Linking syntactic and semantic arguments in a dependency-based formalism
Christian Korthals and Ralph Debusmann
Computational Linguistics
Universit¨at des Saarlandes, Geb. 17
Postfach 15 11 50
66041 Saarb¨ucken, Germany
(cnkortha|rade)@coli.uni-sb.de
Abstract
We propose a formal characterization of variation
in the syntactic realization of semantic arguments,
using hierarchies of syntactic relations and thematic
roles, and a mechanism of lexical inheritance to ob-
tain valency frames from individual linking types.
We embed the formalization in the new lexicalized,
dependency-based grammar formalism of Topologi-
cal Dependency Grammar (TDG) (Duchier and De-
busmann, 2001). We account for arguments that can
be alternatively realized as a NP or a PP, and model
thematic role alternations. We also treat auxiliary
constructions, where the correspondance between
syntactic and semantic argumenthood is indirect.1
1 Introduction
This paper deals with the mapping (or linking) of se-
mantic predicate-argument structure to surface syn-
tactic realizations. We present a formal architecture
in the framework of a multi-dimensional, heavily
lexicalized, efficiently parsable dependency formal-
ism (Duchier and Debusmann, 2001), which uses
lexical inheritance as a means to explicitly model
syntactic variation. We concentrate on variation be-
tween prepositional phrases and nominal phrases
which realize verbal arguments, and remedy prob-
lems that occur with this kind of variation in recent
approaches like the HPSG linking architecture pro-
posed by (Davis, 1998).
Section 2 presents and analyses some of the prob-
lematic data we can model, English dative shift, op-
tional complements and thematic role alternations.
Section 3 compares the HPSG account with less for-
mal valency or dependency based approaches and
comments on the shortcomings, focusing on the
treatment of PP complements. We then present a
new account in the formal framework of Topologi-
1The authors wish to thank Denys Duchier and Geert-Jan
Kruijff for lots of helpful comments on this paper.
cal Dependency Grammar (TDG) by adding a new
representational level (thematic structure addition-
ally to ID structure) to the framework in Section
4.1 and introducing the concept of a valency frame
in the TDG inheritance lexicon (Sections 4.2 and
4.3). We then show how we use syntactic role hi-
erarchies to account for the data in a linguistically
concise way and define admissibility conditions for
a TDG derivation. Section 5 contrasts our analysis
of the dative shift construction with the analysis of
thematic role alternations.
2 Linguistic Data
Insights from corpus studies (e.g. the NEGRA
treebank for German (Skut et al., 1998), or the
material annotated in the Framenet (Baker et al.,
1998) project on the basis of The Bank of English
(Cobuild, 2001) show that the syntactic patterns
specific verbs occur with vary stongly. Not only do
we observe different patterns for different verbs, but
also alternative patterns with the same verbs. (1)
to (6) illustrate the well-known dative shift (Levin,
1993):45 phenomenon, which occurs with a re-
stricted class of verbs only. While the distinction
between (2) and (4) can be explained in terms of
lexical semantics, even semantically closely related
verbs as English give and deliver can differ in their
syntactic behaviour, as the contrast between (1) and
(5) shows.
(1) [The postman] gave [him] [a package].
(2) [The postman] gave [a package] [to him].
(3) [The postman] charged [him] [5 Euros].
(4) *[The postman] charged [5 Euros] [to him].
(5) *[The postman] delivered [him] [a package].
(6) [The postman] delivered [a package] [to him].
In contrast to (Davis, 1998):3:562 for instance,
we do not assume a difference in meaning between
(1) and (2).3 Therefore, in order to compute a se-
mantics from this data without spurious ambigu-
ity, we must be able to express the semantic gen-
eralisation that to him and him realize the same
semantic argument. It is useful to employ the-
matic roles in the lexicon and in grammatical de-
scriptions for this purpose.4 See e.g. (Tarvainen,
1987), (Helbig, 1989), (Sgall et al., 1986) or the
Framenet project (Baker et al., 1998) for differ-
ent application-oriented sets of thematic roles (or
“frame elements”). For discussion see e.g. (Helbig,
1995) or (Kruijff, 2001), for criticism see (Dowty,
1991). We can also use thematic roles to structure
verbs into an ontology, as e.g. attempted by (Helbig,
1995), (Davis, 1998) or (Baker et al., 1998) in order
to make semantic predictions for syntactic valency
patterns. For instance, it is a regularity in English
that verbs of charging do not show the dative shift
