INTRODUCCI

1
INTRODUCCIÓ

• La major part de les aplicacions interessants de
  llenguatge requeririen obtenir la representació
  del significat de les oracions.

2
Estructura predicat argument

• La estructura predicat argument descriu les
  relacions semàntiques que es donen entre les
  entitats que apareixen en la oració.
• -who does what to whom,
• -how, where, why?

3
Estructura predicat argument

• I eat sushi
• PRED eat ARG1 I ARG2 sushi.

4
Un exemple més complex.

• En frases complexes tenim més de una proposició.
• Mary loves the man who bougth the blue car
• P1 Mary loves the man.
• P2 The man bought the car.
• P3 blue car.

5
Perquè serveix la estructura sintàgmatica?

• Per obtenir la estructura predicat argument es
  necessari computar primer la estructura
  sintagmàtica o al menys una estructura de
  dependències.
• La estructura sintagmàtica i les corresponents
  regles son les que permeten que el llenguatge
  sigui composicional.

6
(No Transcript)
7
(No Transcript)
8
Però es cert això?
9
Resum

• Historia (per entendre els objectius de les
  teories)
• Why phrase structures? (problemes)
• Why dependency grammars? (problemes)
• Why a probabilistic approach?(Brute force vs.
  theory)
• Estat actual del nostre model i recerca futura.

10
Una previaCom millorar els resultats?

• Augmentant el training size?
• Mètodes estadístics més eficients?
• O millorant les teories?

11
Historia

• Grammars as computational theories

12
Grammars as computational theories

• Cognition is computation.
• A grammar is a form of computation.

13
Computational theories (Marr 1980)

• What is the goal of the computation?
• Why is it appropriate?
• What is the logic of the strategy by which it can
  be carried out?

14
Chomskys Goal

• A syntactic theory has as a goal to explain the
  capacity of speakers to judge as acceptable (or
  generate) well formed sentences and to rule out
  ill-formed ones.

15
Justification

• Syntax is indepedent of semantics.
• Speakers can judge as ill or well-formed new
  sentences that they have never heard before.

16
Quin es el origin dels sintagmes?

• No es semàntic. Un NP no es un NP perquè es
  correspongui amb un argument semàntic.
• Es un NP en base a trets purament sintàctics.
  Regularitats en la distribució de les paraules en
  les frases.
• Tests que determinen que es un constituïen (un
  sintagma) i que no ho és.

17
Constituency Tests

• Tests of constituency are basic components of
  the syntacticians toolbox. By investigating
  which strings of words can and cannot be moved,
  deleted, coordinated or stand in coreference
  relations, it is possible to draw inferences
  about the internal structure of sentences.
  (Phillips, 1998, p. 1)

18

• Chomsky assumed that, given the independence of
  syntax, a theory of syntax can be developed
  without a semantic theory and ignoring the
  mapping process, following only the
  well-formedness goal.

19
Mapping Goal

• A syntactic theory has as a goal to explain
  the capacity of native speakers to map sentences
  into the corresponding conceptual representations
  and vice versa.
•  

20
Mapping Goal

• The mapping goal tries to figure out how
  linguistic expressions can be mapped in the
  respective propositional representations in in
  the most simple and direct way.

21
Mapping Goal

• (3.a) IBMP gave the company the patent.
• (3.b) PRED gave ARG1 IBMP ARG2 the patent
  ARG3 the company.
• (4.a) Low prices.
• (4.b) PRED low ARG1 prices.

22
Well-Formedness Goal

• (3.a) IBMP gave the company the patent.
• (3.b) IBMP company gave the the patent.
• (4.a) Low prices.
• (4.b) Prices low

23
Direct mapping

• The carpenter gave the nurse the book.

PRED gave ARG1 the carpenter ARG2 the
book ARG3 the nurse.
24
El mapping pot ser directe en expresions simples

• Aixo es cert per oracions simples.
• Culicover, Peter W. and Andrzej Nowak. Dynamical
  Grammar. Volume Two of Foundations of Syntax.
  Oxford University Press.  2003.
• Roger Schank i collaboradors en els anys 70.

