4 Articles On Verb Meaning and Argument Structure

chapter 1
Ergativity and the object-oriented

representation of verb meaning*
Anton Benz
Centre for General Linguistics
Berlin/Germany
There is an interesting parallelism between the representation of one- place

and two-place operations in object-oriented programming and case marking
in ergative languages. The object-oriented approach has proven to be highly
successful in computational system design and analysis. One of its peculiarities
is to define operations within the class of the objects which are their main argu-
ments. More specifically, operations that correspond to intransitive verbs are
encoded in the class which corresponds to the subject NP, and operations that
correspond to transitive verbs are encoded in the class that corresponds to the
NP of the direct object. In ergative languages these NPs are marked by absolu-
tive case. We discuss a number of semantic phenomena related to ergativity
that show that conceptual representation of word meaning follows the object-
oriented paradigm.
1. Introduction
At the centre of our paper lies Keenans (1984) observation that the meaning of
intransitive verbs can be different for different subject arguments, and the mean-
ing of transitive verbs can be different for different direct objects. Keenan supports
his claim by the following examples (1984:201):
(15) a. John / the horse is still running
b. The car / the motor / my watch is still running
c. The tap / my nose is still running
d. My Fair Lady / the Braque exhibition is still running
* The research presented in this article was supported by the German Bundesministerium fr
Bildung und Forschung (BMBF), Grant Nr. 01UG0711.
66 Anton Benz
(16) a. John cut his arm / his foot

b. John cut his nails / his hair / the lawn
c. John cut his cake / the roast
d. John cut a path (through the field) / a tunnel (through the mountain)
e. John cuts his whiskey with water / his marijuana with tea
f. The company cut production quota / prices
The interpretation of is still running differs significantly from (15a) to (15d) but
is constant for the same type of subject. In (16a) to (16f), we find the same varia-
tion with respect to the direct objects of cut. In (16a), the body part is lacerated
but not cut through. In (16b), pieces of the cut entity are severed from the rest of
it. In (16c), the cut entity is divided into several pieces. In (16d), the cutting clears
a path through some material which would otherwise not be penetrable or only
with considerable effort. In (16e), the whiskey and marijuana are mixed with some
other substance. Finally, in (16f), the cutting means a reduction of production
quota and prices.
Keenan (1984) discusses these examples in connection with the phenomenon
of ergativity. He provides a list of properties that are shared by subjects of intran-
sitive verbs and direct objects of transitive verbs. He maintains that the type of
polysemy observed in (15) and (16) is bound to the subject of intransitive verbs,
and the direct object of transitive verbs.
There is no obvious logical reason why this variation of verb meaning should
depend universally on one argument role only, and why this role should be differ-
ent for intransitive and transitive verbs. Although there are exceptions to this rule,
we assume that it is basically correct. In this paper, we speculate that the explana-
tion can be found in the principles of semantic representation which are part of
the object-oriented paradigm in computer science.
A recurrent problem for which the object oriented approach proved to be very
successful is the problem of polymorphisms, i.e., the phenomenon that opera-
tions show different behavior with different arguments. For example, the execution
of the print() operation1 can be quite different mathematically for different
arguments. Printing a line involves different data and calculations than printing
a square or a circle. Polymorphism is an obvious analogue to polysemy in verb
semantics.
1. Sample expressions belonging to a programming language which does not need to be exactly
specified in the present context are presented in typewriter font. In the example above the
empty parentheses indicate that the print-method does not require any argument. In the object-
oriented language Java the drawing method used to render figures actually requires an object
from the class Graphics as an argument. That object may be conceived as a pencil used to
draw the figure at issue.
Chapter 1. Ergativity and object-orientation 67
Object orientation was partly a reaction to the problems that the ever growing
complexity of software imposed on the maintainability and re-usability of code,
see, e.g., Booch et al. (2007). These problems can be much reduced by a modular
software design. This modularity is achieved by packing together data and the
methods operating on them in one unit. These units are called objects. Each object
combines attributes, i.e., data, and the methods which use these data.
In domain modeling, objects represent concrete entities. Methods represent
the actions which can be performed by these entities. From another point of view,
objects are instantiations of classes which define their general properties. In the
following, we assume that classes correspond to the lexicon entries of common
nouns and methods to verb meanings. This approach to lexical semantics has the
interesting consequence that the meaning of verbs is defined within the lexical
entries of their arguments. Furthermore, it provides a cognitive basis for the mor-
phological case marking in ergative languages: the absolutive shows the addressee
in which class to look up the meaning of the verb. If it is a transitive verb, the
addressee has to look it up in the class representing the meaning of the direct
object; if the verb is intransitive, the addressee has to look it up in the class repre-
senting the meaning of the subject. In this paper, we want to show the potential
of the object-oriented approach by exploring semantic phenomena related to the
ergative/absolutive distinction.
The paper is structured as follows: we first introduce the elements of object-
oriented programming which play a role in our model of lexical semantics. We
then discuss the phenomenon of polysemy in the context of the object-oriented
lexicon. Finally, we show how this paradigm might provide a cognitive basis for
the ergative/absolutive distinction.
2. The object-oriented paradigm
What exactly the object-oriented paradigm makes up is a question which has no

