Introducing… The Letter Tt. “Tt” Words table “Tt” Words tack.
Pushing TAG to the Extremesstefan/Bilder/wp... · 3 From TAG to TT-MCTAG 4 Coherent Constructions:...
Transcript of Pushing TAG to the Extremesstefan/Bilder/wp... · 3 From TAG to TT-MCTAG 4 Coherent Constructions:...
Pushing TAG to the Extremes
Coherent Constructions and Ellipsis in German
Timm Lichte
University of Dusseldorf, Germany
Berlin, 19.03.2013
SFB 991
1 / 63
Overview
Two fundamental challenges for syntactic theories that incorporatevalency relations:
Discontinuity
(1) Zu reparieren versprochen hat Peter den VW.
Ellipsis
(2) Peter hat den VW zu reparieren versprochen undSusi hat den Ford zu reparieren versprochen.
⇒ How to model this with TAG and its variants?
2 / 63
Table of contents
1 Tree-Adjoining Grammar
2 Coherent Constructions
3 From TAG to TT-MCTAG
4 Coherent Constructions: Analyses with TT-MCTAG
5 Ellipsis
6 Ellipsis: Analyses with TAG/TT-MCTAG
3 / 63
Tree-Adjoining Grammar – Basics
A Tree Adjoining Grammar (TAG) is a set of elementary trees:
a finite set of initial trees
a finite set of auxiliary trees
E.g.:
VP
ADV VP*
easily
VP
NP↓ VP
V NP↓
repaired
Combinatorial operations:
substitution: replacing a non-terminal leaf with an initial tree
adjunction: replacing an internal node with an auxiliary tree
4 / 63
Tree-Adjoining Grammar – Example
NP
Peter
VP
NP↓ VP
V NP↓
repaired
NP
the fridgeVP
ADV VP*
easily
derived tree derivation treeVP
NP VP
Peter ADV VP
easily V NP
repaired the fridge
repaired
Peter
1
easily
2
the fridge
22
5 / 63
Tree-Adjoining Grammar – As a linguistic theory
⇒ What does an elementary tree stand for, and what is its shape?
Lexicalization
Each elementary tree has at least one non-empty lexical item, itslexical anchor.
Fundamental TAG Hypothesis (FTH, [Frank, 2002])
Every syntactic dependency (valency, binding, marking,modification, . . . ) is expressed locally within an elementary tree.
θ-criterion (TAG version,[Frank, 2002])
a. If H is the lexical head of an elementary tree T, H assigns all ofits θ-roles in T.
b. If A is a frontier non-terminal of elementary tree T, A must beassigned a θ-role in T.
. . .6 / 63
Tree-Adjoining Grammar – Large TAG-based grammars
English and Korean (XTAG, UPenn)
French TAG (Benoit Crabbe’s PhD-thesis)
German (GerTT, with TT-MCTAG)
. . .
7 / 63
Tree-Adjoining Grammar – Formal properties
⇒ How is it possible to follow these design principles?
Elementary trees span over an extended domain of locality
Due to recursive adjunction, nodes of elementary trees can end upat (arbitrarily) distant positions in the derived tree.
TAGs are mildly context-sensitive
1) Polynomial time parsing complexity
2) Generation of limited crossing dependencies
3) Constant growth property (semilinearity)
Mild context-sensitivity characterizes the generative capacityneeded for the analysis of natural language syntax.
8 / 63
Why CFG is not enough – Low expressive power
Example: cross-serial dependencies in Dutch and in SwissGerman [Huybrechts, 1984, Shieber, 1985]
(3) ......
datthat
WimWim
JanJan
MarieMarie
de kinderenthe children
zagsaw
helpenhelp
lerenteach
zwemmenswim
(’... that Wim saw Jan help Marie teach the children to swim’)
(4) ......
merwe
d’chindthe children
em Hansthe Hans
es huusthe house
haendhave
welewanted
laalet
halfehelp
aastriichepaint
(’... we have wanted to let the children help Hans paint the house.’)