25
5941 Mr. NNP - (A0
5941 Nakamur NNP -
) 5941 cites
VBZ cite (V)
5941 the DT - (A1
5941 case NN -
5941 of
IN -
5941 a DT - (A0
(A0 (A0 (A0 lt---------4NLDs5941
custome NN - ) )
) ) 5941 who
WP - (R-A0) (R-A0) (R-A0)
(R-A0)5941 wants VBZ want
(V) 5941 to TO
- (A1 5941
build VB build (V)
5941 a DT -
(A1 5941 giant JJ -
5941
tourism NN -
5941 complex NN -
) 5941 in IN -
(AM-LOC 5941
Baja NN - ) )
5941 and CC -
5941 has VBZ -
5941
been VBN -
5941 trying VBG try
(V) 5941 for IN -
(AM-TMP5941
eight CD -
5941 years NNS -
) 5941 to TO -
(A1 5941
get VB get
(V5941 around IN -
)5941 Mexican NNP
-
(A15941 restric NNS -
5941 on IN -
5941
foreign JJ -
5941 ownersh NN -
5941 of IN -
5941
beachfr JJ -
5941 propert NN - )
) )5941 . .
-
26
Direct mapping

• Per Culicover en frases mes complexes no es
  possible.
• Mary loves the man who bougth the blue car
• P1 Mary loves the man.
• P2 The man bought the car.
• P3 blue car.

27
Direct mapping

• No es possible?
• Mary loves the man who bougth the blue car
• P1 PRED loves ARG1 Mary ARG2 the man.
• P2 PRED bought ARG1the man ARG2 the car..
• P3 PREDblue ARG1 car.

28

• Why phrase structures?
• Why dependency grammars?

29
No son necessaries

• Es pot aconseguir composicionalitat sense
  computar estructura sintagmàtica
• Es pot fer un mapping directe a la estructura
  predicat argument sense computar ni una
  estructura de dependències ni una sintagmàtica.
• Es simplfica considerablement el procés de
  parsing i el tractament de la ambigüitat.

30
Temes per poder entrar a fons

• Why phrase structures?
• Why dependency grammars?
• Why a probabilistic approach?
• (al menys la versió brute-cutre force)

31
D-SemMap

• V1.0

32
Vectors and propositions

• A proposition can be represented by a vector of
  features (Hinton, 1981).
• In order to represent the proposition the vector
  is divided into slots.
• Each element of the proposition is represented in
  one slot.

33
Vectors and propositions
Module 2 Semantic classes
Mary drives a bus
action human artifact entity
SLOT 0 SLOT 1 SLOT 2 SLOT 3
Types Backs
Module 1 POS
Mary drives a bus
V MA N DT N
SLOT 0 SLOT 1 SLOT 2 SLOT 3
Types Backs
34
MODULE 1
Output Layer
Yamada (2003) Nivre (2004) Magerman
(1994) Ratnaparky (1999)
Input Layer
Input Word
Slot 0 Slot 1 Slot 2 Slot 3 Type S Back,
Test Subcat.
35
carpenter
bought
a
shirt
with
Credit-card
The
Output
Hidden
Slot 1
Slot 0
Slot 2
Slot 3
DT
N C
V MA PE PA
DT
N C
IIN
N C
Subcategorization backtracking
36
Module 2 supervises argument position
MODULE 1
MODULE 2
P1) PRED V MA PE PA ( bought) ARG1 N
PR (Mary) ARG2 DT N C ( a shirt ) ARG3
IIN N C (with pockets)
P1) PRED get, transfer, give, pay, ARG1
entity, person ARG2 entity, object, artifact
(shirt) ARG3 artifact, part-of-dress
Subcategorization and Selectional
Restrictions Parsing strategy Attaches first to
the current proposition
37
Binding problem
38
(No Transcript)
39
Mary bought a shirt with pockets
MODULE 1
MODULE 2
P1) PRED V MA PE PA ( bought) ARG1 N
PR (Mary) ARG2 DT N C ( a shirt ) ARG3
IIN N C (with pockets)
P1) PRED get, transfer, give, pay, ARG1
entity, person ARG2 entity, object, artifact
ARG3 artifact part-of-dress
ARG3 artifact part-of-dress
Parsing strategy Attaches first to the current
proposition
40
Mary bought a shirt with pockets
MODULE 1
MODULE 2
P1 PRED V MA PE ( bought) ARG1 N PR
(Mary) ARG2 DT N C ( a shirt ) ARG3
P1 PRED get, transfer,pay, accept ... ARG1
entity, person,... ARG2 entity, object,
artifact, shirt ARG3
P2 PRED ARG1 DT N C (a shirt) ARG2 IIN
N C (with pockets) ARG3
P2 PRED ARG1 entity, object, artifact,
shirt ARG2 artifact part-of-dress ARG3
41
PARSING COMPLEX SENTENCES
42
Elementary expressions
43
a blue shirt
MODULE 1
MODULE 2
PRED JJ ( blue ) (SLOT 0) ARG1 DT N C
( a shirt ) (SLOT 1)
PRED colour, blue (SLOT 0) ARG1 entity,
object, artifact (SLOT 1)
the governements minister
ARG1 entity, person... (SLOT 1) ARG2 entity,
person... (SLOT 2) TYPE POS (SLOT type)
ARG1 DT N C ( the minister ) (SLOT 1) ARG2
N C POS ( goverDTents ) (SLOT 2) TYPE
POS (SLOT type)
44
Complex sentences