generally agreed upon answers. Lists of characteristics of the paradigm can vary
considerably. The object-oriented paradigm resembles a bundle of related concepts
and methodologies of which some core concepts can be identified (Armstrong
2006). We concentrate here on aspects which we think are especially interesting to
lexical semantics. These are class and object, inheritance and polymorphism, and
encapsulation and interfaces. For our paper, we can concentrate on the handling
of polymorphisms (an operation may show different behavior depending on the
type of its arguments) through subtyping (inheritance of properties).
A class is often defined as a set of objects that have a common structure and
behavior (Booch et al. 2007). For example, we may have a class of triangles which
68 Anton Benz
are defined by three distinct points in the Euclidean plane, and which may be
rotated or printed. In object-oriented languages, class definitions always divide
into three parts: the class name, the class attributes, and the class operations.
We can identify a class with an abstract object which contains specifications of
name, attributes, and operations. In case of the Triangle class, let us assume
for simplicity that its elements share one attribute, position, which is an array
of three coordinates, which are again pairs of real numbers, and two operations.
Graphically, we can represent the triangle class as in Figure1. The top cell of the
diagram contains the name of the class, the centre cell the list of attributes, and
the bottom cell the list of methods or operations.
Triangle
position : array of three coordinates
readPosition()
print
rotate
Figure 1. A Triangle class
This named, abstract collection of attributes and operations defines a specific

class from which instantiations of concrete triangles can be derived. These con-
crete instantiations are called objects. An object is an entity which has a defined
state, behavior and an identity (Booch et al. 2007). This means that an object
t of type Triangle has to have a position, and the operations readPosi-
tion(), print(), and rotate() must be defined for it. Using a common
notation, we can write this as follows: If t is of type Triangle (t : Triangle),
then t.position() has a value, and t.readPosition(), t.print(), and
t.rotate() are defined.2 The obvious linguistic semantic analogues to class and
object are predicate and domain entity.
The best known feature of object-orientation is probably inheritance. We are
interested in this feature in connection with polymorphism. To start with, let us
consider how a procedural program would handle the task of printing different
geometrical objects, say triangles, circles, and squares. The data base entries for
each of these figures may be very different. A triangle is defined by three points,
but the circle may be defined by its centre and its radius, and the square by the
coordinates of its lower left corner and the length of its sides. Hence, the actual
implementation of the print procedure is different for each object, and it may be
2. We assume here that rotate() rotates a figure by a certain fixed angle. Alternatively, we
could assume a number argument degree of the number type double (double precision
number): rotate(float degree).
necessary to define a separate drawFigure() procedure for each geometrical

figure. Hence, the print procedure may look as follows:
printImage (object) begin
if object isa circle do drawCircle(object),
else if object isa square do drawSquare(object),
else if object isa triangle do drawTriangle(object),
...
end;
This phenomenon that one operation shows a different behavior depending on its
arguments is called polymorphism.
What if it later turns out that other geometrical figures, say arbitrary quadri-
laterals, have to be printed? In this case, first, new data types for quadrilaterals like
trapezoids and parallelograms have to be defined. Second, the print procedure has
to be extended by new if-else clauses for the drawTrapezoid(), drawParal-
lelogram() etc. procedures. This means that the old code has to be changed
at two different places, new data types have to be added and old procedures have
to be rewritten. In an object-oriented approach, the print command is part of the
operations included in the class definition of the geometrical figures. For example,
the printTriangle() from our procedure would be the implementation of
the print procedure in the Triangle class which we defined before. Hence, if a
programmer has to extend his system by trapezoids and parallelograms, then he
can just add new class definitions as in Figure2.
Triangle Parallelogram
position : array of three coordinates position : array of four coordinates
readPosition() readPosition()
print print
rotate rotate
Figure 2. Two classes of the same type
This means that the code for the old print commands remains untouched when
adding new data types to the system. The advantage of the object-oriented
approach lies in its easy extendibility. Seen from the outside, both programs, the
procedural and the object-oriented, do the same things, they print geometrical fig-
ures, but the object-oriented approach leads to a much greater modularity of rep-
resentation. This becomes important when we apply the object-oriented approach
to the lexicon and natural language semantics. Similarly as the choice between
an object-oriented and a procedural representation makes no difference for the
actual behavior of operations, the choice of an object-oriented representation of
the lexicon has no consequences in terms of truth conditions. One representation
70 Anton Benz
is translatable into the other. But it has consequences for language acquisition and
the handling of polysemy.
So far, we have seen that the object-oriented approach reduces changes to the
old code when extending it to a minimum. Furthermore, all changes are located
in one single object, the definition of the class which represents the new data
type. But still, the programmer has to remember which operations and attributes
have to be defined for the new classes. Probably, he not only wants to print and
rotate geometrical figures but also to shift them, stretch them, or fill them with
color. In a procedural program, the modification has to be repeated for every
procedure that may take an object of the new data type as argument. Hence, the
programmer has to remember which procedures may operate on them. There is
no code-internal means for memorizing this. In an object-oriented approach, this
memorizing is done by the inheritance relation. If operations are applicable to all
geometrical figures, then they are implemented by all objects representing them.
Hence, they belong to the class which defines geometrical figures. The classes of
triangles, squares, and circles are only subclasses of this larger class. Graphically,
this is notated as in Figure3. There, the name of the classes of geometrical fig-
ures and quadrilaterals are set in italics, which says that these classes are abstract
classes. This means that they cannot immediately be instantiated by objects but
only their subclasses.
Geometrical figure
readPosition()
print
rotate
...
Triangle Quadrilaterals
position : three coordinates position : four coordinates
Square Parallelogram
Figure 3. Sub-classes and super-classes in inheritance relation
Inheritance means that all subclasses share the attributes and operations of their
super-classes. In our example, instances of triangles, squares, parallelograms all
have to define the print, rotate and other operations, and all squares and par-
allelograms have to provide four coordinates for their position. If the programmer
now introduces a new class of geometrical figures, e.g., the class of trapezoids, he
can introduce them as a new subclass to the super-class of quadrilaterals. This
will automatically tell him which attributes and operations he has to implement
for trapezoids. Hence, from the perspective of maintainability and extendibility
of programs, the object-oriented approach helps to minimize changes, to bundle
them together in one place, and to memorize which attributes and operations have
to be defined.
The obvious analogues to the inheritance relation between super-class and
sub-class are the semantic relations of hypernymy and hyponymy. The class hier-
archy defined by inheritance is closely related to the notion of an ontology which is
defined by an isa relation. In an ontology, the typical definition of a class contains
attributes and constraints which must be satisfied by each instantiation of the class.
For example, triangles can be defined as triples of points in the Euclidean plane
such that no straight line which connects two of the points also contains the third
point. An equilateral triangle is a triangle for which the distance between all points
is equal. Hence, if we predicate of an entity t that it is of type equilateral triangle,
then it must satisfy the condition for triangle and the condition for equilaterality.
The inheritance relation in the object oriented class hierarchy is also an isa
relation but with an important difference with respect to operations. If we say
that an object t is an instantiation of the triangle class as defined before, then it
follows that there must be an array a of Euclidean points such that a contains the
coordinates of t. Logically, this means that: t triangle(t) implies a[array-of-

