SlideShare une entreprise Scribd logo
1  sur  42
Télécharger pour lire hors ligne
Logical Description Grammar
for sentence and discourse interpretation
Introduction to the grammar system
Noor van Leusen
Radboud University
2012
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 1 / 42
Content
Content
Logical Description Grammar as a method for the description of
sentence syntax and semantics.
Muskens 1996, 2001. /cf. ‘Saarbruecken school’.
Underspecified representations. Syntactic and semantic properties of
linguistic trees. Parsing by deduction. CDRT and local contexts in the
semantic dimension.
vLeusen&Muskens 2003
Discourse processing and the computation of coherence in LDG.
Anaphora and Presupposition.
vLeusen 2007 /cf. Gardent&Webber 1998, Egg&Redeker 2006.
Implicit discourse relations. Inferring discourse meaning.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 2 / 42
LDG framework for Sentence Analysis Introducing Logical Description Grammar
Description Grammar for sentence analysis
LDG combines description theory (Vijay-Shanker c.s.) lexicalised TAG
(Joshi, Webber c.s.) and compositional DRT (Kamp, Muskens).
With a view to ambiguity and implicitness of meaning it generates
‘underspecified representations’.
It describes syntactic, semantic and pragmatic constraints in parallel,
and is specifically tuned to model the interaction of these constraints.
The semantics integrates a context parameter, employed in the
treatment of anaphora, presupposition and accommodation.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 3 / 42
LDG framework for Sentence Analysis Introducing Logical Description Grammar
Discourse contexts and discourse descriptions
Discourse contexts are represented by discourse descriptions.
Discourse descriptions are built up incrementally from sentence descriptions.
r
k′
+
relation
k
Sentence representations are linked to their discourse context by explicit
connectives or implicit discourse relations.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 4 / 42
LDG framework for Sentence Analysis Introducing Logical Description Grammar
Underspecified representations
There are two levels of linguistic analysis:
sentence description (underspecified representation)
?
set of verifying trees (fully specified linguistic repr.)
Sentence descriptions are statements in classical type logic.
They partially describe properties of linguistic trees, constraining
syntactic, semantic and pragmatic features in parallel.
In case of ambiguity, more than one tree class verifies a description.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 5 / 42
LDG framework for Sentence Analysis Linguistic Representations
Linguistic Tree Representations
Sentence and discourse representations are tree structures, decorated with
lexemes, syntactic labels, semantic values, and local contexts.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 6 / 42
LDG framework for Sentence Analysis Linguistic Representations
Example tree, showing just syntax and lexemes:
S
S
DP
D
Max
VP
V
opened
DP
D
a
NP
N
can
In situ analysis of quantifiers (’a can’): lexeme in surface position,
quantification potential is scoped out (at top S).
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 7 / 42
LDG framework for Sentence Analysis Linguistic Representations
Semantic values added:
S
[u2 | wr : can u2 ] ⊕ [ | wr : o0 opened u2]
S
[ |wr : o0 opened u2]
DP
o0
D
o0
Max
VP
λv [ |wr : v opened u2]
V
λv′
λv [ |wr : v opened v′
]
opened
DP
u2
D
u2
a
NP
λv [ |wr : can v]
N
λv [ |wr : can v]
can
The semantic representation language is finegrained compositional DRT, a
Lambda-DRT; [...|.....] is a DRS, ⊕ merges DRSs.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 8 / 42
LDG framework for Sentence Analysis Linguistic Representations
Fine-grained compositional DRT
[..|...] is a DRS, to the left of the | sign is the universe, to the right
the conditions.
uk, wk, ok , . . . are discourse markers;
uk , ok store individuals.
uk represent new referents (generated in the discourse);
ok represent given referents; they are globally available and interpreted
referentially rather than existentially;
wk store worlds.
wr is the marker for the actual world.
⊕ merges DRSs, ⊑ denotes inclusion, |≍ entailment.
‘proper(K)’ defines K as a DRS in which no free discourse markers
occur.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 9 / 42
LDG framework for Sentence Analysis Linguistic Representations
Context Parameter
Each node in a tree structure carries a DRS which is its local context.
(cf. Karttunen 1974)
The compositional semantics refers to this context parameter in the
resolution of anaphora, presupposition satisfaction and projection, and
accommodation in general.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 10 / 42
LDG framework for Sentence Analysis Linguistic Representations
Local contexts and a presuppositional constraint added:
S
[u2 | wr : can u2 ] ⊕ [ | wr : o0 opened u2]
B
S
[ |wr : o0 opened u2]
B ⊕ [u2 | wr : can u2]
DP
o0
B ⊕ [u2 | wr : can u2]
[o1 | wr : Max o0] ⊑ B
D
o0
B ⊕ [u2 | wr : can u2]
Max
VP
λv [ |wr : v opened u2]
B ⊕ [u2 | wr : can u2]
V
λv′λv [ |wr : v opened v′]
B ⊕ [u2 | wr : can u2]
opened
DP
u2
B ⊕ [u2 | wr : can u2]
D
u2
B ⊕ [u2 | wr : can u2]
a
NP
λv [ |wr : can v]
B ⊕ [u2 | wr : can u2]
N
λv [ |wr : can v]
B ⊕ [u2 | wr : can u2]
can
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 11 / 42
LDG framework for Sentence Analysis Linguistic Representations
Local contexts and context-sensitivity
Local contexts ‘collect’ information on the semantic tier of the
discourse context.
They are constructed going top-down and from left to right through a
discourse or sentence tree.
The local context of the root node is B, the implicit background.
Anaphora and projective elements such as presupposition triggers
introduce conditions on local contexts.
Accommodation/projection results from satisfying these conditions in
partially underspecified local backgrounds.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 12 / 42
LDG framework for Sentence Analysis Description Grammar as a reasoning system
Description Grammar, a constraint-based reasoning system
A description grammar G consists of lexical descriptions and some
general axioms.
It generates a sentence description ∆u per processed utterance u.
The grammar is itself a description, a set of statements in type logic.
Given the theory G + ∆u, a language user may reason about and
obtain the possible verifying trees of the sentence description M(∆u).
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 13 / 42
LDG framework for Sentence Analysis Description Grammar as a reasoning system
While G represents a language user’s general linguistic knowledge, a
sentence description ∆u represents his specific linguistic knowledge of
the sentence.
Linguistic knowledge means knowledge in any of the relevant
domains, such as phonology, syntax, semantic and pragmatics.
Sentence interpretation is a reasoning process.
Since descriptions can be partial, G + ∆u can have more than a single
verifying tree, and the syntax or semantics of the sentence can remain
underspecified.
Each verifying tree comes with a possible reading of the sentence.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 14 / 42
LDG framework for Sentence Analysis Lexicon and General Axioms
G consists of lexical descriptions and general axioms
General axioms constrain the general configurational and semantic
properties of linguistic representations.