(Levin, 1993), while verbs of change of possession
sometimes do.
Now consider the set of German examples in (7)
to (11), which all roughly express the proposition
Peter robs her of her money. All of the patterns are
attested in the NEGRA corpus (Skut et al., 1998),
but (10) cannot be found.
(7) [Peter]
Peter
beraubt
robs
[sie]
her+ACC
[ihres
her+GEN
Bargelds].
cash+GEN
“Peter robs her of her cash.”
(8) [Peter]
Peter
beraubt
robs
[sie]
her+ACC
[um
of
ihr
her
Bargeld].
cash
“Peter robs her of her cash.”
(9) [Peter] beraubt [sie].
(10) *[Peter] beraubt.
(11) [Peter]
Peter
raubt.
robs
“Peter steals.”
The data illustrates that it can be a lexical prop-
erty of verbs to allow or prohibit omission of their
2We cite the chapter and the page num-
ber of the online postscript version at
http://www-csli.stanford.edu/ tdavis/
3We expect English give to have at least two separate mean-
ings (a) cause someone to physically have something and (b) to
cause someone trouble, pain, etc. with different lexical entries,
following established lexicographic practice. While the lexical
entry for meaning (b) will exhibit the syntactic pattern illus-
trated by (1) only (*To give headache to someone), the entry
for meaning (a) exhibits both the patterns in (1) and (2).
4Note that we do not commit ourselves to a specific set of
thematic roles in this paper.
complements (Levin, 1993):33, (Helbig, 1995):99.
Therefore, we will analyse syntactic arguments in
terms of optionality and obligatoriness. Note that
this distinction is not predictable from the thematic
roles realized by the syntactic elements (e.g. by
distinguishing inner and outer roles in (Sgall et
al., 1986) and (Tarvainen, 1987)) nor by the syn-
tactic form or even function of the syntactic ele-
ments. Neither is the distinction between obliga-
tory and optional elements the same as the comple-
ment/adjunct distinction.
We analyse (1) to (6) as alternative realizations
of a thematic role, because one semantic argument
(the PATIENT) can either be realized as indirect ob-
ject or PP, while the THEME is always realized as a
direct object NP. Compare this data to alternations
as in (12) and (13). Here, additionally, one syntac-
tic function (direct object) is open for either of two
thematic roles (Levin, 1993).
(12) [He] cleared [the dirt] [from the table].
(13) [He] cleared [the table] [of the dirt].
We will show in Section 4 how we can account
for the data illustrated in this section in a lexical-
ized dependency grammar formalism and show that
the linguistic view taken above helps to reduce re-
dundancy in the lexicon.
3 Alternative approaches
The approach taken in this paper formalizes notions
that have only been informally employed in depen-
dency grammar. (Helbig, 1995):167 defines valency
frames on a formal syntactic and functional syntac-
tic level, a thematic role level and a logical level in
his 6-level-model, but only informally and for the
purpose of learners’ dictionaries. There is a long
tradition in German lexicography which has pro-
duced a number of valency dictionaries (e.g. (Hel-
big and Schenkel, 1975), (Fischer, 1997), (Engel
and Schumacher, 1976)). The syntactic analyses in
these dictionaries are compatible with our model,
but they do not provide a thematic role level.
(Melcuk, 1988) characterizes valency frames in
a similar fashion (94), but uses informal additions
in natural language to constrain the possible pat-
terns. Also (Melcuk, 1988) assumes different levels
of representation. A shortcoming of the syntactic
level in (Melcuk, 1988) is, though, that his syntactic
classes are dependent on the specific lexical item,
and therefore problematic to define. The approach
we will take resembles LFG (Kaplan and Bresnan,
1982) (Bresnan, 2001) in that it assumes syntactic
relations.
(Davis, 1998) has recently proposed a linking the-
ory in the formal framework of HPSG. He sepa-
rates syntax and semantics by postulating thematic
roles under the CONTENT feature of his HPSG ar-
chitecture (Pollard and Sag, 1994), and syntac-
tic characterizations of the arguments under CAT-
EGORYjARG-ST and CATEGORYjSUBCAT. He has
separate hierarchies of syntactic patterns (intrans,
trans, ditrans, 5:32) and semantic classes (subtypes
of RELATION, 5:72). These hierarchies interact by
a set of linking constraints and yield a hierarchy
of predicators (5:41), which specifies possible link-
ings of thematic roles to syntactic arguments. While
(Helbig, 1995) obviously employs a large role set,
(Davis, 1998) has only 6 roles, and moves thematic
roles further down into semantics than we assume
by postulating them on an event level, which “ef-