A formalism that can generate cross-serial dependencies must (?)be able to generate the copy language {ww |w ∈ {a, b}∗}.
But: The copy language is not context-free.
9 / 63
Example for an analysis of cross-serial dependencies
VP
VP V
NP VP* zag
Wim
VP
VP V
NP VP* helpen
Jan
VP
VP V
NP VP* leren
Marie
VP
VP V
NP zwemmen
de kinderen10 / 63
Coherent Constructions – Definition
Coherent Constructions according to [Bech, 1955] in a nutshell
subordinative chain of verbs+ shared access to a topological region (“Koharenzfeld”)= arguments of different verbs can be linearly permuted
(5) Den VWN3
hatV 1
PeterN1
zu reparierenV 3
2
versprochen.V 2
1
F 1 F 3 F 2 (verbal fields)
︸ ︷︷ ︸ ︸ ︷︷ ︸
K 1 = restfield endfield / verbal complex
12 / 63
Coherent Constructions – Harder nuts
3rd construction / partial extraposition:
(6) a. dass Peter ihn versucht, zu reparierenb. dass Peter versucht, ihn zu reparierenc. *dass ihn versucht, Peter zu reparieren
Partial fronting:
(7) a. Zu reparieren versprochen hat ihm das Peter.b. Ein Außenseiter gewonnen hat hier noch nie.
Remote passive:
(8) a. wenn Peter ihn zu reparieren versuchtb. wenn er (*ihn) von Peter zu reparieren versucht wird
13 / 63
Coherent Constructions – Modeling strategies I
discontinuity
indirectsyntactic representation
directsyntactic representation
movement valency merge
“reanalysis” “valency inheritance/transfer” “crossing branches”“restructuring” “θ-grid composition” “tangled trees”
“S/CP/IP deletion” “argument attraction” “tecto- vers. phenogrammar”“base generation” “linearization domains”
e. g. [Evers, 1975], e. g. [Geach, 1970], e. g. [Curry, 1963],[McCawley, 1982],
[Haegeman and van Riemsdijk, 1986], [Steedman, 1985],[Jacobs, 1992], [Zwicky, 1986],[Reape, 1992],
[von Stechow, 1990], [Sabel, 1996], [Hinrichs and Nakazawa, 1989], [Kathol, 1995],[Muller, 1999],
[Meurers, 1999],[Muller, 2002], [Duchier and Debusmann, 2001],
[Sternefeld, 2006] [Daniels, 2005]
14 / 63
Coherent Constructions – Modeling strategies II
Movement [Evers, 1975]:
S-deletion + V-movement NP-movement
S
NP1 S V1
NP2 V2
=⇒
S
NP1 NP2 V
V2 V1
=⇒
S
NP2 NP1 V
V2 V1
Valency merge (mainly CCG, HPSG):
sbc〈 〉
np sbc〈 1np 〉
NP2 np sbc〈 1 ,np〉
NP1 sbc<np> sbc〈 1 ,np,[sbc〈 1 〉]〉
V2 V1
15 / 63
Coherent Constructions – Modeling strategies III
Direct syntactic representation:
S
VP
NP NP V V
NP2 NP1 V2 V1
ID/LP: direct liberation, Mobile Grammars, Generalized ID/LP
yield functions: LCFRS, RCG, . . .
HPSG with Reape’s linearization domains
Dependency Grammar: XDG
. . . and TAG
16 / 63
Coherent Constructions – Modeling strategies IV
Where are the tangled trees in TAG?
γ2 :
S
NP2 VP
V2
γ1 :VP
NP1 VP* V1
Inverted derivation tree, mapped onto the terminal string:
γ1
γ2
NP2 NP1 V2 V1
17 / 63
From TAG to TT-MCTAG – Why not TAG?