∀k ¬[k ≺ k] No node precedes itself
∀k [Γk |≍⊥ ∧ Γr ⊕ σr |≍⊥] Local contexts are consistent
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 15 / 42
LDG framework for Sentence Analysis Lexicon and General Axioms
Lexical descriptions
characterise the syntactic, semantic, and pragmatic properties of
lexemes. They describe ‘little trees’ similar to elementary structures in
TAG.
Lexical description of the noun can:
∀k [can(k) → (cn(k) ∧ σ(k) = λv [ |wr : can v]) ]
∀k [cn(k) → ∃k2(ℓ(k) = n ∧ ℓ(k2) = np ∧ k2 ¡ k∧
σ(k2) = σ(k) ∧ k
+
←֓ {k2} ∧ k
−
←֓ ∅)]
Graphic representation:
NP+
k2
λv[ | wr : can v]
Nk
can
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 16 / 42
LDG framework for Sentence Analysis Example description
Sentence description of ‘Max opened a can’
Sentence descriptions consist of the lexical descriptions of the lexemes
occurring in them, taking into account their surface order.
Graphic representation:
S−
r
Γr = B
DP+
10
o0
[o0 | wr : Max o0] ⊑ B
D0
Max
S+
11
σ13(σ12)
DP−
12 VP−
13
VP+
14
λv[ | wr : v opened σ15]
V1
opened
DP−
15
S+
16
[u2 | ] ⊕ σ19(u2) ⊕ σ17
S−
17
Γ17 = Γ16 ⊕ [u2 | ] ⊕ σ19(u2)
DP+
18
u2
D2
a
NP−
19
NP+
20
λv[ | wr : can v]
N3
can
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 17 / 42
LDG framework for Sentence Analysis Example description
Syntactic properties
Each lexeme enters its own lexical description.
S, VP, DP, NP, D, V are syntactic labels.
Dashed lines stand for underspecified dominance relations, straight
lines for immediate dominance.
The general axioms enforce that the described structures are tree
structures.
+ and - represent anchoring relations. Each node must be both
positively and negatively anchored to some lexical element. The
general axioms enforce that + and - marked nodes are paired off
one-to-one and the pairs identified.
Parsing comes down to reasoning about node identifications.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 18 / 42
LDG framework for Sentence Analysis Example description
Semantic properties
σ associates semantic values with nodes.
σα
k indicates a semantic value of type α.
Semantic values are in ‘finegrained compositional DRT’.
Γ associates Karttunen-style local contexts with nodes.
B, the local context of the root node, represents the implicit, general
background to the discourse.
Indefinite descriptions introduce fresh discourse referents.
Names are presuppositional elements. They introduce a condition on
B which has the effect of resolving or accommodating them in the
main DRS of the resulting discourse meaning.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 19 / 42
LDG framework for Sentence Analysis Reasoning about the description
Parsing as deduction I
S−
r
Γr = B
DP+
10
o0
[o0 | wr : Max o0] ⊑ B
D0
Max
S+
11
σ13(σ12)
DP−
12 VP−
13
VP+
14
λv[ | wr : v opened σ15]
V1
opened
DP−
15
S+
16
[u2 | ] ⊕ σ19(u2) ⊕ σ17
S−
17
Γ17 = Γ16 ⊕ [u2 | ] ⊕ σ19(u2)
DP+
18
u2
D2
a
NP−
19
NP+
20
λv[ | wr : can v]
N3
can
What tree structures fit this description?
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 20 / 42
LDG framework for Sentence Analysis Reasoning about the description
Parsing as deduction II
On the assumption that there are currently no other lexemes than the ones
processed, the language user may reason about verifying trees of the
description.
Parsing The only way to pair of + and - marked nodes that is in
accordance with the grammar is
r = 16 ∧ 11 = 17 ∧ 10 = 12 ∧ 13 = 14 ∧ 15 = 18 ∧ 19 = 20
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 21 / 42
LDG framework for Sentence Analysis Reasoning about the description
Parsing as deduction III
S−
r
Γr = B
DP+
10
o0
[o0 | wr : Max o0] ⊑ B
D0
Max
S+
11
σ13(σ12)
DP−
12 VP−
13
VP+
14
λv[ | wr : v opened σ15]
V1
opened
DP−
15
S+
16
[u2 | ] ⊕ σ19(u2) ⊕ σ17
S−
17
Γ17 = Γ16 ⊕ [u2 | ] ⊕ σ19(u2)
DP+
18
u2
D2
a
NP−
19
NP+
20
λv[ | wr : can v]
N3
can
r = 16 ∧ 11 = 17 ∧ 10 = 12 ∧ 13 = 14 ∧ 15 = 18 ∧ 19 = 20
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 22 / 42
LDG framework for Sentence Analysis Reasoning about the description
If the language user adds this inferred information to the discourse
description, we get
S16
[u2 | wr : o0 opened u2, wr : can u2]
B
[o0 | wr : Max o0] ⊑ B
S11
DP10
D0
Max
VP14
V1
opened
DP18
D2
a
NP20
N3
can
The linguistic tree we saw before is the single verifying tree of this
description; there is no ambiguity.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 23 / 42
LDG framework for Sentence Analysis Reasoning about the description
Inferring Discourse Meaning
The meaning of a discourse given a general background B is
identified with B ⊕ σr
what was ‘taken for granted’ updated with ‘what was said’, while at
the same time all constraints collected in the description are satisfied.
The discourse meaning of ‘Max opened a can’ is
B ⊕ [u2 | wr : o0 opened u2, wr : can u2],
where [wr | ] ⊑ B and [o0 | wr : Max o0] ⊑ B hold.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 24 / 42
LDG framework for Sentence Analysis Reasoning about the description
Sentence meaning
Sentences figure as minimal discourses.
Sentence meaning cannot be computed independently of computing
discourse meaning, unless an out of the blue context is assumed.
Given a discourse description and a sentence whose root node is k,
the sentence meaning ‘in context’ corresponds to Γk ⊕ σk, where all
constraints contributed by elements dominated by k must be satisfied.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 25 / 42
Discourse Processing An example Discourse
Discourse processing: an incrementation step
‘Max opened a can. It contained the money.’
S−
r
B
S+
16
[u2 | wr : o0 opened u2, wr : can u2]
[o0 | wr : Max o0] ⊑ B
Max opened a can
DP+
24
σ5
[σk | ] ⊑ Γ24
σ5 ∈ Tms
24
D5
it
S+
25
σ26(σ27)
DP−
26
VP−
27
VP+
28
λv[ | wr : v contained σ29]
V6
contained
DP−
29
DP+
30
σ7
[σ7 | ] ⊑ Γ30
Γ30 |≍[ | wr : money σ7]
D7
the
NP9
money
‘it’ and ‘the money’ are context dependent items.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 26 / 42
Discourse Processing An example Discourse
Anaphora resolution and presupposition
is the topic of another slide-show.
In the example, the hearer will infer σ5 = u2,
and accommodate/project
[ u7′ | wr : money u7′ ] ⊑ B and [o0 | wr : Max o0] ⊑ B
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 27 / 42
Discourse Processing Further reasoning about the discourse description
Further reasoning about node identifications...
(1) S−
r
B
S+
16
[u2 | wr : o0 opened u2, wr : can u2]
[o0 | wr : Max o0] ⊑ B
Max opened a can
S+
25
[ | wr : σ5 contained σ7 ]
[σ5 | ] ⊑ Γ24, [σ7 | ] ⊑ Γ30
Γ30 |≍[ | wr : money σ7]
σ5 ∈ Tmostsalient
24
It contained the money
Heuristics must guide the reasoning process: ‘safe’ conclusions drawn
earlier must not be lost through the incrementation step.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 28 / 42
Discourse Processing Selecting an appropriate discourse relation
.. and implicit discourse relations
(2) S−
r
B
S+
16
[u2 | wr : o0 opened u2, wr : can u2]
[o0 | wr : Max o0] ⊑ B
Max opened a can
S+
25
[ | wr : σ5 contained σ7 ]
[σ5 | ] ⊑ Γ24, [σ7 | ] ⊑ Γ30
Γ30 |≍[ | wr : money σ7]
σ5 ∈ Tmostsalient
24
It contained the money
The grammar enforces coherence by requiring that every discourse
description describes a discourse parse tree.
So the sentence structure must be attached to the discourse context
structure.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 29 / 42
Discourse Processing Selecting an appropriate discourse relation
There must be an implicit anchor
