The Internal Structure of Coordinate Categories Roger Levy UC San - - PowerPoint PPT Presentation

the internal structure of coordinate categories
SMART_READER_LITE
LIVE PREVIEW

The Internal Structure of Coordinate Categories Roger Levy UC San - - PowerPoint PPT Presentation

Intro Conjoin Likes Model Experiments Discussion References The Internal Structure of Coordinate Categories Roger Levy UC San Diego Department of Linguistics Presented at Structure and Evidence in Linguistics 29 April 2013 Roger Levy


slide-1
SLIDE 1

Intro Conjoin Likes Model Experiments Discussion References

The Internal Structure of Coordinate Categories

Roger Levy

UC San Diego Department of Linguistics Presented at Structure and Evidence in Linguistics

29 April 2013

Roger Levy The Internal Structure of Coordinate Categories

slide-2
SLIDE 2

Intro Conjoin Likes Model Experiments Discussion References

The historical arc of constraint-based grammar

§ Hallmark principles of the Sag tradition of grammatical

analysis:

Roger Levy The Internal Structure of Coordinate Categories

slide-3
SLIDE 3

Intro Conjoin Likes Model Experiments Discussion References

The historical arc of constraint-based grammar

§ Hallmark principles of the Sag tradition of grammatical

analysis:

§ It’s essential to state formal claims about the nature of

grammar precisely

Roger Levy The Internal Structure of Coordinate Categories

slide-4
SLIDE 4

Intro Conjoin Likes Model Experiments Discussion References

The historical arc of constraint-based grammar

§ Hallmark principles of the Sag tradition of grammatical

analysis:

§ It’s essential to state formal claims about the nature of

grammar precisely

§ We must seriously engage with the full range of distributional

generalizations in the data

Roger Levy The Internal Structure of Coordinate Categories

slide-5
SLIDE 5

Intro Conjoin Likes Model Experiments Discussion References

The historical arc of constraint-based grammar

§ Hallmark principles of the Sag tradition of grammatical

analysis:

§ It’s essential to state formal claims about the nature of

grammar precisely

§ We must seriously engage with the full range of distributional

generalizations in the data

§ But we must also be rigorous in determining when to attribute

a distributional generalization to the grammar proper

Roger Levy The Internal Structure of Coordinate Categories

slide-6
SLIDE 6

Intro Conjoin Likes Model Experiments Discussion References

The historical arc of constraint-based grammar

§ Hallmark principles of the Sag tradition of grammatical

analysis:

§ It’s essential to state formal claims about the nature of

grammar precisely

§ We must seriously engage with the full range of distributional

generalizations in the data

§ But we must also be rigorous in determining when to attribute

a distributional generalization to the grammar proper

§ When implemented correctly, these principles are powerful in

identifying both grammatical knowledge and its interface with the rest of cognition

Roger Levy The Internal Structure of Coordinate Categories

slide-7
SLIDE 7

Intro Conjoin Likes Model Experiments Discussion References

The internal structure of coordinate categories

§ Principle of Conjoin Likes (Chomsky, 1965)

X Ñ X Conj X

Roger Levy The Internal Structure of Coordinate Categories

slide-8
SLIDE 8

Intro Conjoin Likes Model Experiments Discussion References

The internal structure of coordinate categories

§ Principle of Conjoin Likes (Chomsky, 1965)

X Ñ X Conj X

§ Empirically false for gross syntactic category (Sag et al.,

1985): Pat is a Republican and proud of it (NP and AdjP)

Roger Levy The Internal Structure of Coordinate Categories

slide-9
SLIDE 9

Intro Conjoin Likes Model Experiments Discussion References

The internal structure of coordinate categories

§ Principle of Conjoin Likes (Chomsky, 1965)

X Ñ X Conj X

§ Empirically false for gross syntactic category (Sag et al.,

1985): Pat is a Republican and proud of it (NP and AdjP)

§ And for case-marking (Przepi´

  • rkowski, 1999; Levy, 2001):

proˇ zdal “waited” governs Acc or Gen

Vˇ cera yesterday vec’ all den’ day

  • n

he proˇ zdal expected.acc or gen [np svoju self’s.acc podrugu girlfriend.acc Irinu] Irina.acc i and [np zvonka call.gen

  • t

from svoego self’s brata brother Grigorija]. Gregory (Russian, Levy, 2001) “Yesterday he waited all day for his girlfriend Irina and for a call from his brother Gregory.”