EucP(a)coordinates-of(t,
a
)]. The same does not hold for operations. If op is an
operation defined in the class or super-class of triangle, e.g., the print operation,
then the existence of an instantiation does not entail the existence of an execution
of op. This is important as operations are the natural analogues to verb meanings.
This also marks an important difference to previous applications of inheritance to
lexical semantics, as we will see soon.
What we have considered so far is single inheritance only. In addition to single
inheritance, object-oriented programming also knows multiple inheritance. This
is typically used for modeling the behavior of a class in different contexts. Multiple
inheritance found significant and wide spread applications in lexical semantics.3
Most important in the context of systematic polysemy is the application of mul-
tiple inheritance for representing context dependent meaning differences in the
nominal domain. An example is the difference between the meanings of newspa-
per in John got angry at the newspaper and spilled coffee over it. The first occur-
rence refers to the content of the newspaper, the second to the physical object.
3. See the collection of Briscoe et al. (1993), and Pustejovsky (1995).

72 Anton Benz
Information PhysObj
Newspaper
Figure 4. The Newspaper class and multiple inheritance
Pustejovsky (1995) introduced dotted types in order to handle such example. If

the dotted type InfoPhysObj is assigned to newspaper, then this means that in
some context newspaper can fill an argument position which requires an argument
of type Info, in another context it can fill an argument position which requires
an argument of type PhysObj, and in yet another context it can fill a position in
which both aspects are interpreted. An example for the latter case is the sentence
John wrote the letter where letter is also of the type InfoPhysObj, and where
Johns writing produced both the letter as a physical object and its content. Dotted
types can be interpreted as a type theoretic equivalent to multiple-inheritance.
Our explanation of the phenomenon observed in the introductory examples
(15) and (16) will only make use of single inheritance. In Pustejovskys generative
lexicon, they are cases for co-composition, which is a generative mechanism which
is in principle independent of dotted types. The meaning differentiations in (15)
and (16) are not an effect of dotted types. Even in a sentence as John spilled coffee
at the newspaper both aspects of the newspaper, physical object and information,
are present. It is only the verb which selects one of them. This means that the exis-
tential quantifier x newspaper(x) in the logical form of the sentence allows us
to infer the existence of a physical object and the information contained in it, but
the predicate spilled-at(coffee,y) takes as argument only the physical substance y
of x. In (15) and (16), however, the different meanings are not different co-existing
aspects of one entity or one event which are selected by the verb but exists inde-
pendently of each other.
We next have a closer look at the dependency between meaning variations and
argument positions of verbs.
3. Polysemy and argument roles
Polymorphism refers to the phenomenon that operations show different behavior

with different arguments. Polysemy means the phenomenon that one word has
several meanings. As mentioned in the introduction, Keenan (1984) observes that
there is a universal regularity to be found in the languages of the world according
to which the meaning of intransitive verbs depends on the subject argument, and
the meaning of transitive verbs depends on the direct object. Examples for the first
rule, we have seen in (15), and for the latter rule in (16). The following German/
English examples confirm this observation:
(17) a. Hans schlgt Peter
John beats Peter
b. Hans schlgt den Esel
John beats the donkey
c. Hans schlgt die Trommel
John beats the drum
d. Hans schlgt die Schlagsahne
John beats the cream
In (17a), a situation in which Hans uses his hands for beating Peter is expected,
whereas in (17b) we expect him to use an instrument as, e.g., a stick. In (17c) and
(17d), we also expect him to use some instrument but of different type. In (17a)
and (17b), it is Hans goal to hurt the beaten; in (17c) his goal is to produce rhyth-
mic sounds, and in (17d) to make the cream fluffy. The following examples show
that, if the arguments are of the same type, the meaning does not change:
(18) a. Hans schlgt Peter / sein Kind / den Schaner / die Gromutter
John beats Peter / his child / the conductor / grandmother
b. Hans schlgt den Esel / das Pferd / das Kamel / die Kuh
John beats the donkey / the horse / the camel / the cow
c. Hans schlgt die Trommel / die Pauke / das Becken
John beats the drum / the timpani / the cymbal
d. Hans schlgt die Schlagsahne / das Eiwei / den Honig
John beats the cream / the egg clear / the honey
Keenan (1984) notes that subjects of transitive verbs have no significant influence
on verb meaning. He provides the following examples:
(19) a. John / the machine / the company / the storm cut the lawn
b. John / the machine / the company / the storm cut the cake
c. John / the machine / the company / the storm cut a path (through the field)
d. John / the machine / the company / the storm cut production quota / prices
This is also confirmed by the following German/English examples:
(20) a. Hans / der Mhdrescher / der Ae schlgt den Mann