Additional structure can only be contributed by a lexical anchor (due
to lexicalisation).
Since no overt lexeme is available, the anchor must be implicit.
The only implicit elements in the grammar that can contribute the
link are discourse relations.
It must be one of the discourse relations in the lexicon.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 30 / 42
Discourse Processing Selecting an appropriate discourse relation
Lexical description for discourse relation Elaboration
S+
k1
σk2 ⊕ σk4 ⊕ [ | wr :εk4 ⊆ εk2 ]
topicsubord(k2, k4)
S−
k2
Γk2 = Γk1
Sk3
Rel⋄
k
elaboration
S−
k4
Γk4 = Γk2 ⊕ σk2
Elaboration expresses that there is a part-whole relation linking the main
eventualities introduced by the two arguments.
There is an additional constraint topicsubord(k2, k4) on the information
structural tier.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 31 / 42
Discourse Processing Selecting an appropriate discourse relation
Discourse topical constraints
This presupposes a discourse topical hierarchy can be constructed on
the basis of the information structure of input clauses and
characteristic constraints contributed by individual discourse relations.
Cf. Asher & Lascarides (2003), B¨uring, Roberts, van Kuppevelt.
topicsubord(k, k′) means the discourse unit headed by node k ‘topic
subordinates’ the discourse unit headed by node k′.
c.commontopic(k, k′) conveys that the discourse units share a
‘contingent common topic’.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 32 / 42
Discourse Processing Selecting an appropriate discourse relation
Lexical description for discourse relation Narration
S+
k1
σk2 ⊕ σk3 ⊕ [ | wr :εk2 < εk3 ]
c.commontopic(k2, k3)
S−
k2
Γk2 = Γk1
Rel⋄
k
narration
S−
k3
Γk3 = Γk2 ⊕ σk2
Narration expresses a temporal sequence: the main eventuality of the
contextual argument temporally precedes the main eventuality of the right
argument.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 33 / 42
Discourse Processing Selecting an appropriate discourse relation
Lexical description for discourse relation Background
S+
k1
σk2 ⊕ σk4 ⊕ [ | wr :εk2 εk4 ]
topicsubord(k1, k4)
S−
k2
Γk2 = Γk1
Sk3
Rel⋄
k
background
S−
k4
Γk4 = Γk2 ⊕ σk2
Background expresses that the main eventualities introduced by the two
arguments are overlapping.
There is a subordinating topic relation (very unclear! Cf. A&L 2003).
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 34 / 42
Discourse Processing Selecting an appropriate discourse relation
There must be a discourse relation (underspecified as yet)
Sr
σ26(σ16)(σ25)
B
S16
[u2 | wr : o0 opened u2, wr : can u2]
B
[o0 | wr : Max o0] ⊑ B
Max opened a can
S
Rel26
S25
[ | wr : σ5 contained σ7 ]
Γ25
[σ5 | ], [σ7 | ] ⊑ Γ25
Γ25 |≍[ | wr : money σ7]
σ5 ∈ Tmostsalient
24
It contained the money
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 35 / 42
Discourse Processing Selecting an appropriate discourse relation
The hearer selects a discourse relation from his lexicon, on the basis of his
linguistic knowledge of this discourse, world knowledge, and common sense
expectations, e.g. Background:
Sr
[u2 | wr : o0 opened u2, wr : can u2] ⊕ [ | wr : σ5 contained σ7] ⊕ [ | wr :εk2
εk4
]
B
S16
[u2 | wr : o0 opened u2, wr : can u2]
B
[o0 | wr : Max o0] ⊑ B
Max opened a can
S
Rel4
background
topicsubord(k1, k4)
S25
[ | wr : σ5 contained σ7 ]
B ⊕ [u2 | wr : o0 opened u2, wr : can u2]
[σ5 | ], [σ7 | ] ⊑ B ⊕ [u2 | wr : o0 opened u2, wr : can u2]
B ⊕ [u2 | wr : o0 opened u2, wr : can u2]|≍[ | wr : money σ7]
σ5 ∈ Tmostsalient
24
It contained the money
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 36 / 42
Discourse Processing Selecting an appropriate discourse relation
Multidimensionality/interacting grammatical levels
The selection of a suitable discourse relation and the resolution of
context dependent elements in the new sentence is interdependent.
This is modelled direcly in that the choice of a suitable relation is the
outcome of constraint satisfaction in all grammatical levels (syntax,
semantics, pragmatics) in parallel.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 37 / 42
Discourse Processing Selecting an appropriate discourse relation
Preferences
A full-fledged discourse grammar must contain means to compute
preferences over interpretations, some form of soft constraints, to
obtain full disambiguation at any stage in an ongoing discourse. cf.
Asher and Lasc. 2003, vLeusen 2007
Given the choice of a specific discourse relation the example
description describes a single verifying tree structure.
Underspecification/ambiguity of interpretation is the case when a
hearer derives two or more equally preferred discourse relations.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 38 / 42
Discourse Processing Computing a Discourse Meaning
Computing a Discourse Meaning
The discourse meaning that can be inferred from the description is an
underspecified DRS:
B ⊕ [ u2 | wr : o0 opened u2, wr : can u2, wr : σ5 contained σ7] ⊕ [ | wr :εk2 εk4 ],
where, among others, the following conditions must be satisfied:
(3) a. [o0 | wr : Max o0] ⊑ B
b. [σ5 | ] ⊑ B ⊕ [u2 | wr : o0 opened u2, wr : can u2]
c. σ5 ∈ Tmostsalient
24
d. B ⊕ [u2 | wr : o0 opened u2, wr : can u2]|≍[ | wr : money σ7]
e. [σ7 | ] ⊑ B ⊕ [u2 | wr : o0 opened u2, wr : can u2]
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 39 / 42
Wrapping up Wrapping up
Some issues w.r.t. current LDG formalism
Processing units are sentences or clauses. What changes if we
implement incrementation per lexeme?
Decidability issue.
Lexical descriptions can contain feature structures, but how do
features interact with semantic composition and local contexts?
Do discourse tree structures represent update history, d-topic
structure, speech acts, or RST-like coherence relations?
How to put a preference system in place?
Can goals, beliefs, intentions, commitments, attitudes of the
discussion participants, dynamics of the utterance situation be
modelled?
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 40 / 42
Wrapping up Wrapping up
Pointer to further topics
Another slide-show explains in more detail
The integration of the context parameter (local contexts) in the
compositional semantics of the grammar.
The specifics of anaphora resolution and presupposition ’by constraint
satisfaction’.
and of accommodation and projection by abduction of underspecified
content or global contextual information.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 41 / 42
Wrapping up References
References
Asher & Lascarides 2003. Logics of Conversation.
Buering 2003. On D-Trees, Beans, and B-Accents.
Gardent & Webber 1998. Describing Discourse Semantics.
Gazdar 1979. Pragmatics. Implicature, presupposition and logical form.
Heim 1982. The Semantics of Definite and Indefinite Noun Phrases.
Karttunen 1974. Presupposition and Linguistic Context.
van Kuppevelt 1995 Discourse Structure, Topicality and Questioning
van Leusen 2007. Description Grammar for Discourse.
van Leusen & Muskens 2003.Construction by Description in Discourse Representation.
Lewis 1979. Score-keeping in a language game.
Muskens 1996. Combining Montague Semantics and Discourse Representation.
Muskens 2001. Talking about Trees and Truth-conditions.
van der Sandt 1992. Presupposition Projection as Anaphora Resolution.
Vijay-Shanker 1992. Using descriptions of trees in a tree-adjoining grammar.
Webber & Joshi 1998. Anchoring a lexicalised tree-adjoining grammar for discourse.
Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 42 / 42

