A Model-Theoretic Reconstruction of Type-Theoretic Semantics for - - PowerPoint PPT Presentation

a model theoretic reconstruction of type theoretic
SMART_READER_LITE
LIVE PREVIEW

A Model-Theoretic Reconstruction of Type-Theoretic Semantics for - - PowerPoint PPT Presentation

A Model-Theoretic Reconstruction of Type-Theoretic Semantics for Anaphora Matthew Gotham University of Oslo 22 nd Conference on Formal Grammar, 22 July 2017 Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 1 / 26 What this


slide-1
SLIDE 1

A Model-Theoretic Reconstruction of Type-Theoretic Semantics for Anaphora

Matthew Gotham

University of Oslo

22nd Conference on Formal Grammar, 22 July 2017

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 1 / 26

slide-2
SLIDE 2

What this talk is about

A framework for the semantics of anaphora and accessibility constraints. Inspired by analyses in type-theoretical approaches to semantics using dependent types, reconstructed in (more or less) simple type theory. We’ll look at a couple of examples of cross-sentential binding and a ‘donkey sentence’, and see how the the system blocks inaccessible antecedents. There are more examples (negation, proportional quantifiers, weak and strong readings) in the paper.

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 2 / 26

slide-3
SLIDE 3

‘Model-Theoretic’?

What I mean is that meanings will be given as expressions of a logical language, which are taken to be dispensable in favour of their interpretations in a model (as in Montague 1973), which is where the ‘real’ semantics is. Expressions of the language of type theory are not understood this way in TTS—see Luo 2014 and Ranta 1994: §2.27. However, I don’t want to lean to heavily on this point from now on.

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 3 / 26

slide-4
SLIDE 4

Pronouns bound outside of scope

(1) A donkey brays. Giles feeds it. ∃x(donkey(x) ∧ bray(x)) ∧ feed(giles, ?) (2) Every farmer who owns a donkey feeds it. ∀y.(farmer(y) ∧ ∃x.donkey(x) ∧ own(y, x)) → feed(y, ?) Various options pursued: ? := x, change the model theory to extend the scope of ∃x ? is a description, possibly indexed to situations ? is a constant manipulated by functions ...etc.

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 4 / 26

slide-5
SLIDE 5

In Type-Theoretic Semantics

(1) A donkey brays. Giles feeds it.

  • Σy : (Σx : DONKEY)(BRAY(x))
  • FEED(giles, π1y)
  • (Sundholm 1986, Ranta 1994)

λc.

  • Σw : (Σu : (Σx : e)(DONKEY(x)))(BRAY(π1u))
  • FEED(giles, π1π1w)
  • (Bekki 2014)

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 5 / 26

slide-6
SLIDE 6

Witnesses

dependent pairs

(1) A donkey brays. Giles feeds it.

  • Σy : (Σx : DONKEY)(BRAY(x))
  • FEED(giles, π1y)
  • The type of ordered pairs a, b , c such that:

a is a donkey, and b is a proof that a brays, and c is a proof that Giles feeds a.

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 6 / 26

slide-7
SLIDE 7

In Type-Theoretic Semantics

(2) Every farmer who owns a donkey feeds it.

  • Πz : (Σx : FARMER)((Σy : DONKEY)(OWN(x, y)))
  • FEED(π1z, π1π2z)
  • (Sundholm 1986, Ranta 1994)

λc.

  • Πu :(Σx : e)

(FARMER(x) × (Σv : (Σy : e)(DONKEY(y)))(OWN(x, π1v)))

  • FEED(π1u, π1π1π2π2u)
  • (Bekki 2014)

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 7 / 26

slide-8
SLIDE 8

Witnesses

dependent functions

(2) Every farmer who owns a donkey feeds it.

  • Πz : (Σx : FARMER)((Σy : DONKEY)(OWN(x, y)))
  • FEED(π1z, π1π2z)
  • The type of functions f such that:

the domain of f is the set of ordered pairs a, b, c such that:

a is a farmer, and b is a donkey, and c is a proof that a owns b, and

f maps every a, b, c in its domain to a proof that a feeds b.

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 8 / 26

slide-9
SLIDE 9

The idea behind this paper

(is very simple)

Formalize those glosses in higher-order logic (Jacobs & Melham (1993) have shown how). Work backwards to the lexical entries we need to derive them compositionally. N.B.: Limited polymorphism required. Event(ualitie)s play the role of proof objects. Discourse-level existential closure plays the role of the non-empty type condition.