• A complex sentence is any sentence that is formed
  by more than one elementary expression
• A complex sentence requires more than one
  proposition for its semantic representation

45
MODULE 1
Output Layer
Input Layer
Input Word
Slot 0 Slot 1 Slot 2 Slot 3 Type S Back,
Test Subcat.
46
Non Invariant Solution
Output Layer
Input Layer
Input Word
COMPLETE SENTENCE STRUCTURE (OPERATIONS WITH
VECTORS)
47
Invariant solution (a kind of shift and reduce
parser)
Output Layer
STACK
Stored Context
Input Layer
Input Word
Slot 0 Slot 1 Slot 2 Slot 3 Type S Back,
Test Subcat.
Focus of attention (current context)
48
MODULE 1
Output Layer
STACK
Stored Context
Input Layer
Input Word
Slot 0 Slot 1 Slot 2 Slot 3 Type S Back,
Test Subcat.
Focus of attention (current context)
49
Modelos concéntricos (Cowan, 1988, 1995, 1999
Oberauer, 2002)
50
A
Neurons whose receptive fields are invariant
(translation and scale), higher visual areas
(inferotemporal cortex)
Covert attention
retinotopic visual neurons (as found in V1 and
V2)
A J
L
A
51
A
Neurons whose receptive fields are invariant
(translation and scale), higher visual areas
(inferotemporal cortex)
Covert attention
A J
L
A
retinotopic visual neurons (as found in V1 and
V2)
52
Attention and invariance

• Shift reduce parser
• Stolke (1990), Sopena (1993), Miikulainen (1996)

53
(No Transcript)
54

• The main manager bought some old cars with three
  wheels.

Input Word The DT OUTPUT M1 PUT1 M2
Current Pred A1 A2 A3 Flags
55
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word The DT M1 NEXT M2
Current Pred A1 The A2 A3 Flags _at_1
56
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word main JJ_PR M1 IZ-IN M2
Current Pred A1 The A2 A3 Flags _at_NEXT _at_1
57
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word main JJ_PR M1 PUT0 M2
Current Pred A1 A2 A3 Flags
Top Pred A1 The A2 A3 Flags _at_1
58
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word main JJ_PR M1 NEXT M2
Current Pred main A1 A2 A3 Flags
Top Pred A1 The A2 A3 Flags _at_1
59
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word manager DT_N M1 PUT1 M2
Current Pred main A1 A2 A3 Flags _at_NEXT
Top Pred A1 The A2 A3 Flags _at_1
60
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word manager DT_N M1 OZ-OUT M2
Current Pred main A1 manager A2 A3 Flags
_at_NEXT
Top Pred A1 The A2 A3 Flags _at_1
61
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word manager DT_N M1 PUT1 M2
PmainA1manager
Current Pred A1 The A2 A3 Flags _at_1 _at_OZ-OUT
62
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word manager DT_N M1 NEXT M2
PmainA1manager
Current Pred A1 The manager A2 A3 Flags _at_1
_at_OZ-OUT
63
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word bought V_MA M1 PUT0 M2
PmainA1manager
Current Pred A1 The manager A2 A3 Flags _at_1
_at_NEXT
64
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word bought V_MA M1 NEXT M2
PmainA1manager
Current Pred bought A1 The manager A2 A3 Flag
s _at_0
65
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word some DT M1 PUT2 M2
PmainA1manager
Current Pred bought A1 The manager A2 A3 Flag
s _at_0 _at_NEXT
66
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word some DT M1 NEXT M2
PmainA1manager
Current Pred bought A1 The manager A2
some A3 Flags _at_2
67
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word old JJ_PR M1 IZ-IN M2
PmainA1manager
Current Pred bought A1 The manager A2
some A3 Flags _at_2 _at_NEXT
68
Generalized Role Labeling using Propositional
Representations