John / the combine harvester / the monkey beats the man
b. Hans / der Mhdrescher / der Ae schlgt den Esel
John / the combine harvester / the monkey beats the donkey
74 Anton Benz
(20) c. Hans / der Mhdrescher / der Ae schlgt die Trommel

John / the combine harvester / the monkey beats the drum
d. Hans / der Mhdrescher / der Ae schlgt die Schlagsahne
John / the combine harvester / the monkey beats the cream
There may be conflicts with selectional restrictions; e.g., a combine harvester has
no arms to beat a man, and it is also not possible to use it for beating cream. This
makes it difficult to interpret the respective sentences, but these violations of selec-
tional restrictions do not create new meanings.
If we assume that the lexicon is organized along object-oriented lines, it fol-
lows immediately that the meaning of verbs can only depend on one argument,
namely the argument in the class of which they are defined. When reading a verb,
the addressee therefore must know by which argument it is defined. We call the
position in which the defining argument has to be found the defining core role of
the verb.
As in a standard lexicon, we can assume that each verb has a lexical entry
on its own. But this entry only needs to provide the information necessary for
identifying the defining core role. This entails that the specification of optional
arguments is not necessary. If the verb has one obligatory argument only, then this
argument must show the defining core role. If it has two or more, then there is an
option. In line with Keenans Observation, we can hypothesize that the class of the
affected object defines the meaning of the verb. For transitive verbs, the affected
object typically is the direct object, whereas the subject is typically filled by the
entity which initiates the action which affects the direct object. This leads to the
following working hypothesis about verbs with an agentive argument:
1. Transitive verbs (cut, hit, see, move to, carry) correspond to procedures where
the initiating role is different from the defining core role.
2. Intransitive verbs (grow, turn, walk, sleep, look around) correspond to proce-
dures where the initiating role is the same as the defining core role.
The initiating and the defining core role do not need to be disjoint. If a process
only manipulates or reads the data that belong to the initiating object, then both
roles, the initiating and the defining, are filled by the same object. This is the case
for intransitive verbs, but also for reflexive transitive sentences:
(21) a. The woman shaves her husband
b. The husband shaves himself
c. The husband shaves
In all three cases, the default interpretation of the shaving event is the same: the
husband gets his facial hair removed by use of a razor. This meaning depends on
the argument which received the shaving, as can be seen from the sentence The
woman shaved herself in which shave has a quite different interpretation.
We use Dixons (1994) terminology, and distinguish between the argument
roles S for subjects of intransitive verbs, A for subjects of transitive verbs, and O
for direct objects of transitive verbs. Keenan notices that arguments in S and O
relation are commonly patients. This means that they are affected by the event
denoted by the sentence as in the following examples:
(22) a. The car exploded / broke down
b. The milk evaporated / spilled
c. Freds argument collapsed / fell apart
d. John blew up the car
e. John spilled the milk
f. John destroyed Freds argument
In contrast, arguments in A relation rarely are patients. Instead, they are typically
agents, instruments, or experiencers.4
Although there is a strong statistical bias, this regularity is not a hard universal
constraint as the following examples show:
(23) a. John jumped
b. John bores
c. The bridge crosses the river
d. The police surrounded the house
In (23a) the subject is an agent, and in (23b) something we might call a cause; in
(23c) and (23d) the direct object is assigned locative role.
Our hypothesis about the existence of a unique defining core role and its
connection to S and O relations carries no implications about the thematic roles
assigned to these relations. In so far, it is in full agreement with the observation
that there are only statistical preferences for the kind of thematic role but no hard
universal constraints. However, the defining core role is not independent of the
thematic properties of the arguments as the following variations of Examples (23c)
and (23d) show (Keenan 1984: fn. 4):
4. Keenan also discusses the case of direct objects which are assigned the role of themes with
verbs of motion, i.e., they are objects which move from one location to another location as in
Bill threw the log into the pool.
76 Anton Benz
(24) a. The soldiers cross the river