Roger Levy The Internal Structure of Coordinate Categories

slide-10
SLIDE 10

Intro Conjoin Likes Model Experiments Discussion References

What’s left for grammar?

§ Generalization: a coordination is categorically

grammatical iff it satisfies all the extrinsic constraints on its well-formedness (Ingria, 1990; Bayer and Johnson, 1995; Bayer,

1996; Dalrymple and Kaplan, 2000; Daniels, 2001; Levy, 2001; Levy and Pollard, 2001; Sag, 2003)

Roger Levy The Internal Structure of Coordinate Categories

slide-11
SLIDE 11

Intro Conjoin Likes Model Experiments Discussion References

What’s left for grammar?

§ Generalization: a coordination is categorically

grammatical iff it satisfies all the extrinsic constraints on its well-formedness (Ingria, 1990; Bayer and Johnson, 1995; Bayer,

1996; Dalrymple and Kaplan, 2000; Daniels, 2001; Levy, 2001; Levy and Pollard, 2001; Sag, 2003)

§ So. . . was “Conjoin Likes” just wrong?

Roger Levy The Internal Structure of Coordinate Categories

slide-12
SLIDE 12

Intro Conjoin Likes Model Experiments Discussion References

What’s left for grammar?

§ Generalization: a coordination is categorically

grammatical iff it satisfies all the extrinsic constraints on its well-formedness (Ingria, 1990; Bayer and Johnson, 1995; Bayer,

1996; Dalrymple and Kaplan, 2000; Daniels, 2001; Levy, 2001; Levy and Pollard, 2001; Sag, 2003)

§ So. . . was “Conjoin Likes” just wrong? § Is there anything left for grammar to say about a “tendency”

for coordinated categories to be like one another?

Roger Levy The Internal Structure of Coordinate Categories

slide-13
SLIDE 13

Intro Conjoin Likes Model Experiments Discussion References

Corpus data

§ Unlike-category coordinations are easy to find in corpora

His son had been friendly, a big fellow of fifty or more, a fishing-boat captain and powerful like the sea

(Parsed Brown corpus)

Roger Levy The Internal Structure of Coordinate Categories

slide-14
SLIDE 14

Intro Conjoin Likes Model Experiments Discussion References

Corpus data

§ Unlike-category coordinations are easy to find in corpora

His son had been friendly, a big fellow of fifty or more, a fishing-boat captain and powerful like the sea

§ But there is a huge quantitative tendency for coordination to

be of like categories in corpora Right-hand conjunct NP AdjP Left-hand Conjunct NP 1308 8 AdjP 6 114

(Parsed Brown corpus)

Roger Levy The Internal Structure of Coordinate Categories

slide-15
SLIDE 15

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes as a gradient grammatical constraint?

§ It is tempting to claim immediately that this pattern

illustrates a “soft constraint” (one of Miller’s “Usage Preferences”) toward Conjoin Likes

Roger Levy The Internal Structure of Coordinate Categories

slide-16
SLIDE 16

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes as a gradient grammatical constraint?

§ It is tempting to claim immediately that this pattern

illustrates a “soft constraint” (one of Miller’s “Usage Preferences”) toward Conjoin Likes

§ But should we really attribute this to the grammar proper?

Roger Levy The Internal Structure of Coordinate Categories

slide-17
SLIDE 17

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes as a gradient grammatical constraint?

Critical difference between nature of evidence for categorical versus probabilistic/gradient grammatical theories:

Roger Levy The Internal Structure of Coordinate Categories

slide-18
SLIDE 18

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes as a gradient grammatical constraint?

Critical difference between nature of evidence for categorical versus probabilistic/gradient grammatical theories:

§ Categorical: the possibility of a string is sufficient to demand

the grammar account for it, regardless of the extralinguistic circumstances required

Roger Levy The Internal Structure of Coordinate Categories

slide-19
SLIDE 19

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes as a gradient grammatical constraint?

Critical difference between nature of evidence for categorical versus probabilistic/gradient grammatical theories:

Roger Levy The Internal Structure of Coordinate Categories

slide-20
SLIDE 20

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes as a gradient grammatical constraint?