⇒ No sufficient derivational capacity:
(9) daß des VerbrechensNP2
der DetektivNP1
den VerdachtigenNP2
dem KlientenNP1
zu uberfuhrenV2
versprochen hatV1 [Becker et al., 1991]
(10) dass den MannNP2
den TeichNP3
niemandNP1
saubernV3
helfenV2
sahV1
(11) SCRind = {σ(NP1 , . . . ,NPk )V 1 . . .V k |k ≥ 1 and σ is a permutation}
⇒ Many and sometimes awkward elementary trees
19 / 63
From TAG to TT-MCTAG – Why not tree-local MCTAG?
Tree-Local Multi-Component TAG (TL-MCTAG) [Weir, 1988]
The grammar consists of sets of elementary trees.
(MC): using an elementary tree set implies the use of all members.
(SIM): all members must be used simultaneously.
(TL): all members must attach to the same elementary tree.
VP
NP1 VP* V 1
VP
NP2 VP*,
VP
VP* V 2
VP
NP3 VP
V 3
⇒ Can handle NP2 NP1 NP2 NP1 V2 V1 and NP2 NP3 NP1 V3 V2 V1 .
⇒ However, cannot cope with NP2 NP4 NP3 NP1 V4 V3 V2 V1 !
(12) weil [der Frau]2 [diesen Teich]4 [den Mann]3 niemand1 leer4fischen3 helfen2 sah1 [Muller, 2010, 270]
20 / 63
From TAG to TT-MCTAG – Beyond TL-MCTAG?
Vector MCTAG (V-TAG) [Rambow, 1994]⇒ “limit non-locality, local domain is marked up explicitly”
MCTAG with shared nodes (SN-MCTAG) [Kallmeyer, 2005]⇒ “extend tree-locality by node sharing, movement analysis”
MCTAG with tree tuples (TT-MCTAG) [Lichte, 2007]⇒ “extend tree-locality by node sharing, no movement analysis”
21 / 63
From TAG to TT-MCTAG – Definition of TT-MCTAG
MCTAG with shared nodes and tree tuples (TT-MCTAG)
Elementary structures are pairs 〈α, {β1 , . . . , βn}〉
α is the head tree.β1 , . . . , βn are the argument trees.
Restrictions on use:
(MC), not (SIM)(SH): Argument trees must attach to their head tree
directly or within node sharing locality.
Node sharing locality of a tree includes:
the root node of each of the adjoining treesthe node sharing locality of each of the adjoining trees
⟨VP
V
repariert
,
VP
NPnom ↓ VP*,
VP
NPacc ↓ VP*
⟩
22 / 63
From TAG to TT-MCTAG – Definition of TT-MCTAG
(13) dass ihn heute Peter repariert(“that Peter repairs it today”)
VP
ADV VP*
heute
⟨
VP
V
repariert
,
VP
NPnom ↓ VP*,
VP
NPacc ↓ VP*
⟩
NP
Peter
NP
ihn
repariert
NPnom
a.1
Peter
s.1
heute
a.ǫ
NPacc
a.ǫ
ihn
s.1
23 / 63
Analyses with TT-MCTAG – Basics
prefield phrase VP prefield
XP VP
V VP middlefield
repariertleft bracket VP
V right bracket
ǫ
25 / 63
Analyses with TT-MCTAG – Basics
⟨
VP[
vf +
stype v12
]
[
vf 1]
VNAVP[
vf -]
[
vf -]
repariert V
ε
,
VP[
vf 2]
[
vf 2]
NPnom ↓VP*[
vf −
]
[
vf −
]
,
VP[
vf 3]
[
vf 3]
NPacc ↓VP*[
vf −
]
[
vf −
]
⟩
⟨
VP[
vf 1
stype v3
]
[
vf −
]
V
repariert
,
VP[
vf 2]
[
vf 2]
NPnom ↓VP*[
vf −
]
[
vf −
]
,
VP[
vf 3]
[
vf 3]
NPacc ↓VP*[
vf −
]
[
vf −
]
⟩
26 / 63
Analyses with TT-MCTAG – Intraposition I
(14) dass ihn Peter zu reparieren versucht
⟨
V[
phrase +]
V*[
phrase -] V[
phrase -]
versucht
,
V[
phrase +]
NPnom ↓ V*[
phrase +]
⟩
⟨V[
phrase 1]
zu reparieren
,
V[
phrase +]
NPacc ↓ V*[
phrase +]
⟩
zu reparieren
versucht
a.ǫ
NPnom
a.ǫ
NPacc
a.ǫ
dass
a.ǫ
ihn
s.1
peter
s.1
27 / 63
Analyses with TT-MCTAG – Intraposition II
(15) Ihn versucht Peter zu reparieren.