The bridge crosses the river
b. The police surrounded the house
The moat surrounded the house
The meaning of cross and surround varies with the subjects in these examples. In
contrast to previous examples, the direct object provides a location, i.e., it does
not denote an object which is affected by an event or state. This seems to indicate
that meaning variation depends on the type of thematic role which is assigned to
an argument rather than its syntactic role.
Keenan acknowledges that meaning of verbs can vary with general prop-
erties such as animacy and agenthood of their A argument, but he argues that
the differences in examples like (24) are of a more systematic nature than the
ad hoc meaning differences generated by different direct objects. However, if we
assume that the defining core role is always the direct object, then these examples
pose a problem which cannot be remedied so easily. Let us discuss some possible
explanations.
A first solution is to assume that a hidden ambiguity of the direct object is
responsible for the deviations exhibited by the examples in (24). The event type in
(24a) and (24b) changes from accomplishment to state, whereas it always remains
the same in (15) and (16). If meaning is inherited and modified from a super-
class, then the latter regularity is to be expected, and we may speculate that, e.g.,
the river in (24a) figures as an obstacle in the first sentence, and as a landmark
in the second. Hence, the river shows here a similar behavior as dotted objects
do, and we could explain the examples in (24) by assuming that river is of type
obstaclelandmark, and cross is defined one time in the super-class obstacle,
and the other time in the super-class landmark. Against this analysis stand exam-
ples like The plane / the bird / the balloon crossed the river in which the river does
not figure as an obstacle, the event type is still that of accomplishment, and the
subject moves over the river along a path through air. The meaning differentia-
tions clearly depend on the semantic properties of the subject, and not on a hidden
polysemy of the direct object.
Another solution is to assume that the defining core role is associated with a
thematic role and not with a syntactic role. The defining core role cannot simply
be identified with the patient or theme as the subject of intransitive verbs may take
on all kinds of thematic roles. But as intransitive verbs have only one argument,
they may be left out of consideration. However, there is an argument connected
to selectional restrictions which tells against this solution.
Let us reconsider the examples in (18). What the direct objects have in com-
mon is the fact that they receive a blow from some object, but there is no common
super-class for which the verb schlagen (beat) is defined. For example in (18c),
schlagen can only be applied to drum like music instruments. If they are not drum
like, selectional restrictions are violated as in Hans schlgt die Flte (John beats
the flute). Therefore, the four types of direct objects in (18) cannot be subsumed,
e.g., under the general super-class of physical objects. Hence, the defining core role
cannot be identified with a general super-class. It can be argued now that thematic
roles do not define classes in the object-oriented sense, but have to be identified
with a pair consisting of a class together with an abstract active operation. For
example, an instance of the patient role may be defined as an instance of the physi-
cal object class which is affected by some force, a prototypical agent an animated
being which performs some action, an experiencer an animated being which expe-
riences some sensation, a result some entity which is produced by an action, etc.
In this sense, all direct objects in (18) can be said to be patients. But the subjects
in (24) fill different thematic roles, and even for schlagen we can find examples as
Die Stadt schlug Mnzen mit dem Portrait des Kaisers (The city struck (beat) coins
showing the portrait of the emperor) in which the direct object is a result and not
a patient. A well known example with the same alternation of the thematic role
of its direct object is bake a cake vs. bake a potato (Pustejovsky 1995). It would
be possible to define roles even more abstractly, but then their function becomes
indistinguishable from signaling the syntactic role of the argument. Hence, we
conclude that the defining core role is always identified with a syntactic role, that
this syntactic role is the role of the direct object for transitive verbs but can be the
subject role in exceptional cases. The exceptional cases arise with direct objects
which are not affected by the event or state denoted by the sentence. Hence, the
defining core role cannot be identified with a specific thematic role but it can be
excluded from some thematic roles.
4. Ergativity
It is a tempting idea that the object-oriented principles for system design are
connected to the ergative/absolutive case marking systems. Speakers want to be
understood, hearers to understand. If the speaker says that A man killed a lion,
he does not want that the addressee understands this as A lion killed a man. This
means that the speaker has to make clear which noun phrase fills which argu-
ment position of the verb. Languages offer a variety of solutions to this problem,
morphological case marking is one of them. We already introduced Dixons (1994)
notation for the argument roles of subjects of intransitive verbs S, for subjects of
transitive verbs A, and for direct objects of transitive verbs O. Every language has
to make clear which NP in a transitive sentence is in A relation and which is in O
relation. According to Dixon, there are three principal possibilities to do this: case
78 Anton Benz
marking a NP by affixation, by means of adpositions and particle, and by cross-

referencing from the verb.
Among the successful case marking systems, there are the following two solu-
tions: Either mark the O-role, and let the other roles be unmarked, or mark the
A-role, and let S- and O-role be unmarked. If a language employs the former solu-
tion, then it has nominative/accusative case marking, if the latter, it has ergative/
absolutive case marking, with accusative and ergative being the marked forms.
From a purely functional perspective, this may be all that there is to say about the
difference between ergative/absolutive and nominative/accusative case marking
systems. Both are efficient in the sense that they uniquely identify the NPs in S,
A, and O relation:
Nominative/accusative case marking: NPs in S and A role are treated the same.
NPs in O role are marked differently from those in S, A position. The case of
S and A is called nominative, the case of O is called accusative.
Ergative/absolutive case marking: NPs in S and O role are treated the same. NPs
in A role are marked differently from those in S, O position. The case of S and
O is called absolutive, the case of A is called ergative.
The following examples are from German which shows a nominative/accusative
distinction:
(25) a. Der Mann lchelt
the.nom man smiles-3rd.sing.pres
The man smiles
b. Der Mann fttert den Hund
the.nom man feed-3rd.sing.pres the.acc dog
The man feeds his dog
The determiner of the NP in S and A relation shows nominative marking (der), the
determiner of the NP in O relation shows accusative marking (den). The subject is
cross-referenced from the verb (-t, 3rd per. sing.).
An example of a language with ergative/absolutive distinction is Dyrbal, spo-
ken in Queensland, Australia. The NP in S and O relation shows absolutive case,
the NP in A relation shows ergative marking (-gu) (Dixon 1994:10):
(26) a. uma banaga-nyu