Critical difference between nature of evidence for categorical versus probabilistic/gradient grammatical theories:

§ Probabilistic/gradient theories: the data currency is relative

prevalence, and one must carefully disentangle the contributions of grammar and extralinguistic circumstances

Roger Levy The Internal Structure of Coordinate Categories

slide-21
SLIDE 21

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes as a gradient grammatical constraint?

Critical difference between nature of evidence for categorical versus probabilistic/gradient grammatical theories:

§ Probabilistic/gradient theories: the data currency is relative

prevalence, and one must carefully disentangle the contributions of grammar and extralinguistic circumstances

Roger Levy The Internal Structure of Coordinate Categories

slide-22
SLIDE 22

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes as a gradient grammatical constraint?

Critical difference between nature of evidence for categorical versus probabilistic/gradient grammatical theories:

§ Probabilistic/gradient theories: the data currency is relative

prevalence, and one must carefully disentangle the contributions of grammar and extralinguistic circumstances

S VP X Z and Y Vtrans NP S VP X Z and Y be NP NP N X Z and Y

Roger Levy The Internal Structure of Coordinate Categories

slide-23
SLIDE 23

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes as a gradient grammatical constraint?

Critical difference between nature of evidence for categorical versus probabilistic/gradient grammatical theories:

§ Probabilistic/gradient theories: the data currency is relative

prevalence, and one must carefully disentangle the contributions of grammar and extralinguistic circumstances

S VP X Z and Y Vtrans NP S VP X Z and Y be NP NP N X Z and Y

Ó Ó Ó NP and NP Uncorrelated Mixture AdjP and AdjP

Roger Levy The Internal Structure of Coordinate Categories

slide-24
SLIDE 24

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes as a gradient grammatical constraint?

Critical difference between nature of evidence for categorical versus probabilistic/gradient grammatical theories:

§ Probabilistic/gradient theories: the data currency is relative

prevalence, and one must carefully disentangle the contributions of grammar and extralinguistic circumstances

S VP X Z and Y Vtrans NP S VP X Z and Y be NP NP N X Z and Y

Ó Ó Ó NP and NP Uncorrelated Mixture AdjP and AdjP Œ Ó Ö Tendency for Conjoin Likes

Roger Levy The Internal Structure of Coordinate Categories

slide-25
SLIDE 25

Intro Conjoin Likes Model Experiments Discussion References

Directed Acyclic Graphical Models (“Bayes Nets”)

Earthquake? Burglary? Burglar Alarm? Phone call

(Example due to Russell and Norvig, 2003) Bayes Nets specify:

§ Probabilistic conditional independencies:

X and Y are conditionally independent given known variables iff every path between X and Y is blocked by:

Roger Levy The Internal Structure of Coordinate Categories

slide-26
SLIDE 26

Intro Conjoin Likes Model Experiments Discussion References

Directed Acyclic Graphical Models (“Bayes Nets”)

Earthquake? Burglary? Burglar Alarm? Phone call

(Example due to Russell and Norvig, 2003) Bayes Nets specify:

§ Probabilistic conditional independencies:

X and Y are conditionally independent given known variables iff every path between X and Y is blocked by:

§ an unknown variable with “converging

arrows”; or

Roger Levy The Internal Structure of Coordinate Categories

slide-27
SLIDE 27

Intro Conjoin Likes Model Experiments Discussion References

Directed Acyclic Graphical Models (“Bayes Nets”)

Earthquake? Burglary? Burglar Alarm? Burglar Alarm? Burglar Alarm? Phone call

(Example due to Russell and Norvig, 2003) Bayes Nets specify:

§ Probabilistic conditional independencies:

X and Y are conditionally independent given known variables iff every path between X and Y is blocked by:

§ an unknown variable with “converging

arrows”; or

Roger Levy The Internal Structure of Coordinate Categories

slide-28
SLIDE 28

Intro Conjoin Likes Model Experiments Discussion References

Directed Acyclic Graphical Models (“Bayes Nets”)

Earthquake? Burglary? Burglar Alarm? Phone call

(Example due to Russell and Norvig, 2003) Bayes Nets specify:

§ Probabilistic conditional independencies:

X and Y are conditionally independent given known variables iff every path between X and Y is blocked by:

§ an unknown variable with “converging

arrows”; or

§ a known variable without “converging

arrows”

Roger Levy The Internal Structure of Coordinate Categories

slide-29
SLIDE 29

Intro Conjoin Likes Model Experiments Discussion References

Directed Acyclic Graphical Models (“Bayes Nets”)

Earthquake? Burglary? Burglar Alarm? Phone call

(Example due to Russell and Norvig, 2003) Bayes Nets specify:

§ Probabilistic conditional independencies:

X and Y are conditionally independent given known variables iff every path between X and Y is blocked by:

§ an unknown variable with “converging

arrows”; or

§ a known variable without “converging

arrows”

Roger Levy The Internal Structure of Coordinate Categories

slide-30
SLIDE 30

Intro Conjoin Likes Model Experiments Discussion References

Directed Acyclic Graphical Models (“Bayes Nets”)

Earthquake? Burglary? Burglar Alarm? Phone call

(Example due to Russell and Norvig, 2003) Bayes Nets specify:

§ Probabilistic conditional independencies:

X and Y are conditionally independent given known variables iff every path between X and Y is blocked by:

§ an unknown variable with “converging

arrows”; or

§ a known variable without “converging

arrows”

§ The basic units of probabilistic

(=gradient) knowledge, Ppchild|parentsq: PpAlarm|Earthquake, Burglaryq PpCall|Alarmq

Roger Levy The Internal Structure of Coordinate Categories

slide-31
SLIDE 31

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes in a probabilistic grammar

M1 M2 F1 F2 O M1, M2 Intended conjunct meanings and extrinsic constraints F1, F2 Realized linguistic forms of the conjuncts O Ordering decision

(NB: Connections from Mi to O are necessary to account for semantic interpretive constraints pertaining to order, e.g., eat and run‰run and eat; Cooper and Ross, 1975)

Roger Levy The Internal Structure of Coordinate Categories

slide-32
SLIDE 32

Intro Conjoin Likes Model Experiments Discussion References

Conjoin Likes in a probabilistic grammar

M1 M2 F1 F2 O ? M1, M2 Intended conjunct meanings and extrinsic constraints F1, F2 Realized linguistic forms of the conjuncts O Ordering decision

(NB: Connections from Mi to O are necessary to account for semantic interpretive constraints pertaining to order, e.g., eat and run‰run and eat; Cooper and Ross, 1975)

Roger Levy The Internal Structure of Coordinate Categories

slide-33
SLIDE 33

Intro Conjoin Likes Model Experiments Discussion References

Model

What “gradient coordination of like categories” means: PpF1, F2|M1, M2q is especially high when F1 and F2 are “like” in the traditional sense of XÑX and X Fully technically: pMIpF1, F2|M1, M2q “ log PpF1, F2|M1, M2q PpF1|M1qPpF2|M2q is monotonically increasing in the structural similarity of F1 and F2

Roger Levy The Internal Structure of Coordinate Categories

slide-34
SLIDE 34

Intro Conjoin Likes Model Experiments Discussion References

Empirical prediction

If forms are gradiently “more grammatical” to the native speaker when they are more probable. . .

Roger Levy The Internal Structure of Coordinate Categories

slide-35
SLIDE 35

Intro Conjoin Likes Model Experiments Discussion References

Empirical prediction

If forms are gradiently “more grammatical” to the native speaker when they are more probable. . . . . . then like-category coordinations should be judged to be more natural, or acceptable, than unlike-category coordinations

Roger Levy The Internal Structure of Coordinate Categories

slide-36
SLIDE 36

Intro Conjoin Likes Model Experiments Discussion References

Experiment 1

Acceptability judgment study (scale of 1–9): Pat is a Republican and a freak. [Noun Noun] Pat is a Republican and freaky. [Noun Adj ] Pat is Republican and a freak. [Adj Noun] Pat is Republican and freaky. [Adj Adj ]

(Baseline: The children decorated the sparkling ornaments onto the tree was a 4.)

Roger Levy The Internal Structure of Coordinate Categories

slide-37
SLIDE 37

Intro Conjoin Likes Model Experiments Discussion References

Experiment 1: Results

Noun Noun Noun Adj Adj Noun Adj Adj Naturalness Rating (1−9) 2 4 6 8

The gradient preference for coordination of unlike categories is pretty strong!