⟨
VP
V VP
versucht V* V
ε
,
VP
NPnom ↓ VP*
⟩
⟨
V(P)
zu reparieren,
VP
NPacc ↓ VP*
⟩
zu reparieren
versucht
a.ǫ
NPacc
a.ǫ
ihn
s.1
NPnom
a.1
peter
s.1
28 / 63
Analyses with TT-MCTAG – 3rd construction
(16) a. dass Peter ihn versucht, zu reparierenb. dass Peter versucht, ihn zu reparierenc. *dass ihn versucht, Peter zu reparieren
⟨VP
V VP*
versucht
,
VP
NPnom ↓ VP*
⟩
⟨V(P)
zu reparieren,
VP
NPacc ↓ VP*
⟩
29 / 63
Analyses with TT-MCTAG – Partial fronting I
(17) a. Ihn zu reparieren versucht Peter.b. Zu reparieren versucht ihn Peter.c. *Peter zu reparieren versucht ihn.
⟨
VP[
vf +]
VP* V VP
versucht V
ǫ
,
VP
NPnom ↓ VP*
⟩
⟨V(P)
zu reparieren,
VP
NPacc ↓ VP*
⟩
30 / 63
Analyses with TT-MCTAG – Partial fronting II
(18) Ein Außenseiter zu gewinnen scheint hier eigentlich nie.
⟨
V(P)[ ]
[
status 2]
zu gewinnen
,
VP[
assign-case 1
agr 2
]
NP[
cas 1
agr 2
] VP*[
assign-case 1
agr 2
]
⟩
⟨
VP
VP*
assign-case nom
agr
[
num sg
pers 3
]
V VP
scheint V
ǫ
,
{ }
⟩
31 / 63
Analyses with TT-MCTAG – Partial fronting III
(19) Zu reparieren versprochen hat ihm das Peter.
V(P)
zu reparieren
VP
VP* V
versprochen
VP
VP* V VP
hat V
ǫ
32 / 63
Analyses with TT-MCTAG – Remote passive
(20) wenn er von Peter zu reparieren versucht wird
⟨
V(P)[ ]
[
assign-case acc
stat 2
]
zu reparieren
,
VP[
assign-case 1
agr 2
]
NP[
cas 1
agr 2
] VP*[
assign-case 1
agr 2
]
⟩
⟨
V(P)[ ]
[
passive +]
V*[
stat 2]
V
versucht
,
VP[
assign-case 1
agr 2
]
PPvon
VP*[
assign-case 1
agr 2
]
⟩
⟨
V(P)
agr
[
num sg
pers 3
]
assign-case nom
V*[
passive +]
V
wird
,
{ }
⟩
33 / 63
Analyses with TT-MCTAG – Intermediate conclusion
Node Sharing does not suffice; we need tree sharing.
Remote passives require the percolation of special featuresin the syntactic structures.
direct representation of discontinuity, but indirect case markingcomparable to Reape’s HPSG accountother than valency merge accounts
Each construction type calls for its own tree template.
impact on grammar sizemissed generalizations?burden on parsing efficiency?
34 / 63
Overview again
Two fundamental challenges for syntactic theories that incorporatevalency relations:
Discontinuity
(21) Zu reparieren versprochen hat Peter den VW. XEllipsis
(22) Peter hat den VW zu reparieren versprochen undSusi hat den Ford zu reparieren versprochen.
⇒ How to model this with TAG and its variants?
35 / 63
Aelbrecht, L. (2010).
The Syntactic Licensing of Ellipsis.John Benjamins, Amsterdam/New York.
Bech, G. (1955).