fectively amounts to a limited amount of seman-
tic decomposition” (5:39). The shortcoming of the
model is that the syntactic patterns assumed are very
sparse indeed with only three transitivity classes.
Due to this, semantic predictions can be made only
for NP-complements, while PPs must be treated by
a separate mechanism (“content sharing account”).
Thus, there is no specific prediction for the preposi-
tional complement in English dative shift construc-
tions. The advantage of Davis’s model, in contrast,
is the lexical inheritance architecture which is a for-
mal means to capture generalizations.
4 Formalization
We formalize our idea of linking and valency frames
as an extension of the new lexicalized, dependency-
based grammar formalism of Topological Depen-
dency Grammar (TDG) (Duchier and Debusmann,
2001), (Debusmann, 2001). So far, TDG is only
concerned with syntax: every TDG analysis con-
sists of an unordered dependency tree (ID tree) and
an ordered and projective topology tree (LP tree).
We only describe a subset of the full TDG grammar
formalism (e.g. completely ignoring any issues con-
cerning word order) and extend it with the notion of
a thematic graph (TH graph). We call the version of
TDG described in this paper TDGTH.
4.1 Thematic graphs
Peter will live in Taipei
subj
vbse
ppin
pcomp
(14)
What is a “thematic graph”? We illustrate this
notion by an example. (14) is an ID tree analysis for
the sentence Peter will live in Taipei: We show the
corresponding TH graph in (15). Here, Peter is the
patient of will live and in Taipei the locative. Note
that we collapsed the auxiliary will and its verbal
complement live into a single node, and also the PP
in Taipei:
Peter will live in Taipei
th loc (15)
4.2 The lexicon
This section deals with the TDGTH-lexicon. We as-
sume a finite set of syntactic roles R and a finite set
of thematic roles T . We write ρ for a syntactic role
in R and θ for a thematic role in T . Π =f!;?g is the
set of optionality flags pi. A = fvalID;valTH;linkg is
the set of lexical features α, and E the set of lexical
entries e, having the following signature:5
2
4
valID : 2R  Π
valTH : 2T  Π
link : 2T  R
3
5
E is a lattice of TDGTH-lexicon entries; lexical en-
tries either correspond to words or to lexical types
which can be inherited (see below).
The value of feature valID is a set of pairs (ρ;pi)
of syntactic roles and an optionality flag modeling
the concept of syntactic valency. The value of valTH
a set of pairs (θ;pi) of thematic roles and an option-
ality flag (thematic valency). For convenience, we
write ρpi for (ρ;pi), and θpi for (θ;pi). The value of
link is a set of pairs (θ;ρ) of thematic and syntactic
roles, expressing the mapping between them. We
call a pair in this set a linking.
5We write α(e) to denote the value of feature α at lexical
entry e.
eat :
2
64
valID : fsubj!;objd?g
valTH : fag!;th?g
link : f(ag;subj);
(th;objd)g
3
75 (16)
As an example, (16) is a lexical entry for finite
eat: eat has an obligatory subject (subj) and an op-
tional direct object (objd) in its syntactic valency. Its
thematic valency contains an obligatory AGENT and
an optional THEME. The link-feature defines two
linkings: one links the AGENT to the subject and
the THEME to the direct object.
4.3 Lexical inheritance
We introduce a mechanism of lexical inheritance.
We write e = e1 u:::uen for ‘lexical entry e in-
herits from lexical entries e1;:::;en’, and define in-
heritance as the set union of the individual features’
values:
e1 u:::uen =2
4
valID : valID(e1)[:::[valID(en)
valTH : valTH(e1)[:::[valTH(en)
link : link(e1)[:::[link(en)
3
5
We can now use lexical inheritance to model our
notion of valency frames. We introduce the notion
of a linking type as a lexical entry that does not spec-
ify any other lexical attributes besides valID, valTH
and link. Such linking types specify a partial va-
lency frame from which we can build complete va-
lency frames by inheritance. For instance, consider
the following two linking types:
l ag subj :
2
4
valID : fsubj!g
valTH : fag!g
link : f(ag;subj)g
3
5 (17)
l th objd :
2
4
valID : fobjd?g
valTH : fth?g
link : f(th;objd)g
3
5 (18)
The linking type l ag subj maps the agent to the
subject, and l th objd the theme to the direct object.
Out of the two, we can construct our lexical entry
for eat by lexical inheritance:
eat = l ag subj u l th objd (19)
which amounts precisely to the lexical entry dis-
played in (16) above. We call the values of the three
features valID, valTH and link in a lexical entry ob-
tained by inheriting from linking types a valency
frame.
4.4 Role hierarchies
We arrange the set R of syntactic roles in a role