Roger Levy The Internal Structure of Coordinate Categories

slide-38
SLIDE 38

Intro Conjoin Likes Model Experiments Discussion References

Greater explanatory power of gradient constraints

§ We saw that “Conjoin Likes” is categorically false, but

“probabilistically” true

Roger Levy The Internal Structure of Coordinate Categories

slide-39
SLIDE 39

Intro Conjoin Likes Model Experiments Discussion References

Greater explanatory power of gradient constraints

§ We saw that “Conjoin Likes” is categorically false, but

“probabilistically” true

§ But why stop at major syntactic categories—what about

category-internal structure (Johnson, 1998; Klein and Manning, 2003)?

Roger Levy The Internal Structure of Coordinate Categories

slide-40
SLIDE 40

Intro Conjoin Likes Model Experiments Discussion References

Greater explanatory power of gradient constraints

§ We saw that “Conjoin Likes” is categorically false, but

“probabilistically” true

§ But why stop at major syntactic categories—what about

category-internal structure (Johnson, 1998; Klein and Manning, 2003)?

§ Such a grammatical preference has previously been explored

under the rubric of parallelism (Frazier et al., 1984; Hale et al., 2006; Dubey et al., 2008) NP NP . . . α . . . Conj and NP . . . α . . .

Roger Levy The Internal Structure of Coordinate Categories

slide-41
SLIDE 41

Intro Conjoin Likes Model Experiments Discussion References

NP-internal parallelism: the genitive alternation

Postnominal Prenominal The future of our country „ Our country’s future The base of the lamp „ The lamp’s base The tail of a cat „ A cat’s tail

NP NP PPof NP Conj and NP PPof NP NP NP NP PossP Conj and NP PPof NP NP NP NP PPof Conj and NP NP PossP NP NP NP PossP Conj and NP NP PossP

Roger Levy The Internal Structure of Coordinate Categories

slide-42
SLIDE 42

Intro Conjoin Likes Model Experiments Discussion References

Corpus data on genitive alternation parallelism

Right Conjunct Post Pre Left Conjunct Post 77 15 Pre 20 39

§ There is also strong evidence for a parallelism preference in

the genitive alternation. . .

Roger Levy The Internal Structure of Coordinate Categories

slide-43
SLIDE 43

Intro Conjoin Likes Model Experiments Discussion References

Corpus data on genitive alternation parallelism

Right Conjunct Post Pre Left Conjunct Post 77 15 Pre 20 39

§ There is also strong evidence for a parallelism preference in

the genitive alternation. . .

§ . . . but once again this analysis fails to control for conjunct

meanings M1, M2

Roger Levy The Internal Structure of Coordinate Categories

slide-44
SLIDE 44

Intro Conjoin Likes Model Experiments Discussion References

Corpus data on genitive alternation parallelism

Right Conjunct Post Pre Left Conjunct Post 77 15 Pre 20 39

§ There is also strong evidence for a parallelism preference in

the genitive alternation. . .

§ . . . but once again this analysis fails to control for conjunct

meanings M1, M2

§ We can control this more tightly with an experiment

Roger Levy The Internal Structure of Coordinate Categories

slide-45
SLIDE 45

Intro Conjoin Likes Model Experiments Discussion References

Experiment 2

Acceptability judgment study (scale of 1–9):

Terry assembled. . . . . . the frame of the chair and the base of the lamp. [Post Post] . . . the frame of the chair and the lamp’s base. [Post Pre ] . . . the chair’s frame and the base of the lamp. [Pre Post] . . . the chair’s frame and the lamp’s base. [Pre Pre ]

(Baseline: The children decorated the sparkling ornaments onto the tree was a 4.)

Roger Levy The Internal Structure of Coordinate Categories

slide-46
SLIDE 46

Intro Conjoin Likes Model Experiments Discussion References

Experiment 2: Results

Post Post Post Pre Pre Post Pre Pre Naturalness Rating (1−9) 2 4 6 8

There is also a preference for parallelism among realizations of the genitive alternation!

Roger Levy The Internal Structure of Coordinate Categories

slide-47
SLIDE 47

Intro Conjoin Likes Model Experiments Discussion References

Comparison of the parallelism effects

But Conjoin Likes ą genitive parallelism!