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 9 / 26

slide-10
SLIDE 10

Types

Base types 1 unit e entities v eventualities t booleans Binary type constructors

  • functional types

× product types ( and × associate to the right, and × binds more tightly than )

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 10 / 26

slide-11
SLIDE 11

Terms

∗ : 1 unit f αβ(aα) : β application λvα(φβ) : αβ abstraction (aα, bβ) : α×β pairing [cα×β]0 : α lef projection [cα×β]1 : β right projection

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 11 / 26

slide-12
SLIDE 12

Example lexical entries

and ✱ ; →λpαβt.λqα×βγt.λiα.λaβ×γ.p(i)([a]0) ∧ q(i, [a]0)([a]1) . →λp1αt.∃zα.p(∗)(z) a →λPe×αβt.λV.λiβ.λa(e×α)×γ.P([a]0)(i) ∧ V([[a]0]0)(i, [a]0)([a]1) where V : eβ×e×αγt donkey →λae×1.λiα.donkey([a]0) brays →λxe.λiα.λev.bray(x, e) Giles →λPeα×eβt.λiα.λae×β.P([a]0)(i, [a0])([a]1) ∧ [a]0 = giles

  • wns →λD(eαvt)βγt.λxe.D
  • λye.λaα.λev.own(x, y, e)
  • it →λVαβγt.λiβ.V(gβα(i))(i)

where g stands for an arbitrarily-chosen free variable

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 12 / 26

slide-13
SLIDE 13

Instantiated lexical entries

a → λPe×11t.λVe1×e×1vt.λi1.λa(e×1)×v.P([a]0)(i) ∧ V([[a]0]0)(i, [a]0)([a]1) donkey → λae×1.λi1.donkey([a]0) brays → λxe.λi1×e×1.λev.bray(x, e) lef context for the whole sentence NP witness, part of the lef context for the VP VP witness a donkey brays → λi1.λa(e×1)×v.donkey([[a]0]0) ∧ bray([[a]0]0, [a]1)

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 13 / 26

slide-14
SLIDE 14

Instantiated lexical entries

Giles → λPe(1×(e×1)×v)×evt.λi1×(e×1)×v.λae×v.P([a]0)(i, [a]0)([a]1) ∧ [a]0 = giles

  • wns → λD.λxe.D
  • λye.λa(1×(e×1)×v)×e.λev.own(x, y, e)
  • where D : (e(1×(e×1)×v)×evt)(1×(e×1)×v)×evt

it → λVe(1×(e×1)×v)×evt.λi(1×(e×1)×v)×e.V(g(1×(e×1)×v)×ee(i))(i) lef context NP witness VP witness Giles owns it → λi1×(e×1)×v.λae×v.own([a]0, g(1×(e×1)×v)×ee(i, [a]0), [a]1) ∧ [a]0 = giles

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 14 / 26

slide-15
SLIDE 15

Instantiated lexical entries

; → λp1(e×1)×vt.λq1×(e×1)×ve×vt.λi1.λa((e×1)×v)×e×v.p(i)([a]0) ∧ q(i, [a]0)([a]1) . → λp1((e×1)×v)×e×vt.∃z((e×1)×v)×e×v.p(∗)(z) lef context first sentence witness, part of the lef context for the second sentence second sentence witness A donkey brays; Giles owns it. → ∃z((e×1)×v)×e×v.

  • donkey([[[z]0]0]0) ∧ bray([[[z]0]0]0, [[z]0]1)
  • wn
  • [[z]1]0, g(1×(e×1)×v)×ee((∗, [z]0), [[z]1]0), [[z]1]1
  • ∧ [[z]1]0 = giles
  • Matthew Gotham (Oslo)

Semantics for Anaphora Formal Grammar 2017 15 / 26

slide-16
SLIDE 16

Resolution of the free variable

g(1×(e×1)×v)×ee Natural resolution: a function that selects an element of (an element of...) a tuple (of tuples...) For any types α, β and γ: λbα.b is a natural resolution function (NRF). λbα×β.[b]0 is an NRF. λbα×β.[b]1 is an NRF. For any terms F : βγ and G : αβ, λbα.F(G(b)) is an NRF if F and G are NRFs. In this case, the resolution that we want gives us g := λb(1×(e×1)×v)×e.[[[[b]0]1]0]0

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 16 / 26

slide-17
SLIDE 17

With the pronoun resolution

g := λb(1×(e×1)×v)×e.[[[[b]0]1]0]0

⇒β ∃z((e×1)×v)×e×v.