hierarchy modeled as a meet semi-lattice. Here is
an example cut-out of the role hierarchy:
dativeshift ppdirectional ppspatial
obji ppto ppinto ppin ppunder
(20)
We write ρ v ρ0 for ρ is a specialization of ρ0 (i.e.
ρ is below ρ0 in the hierarchy).
We employ the role hierarchy to model alterna-
tive realizations in the sense of section 2: e.g. us-
ing the hierarchy above, dativeshift can be realized
as either obji or ppto but not by either ppdirectional,
ppinto or ppin. Note that certain roles (ppto, ppinto,
etc.) will be realized by only two lexical entries,
viz. the prepositions to and into respectively, while
other roles like subj, obji or objd can be realized by
a large set of lexical entries.
In the same fashion, we arrange the set T of the-
matic roles in a role hierarchy, but in this article we
keep this hierarchy completely flat.
Lexical entry constraint. To forbid that different
thematic roles are mapped to the same syntactic role
realization, we add a condition for well-formed lex-
ical entries: for every lexical entry e, the value of
its link-feature, link(e) =f(θ1;ρ1);:::(θn;ρn)g must
not include two syntactic roles ρi, ρ j (1 i6= j  n)
such that ρi v ρ j.
4.5 TDGTH analyses
We can now define a TDGTH-analysis as a tuple
(V;EID;ε; ;ETH). It consists of the finite set V
of nodes w, the finite set EID of ID edges EID  
V  V  R , where R is the set of syntactic roles, and
the lexical assignment function ε : V ! E assigning
lexical entries to nodes. We write w1 ρ!IDw2 for
an ID edge from node w1 to node w2 labeled with
syntactic role ρ.
Collapsing nodes. As in the example ID tree and
TH graph-analyses in (14) and (15) above, we
would like to be able to collapse sets of nodes in
the ID tree into single nodes in the TH graph. We
introduce a collapsing principle into the grammar
formalism, according to which the node of an auxil-
iary verb, a preposition or a determiner will collapse
with its daughter.
To capture this idea, we posit an equivalence re-
lation  , and write V= for the set of equivalence
classes over V . An equivalence class directly cor-
responds to a “collapsed node” in the TH graph.
ETH  (V= ) (V= ) T is the finite set of of TH
edges, and we write w for the equivalence class con-
taining w. w1 θ!THw2 is a TH edge from node w1
to node w2 labeled with θ.
When we collapse nodes, we also collapse their
lexical entries: the value of the lexical feature α of
a collapsed node w1 = fw1;:::;wng is the set union
of the values assigned to the individual nodes:
α(w1) = α(w1)[:::[α(wn) (21)
In the example TH graph in (15) above, the two
nodes will and live are collapsed into the equiva-
lence class will live. We assume that will is mapped
to the following lexical entry:
will :
2
4
valID : fsubj!;vbse!g
valTH : fg
link : fg
3
5 (22)
Here, we use the independent definition of the valID,
valTH and link features in order to express that func-
tion words like auxiliaries or prepositions realize
syntactic arguments which are semantically depen-
dent on different lexical items. This also allows for
an elegant treatment of semantically void syntactic
arguments as fake-reflexives and non-referential it.6
Infinitive live has this lexical entry:
live :
2
64
valID : fppspatial!g
valTH : fag!;loc!g
link : f(ag;subj)
(loc;ppspatial)g
3
75 (23)
When the two nodes collapse into one (will live),
these are the values of valTH and link:7
valTH(will live) = fag!;loc!g
link(will live) = f(ag;subj);(loc;ppspatial)g
Valency constraints. The well-formedness con-
ditions of a TDGTH-analysis are stated in terms
of lexicalized constraints of syntactic valency, the-
matic valency and linking. The syntactic valency
6Fake-reflexives occur in German and are reflexive pro-
nouns that do not have any semantic content, as e.g. the Ger-
man verb sich f¨urchten (be afraid, sich is the reflexive pro-
noun). The linking type for a fake reflexive will only specify
valID : reflpronoun, but empty valTH and link.
7We omit those features that are not relevant to the TH graph
according to its well-formedness conditions. In particular, the
value of the valID feature is not relevant to the TH graph.
constraint restricts the number of outgoing edges of
each node in the ID tree: if (ρ;!) 2 valID(w), then w
must have precisely one outgoing edge labeled with
ρ0 v ρ, at most one if (ρ;?) 2 valID(w), and none
otherwise. Thus, (ρ;!) stands for an obligatory ρ-
complement of w and (ρ;?) for an optional one. The
thematic valency constraint is defined analogously.
Linking constraint. An edge in the TH graph is
only licensed if it satisfies the lexicalized linking
constraint. It states that an edge w1 θ!THw2 in
