations it satisfies.

Page 62: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Are the universals satisfied by models fit to the actual orderings for our 50 languages?

%

Page 63: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Are the universals satisfied by models fit to the actual orderings for our 50 languages?

%

Page 64: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Are the universals satisfied by models fit to the actual orderings for our 50 languages?

Prevalence of SVO (Dryer 1992)

Limitation of formalisation

%

Page 65: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 66: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Percentage of grammars optimized for each objective satisfying the universal

Page 67: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Percentage of grammars optimized for each objective satisfying the universal

Assessing Significance:X = “Object precedes verb”Y = “Object-patterner precedes verb-patterner”

Logistic model:Y ~ X + (1+X|family) + (1+X|language)

Page 68: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 69: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 70: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 71: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 72: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Predictions largely complementary

Page 73: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Predictions mostly agree

Page 74: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Predictions mostly agree

Communicative Utility replicates predictions of Dependency Length Minimization.

Page 75: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Predictions mostly agree

Communicative Utility replicates predictions of Dependency Length Minimization.Both measures predict most of the correlation universals.

Page 76: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Conclusion

● Tested explanations of Greenberg correlation universals in terms of efficiency of human processing and communication

● Using corpora from 50 languages, constructed counterfactual optimized languages

● Most of the correlations can be derived from pressure to shorten dependencies, decrease surprisal, or increase parsability

● Clear evidence for functional explanations of word order universals

Page 77: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 78: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Optimized grammars are easier to parse even when sentences are presented in orders very different from natural language

ACEBD ADBEC ACEDBABCDE

Page 79: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Random grammarOptimized grammar

Random grammars remain hard to parse even as training data increases.

Page 80: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 81: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Formalizing ParsabilityNeural parser (Dozat and Manning 2017):

Mary met John

R

Mar

y

met

Jo

hn 1. BiLSTM reads the sentence2. Identify heads by

computing score for each pair of words

Generic architecture, no assumptions beyond sequential nature of input.

Page 82: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Formalizing ParsabilityInformation about syntactic tree that can be extracted from sentence:

Mary met John

R

Mar

y

met

Jo

hn

Page 83: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Formalizing Dependency Length

Distance between word and its syntactic head

summing over all words in sentence

sentence w = w1...wn

Page 84: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Formalizing Surprisal

Page 85: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Formalizing Surprisal

summing over all words in sentence

per-word surprisal

Page 86: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Formalizing Surprisal

Surprisal depends on the probability model P.

Page 87: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Formalizing Surprisal

Surprisal depends on the probability model P.

Right choice of P depends on the entire language!

Page 88: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Formalizing Surprisal

Given a word order grammar θd choose the model that minimizes surprisal on the resulting sentences.

Page 89: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Formalizing Surprisal

Use LSTM recurrent neural networks, the SOTA in probabilistic modelling of natural language and predicting reading times.Very general sequence models, arguably minimizing architectural biases.

Page 90: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Formalizing Informativity

Information about the syntactic tree that can be extracted from the sentence:

Page 91: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Formalizing Informativity

Information about the syntactic tree that can be extracted from the sentence:

Use a recent neural model (Dozat and Manning 2017) with generic architecture and SOTA performance on many languages.

Page 92: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Word Order Grammars

For each dependency type, there are two parameters:a. α: probability that whether dependent precede headb. β: determines distance

Page 93: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary

has

two green

booksαverb-object = 0.1

αverb-subject = 0.95

αnoun-numeral = 0.99 αnoun-adjective = 0.8

Page 94: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary

has

two green

booksαverb-object = 0.1

αverb-subject = 0.95

αnoun-numeral = 0.99 αnoun-adjective = 0.8

Page 95: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Maryhas

two

green

books

Page 96: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Word Order Grammars

For each dependency type, there are two parameters:a. α: probability that dependent precede headb. β: determines distance

Page 97: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary

has

two

green

books

βNoun-Adjective = -0.3

βNoun-Numeral = 0.8

Page 98: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary

has

two

green

books

βNoun-Adjective = -0.3

βNoun-Numeral = 0.8

softmax(βNoun-Adjective , βNoun-Numeral ) ~ (0.1, 0.9)

adjective first

numeral first

Page 99: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary

has

two

green

books

βNoun-Adjective = -0.3

βNoun-Numeral = 0.8

softmax(βNoun-Adjective , βNoun-Numeral ) ~ (0.1, 0.9)

adjective first

numeral first

Page 100: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Maryhas

twogreen

books

Page 101: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Maryhastwogreenbooks

Page 102: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Word Order Grammars

For each dependency type, there are two parameters:a. α: probability that dependent precede headb. β: determines distance

This specifies the space of possible grammars, within which we optimize.