Contenu connexe

Tendances

Construction Grammar
Construction GrammarConstruction Grammar
Construction Grammar
maricell095
 
Simple effective decipherment via combinatorial optimization
Simple effective decipherment via combinatorial optimizationSimple effective decipherment via combinatorial optimization
Simple effective decipherment via combinatorial optimization
Attaporn Ninsuwan
 
A Constructive Mathematics approach for NL formal grammars
A Constructive Mathematics approach for NL formal grammarsA Constructive Mathematics approach for NL formal grammars
A Constructive Mathematics approach for NL formal grammars
Federico Gobbo
 
A supervised word sense disambiguation method using ontology and context know...
A supervised word sense disambiguation method using ontology and context know...A supervised word sense disambiguation method using ontology and context know...
A supervised word sense disambiguation method using ontology and context know...
Alexander Decker
 
Utilising wordsmith and atlas to explore, analyse and report qualitative data
Utilising wordsmith and atlas to explore, analyse and report qualitative dataUtilising wordsmith and atlas to explore, analyse and report qualitative data
Utilising wordsmith and atlas to explore, analyse and report qualitative data
Merlien Institute
 

Tendances (20)

Distributional semantics
Distributional semanticsDistributional semantics
Distributional semantics
 
Construction Grammar
Construction GrammarConstruction Grammar
Construction Grammar
 
Simple effective decipherment via combinatorial optimization
Simple effective decipherment via combinatorial optimizationSimple effective decipherment via combinatorial optimization
Simple effective decipherment via combinatorial optimization
 
A Constructive Mathematics approach for NL formal grammars
A Constructive Mathematics approach for NL formal grammarsA Constructive Mathematics approach for NL formal grammars
A Constructive Mathematics approach for NL formal grammars
 
Exempler approach
Exempler approachExempler approach
Exempler approach
 
RuleML2015 The Herbrand Manifesto - Thinking Inside the Box
RuleML2015 The Herbrand Manifesto - Thinking Inside the Box RuleML2015 The Herbrand Manifesto - Thinking Inside the Box
RuleML2015 The Herbrand Manifesto - Thinking Inside the Box
 
Improving Robustness and Flexibility of Concept Taxonomy Learning from Text
Improving Robustness and Flexibility of Concept Taxonomy Learning from Text Improving Robustness and Flexibility of Concept Taxonomy Learning from Text
Improving Robustness and Flexibility of Concept Taxonomy Learning from Text
 
Learning to summarize using coherence
Learning to summarize using coherenceLearning to summarize using coherence
Learning to summarize using coherence
 
A supervised word sense disambiguation method using ontology and context know...
A supervised word sense disambiguation method using ontology and context know...A supervised word sense disambiguation method using ontology and context know...
A supervised word sense disambiguation method using ontology and context know...
 
Word Embedding to Document distances
Word Embedding to Document distancesWord Embedding to Document distances
Word Embedding to Document distances
 
semeval2016
semeval2016semeval2016
semeval2016
 
Nakov S., Nakov P., Paskaleva E., Improved Word Alignments Using the Web as a...
Nakov S., Nakov P., Paskaleva E., Improved Word Alignments Using the Web as a...Nakov S., Nakov P., Paskaleva E., Improved Word Alignments Using the Web as a...
Nakov S., Nakov P., Paskaleva E., Improved Word Alignments Using the Web as a...
 
Syntactic aggregation
Syntactic aggregationSyntactic aggregation
Syntactic aggregation
 
A COMPUTATIONAL APPROACH FOR ANALYZING INTER-SENTENTIAL ANAPHORIC PRONOUNS IN...
A COMPUTATIONAL APPROACH FOR ANALYZING INTER-SENTENTIAL ANAPHORIC PRONOUNS IN...A COMPUTATIONAL APPROACH FOR ANALYZING INTER-SENTENTIAL ANAPHORIC PRONOUNS IN...
A COMPUTATIONAL APPROACH FOR ANALYZING INTER-SENTENTIAL ANAPHORIC PRONOUNS IN...
 
Automatic Identification of False Friends in Parallel Corpora: Statistical an...
Automatic Identification of False Friends in Parallel Corpora: Statistical an...Automatic Identification of False Friends in Parallel Corpora: Statistical an...
Automatic Identification of False Friends in Parallel Corpora: Statistical an...
 