father.abs return-nonfut
father returned
b. yabu banaga-nyu
mother.abs return-nonfut
mother returned
c. uma yabu-gu bura-n

father.abs mother-erg see-nonfut
mother saw father
d. yabu uma-gu bura-n
mother.abs father-erg see-nonfut
father saw mother
In general, nominative and absolutive are the unmarked cases, accusative and
ergative the marked ones. Efficiency of linguistic code can be increased by, e.g.,
taking features like animacy or gender into account. In German for example, only
a direct object with male grammatical gender is morphologically marked as accu-
sative (den). If a direct objects grammatical gender is female or neuter, its accusa-
tive form is identical to the nominative form. This can be explained by the fact
that female or neuter direct objects are very frequent, whereas male objects are
infrequent. Hence, the strife for efficient encoding of information should favor
case marking that takes into account these different frequencies.5
The examples we presented all show morphological ergativity. The term erga-
tivity is also used for syntactic regularities connected to the S, A, and O relation.6
As we are only interested in semantic patterns which are related to ergativity, i.e.,
semantic patterns which treat NPs in S and O relation in the same way, we leave
syntactic ergativity out of consideration.
The picture is made more complicated by the fact that in one and the same
languages we may find both nominative and ergative patterns. Many ergative lan-
guages show a split between an ergative and an accusative pattern according to
some grammatical category. There are the following possibilities (Dixon 1994:
Chapter 4):
1. A split conditioned by the semantic nature of the verb. There may be a split
in the class of intransitive verbs depending on, e.g., whether the described
eventuality is an activity or done voluntarily.
2. A split conditioned by the semantic nature of the NPs.
3. A split conditioned by tense / aspect / mood.
4. A split between main and subordinated clauses.
The split conditioned by tense / aspect / mood distinguishes often between past/
perfect and present/future. According to Trask (1979), the languages with a split
in the NPs have a different historic origin than languages with a split according
5. This is the basis for evolutionary and diachronic learning models for differential object
marking as proposed by Jger (2004, 2007).
6. See for example Dixon (1994:13).
80 Anton Benz
to tense / aspect / mood. In languages of the latter type, the effect of an ergative
seems to be similar to a perfective emphasizing the result of an event. According
to Trask (1979), this split is restricted to languages which do not have an auxiliary
for have.
We have already noticed that explicit morphological marking of accusative
case may be conditioned by the grammatical gender of an NP. Since Silverstein
(1976) it is well known that the distribution of ergative marking in a language
may be conditioned by an animacy scale. In the following scale, the NPs to the left
are more likely to follow an accusative pattern, the NPs to the right more likely to
follow an ergative pattern. Where exactly the split occurs is language dependent.
1st pers. 2nd pers. Demonstratives Proper Common nouns

pronouns pronouns 3rd pers. nouns human animate inanimate
pronouns
Trask (1979) claims that this type of split only occurs in languages in which the
ergative originated from a reanalysis of a passive construction. Another obvious
theoretical explanation of the origin of this type of split is to assume a nomina-
tive/accusative system at the beginning for which explicit accusative case marking
gets restricted to animate objects. This leads to an ambiguity if inanimate NPs
occur in both A and S relation. This situation can be disambiguated by marking
the inanimate subject, which leads to an ergative language with a split along the
animacy scale.
There is an obvious motivation for the ergative/absolutive case distinction if
we assume that the mental lexicon is organized along object-oriented lines. In this
case, the addressee not only has to know which NP fills which argument position,
but also which noun phrase defines the meaning of the verb. In an ergative/abso-
lutive case marking system, this noun phrase is always the absolutive (un)marked
NP. The existence of splits in the case marking system of a language does not pose
any problem to this explanation as the objective of indicating the defining core
role may only be one among other competing objectives.
This short overview of grammatical phenomena related to ergativity can only
mention the most important features. All of them can be explained along purely
functional lines, i.e., by the fact that languages have to solve the task of assign-
ing NPs to argument roles and do it in an efficient way. Hence, our hypothesis
that ergative/absolutive case marking is related to object-oriented principles of
organizing the lexicon cannot be directly tested. What is of interest to us are,
therefore, not the morpho-syntactic patterns associated with ergativity but the
semantic regularities observed by Keenan (1984). We turn to these regularities in
the next section.
5. Correlates of the ergative/absolutive distinction
Keenan (1984) lists a number of grammatical phenomena that link the S role
together with the O role. We are interested whether these regularities support the
theses that there exist defining core roles, and that these roles are the S and the O
role. Keenan divides the phenomena into three groups:
1. Bondedness to the verb: under this heading, Keenan lists existence depen-
dencies, semantic polysemy, selectional restrictions and verbal classifiers, and
noun incorporation.
2. Thematic role properties: Keenan notes that NPs in S and O relation are often
assigned the thematic roles of patient or theme.
3. Control phenomena: under this heading he addresses expressions such as
adjectives and infinitival phrases predicating something about the arguments
of the verb. For example, he notes that adjectives within a predicate are nor-
mally denoting a property of the object in S or O relation.
We have already discussed the thematic role properties and found that the defining
core role cannot be identified with a thematic role but that some thematic roles are
excluded from being the defining core role. We are now going to discuss the other
regularities one by one with relation to our hypothesis that each verb has exactly
one defining core role. We are especially interested in putative counter-examples.
5.1 Bondedness to the verb
There is a large class of verbs which refer to events which involve the coming into
existence of an object. These verbs may be intransitive (A crowd gathered around
John) as well as transitive:
(27) a. A student lit a fire in the basement
b. He committed a crime / made a mistake
As Keenan notes, the object coming into existence is always the object appearing
in O relation.7 Given our assumption that for each verb there must be a defining
core role, it is not surprising that this core role is the role of the object created
by an activity. What is involved in creating an entity much more depends on the
type of entity created than on the creator. For our argument it is important that
we again find that the meaning differs with different direct objects but not with
different subjects:
7. We concentrate on transitive verbs in this section as they provide the crucial data.
82 Anton Benz
(28) a. John painted a picture / the wall