Page 103: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

dobj

nummod

amod

Mary

hastwo

greenbooks

Tree Topologies

Maryhastwogreenbooks

Counterfactual Corpus

Dependency Corpus Ordering GrammarNOUN ADJamod

0.3

NOUN NUMnummod

VERB NOUNnsubj

VERB NOUNdobj

...

0.7

-0.2

0.8

Page 104: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

obj

nummod

amod

Will be working with trees in the Universal Dependencies format:

Page 105: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Mary has two green books

nsubj

obj

nummod

amod

Will be working with trees in the Universal Dependencies format:

To optimize grammars, we need a space of possible grammars.

Page 106: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 107: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 108: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN
Page 109: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

SOV

SVO

SOV and VSO support correlationSVO does not

VSO

Page 110: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Support SVO(Gibson et al 2013)

Page 111: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Dependency Length MinimizationShort syntactic dependencies ease processing (Gibson, 1998; Grodner and Gibson, 2005; Demberg and Keller, 2008; Bartek et al., 2011)

Page 112: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Dependency Length MinimizationShort syntactic dependencies ease processing (Gibson, 1998; Grodner and Gibson, 2005; Demberg and Keller, 2008; Bartek et al., 2011)

Quantitative corpus evidence from many languages confirms that languages have shorter dependencies than would be expected at random (Futrell et al., 2015).

Page 113: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Dependency Length MinimizationShort syntactic dependencies ease processing (Gibson, 1998; Grodner and Gibson, 2005; Demberg and Keller, 2008; Bartek et al., 2011)

Quantitative corpus evidence from many languages confirms that languages have shorter dependencies than would be expected at random (Futrell et al., 2015).

Argued to explain several of the Greenberg correlations (Rijkhoff, 1986; Hawkins, 1994, 2003)

Page 114: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

21 1

Page 115: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Two Objectives for Optimization

Dependency Length Minimization

Communicative Utility

Page 116: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Two Objectives for Optimization

Communicative Utility

Utility Informativity Cost-=

Amount of Meaning that can be extracted from utterance

Cost of processing utterance

λ

Long tradition as an explanation of language (Gabelentz 1903, Zipf 1949, Horn 1984, …)

Page 117: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Two Objectives for Optimization

Communicative Utility

Utility Informativity Cost-= λ

Page 118: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Two Objectives for OptimizationCommunicative Utility

Utility Informativity Cost-= λ

Mary has two green books.

Page 119: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Two Objectives for OptimizationCommunicative Utility

Utility Informativity Cost-= λ

Mary has two green books.

Informativity(utterance) := log P(tree | utterance) - log P(tree)

Page 120: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Two Objectives for OptimizationCommunicative Utility

Utility Informativity Cost-= λ

Mary has two green books.

Informativity(utterance) := log P(tree | utterance) - log P(tree)We use a neural network model (Dozat and Manning 2017) with extremely generic architecture.

Page 121: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Two Objectives for OptimizationCommunicative Utility

Utility Informativity Cost-= λ

Page 122: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Two Objectives for OptimizationCommunicative Utility

Utility Informativity Cost-= λ

Surprisal(wi|w1...wi-1) = -log P(wi|w1...wi-1)

Page 123: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Two Objectives for OptimizationCommunicative Utility

Utility Informativity Cost-= λ

Surprisal(wi|w1...wi-1) = -log P(wi|w1...wi-1)

We use recurrent neural networks, the SOTA in probabilistic modelling of natural language and predicting reading times.

Page 124: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Dependency Length Surprisal

Parsability

2.35.81.8

(1) For each objective, find parameters that optimise it.

(2) Which universals do the resulting counterfactual languages satisfy?

Page 125: Testing Functional Explanations of Word Order Universalsstanford.edu/~mhahn2/cgi-bin/files/CAMP2018.pdf · 2019. 2. 2. · Dependency Corpus Ordering Grammar NOUN ADJ amod 0.1 NOUN

Dependency Length Surprisal

Parsability

2.35.81.8

(1) For each objective, find parameters that optimise it.

(2) Which universals do the resulting counterfactual languages satisfy?