Studien uber das deutsche Verbum infinitum.2nd unrevised edition published 1983 by Max Niemeyer Verlag, Tubingen.
Becker, T., Joshi, A. K., and Rambow, O. (1991).
Long-distance scrambling and tree adjoining grammars.In Proceedings of the 5th Conference of the European Chapter of the Association for ComputationalLinguistics.
Blevins, J. (1990).
Syntactic Complexity: Evidence for Discontinuity and Multidomination.PhD thesis, University of Massachusetts, Amherst, MA.
Curry, H. B. (1963).
Some logical aspects of grammatical structure.In Structure of Language and its Mathematical Aspects: Proceedings of the Twelfth Symposium in AppliedMathematics, pages 56–68.
Daniels, M. W. (2005).
Generalized ID/LP Grammar: A Formalism for Parsing Linearization-Based HPSG Grammars.PhD thesis, The Ohio State University.
Duchier, D. and Debusmann, R. (2001).
Topological dependency trees: A constraint-based account of linear precedence.In Proceedings of 39th Annual Meeting of the Association for Computational Linguistics, pages 180–187,Toulouse, France.
Evers, A. (1975).
The Transformational Cycle in Dutch and German.PhD thesis, University of Utrecht.
Frank, R. (2002).
Phrase Structure Composition and Syntactic Dependencies.MIT Press, Cambridge,MA.
Geach, P. T. (1970).
A program for syntax.Synthese, 22:3–17.
Haegeman, L. and van Riemsdijk, H. (1986).
Verb projection raising, scope, and the typology of rules affecting verbs.Linguistic Inquiry, 17:417–466.
Haider, H. (1993).
Deutsche Syntax - generativ. Vorstudien zu einer projektiven Grammatik.Gunter Narr Verlag, Tubingen.
Hartmann, K. (2000).
Right Node Raising and Gapping.John Benjamins, Amsterdam.
Hinrichs, E. and Nakazawa, T. (1989).
Flipped out: Aux in German.In Papers from the 25th Regional Meeting of the Chicago Linguistics Society, pages 193–202, Chicago,Illinois.
Hohle, T. (1990).
Assumptions about asymmetric coordination in German.In Mascaro, J. and Nespor, M., editors, Grammar in Progress. Glow Essays for Henk van Riemsdijk, pages221–235. Foris, Dordrecht.
Hohle, T. N. (1991).
On reconstruction and coordination.In Haider, H., Netter, K., Haegeman, L., Maling, J., and McCloskey, J., editors, Representation andDerivation in the Theory of Grammar, volume 22 of Studies in Natural Language and Linguistic Theory,pages 139–197. Springer.
Huybrechts, R. A. C. (1984).
The weak inadequacy of context-free phrase structure grammars.In de Haan, G., Trommelen, M., and Zonneveld, W., editors, Van Periferie naar Kern, pages 81–99. Foris,Dordrecht.
Jacobs, J. (1992).
Bewegung als Valenzvererbung - Teil 1.Linguistische Berichte, 138:85–121.
Kallmeyer, L. (2005).
Tree-local multicomponent tree adjoining grammars with shared nodes.Computational Linguistics, 31:2:187–225.
Kathol, A. (1995).
Linearization-Based German Syntax.PhD thesis, Ohio State University.
Kindt, W. (1985).
Grammatische Prinzipien sogenannter Ellipsen und ein neues Syntaxmodell.In Meyer-Hermann, R. and Rieser, H., editors, Ellipsen und fragmentarische Ausdrucke, volume 1, pages161–290. Niemeyer, Tubingen.
Klein, W. (1981).
Some rules of regular ellipsis in German.In Klein, W. and Levelt, W. J., editors, Crossing the Boundaries in Linguistics. Studies Presented toManfred Bierwisch, pages 51–78. Reidel, Dordrecht.
Lichte, T. (2007).
An MCTAG with tuples for coherent constructions in German.In Proceedings of the 12th Conference on Formal Grammar, Dublin, Ireland.4-5 August 2007.
Lichte, T. and Kallmeyer, L. (2010).