Unlike categories Genitive Alternation Parallelism effect size 0.0 0.2 0.4 0.6 0.8

Roger Levy The Internal Structure of Coordinate Categories

slide-48
SLIDE 48

Intro Conjoin Likes Model Experiments Discussion References

Discussion

M1 M2 F1 F2 O ?

§ Grammar has very little to say about categorical constraints

  • n the relation between conjuncts

Roger Levy The Internal Structure of Coordinate Categories

slide-49
SLIDE 49

Intro Conjoin Likes Model Experiments Discussion References

Discussion

M1 M2 F1 F2 O ?

§ Grammar has very little to say about categorical constraints

  • n the relation between conjuncts

§ But corpus data suggest there’s much more to say about the

gradient constraints on their relation

Roger Levy The Internal Structure of Coordinate Categories

slide-50
SLIDE 50

Intro Conjoin Likes Model Experiments Discussion References

Discussion

M1 M2 F1 F2 O ?

§ Grammar has very little to say about categorical constraints

  • n the relation between conjuncts

§ But corpus data suggest there’s much more to say about the

gradient constraints on their relation

§ We now have the technical tools to formally characterize

these gradient constraints

Roger Levy The Internal Structure of Coordinate Categories

slide-51
SLIDE 51

Intro Conjoin Likes Model Experiments Discussion References

Discussion

M1 M2 F1 F2 O ?

§ Grammar has very little to say about categorical constraints

  • n the relation between conjuncts

§ But corpus data suggest there’s much more to say about the

gradient constraints on their relation

§ We now have the technical tools to formally characterize

these gradient constraints

§ This formalization revealed a weakness of (sparse) corpus data

and guided experiments to test for and quantify the strength

  • f these constraints

Roger Levy The Internal Structure of Coordinate Categories

slide-52
SLIDE 52

Intro Conjoin Likes Model Experiments Discussion References

Discussion

M1 M2 F1 F2 O ?

§ Grammar has very little to say about categorical constraints

  • n the relation between conjuncts

§ But corpus data suggest there’s much more to say about the

gradient constraints on their relation

§ We now have the technical tools to formally characterize

these gradient constraints

§ This formalization revealed a weakness of (sparse) corpus data

and guided experiments to test for and quantify the strength

  • f these constraints

§ We found that gradient “Conjoin Likes” is real, and has

greater explanatory reach than was ever claimed for the categorical version!

Roger Levy The Internal Structure of Coordinate Categories

slide-53
SLIDE 53

Intro Conjoin Likes Model Experiments Discussion References

Discussion

But why should gradient “Conjoin Likes” exist in the first place?

Roger Levy The Internal Structure of Coordinate Categories

slide-54
SLIDE 54

Intro Conjoin Likes Model Experiments Discussion References

Discussion

But why should gradient “Conjoin Likes” exist in the first place?

§ Stylistic preference?

Roger Levy The Internal Structure of Coordinate Categories

slide-55
SLIDE 55

Intro Conjoin Likes Model Experiments Discussion References

Discussion

But why should gradient “Conjoin Likes” exist in the first place?

§ Stylistic preference? § Psychological mechanism

(priming)?

Roger Levy The Internal Structure of Coordinate Categories

slide-56
SLIDE 56

Intro Conjoin Likes Model Experiments Discussion References

Discussion

But why should gradient “Conjoin Likes” exist in the first place?

§ Stylistic preference? § Psychological mechanism

(priming)?

§ Or is there a deeper source of

explanation in the nature of probabilistic grammatical knowledge? M1 M2 F1 F2 O

Roger Levy The Internal Structure of Coordinate Categories

slide-57
SLIDE 57

Intro Conjoin Likes Model Experiments Discussion References

Discussion

But why should gradient “Conjoin Likes” exist in the first place?

§ Stylistic preference? § Psychological mechanism

(priming)?

§ Or is there a deeper source of

explanation in the nature of probabilistic grammatical knowledge? M1 M2 F1 F2 O G

Roger Levy The Internal Structure of Coordinate Categories

slide-58
SLIDE 58

Intro Conjoin Likes Model Experiments Discussion References

Discussion

But why should gradient “Conjoin Likes” exist in the first place?

§ Stylistic preference? § Psychological mechanism

(priming)?