the TH graph is licensed only if there is a linking
(θ0;ρ0) 2 link(w1) and an edge w01 ρ!IDw02 in the
ID tree such that w10 = w1, w20 = w2, θ v θ0 and
ρ v ρ0.
Consider the example ID tree and TH graph
analyses in (14) and (15) above. The edge
will live loc!THin Taipei is mandated by the the-
matic valency of will live, but it must also be li-
censed by the linking principle: indeed there is a
linking (loc;ppspatial) in link(will live) and an ID
edge live ppin!IDin such that live = will live and
in = in Taipei, and loc v loc and ppin v ppspatial.
5 Application
This section describes the linguistic examples from
section 2 within the framework developed above.
We define a linking type for the English dative
shift construction as follows, realizing our notion of
an alternative realization from section 2.
l pat dativeshift :2
4
valID : fdativeshift!g
valTH : fpat!g
link : f(pat;dativeshift)g
3
5 (24)
As can be seen from the role hierarchy in (20),
the syntactic role dativeshift can either be realized as
obji or ppto. This linking type will be inherited by all
English verbs that exhibit the dative-shift alternative
realization. For instance, it is inherited by the lexi-
cal entry for the English verb give (cf. examples (1)
and (2)). Additionally, give inherits from the link-
ing type l ag subj defined in (17). l ag subj will be
shared among a large set of other verbs which re-
alize their agents as subjects, thus reducing redun-
dancy in the lexicon.
gives = l ag subj u
l pat dativeshift u
l th objd
(25)
The lexicon entry for deliver ((5) and (6)) will
differ from the one for give by inheriting from
l pat obji instead of l pat dativeshift and thus not
allow for a dative shift.
For the German data in examples (7) to (11) we
define a syntactic role hierarchy in the same fashion,
where genitive and ppum are below a role ppumgen.
Then, the lexical entry for berauben inherits from a
linking type l theme ppumgen.
In contrast to alternative realizations, alternations
as in examples (12) and (13) realize two different
thematic roles (pat and th) as the same syntactic role
(objd). By the lexical entry constraint in section 4.4,
there cannot be a single lexical entry for both alter-
nants of clear. We therefore model finite forms of
clear by two separate valency frames (we skip the
definitions of some of the linking types).
clears = l ag subj u
l pat objd u
l th ppof
(26)
clears = l ag subj u
l th objd u
l pat ppfrom
(27)
6 Conclusion
We proposed an architecture that abstractly cap-
tures reocurring patterns in the way in which dif-
ferent lexical items syntactically realize their se-
mantic arguments. We focused on interchange-
ability between prepositional phrases and nominal
phrases. We therefore defined a hierarchy of the-
matic roles, and a separate hierarchy of syntactic
functions, clearly separating between syntax and se-
mantics.
We enriched the framework of Topological De-
pendency Grammar (Duchier and Debusmann,
2001) (TDG) with a third level of representation,
thematic structure, and defined well formedness
conditions on the thematic structure and on the re-
lation between thematic structure (TH) and syntac-
tic dominance (ID) structure. This enabled us to
present a formal definition of the concept of multi-
level valency frames. We demonstrated how such
valency frames can be assembled using the lexical
inheritance model of TDG, in order to reduce lex-
ical redundancy. We also proposed a treatment of
auxiliary constructions using a notion of node col-
lapsing.
We applied our formalism to English dative shift
constructions, variation between NP and PP in Ger-
man, optional complements, semantically empty el-
ements, and thematic role alternations.
Our approach makes weaker theoretical predic-
tions about “alternations” as opposed to “alternative
realizations” than the approach in (Davis, 1998), but
is more powerful in the treatment of PPs. This is
partly due to the choice of an underlying depen-
dency formalism, because dependency structures
lead to concise and semantically flat parse trees.
Our approach is data-driven in the sense that a wider
range of syntactic patterns can be mapped to se-
mantic arguments. Thus, it lends itself easy to
techniques of automatically acquiring lexica from
syntactically and semantically annotated corpora,
which start developing at present ((Baker et al.,
1998), (Skut et al., 1998)).
It is possible to include a mechanism that makes
stronger predictions about alternations of the clear
type, or about passivization. Another phenomenon
we would like to tackle in this framework are raising
and control constructions. We think we can use a
concept similar to node collapsing for them.