Supporting language learners with the
Supporting language learners with theSupporting language learners with the
Supporting language learners with the
 
Utilising wordsmith and atlas to explore, analyse and report qualitative data
Utilising wordsmith and atlas to explore, analyse and report qualitative dataUtilising wordsmith and atlas to explore, analyse and report qualitative data
Utilising wordsmith and atlas to explore, analyse and report qualitative data
 
Topic model an introduction
Topic model an introductionTopic model an introduction
Topic model an introduction
 
Role of unification and realization in natural language generation
Role of unification and realization in natural language generationRole of unification and realization in natural language generation
Role of unification and realization in natural language generation
 
FUZZY LOGIC IN NARROW SENSE WITH HEDGES
FUZZY LOGIC IN NARROW SENSE WITH HEDGESFUZZY LOGIC IN NARROW SENSE WITH HEDGES
FUZZY LOGIC IN NARROW SENSE WITH HEDGES
 

Similaire à LDG-basic-slides

Improvement wsd dictionary using annotated corpus and testing it with simplif...
Improvement wsd dictionary using annotated corpus and testing it with simplif...Improvement wsd dictionary using annotated corpus and testing it with simplif...
Improvement wsd dictionary using annotated corpus and testing it with simplif...
csandit
 
Csr2011 june17 15_15_kaminski
Csr2011 june17 15_15_kaminskiCsr2011 june17 15_15_kaminski
Csr2011 june17 15_15_kaminski
CSR2011
 
Ldml - public
Ldml - publicLdml - public
Ldml - public
seanscon
 
Tensor-based Models of Natural Language Semantics
Tensor-based Models of Natural Language SemanticsTensor-based Models of Natural Language Semantics
Tensor-based Models of Natural Language Semantics
Dimitrios Kartsaklis
 
Discourse analysis by gillian brown & george yule
Discourse analysis   by gillian brown & george yuleDiscourse analysis   by gillian brown & george yule
Discourse analysis by gillian brown & george yule
John Ykaz
 
Merging controlled vocabularies through semantic alignment based on linked data
Merging controlled vocabularies through semantic alignment based on linked dataMerging controlled vocabularies through semantic alignment based on linked data
Merging controlled vocabularies through semantic alignment based on linked data
John Pap
 

Similaire à LDG-basic-slides (20)

A New Approach For Paraphrasing And Rewording A Challenging Text
A New Approach For Paraphrasing And Rewording A Challenging TextA New Approach For Paraphrasing And Rewording A Challenging Text
A New Approach For Paraphrasing And Rewording A Challenging Text
 
Designing, Visualizing and Understanding Deep Neural Networks
Designing, Visualizing and Understanding Deep Neural NetworksDesigning, Visualizing and Understanding Deep Neural Networks
Designing, Visualizing and Understanding Deep Neural Networks
 
Latent dirichletallocation presentation
Latent dirichletallocation presentationLatent dirichletallocation presentation
Latent dirichletallocation presentation
 
Effective Approach for Disambiguating Chinese Polyphonic Ambiguity
Effective Approach for Disambiguating Chinese Polyphonic AmbiguityEffective Approach for Disambiguating Chinese Polyphonic Ambiguity
Effective Approach for Disambiguating Chinese Polyphonic Ambiguity
 
Discourse annotation
Discourse annotationDiscourse annotation
Discourse annotation
 
Word sense dissambiguation
Word sense dissambiguationWord sense dissambiguation
Word sense dissambiguation
 
Improvement wsd dictionary using annotated corpus and testing it with simplif...
Improvement wsd dictionary using annotated corpus and testing it with simplif...Improvement wsd dictionary using annotated corpus and testing it with simplif...
Improvement wsd dictionary using annotated corpus and testing it with simplif...
 
Csr2011 june17 15_15_kaminski
Csr2011 june17 15_15_kaminskiCsr2011 june17 15_15_kaminski
Csr2011 june17 15_15_kaminski
 
Ldml - public
Ldml - publicLdml - public
Ldml - public
 
Knowledge of meaning an introduction-to_semantic_theory-buku
Knowledge of meaning   an introduction-to_semantic_theory-bukuKnowledge of meaning   an introduction-to_semantic_theory-buku
Knowledge of meaning an introduction-to_semantic_theory-buku
 
Argumentation And Tensiveness. A Semiotic Interpretation Of Ducrot S Argument...
Argumentation And Tensiveness. A Semiotic Interpretation Of Ducrot S Argument...Argumentation And Tensiveness. A Semiotic Interpretation Of Ducrot S Argument...
Argumentation And Tensiveness. A Semiotic Interpretation Of Ducrot S Argument...
 
L3 v2
L3 v2L3 v2
L3 v2
 
Tensor-based Models of Natural Language Semantics
Tensor-based Models of Natural Language SemanticsTensor-based Models of Natural Language Semantics
Tensor-based Models of Natural Language Semantics
 
A CORPUS-DRIVEN DESIGN OF A TEST FOR ASSESSING THE ESL COLLOCATIONAL COMPETEN...
A CORPUS-DRIVEN DESIGN OF A TEST FOR ASSESSING THE ESL COLLOCATIONAL COMPETEN...A CORPUS-DRIVEN DESIGN OF A TEST FOR ASSESSING THE ESL COLLOCATIONAL COMPETEN...
A CORPUS-DRIVEN DESIGN OF A TEST FOR ASSESSING THE ESL COLLOCATIONAL COMPETEN...
 
Discourse analysis-by-gillian-brown-george-yule
Discourse analysis-by-gillian-brown-george-yuleDiscourse analysis-by-gillian-brown-george-yule
Discourse analysis-by-gillian-brown-george-yule
 
Discourse analysis by gillian brown & george yule
Discourse analysis   by gillian brown & george yuleDiscourse analysis   by gillian brown & george yule
Discourse analysis by gillian brown & george yule
 
Nlp ambiguity presentation
Nlp ambiguity presentationNlp ambiguity presentation
Nlp ambiguity presentation
 
AINL 2016: Eyecioglu
AINL 2016: EyeciogluAINL 2016: Eyecioglu
AINL 2016: Eyecioglu
 
Merging controlled vocabularies through semantic alignment based on linked data
Merging controlled vocabularies through semantic alignment based on linked dataMerging controlled vocabularies through semantic alignment based on linked data
Merging controlled vocabularies through semantic alignment based on linked data
 
Multilingual Text Classification using Ontologies
Multilingual Text Classification using OntologiesMultilingual Text Classification using Ontologies
Multilingual Text Classification using Ontologies
 