  • donkey([[[z]0]0]0) ∧ bray([[[z]0]0]0, [[z]0]1)
  • wn
  • [[z]1]0, ([[[[((∗, [z]0), [[z]1]0)]0]1]0]0), [[z]1]1
  • ∧ [[z]1]0 = giles
  • ⇒β ∃z((e×1)×v)×e×v.
  • donkey([[[z]0]0]0) ∧ bray([[[z]0]0]0, [[z]0]1)
  • wn
  • [[z]1]0, [[[z]0]0]0, [[z]1]1
  • ∧ [[z]1]0 = giles

∃xe.∃ev.∃ye.∃dv.

  • donkey(x) ∧ bray(x, e)
  • wn(y, x, d) ∧ y = giles
  • Matthew Gotham (Oslo)

Semantics for Anaphora Formal Grammar 2017 17 / 26

slide-18
SLIDE 18

More lexical entries

every → λPe×αβt.λVeβ×e×αγt.λiβ.λf e×αγ.∀ae×α.P(a)(i) → V([a]0)(i, a)(f(a)) who → λVeβ×e×αγt.λPe×αβt.λae×α×γ.λiβ.P([a]0, [[a1]0])(i) ∧ V([a]0)(i, [a]0, [[a]1]0)([[a]1]1)

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 18 / 26

slide-19
SLIDE 19

A donkey sentence

Every farmer who owns a donkey feeds it. → ∃f e×1×(e×1)×vv.∀ae×1×(e×1)×v.

  • farmer([a]0) ∧ donkey([[[[a]1]1]0]0)

∧ own([a]0, [[[[a]1]1]0]0, [[[a]1]1]1)

  • → feed([a]0, g1×e×1×(e×1)×ve(∗, a), f(a))

(empty) lef context NP witness The resolution we want: g := λb1×e×1×(e×1)×v.[[[[[b]1]1]1]0]0

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 19 / 26

slide-20
SLIDE 20

Resolved

g := λb1×e×1×(e×1)×v.[[[[[b]1]1]1]0]0

⇒β ∃f e×1×(e×1)×vv.∀ae×1×(e×1)×v.

  • farmer([a]0) ∧ donkey([[[[a]1]1]0]0)

∧ own([a]0, [[[[a]1]1]0]0, [[[a]1]1]1)

  • → feed([a]0, [[[[a]1]1]0]0, f(a))

≡ ∀xe.∀ye.∀ev.

  • farmer(x)∧donkey(y)∧own(x, y, e)
  • → ∃dv.feed(x, y, d)

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 20 / 26

slide-21
SLIDE 21

Accessibility

Every donkey brays; Giles owns it. → ∃a(e×1v)×e×v.∀xe×1 donkey([x]0) → bray([x]0, [a]0(x))

  • wn([[a]1]0, g1×(e×1v)×ee(∗, [a]0), [[a]1]1)

∧ [[a]1]0 = giles

  • first sentence witness, part of the lef context for the second sentence

second sentence witness Given the type of g, there is no way for the pronoun to be bound to donkeys.

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 21 / 26

slide-22
SLIDE 22

Plurals

two → λP.λV.λiβ.λX.two

  • λxe.∃aα.∃dγ.X((x, a), d)
  • ∧ ∀be×α.∀cγ.X(b, c) →
  • P(b)(i) ∧ V([b]0)(i, b)(c)
  • where P : e×αβt, V : eβ×e×αγt and X, Y : (e×α)×γt

them → λVαβγt.λiβ.λcγ.∀aα.Gβαt(i) → V(a)(i)(c) where G stands for an arbitrarily-chosen free variable

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 22 / 26

slide-23
SLIDE 23

Instantiated

Two donkeys bray; Giles owns them. → ∃a((e×1)×vt)×e×v.two

  • λxe.∃y1.∃dv.[a]0((x, y), d)
  • ∧ ∀be×1

∀cv.[a]0(b, c) →

  • donkey([b]0) ∧ bray([b]0, c)
  • ∧ ∀ye.G(1×((e×1)×vt))×eet((∗, [a]0), [[a]1]0)(y)