b. John / God / the machine painted a picture
c. John / the machine painted the wall
Whether an object comes into existence depends on the type of direct object: paint-
ing a picture creates a picture but not painting a wall.
There are, however, examples in which the verb meaning appears to depend
on the subject:
(29) a. John painted a wall
b. Rembrandt painted a wall
Sentence (29b) will normally be understood as meaning that Rembrandt painted a

picture of a wall. This putative counter-example can be explained if we assume that
the literal content of the sentence is pragmatically enriched to Rembrandt painted
a picture of a wall. We only have to assume that the specific interpretation of the
verb is calculated after enrichment.8 In this case, the meaning difference is caused
by a difference in the types of direct objects.
Under the heading of multiple senses, Keenan discusses the meaning variation
which we find in our core examples (15) and (16), and which we called Keenans
Observation.
In (24), we have already seen examples which seem to contradict his observa-
tion. There are also other putative counter-examples which seem to show that the
subject in A role changes the meaning of a verb (Benz 2006):
(30) a. The monk killed the patient
b. The doctor killed the patient
c. The psychologist killed the patient
d. The sadist killed the patient
In all examples, we find some slight difference in interpretation. In the second,

third and fourth example, we would expect that the killing is indirect and in some
way connected to the profession of the subject and the role of the victim. In The
doctor/the psychologist/the pharmacist killed the butterfly a much more direct kill-
ing could be expected. In (30d) the expected killing may again be more direct but
in addition also extended and torturousin contrast to (30a). These differences
do not affect the semantics of the event type; they are rather related to different
modes of the killing, e.g., they may carry special expectations about the instru-
ment (pharmacist), and are different connotations rather than different meanings
of the sentence.
8. Pragmatic enrichment here refers to pragmatic strengthening of meanings as postulated by

relevance theorists, specifically to free enrichment; see, e.g., Carston (2004).
Another property which is common to S and O relation but not to A relation

is the existence of rather strict selectional restrictions. Keenan mentions, e.g., the
case of the ambitransitive verb spill: The objects which can spill or which can be
spilled by someone seem to be restricted to liquids and finely granular objects like
coffee grounds. In contrast, he notes that he knows of no verb which would impose
a similar restriction to the A role. Furthermore, things that shatter or can be shat-
tered must have a special physical property. Keenan claims that there seem to be
no transitive verbs that impose restrictions of the same kind on their subjects (A).
We have already discussed the Schlagen-examples in (18). We have seen that
there is no common super-class which would be identical to the set of all possible
direct objects. We found, for example, a restriction to drum-like music instru-
ments (18c), or to human beings in (18a). The different types of direct objects
in (18) can, in particular, not be subsumed, e.g., under the general super-class of
physical objects, as there are physical objects which cannot appear in direct object
position of schlagen. Hence, we find also in this example an instance of very fine-
grained selectional restrictions in the defining core role.
The assumption that the meaning of verbs is defined in the class of its defining
core role poses no restrictions on the fine-grainedness of the selectional restric-
tions for this role. It is not necessary that the defining classes combine to a single
super-class, or that the defining classes are high up in the class hierarchy. The
further down in the class hierarchy they are, the more fine-grained the restriction
on the argument type will be. In contrast, the restrictions for the other arguments
must be provided together with the definition of the verbs meaning. Once the
defining core role is filled, the general selectional restriction and the theta role of
the remaining argument must be defined. Hence, variation of thematic roles, if it
exists, should always be determined by the class of the direct object.
A similar pattern as with selectional restrictions can be found in the restric-
tions imposed by object classifiers. These classifiers are affixes to the verb root and
require that certain arguments satisfy specific semantic constraints. In Navajo, e.g.,
there exists a class of six affixes; if they are attached to a verb, then they indicate
that an argument is a round solid object, a long slender object, a wool-like mass, a
mud-like mass, etc.9 The available data seem to indicate that these restrictions only
apply to NPs in S and O relation but not to those in A relation. Both phenomena
lend strong support to the assumption that verb meaning is defined in the class
denoted by the NP in S or O relation.
The final regularity which Keenan discusses under the heading of bonded-
ness to the verb is noun incorporation. Noun incorporation is a process by which
a noun is incorporated in a verbal structure as, e.g., in German staubsaugen
9. See Keenan (1984: Section A.3) for references.