LDG-basic-slides

  • 1. Logical Description Grammar for sentence and discourse interpretation Introduction to the grammar system Noor van Leusen Radboud University 2012 Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 1 / 42
  • 2. Content Content Logical Description Grammar as a method for the description of sentence syntax and semantics. Muskens 1996, 2001. /cf. ‘Saarbruecken school’. Underspecified representations. Syntactic and semantic properties of linguistic trees. Parsing by deduction. CDRT and local contexts in the semantic dimension. vLeusen&Muskens 2003 Discourse processing and the computation of coherence in LDG. Anaphora and Presupposition. vLeusen 2007 /cf. Gardent&Webber 1998, Egg&Redeker 2006. Implicit discourse relations. Inferring discourse meaning. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 2 / 42
  • 3. LDG framework for Sentence Analysis Introducing Logical Description Grammar Description Grammar for sentence analysis LDG combines description theory (Vijay-Shanker c.s.) lexicalised TAG (Joshi, Webber c.s.) and compositional DRT (Kamp, Muskens). With a view to ambiguity and implicitness of meaning it generates ‘underspecified representations’. It describes syntactic, semantic and pragmatic constraints in parallel, and is specifically tuned to model the interaction of these constraints. The semantics integrates a context parameter, employed in the treatment of anaphora, presupposition and accommodation. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 3 / 42
  • 4. LDG framework for Sentence Analysis Introducing Logical Description Grammar Discourse contexts and discourse descriptions Discourse contexts are represented by discourse descriptions. Discourse descriptions are built up incrementally from sentence descriptions. r k′ + relation k Sentence representations are linked to their discourse context by explicit connectives or implicit discourse relations. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 4 / 42
  • 5. LDG framework for Sentence Analysis Introducing Logical Description Grammar Underspecified representations There are two levels of linguistic analysis: sentence description (underspecified representation) ? set of verifying trees (fully specified linguistic repr.) Sentence descriptions are statements in classical type logic. They partially describe properties of linguistic trees, constraining syntactic, semantic and pragmatic features in parallel. In case of ambiguity, more than one tree class verifies a description. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 5 / 42
  • 6. LDG framework for Sentence Analysis Linguistic Representations Linguistic Tree Representations Sentence and discourse representations are tree structures, decorated with lexemes, syntactic labels, semantic values, and local contexts. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 6 / 42
  • 7. LDG framework for Sentence Analysis Linguistic Representations Example tree, showing just syntax and lexemes: S S DP D Max VP V opened DP D a NP N can In situ analysis of quantifiers (’a can’): lexeme in surface position, quantification potential is scoped out (at top S). Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 7 / 42
  • 8. LDG framework for Sentence Analysis Linguistic Representations Semantic values added: S [u2 | wr : can u2 ] ⊕ [ | wr : o0 opened u2] S [ |wr : o0 opened u2] DP o0 D o0 Max VP λv [ |wr : v opened u2] V λv′ λv [ |wr : v opened v′ ] opened DP u2 D u2 a NP λv [ |wr : can v] N λv [ |wr : can v] can The semantic representation language is finegrained compositional DRT, a Lambda-DRT; [...|.....] is a DRS, ⊕ merges DRSs. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 8 / 42
  • 9. LDG framework for Sentence Analysis Linguistic Representations Fine-grained compositional DRT [..|...] is a DRS, to the left of the | sign is the universe, to the right the conditions. uk, wk, ok , . . . are discourse markers; uk , ok store individuals. uk represent new referents (generated in the discourse); ok represent given referents; they are globally available and interpreted referentially rather than existentially; wk store worlds. wr is the marker for the actual world. ⊕ merges DRSs, ⊑ denotes inclusion, |≍ entailment. ‘proper(K)’ defines K as a DRS in which no free discourse markers occur. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 9 / 42
  • 10. LDG framework for Sentence Analysis Linguistic Representations Context Parameter Each node in a tree structure carries a DRS which is its local context. (cf. Karttunen 1974) The compositional semantics refers to this context parameter in the resolution of anaphora, presupposition satisfaction and projection, and accommodation in general. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 10 / 42
  • 11. LDG framework for Sentence Analysis Linguistic Representations Local contexts and a presuppositional constraint added: S [u2 | wr : can u2 ] ⊕ [ | wr : o0 opened u2] B S [ |wr : o0 opened u2] B ⊕ [u2 | wr : can u2] DP o0 B ⊕ [u2 | wr : can u2] [o1 | wr : Max o0] ⊑ B D o0 B ⊕ [u2 | wr : can u2] Max VP λv [ |wr : v opened u2] B ⊕ [u2 | wr : can u2] V λv′λv [ |wr : v opened v′] B ⊕ [u2 | wr : can u2] opened DP u2 B ⊕ [u2 | wr : can u2] D u2 B ⊕ [u2 | wr : can u2] a NP λv [ |wr : can v] B ⊕ [u2 | wr : can u2] N λv [ |wr : can v] B ⊕ [u2 | wr : can u2] can Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 11 / 42
  • 12. LDG framework for Sentence Analysis Linguistic Representations Local contexts and context-sensitivity Local contexts ‘collect’ information on the semantic tier of the discourse context. They are constructed going top-down and from left to right through a discourse or sentence tree. The local context of the root node is B, the implicit background. Anaphora and projective elements such as presupposition triggers introduce conditions on local contexts. Accommodation/projection results from satisfying these conditions in partially underspecified local backgrounds. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 12 / 42
  • 13. LDG framework for Sentence Analysis Description Grammar as a reasoning system Description Grammar, a constraint-based reasoning system A description grammar G consists of lexical descriptions and some general axioms. It generates a sentence description ∆u per processed utterance u. The grammar is itself a description, a set of statements in type logic. Given the theory G + ∆u, a language user may reason about and obtain the possible verifying trees of the sentence description M(∆u). Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 13 / 42
  • 14. LDG framework for Sentence Analysis Description Grammar as a reasoning system While G represents a language user’s general linguistic knowledge, a sentence description ∆u represents his specific linguistic knowledge of the sentence. Linguistic knowledge means knowledge in any of the relevant domains, such as phonology, syntax, semantic and pragmatics. Sentence interpretation is a reasoning process. Since descriptions can be partial, G + ∆u can have more than a single verifying tree, and the syntax or semantics of the sentence can remain underspecified. Each verifying tree comes with a possible reading of the sentence. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 14 / 42
  • 15. LDG framework for Sentence Analysis Lexicon and General Axioms G consists of lexical descriptions and general axioms General axioms constrain the general configurational and semantic properties of linguistic representations. ∀k ¬[k ≺ k] No node precedes itself ∀k [Γk |≍⊥ ∧ Γr ⊕ σr |≍⊥] Local contexts are consistent Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 15 / 42