• The main manager bought some old cars with three
  wheels.

Input Word old JJ_PR M1 PUT0 M2
PmainA1manager
Current Pred A1 A2 A3 Flags _at_IZ-IN
Top Pred bought A1 The manager A2
some A3 Flags _at_2
69
Compositionality

• We now turn to what I think was an important
  mistake at the core of generative grammar, one
  that in retrospect lies behind much of the
  alienation of linguistic theory from the
  cognitive sciences. Chomsky did demonstrate that
  language requires a generative system that makes
  possible an infinite variety of sentences.
  However, he explicitly assumed, without argument
  (1965 16, 17, 75, 198), that generativity is
  localized in the syntactic component of the
  grammar (Jackendoff, 2002)

70
Compositionality

• The fact that semantics is purely interpretive
  has as a consequence that thought has no
  independent status and it cannot be creative or
  have an independent capacity of combinatoriality
  outside of language.
• As Phillips (2004) points out, that thought is
  purely interpretative and not creative is a
  consequence that is likely to be uncomfortable
  for many, including Jackendoff (Phillips, 2004
  p. 574).

71
Compositionality

• Semantic (or thought) is purely interpretative.
  Only syntax is creative.
• I think that to believe in absolute free will
  is the main cause of all types of fundamentalism,

72
Training minimalista

• SS-1-1- (DT The WAIT NEXT)
• SS-1-2- (DT_N man PUT1 NEXT)
• SS-1-3- (V_MA sold PUT0 NEXT)
• SS-1-4- (DT some WAIT NEXT)
• SS-1-5- (DT_N offerings PUT2 NEXT)
• SS-1-6- (IIN_DT to WAIT NEXT)
• SS-1-7- (DT the WAIT NEXT)
• SS-1-8- (DT_N president PUT3 NEXT)
• SS-1-9- (. . OZ-OUT NEXT)
• SS-1-10- (FIN)

73
Training minimalista(8-10 paraules maxim)

• RL-22-1- (DT a NADA NEXT)
• RL-22-2- (DT_N land PUT1 NEXT)
• RL-22-3- (CC , PUTtypeCC NEXT)
• RL-22-4- (WP2 where TESTARG NOTEST CLEARmodeCC
  IZ-IN2 PUTtypeWDT NEXT)
• RL-22-5- (DT a NADA NEXT)
• RL-22-6- (DT_N saying PUT1 NEXT)
• RL-22-7- (V_MA says PUT0 BACK2 MV23 BACK_ADJ
  IZ-INE2 PUTtypeADJ OZ-OUT NEXT)
• RL-22-8- (. . OZ-OUT OZ-OUT NEXT)
• RL-22-9- (FIN)

74
Test real, PTBII (55-26)

• s5974 But predictions that central banks of the
  Group of Seven - G-7 - major industrial nations
  would continue their massive dollar sales went
  astray , as the market drove the dollar downward
  on its own , reacting to Wall Street 's plunge
  and subsequent price volatility , lower U.S.
  interest rates and signs of a slowing U.S.
  economy .

75
Test

• Ho hem probat amb 254 frases del PTBII
• Els resultats son molt bons. La idea es tenir 0
  errors i crec que es pot conseguir.
• Quin es le problema?