References

Collin F. Baker, Charles J. Fillmore, and John B.
Lowe. 1998. The Berkeley FrameNet project.
In Christian Boitet and Pete Whitelock, editors,
36th ACL and 17th ICCL Proceedings, pages 86–
90, San Francisco, California. Morgan Kaufmann
Publishers.

Joan Bresnan. 2001. Lexical Functional Syntax.
Blackwell.

Cobuild. 2001. The bank of english.
http://titania.cobuild.collins.co.uk/
boe info.html.

Tony Davis. 1998. Linking as constraints in the hi-
erarchical lexicon. Chicago UP.

Ralph Debusmann. 2001. A declarative grammar
formalism for dependency grammar. Master’s
thesis, University of the Saarland.

David Dowty. 1991. Thematic proto-roles and ar-
gument selection. Language, 67(547–619).

Denys Duchier and Ralph Debusmann. 2001.
Topological dependency trees: A constraint-
based account of linear precedence. In ACL 2001
Proceedings, Toulouse, France.

Ulrich Engel and Helmut Schumacher. 1976.
Kleines Valenzlexikon deutscher Verben.
Forschungsbericht IDS Mannheim.

Klaus Fischer. 1997. German-English Verb Va-
lency. Narr.

Gerhard Helbig and Wolfgang Schenkel. 1975.
W¨orterbuch zur Valenz und Distribution
deutscher Verben. VEB Bibliographisches
Institut.

Gerhard Helbig. 1989. Deutsche Grammatik. VEB
Enzyklop¨adie.

Gerhard Helbig. 1995. Probleme der Valenz- und
Kasustheorie. Narr.

Ron Kaplan and Joan Bresnan. 1982. Lexical func-
tional grammar: A formal system for grammati-
cal representation. In The mental representation
of grammatical relations. MIT Press.

Geert-Jan Kruijff. 2001. A categorial-modal log-
ical architecture of informativity. Ph.D. thesis,
Charles University, Prague.

Beth Levin. 1993. English verb classes and alter-
nations. Chicago UP.

Igor Melcuk. 1988. Dependency syntax: Theory
and practice. Albany: State Univ of NY.
Carl Pollard and Ivan A. Sag. 1994. Head Driven
Phrase Structure Grammar. Chicago UP.

Petr Sgall, Eva Hajicova, and Jarmila Panenova.
1986. The Meaning of the Sentence in its Seman-
tic and Pragmatic Aspects. Reidel.

Wojciech Skut, Thorsten Brants, Brigitte Krenn,
and Hans Uszkoreit. 1998. A linguistically in-
terpreted corpus of German newspaper text. In
Proceedings of the ESSLLI Workshop on Recent
Advances in Corpus Annotation.

Kalevi Tarvainen. 1987. Cases in the framework of
dependency grammar. In Concepts of Case. Narr.