Gapping through TAG derivation trees.In Proceedings of the 10th International Workshop on Tree-Adjoining Grammar and Related Formalisms(TAG+10), pages 93–100, New Haven, CT.
Lyons, J. (1968).
Introduction to Theoretical Linguistics.Cambridge University Press.
McCawley, J. D. (1982).
Parentheticals and discontinuous constituent structure.Linguistic Inquiry, 13(1):91–106.
Meurers, W. D. (1999).
Lexical Generalizations in the Syntax of German Non-Finite Constructions.PhD thesis, Seminar fur Sprachwissenschaft, Universitat Tubingen, Tubingen, Germany.published 2000 as Volume 145 in Arbeitspapiere des SFB 340, ISSN 0947-6954/00.
Muller, S. (1999).
Deutsche Syntax deklarativ. Head-Driven Phrase Structure Grammar fur das Deutsche.Number 394 in Linguistische Arbeiten. Max Niemeyer Verlag, Tubingen.
Muller, S. (2002).
Complex Predicates. Verbal Complexes, Resultative Constructions, and Particle Verbs in German.Studies in Constraint-Based Lexicalism. CSLI Publications, Stanford.
Muller, S. (2010).
Grammatiktheorie.Number 20 in Stauffenburg Einfuhrungen. Stauffenburg Verlag, Tubingen.
Osborne, T. (2008).
Major constituents and two dependency grammar constraints on sharing in coordination.Linguistics, 46:1109–1165.
Rambow, O. (1994).
Formal and Computational Aspects of Natural Language Syntax.PhD thesis, University of Pennsylvania, Philadelphia.IRCS Report 94-08.
Reape, M. (1992).
A Formal Theory of Word Order: A Case Study in West Germanic.PhD thesis, University of Edinburgh.
Sabel, J. (1996).
Restrukturierung und Lokalitat. Universelle Beschrankungen fur Wortstellungsvarianten, volume 42 ofStudia grammatica.Akademie-Verlag, Berlin.
Sarkar, A. (1997).
Seperating dependency from constituency in a tree rewriting system.In Becker, T. and Krieger, H.-U., editors, Proceedings of the Fifth Meeting on Mathematics of Language,pages 153–160, Saarbrucken.
Sarkar, A. and Joshi, A. (1997).
Handling coordination in a tree adjoining grammar.Longer version of paper in Proceedings of COLING 1996. Draft of August 19, 1997.
Seddah, D. and Sagot, B. (2006).
Modeling and analysis of elliptic coordination by dynamic exploitation of derivation forests in LTAG parsing.In Proceedings of The Eigth International Workshop on Tree Adjoining Grammars and Related Formalisms(TAG+8), pages 147–152, Syndey.
Shieber, S. (1985).
Evidence against the context-freeness of natural language.Linguistics and Philosophy, 8:333–343.
Steedman, M. (1985).
Dependency and coordination in the grammar of Dutch and English.Language, 61(3):523–568.
Sternefeld, W. (2006).
Eine morphologisch motivierte generative Beschreibung des Deutschen.Stauffenburg Verlag, Tbingen.2 Bande.
van Oirsouw, R. R. (1987).
The Syntax of Coordination.Croom Helm Linguistics Series. Croom Helm, London.
von Stechow, A. (1990).
Status government and coherence in German.In Grewendorf, G. and Sternefeld, W., editors, Scrambling and Barriers. John Benjamins, Amsterdam.
Weir, D. J. (1988).
Characterizing Mildly Context-Sensitive Grammar Formalisms.PhD thesis, University of Pennsylvania.
Winkler, S. and Schwabe, K. (2003).
Exploring the interfaces from the perspective of omitted structures.In Schwabe, K. and Winkler, S., editors, The Interfaces. Deriving and Interpreting Omitted Structures,pages 1–26. John Benjamins, Amsterdam.
Zwicky, A. (1986).
Concatenation and liberation.In Proceedings of the Twenty-Second Regional Meeting (General Session) of the Chicago LinguisticsSociety, pages 65–74.