  • 16. LDG framework for Sentence Analysis Lexicon and General Axioms Lexical descriptions characterise the syntactic, semantic, and pragmatic properties of lexemes. They describe ‘little trees’ similar to elementary structures in TAG. Lexical description of the noun can: ∀k [can(k) → (cn(k) ∧ σ(k) = λv [ |wr : can v]) ] ∀k [cn(k) → ∃k2(ℓ(k) = n ∧ ℓ(k2) = np ∧ k2 ¡ k∧ σ(k2) = σ(k) ∧ k + ←֓ {k2} ∧ k − ←֓ ∅)] Graphic representation: NP+ k2 λv[ | wr : can v] Nk can Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 16 / 42
  • 17. LDG framework for Sentence Analysis Example description Sentence description of ‘Max opened a can’ Sentence descriptions consist of the lexical descriptions of the lexemes occurring in them, taking into account their surface order. Graphic representation: S− r Γr = B DP+ 10 o0 [o0 | wr : Max o0] ⊑ B D0 Max S+ 11 σ13(σ12) DP− 12 VP− 13 VP+ 14 λv[ | wr : v opened σ15] V1 opened DP− 15 S+ 16 [u2 | ] ⊕ σ19(u2) ⊕ σ17 S− 17 Γ17 = Γ16 ⊕ [u2 | ] ⊕ σ19(u2) DP+ 18 u2 D2 a NP− 19 NP+ 20 λv[ | wr : can v] N3 can Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 17 / 42
  • 18. LDG framework for Sentence Analysis Example description Syntactic properties Each lexeme enters its own lexical description. S, VP, DP, NP, D, V are syntactic labels. Dashed lines stand for underspecified dominance relations, straight lines for immediate dominance. The general axioms enforce that the described structures are tree structures. + and - represent anchoring relations. Each node must be both positively and negatively anchored to some lexical element. The general axioms enforce that + and - marked nodes are paired off one-to-one and the pairs identified. Parsing comes down to reasoning about node identifications. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 18 / 42
  • 19. LDG framework for Sentence Analysis Example description Semantic properties σ associates semantic values with nodes. σα k indicates a semantic value of type α. Semantic values are in ‘finegrained compositional DRT’. Γ associates Karttunen-style local contexts with nodes. B, the local context of the root node, represents the implicit, general background to the discourse. Indefinite descriptions introduce fresh discourse referents. Names are presuppositional elements. They introduce a condition on B which has the effect of resolving or accommodating them in the main DRS of the resulting discourse meaning. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 19 / 42
  • 20. LDG framework for Sentence Analysis Reasoning about the description Parsing as deduction I S− r Γr = B DP+ 10 o0 [o0 | wr : Max o0] ⊑ B D0 Max S+ 11 σ13(σ12) DP− 12 VP− 13 VP+ 14 λv[ | wr : v opened σ15] V1 opened DP− 15 S+ 16 [u2 | ] ⊕ σ19(u2) ⊕ σ17 S− 17 Γ17 = Γ16 ⊕ [u2 | ] ⊕ σ19(u2) DP+ 18 u2 D2 a NP− 19 NP+ 20 λv[ | wr : can v] N3 can What tree structures fit this description? Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 20 / 42
  • 21. LDG framework for Sentence Analysis Reasoning about the description Parsing as deduction II On the assumption that there are currently no other lexemes than the ones processed, the language user may reason about verifying trees of the description. Parsing The only way to pair of + and - marked nodes that is in accordance with the grammar is r = 16 ∧ 11 = 17 ∧ 10 = 12 ∧ 13 = 14 ∧ 15 = 18 ∧ 19 = 20 Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 21 / 42
  • 22. LDG framework for Sentence Analysis Reasoning about the description Parsing as deduction III S− r Γr = B DP+ 10 o0 [o0 | wr : Max o0] ⊑ B D0 Max S+ 11 σ13(σ12) DP− 12 VP− 13 VP+ 14 λv[ | wr : v opened σ15] V1 opened DP− 15 S+ 16 [u2 | ] ⊕ σ19(u2) ⊕ σ17 S− 17 Γ17 = Γ16 ⊕ [u2 | ] ⊕ σ19(u2) DP+ 18 u2 D2 a NP− 19 NP+ 20 λv[ | wr : can v] N3 can r = 16 ∧ 11 = 17 ∧ 10 = 12 ∧ 13 = 14 ∧ 15 = 18 ∧ 19 = 20 Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 22 / 42
  • 23. LDG framework for Sentence Analysis Reasoning about the description If the language user adds this inferred information to the discourse description, we get S16 [u2 | wr : o0 opened u2, wr : can u2] B [o0 | wr : Max o0] ⊑ B S11 DP10 D0 Max VP14 V1 opened DP18 D2 a NP20 N3 can The linguistic tree we saw before is the single verifying tree of this description; there is no ambiguity. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 23 / 42
  • 24. LDG framework for Sentence Analysis Reasoning about the description Inferring Discourse Meaning The meaning of a discourse given a general background B is identified with B ⊕ σr what was ‘taken for granted’ updated with ‘what was said’, while at the same time all constraints collected in the description are satisfied. The discourse meaning of ‘Max opened a can’ is B ⊕ [u2 | wr : o0 opened u2, wr : can u2], where [wr | ] ⊑ B and [o0 | wr : Max o0] ⊑ B hold. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 24 / 42
  • 25. LDG framework for Sentence Analysis Reasoning about the description Sentence meaning Sentences figure as minimal discourses. Sentence meaning cannot be computed independently of computing discourse meaning, unless an out of the blue context is assumed. Given a discourse description and a sentence whose root node is k, the sentence meaning ‘in context’ corresponds to Γk ⊕ σk, where all constraints contributed by elements dominated by k must be satisfied. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 25 / 42
  • 26. Discourse Processing An example Discourse Discourse processing: an incrementation step ‘Max opened a can. It contained the money.’ S− r B S+ 16 [u2 | wr : o0 opened u2, wr : can u2] [o0 | wr : Max o0] ⊑ B Max opened a can DP+ 24 σ5 [σk | ] ⊑ Γ24 σ5 ∈ Tms 24 D5 it S+ 25 σ26(σ27) DP− 26 VP− 27 VP+ 28 λv[ | wr : v contained σ29] V6 contained DP− 29 DP+ 30 σ7 [σ7 | ] ⊑ Γ30 Γ30 |≍[ | wr : money σ7] D7 the NP9 money ‘it’ and ‘the money’ are context dependent items. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 26 / 42
  • 27. Discourse Processing An example Discourse Anaphora resolution and presupposition is the topic of another slide-show. In the example, the hearer will infer σ5 = u2, and accommodate/project [ u7′ | wr : money u7′ ] ⊑ B and [o0 | wr : Max o0] ⊑ B Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 27 / 42
  • 28. Discourse Processing Further reasoning about the discourse description Further reasoning about node identifications... (1) S− r B S+ 16 [u2 | wr : o0 opened u2, wr : can u2] [o0 | wr : Max o0] ⊑ B Max opened a can S+ 25 [ | wr : σ5 contained σ7 ] [σ5 | ] ⊑ Γ24, [σ7 | ] ⊑ Γ30 Γ30 |≍[ | wr : money σ7] σ5 ∈ Tmostsalient 24 It contained the money Heuristics must guide the reasoning process: ‘safe’ conclusions drawn earlier must not be lost through the incrementation step. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 28 / 42
  • 29. Discourse Processing Selecting an appropriate discourse relation .. and implicit discourse relations (2) S− r B S+ 16 [u2 | wr : o0 opened u2, wr : can u2] [o0 | wr : Max o0] ⊑ B Max opened a can S+ 25 [ | wr : σ5 contained σ7 ] [σ5 | ] ⊑ Γ24, [σ7 | ] ⊑ Γ30 Γ30 |≍[ | wr : money σ7] σ5 ∈ Tmostsalient 24 It contained the money The grammar enforces coherence by requiring that every discourse description describes a discourse parse tree. So the sentence structure must be attached to the discourse context structure. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 29 / 42