76
NLDs, coordination, comparatives, puntuació

• Dependency grammars and parsers often ignore some
  classes of dependencies
• Puntuació (guionets, parentesis, comes, dos
  punts, .....)

77
NLDs
78
5941 Mr. NNP - (A0
5941 Nakamur NNP -
) 5941 cites
VBZ cite (V)
5941 the DT - (A1
5941 case NN -
5941 of
IN -
5941 a DT - (A0
(A0 (A0 (A0 lt---------4NLDs5941
custome NN - ) )
) ) 5941 who
WP - (R-A0) (R-A0) (R-A0)
(R-A0)5941 wants VBZ want
(V) 5941 to TO
- (A1 5941
build VB build (V)
5941 a DT -
(A1 5941 giant JJ -
5941
tourism NN -
5941 complex NN -
) 5941 in IN -
(AM-LOC 5941
Baja NN - ) )
5941 and CC -
5941 has VBZ -
5941
been VBN -
5941 trying VBG try
(V) 5941 for IN -
(AM-TMP5941
eight CD -
5941 years NNS -
) 5941 to TO -
(A1 5941
get VB get
(V5941 around IN -
)5941 Mexican NNP
-
(A15941 restric NNS -
5941 on IN -
5941
foreign JJ -
5941 ownersh NN -
5941 of IN -
5941
beachfr JJ -
5941 propert NN - )
) )5941 . .
-
79
5961 Fed NNP -
5961 funds NNS -
5961 is VBZ -
5961 the DT
- (A1 5961 rate
NN - ) 5961
banks NNS - (A0)
5961 charge VBP charge (V)
5961 each DT - (A2
5961 other JJ -
) 5961 on IN
- (A3 5961 overnig
JJ - 5961
loans NNS - )
5961 -
5961 the DT -
(A0 (A0 (A05961 Fed NNP -
) ) )5961 influen VBZ
influenc (V) 5961 the
DT - (A1
5961 rate NN - )
5961 by IN -
(AM-MNR 5961 adding VBG add
(V) 5961 or CC
- 5961 drainin
VBG drain (V)5961
reserve NNS - (A1)
(A2)5961 from IN -
(AM-MNR(A15961 the DT -
5961 banking NN
- 5961 system
NN - ) ) )5961
. . -

80
5920 For IN -
(AM-PNC 5920 the DT
- (A0
5920 PRI NNP - )
5920 to TO
-
5920 stand VB stand (V)
5920 a DT
- (A1
5920 chance NN - )
) 5920 , ,
-
5920 Mr. NNP -
(A0 5920 Salinas NNP
- )
5920 has VBZ have (V)
(AM-MOD 5920 to TO
-
5920 press VB press
(V 5920 on RP
- )
5920 with IN -
(A1 5920 an DT
- (A0 (A0
(A05920 economi JJ -
5920 program NN
- ) )
)5920 that WDT -
(R-A0) (R-A0) (R-A0) lt---------
3NLDs5920 so RB -
(AM-TMP 5920 far RB
- )
5920 has VBZ -
5920 succeed VBN
succeed (V)
5920 in IN -
5920 lowerin VBG
lower (V)
5920 inflati NN -
(A1) 5920 and CC
-
5920 providi VBG provide
(V)5920 moderat JJ
-
(A15920 economi JJ -
5920 growth NN
- ) )
)5920 . . -

81
NLDs (Johnson 2002)

• Broad coverage syntactic parsers withgood
  performance have recently become available
  (Charniak, Collins), but these typically produce
  as output a parse tree that only encodes local
  syntactic information, i.e., a tree that does not
  include any "empty nodes".

82
NLDs (Dienes 2003)

• Intuitively, the problem of parsing with NLDs is
  that the empty elements (EEs) representing these
  dependencies are not in the input.
• Therefore, the parser has to hypothesise where
  these EEs might occur in the worst case, it
  might end up suggesting exponentially many
  traces, rendering parsing infeasible (Johnson and
  Kay1994).

83
NLDs

• From the point of view of the dependency
  structure, NLDs are dif?cult because they
  violate the assumption that dependency structures
  are represented as directedtrees. Speci?cally,
  NLDs give rise to re-entrancies in the dependency
  graph, i.e., itis no longer a directed tree but
  a directed graph, with nodes possibly having
  multi- ple parents (e.g. apple in Figure 2.2).
  Now, the parser has to explore a much
  largersearch space.