  • wn([[a]1]0, y, [[a]1]1) ∧ [[a]1]0 = giles
  • first sentence witness, part of the lef context for the second sentence

second sentence witness

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 23 / 26

slide-24
SLIDE 24

Additional resolution conventions for plural (set) entities

For any types α, β and γ: λbα×βt.λYα.∃Zβ.b(Y, Z) is an NRF. λbα×βt.λYα.∃Zβ.b(Z, Y) is an NRF. So G(1×((e×1)×vt))×eet can be resolved to λb(1×((e×1)×vt))×e.λxe.∃n1.∃ev.[[b]0]1((x, n), e)

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 24 / 26

slide-25
SLIDE 25

Resolved

G := λb(1×((e×1)×vt))×e.λxe.∃n1.∃ev.[[b]0]1((x, n), e)

∃a((e×1)×vt)×e×v.two

  • λxe.∃y1.∃dv.[a]0((x, y), d)
  • ∧ ∀be×1

∀cv.[a]0(b, c) →

  • donkey([b]0) ∧ bray([b]0, c)
  • ∧ ∀ye.∃n1

∃ev.[a]0((y, n), e)

  • wn([[a]1]0, y, [[a]1]1) ∧ [[a]1]0 = giles

∃Re×vt.∃ze.∃ev.two(λxe.∃dv.R(x, d)

  • ∧ ∀ve

∀cv.R(v, c) → (donkey(v) ∧ bray(v, c))

  • ∧ ∀ye.∃bv(R(y, b)) → (own(z, y, e) ∧ z = giles)

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 25 / 26

slide-26
SLIDE 26

Conclusion

Reconstruction of type-theoretic treatment of anaphora in (more or less) simple type theory. Similarity to list-, or stack-based approaches to dynamic semantics (Dekker 1994, van Eijck 2001, de Groote 2006, Nouwen 2007). A semantic account of pronoun accessibility. Not much yet to say about:

Anti-locality effects (‘Principle B’). Crossover. Quantificational/modal subordination. Many other issues.

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 26 / 26

slide-27
SLIDE 27

References I

Bekki, Daisuke. 2014. Representing anaphora with dependent types. In Nicholas Asher & Sergei Soloviev (eds.), Logical aspects of computational linguistics (Lecture Notes in Computer Science 8535), 14–29. Berlin, Heidelberg: Springer. de Groote, Philippe. 2006. Towards a Montagovian account of dynamics. In Masayuki Gibson & Jonathan Howell (eds.), Proceedings of Semantics and Linguistic Theory, vol. 16, 1–16. Dekker, Paul. 1994. Predicate logic with anaphora. In Mandy Harvey & Lynn Samelmann (eds.), Proceedings of Semantics and Linguistic Theory,

  • vol. 4, 79–95.

van Eijck, Jan. 2001. Incremental dynamics. Journal of Logic, Language and Information 10. 319–351.

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 27 / 26

slide-28
SLIDE 28

References II

Jacobs, Bart & Tom Melham. 1993. Translating dependent type theory into higher order logic. In TLCA 1993: International conference on typed lambda calculi and applications (Lecture Notes in Computer Science 664). Berlin, Heidelberg: Springer. Luo, Zhaohui. 2014. Formal semantics in modern type theories: is it model-theoretic, proof-theoretic, or both?. In Nicholas Asher & Sergei Soloviev (eds.), Logical aspects of computational linguistics: 8th international conference (Lecture Notes in Computer Science 8535), 177–188. Berlin, Heidelberg: Springer. Montague, Richard. 1973. The proper treatment of quantification in ordinary

  • English. In Patrick Suppes, Julius Moravcsik & Jaakko Hintikka (eds.),

Approaches to natural language, 221–242. Dordrecht: D. Reidel. Nouwen, Rick. 2007. On dependent pronouns and dynamic semantics. Journal of Philosophical Logic 36(2). 123–154.

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 28 / 26

slide-29
SLIDE 29

References III

Ranta, Aarne. 1994. Type-theoretical grammar. (Indices 1). Oxford: Oxford University Press. Sundholm, Göran. 1986. Proof theory and meaning. In Dov Gabbay & Franz Guenther (eds.), Handbook of philosophical logic, vol. 3, 471–506. Dordrecht: D. Reidel.

Matthew Gotham (Oslo) Semantics for Anaphora Formal Grammar 2017 29 / 26