  • 30. Discourse Processing Selecting an appropriate discourse relation There must be an implicit anchor Additional structure can only be contributed by a lexical anchor (due to lexicalisation). Since no overt lexeme is available, the anchor must be implicit. The only implicit elements in the grammar that can contribute the link are discourse relations. It must be one of the discourse relations in the lexicon. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 30 / 42
  • 31. Discourse Processing Selecting an appropriate discourse relation Lexical description for discourse relation Elaboration S+ k1 σk2 ⊕ σk4 ⊕ [ | wr :εk4 ⊆ εk2 ] topicsubord(k2, k4) S− k2 Γk2 = Γk1 Sk3 Rel⋄ k elaboration S− k4 Γk4 = Γk2 ⊕ σk2 Elaboration expresses that there is a part-whole relation linking the main eventualities introduced by the two arguments. There is an additional constraint topicsubord(k2, k4) on the information structural tier. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 31 / 42
  • 32. Discourse Processing Selecting an appropriate discourse relation Discourse topical constraints This presupposes a discourse topical hierarchy can be constructed on the basis of the information structure of input clauses and characteristic constraints contributed by individual discourse relations. Cf. Asher & Lascarides (2003), B¨uring, Roberts, van Kuppevelt. topicsubord(k, k′) means the discourse unit headed by node k ‘topic subordinates’ the discourse unit headed by node k′. c.commontopic(k, k′) conveys that the discourse units share a ‘contingent common topic’. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 32 / 42
  • 33. Discourse Processing Selecting an appropriate discourse relation Lexical description for discourse relation Narration S+ k1 σk2 ⊕ σk3 ⊕ [ | wr :εk2 < εk3 ] c.commontopic(k2, k3) S− k2 Γk2 = Γk1 Rel⋄ k narration S− k3 Γk3 = Γk2 ⊕ σk2 Narration expresses a temporal sequence: the main eventuality of the contextual argument temporally precedes the main eventuality of the right argument. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 33 / 42
  • 34. Discourse Processing Selecting an appropriate discourse relation Lexical description for discourse relation Background S+ k1 σk2 ⊕ σk4 ⊕ [ | wr :εk2 εk4 ] topicsubord(k1, k4) S− k2 Γk2 = Γk1 Sk3 Rel⋄ k background S− k4 Γk4 = Γk2 ⊕ σk2 Background expresses that the main eventualities introduced by the two arguments are overlapping. There is a subordinating topic relation (very unclear! Cf. A&L 2003). Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 34 / 42
  • 35. Discourse Processing Selecting an appropriate discourse relation There must be a discourse relation (underspecified as yet) Sr σ26(σ16)(σ25) B S16 [u2 | wr : o0 opened u2, wr : can u2] B [o0 | wr : Max o0] ⊑ B Max opened a can S Rel26 S25 [ | wr : σ5 contained σ7 ] Γ25 [σ5 | ], [σ7 | ] ⊑ Γ25 Γ25 |≍[ | wr : money σ7] σ5 ∈ Tmostsalient 24 It contained the money Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 35 / 42
  • 36. Discourse Processing Selecting an appropriate discourse relation The hearer selects a discourse relation from his lexicon, on the basis of his linguistic knowledge of this discourse, world knowledge, and common sense expectations, e.g. Background: Sr [u2 | wr : o0 opened u2, wr : can u2] ⊕ [ | wr : σ5 contained σ7] ⊕ [ | wr :εk2 εk4 ] B S16 [u2 | wr : o0 opened u2, wr : can u2] B [o0 | wr : Max o0] ⊑ B Max opened a can S Rel4 background topicsubord(k1, k4) S25 [ | wr : σ5 contained σ7 ] B ⊕ [u2 | wr : o0 opened u2, wr : can u2] [σ5 | ], [σ7 | ] ⊑ B ⊕ [u2 | wr : o0 opened u2, wr : can u2] B ⊕ [u2 | wr : o0 opened u2, wr : can u2]|≍[ | wr : money σ7] σ5 ∈ Tmostsalient 24 It contained the money Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 36 / 42
  • 37. Discourse Processing Selecting an appropriate discourse relation Multidimensionality/interacting grammatical levels The selection of a suitable discourse relation and the resolution of context dependent elements in the new sentence is interdependent. This is modelled direcly in that the choice of a suitable relation is the outcome of constraint satisfaction in all grammatical levels (syntax, semantics, pragmatics) in parallel. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 37 / 42
  • 38. Discourse Processing Selecting an appropriate discourse relation Preferences A full-fledged discourse grammar must contain means to compute preferences over interpretations, some form of soft constraints, to obtain full disambiguation at any stage in an ongoing discourse. cf. Asher and Lasc. 2003, vLeusen 2007 Given the choice of a specific discourse relation the example description describes a single verifying tree structure. Underspecification/ambiguity of interpretation is the case when a hearer derives two or more equally preferred discourse relations. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 38 / 42
  • 39. Discourse Processing Computing a Discourse Meaning Computing a Discourse Meaning The discourse meaning that can be inferred from the description is an underspecified DRS: B ⊕ [ u2 | wr : o0 opened u2, wr : can u2, wr : σ5 contained σ7] ⊕ [ | wr :εk2 εk4 ], where, among others, the following conditions must be satisfied: (3) a. [o0 | wr : Max o0] ⊑ B b. [σ5 | ] ⊑ B ⊕ [u2 | wr : o0 opened u2, wr : can u2] c. σ5 ∈ Tmostsalient 24 d. B ⊕ [u2 | wr : o0 opened u2, wr : can u2]|≍[ | wr : money σ7] e. [σ7 | ] ⊑ B ⊕ [u2 | wr : o0 opened u2, wr : can u2] Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 39 / 42
  • 40. Wrapping up Wrapping up Some issues w.r.t. current LDG formalism Processing units are sentences or clauses. What changes if we implement incrementation per lexeme? Decidability issue. Lexical descriptions can contain feature structures, but how do features interact with semantic composition and local contexts? Do discourse tree structures represent update history, d-topic structure, speech acts, or RST-like coherence relations? How to put a preference system in place? Can goals, beliefs, intentions, commitments, attitudes of the discussion participants, dynamics of the utterance situation be modelled? Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 40 / 42
  • 41. Wrapping up Wrapping up Pointer to further topics Another slide-show explains in more detail The integration of the context parameter (local contexts) in the compositional semantics of the grammar. The specifics of anaphora resolution and presupposition ’by constraint satisfaction’. and of accommodation and projection by abduction of underspecified content or global contextual information. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 41 / 42
  • 42. Wrapping up References References Asher & Lascarides 2003. Logics of Conversation. Buering 2003. On D-Trees, Beans, and B-Accents. Gardent & Webber 1998. Describing Discourse Semantics. Gazdar 1979. Pragmatics. Implicature, presupposition and logical form. Heim 1982. The Semantics of Definite and Indefinite Noun Phrases. Karttunen 1974. Presupposition and Linguistic Context. van Kuppevelt 1995 Discourse Structure, Topicality and Questioning van Leusen 2007. Description Grammar for Discourse. van Leusen & Muskens 2003.Construction by Description in Discourse Representation. Lewis 1979. Score-keeping in a language game. Muskens 1996. Combining Montague Semantics and Discourse Representation. Muskens 2001. Talking about Trees and Truth-conditions. van der Sandt 1992. Presupposition Projection as Anaphora Resolution. Vijay-Shanker 1992. Using descriptions of trees in a tree-adjoining grammar. Webber & Joshi 1998. Anchoring a lexicalised tree-adjoining grammar for discourse. Noor van Leusen (Radboud University) Logical Description Grammar for sentence and discourse interpretation 2012 42 / 42