84
NLDs

• Arguably, the search space is much more
  restricted by an actual grammar that exploits,
  for instance, the knowledge that buy is a
  transitive verb and thus requires a direct
  object.
• Nevertheless, the problem does not disappear.
  Consider the following example When demand is
  stronger than suppliers can handle and delivery
  times lengthen, prices tend to rise.
  (wsj_0036.mrg)

85
NLDs

• Non-local dependencies and displacement phenomena
  have been a central topic of generative
  linguistics since its inception half a century
  ago. However
• Many current linguistic theories of non-local
  dependencies are extremely complex, and would be
  difficult to apply with the kind of broad
  coverage described here.

86
Why a probabilistic approach?

• Ambiguity and underspecification are ubiquitous
  in human language utterances, at all levels
  (lexical, syntactic, semantic, etc.), and how to
  resolve these ambiguities is a key communicative
  task for both human and computer natural language
  understanding (Manning, 2003).

87

• At the highest level, the probabilistic approach
  to natural language understanding is to view the
  task as trying to learn the probability
  distribution
• P(meaning utterance context)
• A mapping from form to meaning conditioned by
  context.

88
Why a probabilistic approach

• quantum mechanics (uncertainty)
• Classical Physics (underspecification)

89
Why a probabilistic approach?

• Collins (1996)
• Ambiguity
• PP-attachment
• Coordination

90
PP-attachment

• Pierre Vinken, 61 years old, joined the board as
  a nonexecutive director.
• Pierre Vinken, 61 years old, joined the board as
  a nonexecutive director.

91
PP-attachment

• 4-tuple joined board as director
• V joined, N1 board, P as, and N2
  director.
• p (A l Vv, N1n1, Pp, N2n2)
• p(l v, n1,p, n2)

92
Results PP attachment ordered by Accuracy
Method
Accuracy
Ratnaparkhi (1994) ID3
77.70
Zavrel et al. (1996) Neural Networks
80.00

Ratnaparkhi (1994) Maximum Entropy Model
81.60
Takahashi (2001) Neural Networks
83.10
Yeh and Vilain (1998) Error-driven learning
83.10

Abney et al. (1999) Boosting
84.40

Zavrel et al. (1997) Memory-Based Learning
84.40
Collins and Brooks (1995) Backed-Off Model
84.50
Krymolowski and Rooth (1998) SNOW
84.80
Committee Machines 1 (Alegre et al, 1999)
86.08
88.01
Committee Machines 2 (Alegre, 2004)
88.01
Average Human Expert (Ratnaparkhi, 1994)
88.20
93
Mary bought a shirt with pockets
MODULE 1
MODULE 2
P1) PRED V MA PE PA ( bought) ARG1 N
PR (Mary) ARG2 DT N C ( a shirt ) ARG3
IIN N C (with pockets)
P1) PRED get, transfer, give, pay, ARG1
entity, person ARG2 entity, object, artifact
ARG3 artifact part-of-dress
ARG3 artifact part-of-dress
Parsing strategy Attaches first to the current
proposition
94
Mary bought a shirt with pockets
MODULE 1
MODULE 2
P1 PRED V MA PE ( bought) ARG1 N PR
(Mary) ARG2 DT N C ( a shirt ) ARG3
P1 PRED get, transfer,pay, accept ... ARG1
entity, person,... ARG2 entity, object,
artifact, shirt ARG3
P2 PRED ARG1 DT N C (a shirt) ARG2 IIN
N C (with pockets) ARG3
P2 PRED ARG1 entity, object, artifact,
shirt ARG2 artifact part-of-dress ARG3
95
Millors resultats

• Alegre (2002)
• PTBI 89.8 (PTBII 92.3)
• Olteanu and Modovan (2005)
• PTBII 94

96
Parsers de dependencies?
97
Ambigüitat

• Tota la artilleria que cal per resoldre el
  PP-attachment shauria de fer servir per resoldre
  com col.locar cada paraula de la frase en la
  estructura?
• En quins fenòmens caldria?
• La resposta es en molt pocs.