§ Or is there a deeper source of

explanation in the nature of probabilistic grammatical knowledge? M1 M2 F1 F2 O G

Roger Levy The Internal Structure of Coordinate Categories

slide-59
SLIDE 59

Intro Conjoin Likes Model Experiments Discussion References

Discussion

But why should gradient “Conjoin Likes” exist in the first place?

§ Stylistic preference? § Psychological mechanism

(priming)?

§ Or is there a deeper source of

explanation in the nature of probabilistic grammatical knowledge?

§ The Sag tradition of precise formal

claims, serious engagement with data, and rigor in assigning credit for distributional generalizations will be essential to working this out M1 M2 F1 F2 O G

Roger Levy The Internal Structure of Coordinate Categories

slide-60
SLIDE 60

Intro Conjoin Likes Model Experiments Discussion References

Thank you, Ivan!

Roger Levy The Internal Structure of Coordinate Categories

slide-61
SLIDE 61

Intro Conjoin Likes Model Experiments Discussion References

References I

Bayer, S. (1996). The coordination of unlike categories. Language, 72(3):579–616. Bayer, S. and Johnson, M. (1995). Features and agreement. In Proceedings of the 1995 ACL, pages 70–76. Association of Computational Linguistics. Chomsky, N. (1965). Aspects of the Theory of Syntax. Cambridge, MA: MIT Press. Cooper, W. and Ross, J. (1975). World order. In Proceedings of the Chicago Linguistic Society, pages 63–111. Dalrymple, M. and Kaplan, R. (2000). Feature indeterminacy and feature

  • resolution. Language, 76(4):759–798.

Daniels, M. W. (2001). On a type-based analysis of feature neutrality and the coordination of unlikes. In Proceedings of HPSG, volume 8. Dubey, A., Keller, F., and Sturt, P. (2008). A probabilisitic corpus-based model of syntactic parallelism. Cognition, 109(3):326–344.

Roger Levy The Internal Structure of Coordinate Categories

slide-62
SLIDE 62

Intro Conjoin Likes Model Experiments Discussion References

References II

Frazier, L., Taft, L., Roeper, T., Clifton, C., and Ehrlich, K. (1984). Parallel structure: A source of facilitation in sentence comprehension. Memory & Cognition, 12(5):421–430. Hale, J., Shafran, I., Yung, L., Dorr, B., Harper, M., Krasnyanskaya, A., Lease, M., Liu, Y., Roark, B., Snover, M., et al. (2006). Pcfgs with syntactic and prosodic indicators of speech repairs. In Proceedings of the 21st International Conference on Computational Linguistics and the 44th annual meeting of the Association for Computational Linguistics, pages 161–168. Association for Computational Linguistics. Ingria, R. J. P. (1990). The limits of unification. In Proceedings of the 28th Annual Meeting of the ACL, pages 194–204. Association for Computational Lingusitics. Johnson, M. (1998). PCFG models of linguistic tree representations. Computational Linguistics, 24(4):613–632. Klein, D. and Manning, C. D. (2003). Accurate unlexicalized parsing. In Proceedings of ACL.

Roger Levy The Internal Structure of Coordinate Categories

slide-63
SLIDE 63

Intro Conjoin Likes Model Experiments Discussion References

References III

Levy, R. (2001). Feature indeterminacy and the coordination of unlikes in a totally well-typed HPSG. MS., Stanford University. Levy, R. and Pollard, C. (2001). Coordination and neutralization in

  • HPSG. In Proceedings of the 8th International Conference on

Head-Driven Phrase Structure Grammar. CSLI Publications. Przepi´

  • rkowski, A. (1999). Case Assignment and the

Complement-Adjunct Dichotomy: A Non-Configurational Constraint-Based Approach. PhD thesis, Universit¨ at T¨ ubingen, Germany. Russell, S. and Norvig, P. (2003). Artificial Intelligence: a Modern

  • Approach. Prentice Hall, second edition.

Sag, I. (2003). Coordination and underspecification. In Kim, J.-B. and Wechsler, S., editors, Proceedings of HPSG, volume 10. Sag, I. A., Gazdar, G., Wasow, T., and Weisler, S. (1985). Coordination and how to distinguish categories. Natural Language and Linguistic Theory, 3:117–171.

Roger Levy The Internal Structure of Coordinate Categories