84 Anton Benz
(dust-sucking, i.e., vacuum cleaning). Keenan uses a survey by Mardirussian (1975)

which has only examples for the incorporation of nouns from S or from O relation
but no nouns that come from A position. This phenomenon is also in line with
the assumption that verb meaning is defined in the arguments in S and O relation.
5.2 Control phenomena
As mentioned before, Keenan considers here expressions such as adjectives and

infinitival phrases which occur as arguments of a verb and predicate some property
of another NP argument. Examples containing adjectives and one subject NP are:
(31) a. John looks smart / angry / lazy
b. The meeting sounds noisy / calm
c. The milk turned sour / green
d. The door flew open
The following examples also contain a direct object:
(32) a. John saw Bill angry

b. John considers Bill smart
c. John packed the meat raw
d. The sun turned the milk sour
e. John drove Bill insane
Keenan notes that the adjective either predicate about arguments in S or O rela-
tion but not about those in A relation. This is in line with our hypothesis. It is to
be expected that if the direct object defines the specific meaning of the verb, it
also defines the specific meaning of the adjective which occurs in an argument
position. There are also examples in which the meaning varies with the argument
in S and O position:
(33) a. John turned sour
b. The milk turned sour
c. Supposedly safe investments turned sour
d. Relations turn sour
e. John considers Bill smart
f. John considers the plan smart
However, it is not obvious whether these examples really support our claim. The
examples in (33c) and (33d) are metaphorical, and the smart in (33e) may simply
refer to a special aspect of Bill namely his ideas. The meaning of sour in (33c) is
not lexicalized, hence, this example should better be explained by some productive
pragmatic mechanism.
Keenans examples of infinitive phrases exhibit the same pattern as the exam-
ples with adjectives. The following may serve as illustration:
(34) a. John needs to arrive on time
b. John started to study law
c. John asked to leave the room
d. John asked Bill to leave early
e. John obliged Bill to review the proposal
f. The incident caused Bill to lock his door at night
From these examples Keenan infers that the infinitive phrase predicates something
about either the argument in S or O relation. Again, these examples are not con-
vincing. For example, (34c) does not fit into this pattern, and the examples with
transitive verbs report speech acts. If a commissive speech act is reported then the
infinitive phrase predicates a future act of the A argument, as in John promised Bill
to leave early. It is not a marginal exception, as Keenan indicates, but well in line
with the predictions of speech act theory.
We may conclude that the examples under the heading of bondedness to the
verb are the most interesting to our hypothesis, and the only ones which really
allow it to be tested against the data. The most problematic examples are in (24),
repeated as (35):
(35) a. The soldiers cross the river
The bridge crosses the river
b. The police surrounded the house
The moat surrounded the house
All other examples are in accordance with the assumption that the defining core
role is either the S or the O role but never the A role.
6. Conclusion
We started out with Keenans Observation that intransitive verbs can receive differ-
ent interpretations for different subjects, and that transitive verbs can receive dif-
ferent interpretations for different objects but not for different subjects. The main
objective of this paper was to point out an interesting motivation for this regularity
in the representation of one-place and two-place operations in object-oriented
programming. Operations that correspond to intransitive verbs are encoded in
the class which corresponds to the subject NP, and operations that correspond to
transitive verbs are encoded in the class that corresponds to the NP of the direct
object. This lead us to the hypothesis that ergative case marking is conceptually
86 Anton Benz
connected with the object-oriented structure of the lexicon. We discussed a num-

ber of semantic phenomena related to ergativity following Keenan (1984) which
show that the semantic meaning of the verb is determined by subjects of intransi-
tive, or by objects of transitive verbs.
We can summarize the discussion by the following hypotheses:
1. For each verb there exists exactly one defining core role. This means:
a. If we fix all other NPs, then replacing the NP in the defining core role may
lead to a different interpretation of the verb.
b. If we fix the NP in the defining core role, then replacing other NP arguments
will not change the interpretation of the verb. (Keenans Observation).
2. The selectional restrictions on the arguments of the defining core role can be
arbitrarily fine-grained. Moreover,
a. Semantic restrictions on the core role can be identified with the set of
classes in which the verb is defined.
b. Possible selectional restrictions on the other arguments are determined
by the class of the argument in the defining core role.
3. A verb cannot be interpreted if the object in defining core role is not known.
More specifically:
a. If the NP in the non-core role is omitted, then an interpretation is still
possible.
b. If the NP in the defining core role is omitted, then the argument type must
be inferable from context.
Bibliography
Armstrong, Deborah J. (2006): The quarks of object-oriented development. Communications

of the ACM 49(2): 123128.
Benz, Anton (2006): Partial blocking and associative learning. Linguistics and Philosophy 29:
587615.
Booch, Grady, Maksimchuk, Robert A., Engle, Michael W., Young, Bobbi, Conallen, Jim &
Houston, Kelli A. (2007): Object-Oriented Analysis and Design with Applications. Amster-
dam: Addison-Wesley Longman.
Briscoe, Ted, de Paiva, Valeria & Copestake, Ann (eds.) (1993): Inheritance, Defaults, and the
Lexicon. Cambridge: Cambrige University Press.
Carston, Robyn (2004): Relevance theory and the saying/implicating distinction. In: Lawrence
Horn & Gregory Warden (eds.) The Handbook of Pragmatics, Oxford: Blackwell Publish-
ing. 632656.
Dixon, Robert Malcolm Ward (1994): Ergativity, volume 69 of Cambridge Studies in Linguistics.
Cambridge: Cambridge University Press.
Jger, Gerhard (2004): Learning constraint subhierarchies: The bidirectional gradual learning
algorithm. In: Reinhard Blutner & Henk Zeevat (eds.) Optimality Theory and Pragmatics,
Basingstoke: Palgrave Macmillan. 251287.
Jger, Gerhard (2007): Evolutionary game theory and typology. A case study. Language 83:
74109.
Keenan, Edward J. (1984): Semantic correlates of the ergative/absolutive distinction. Linguistics
22: 197223.
Mardirussian, Galust (1975): Noun incorporation in universal grammar. Chicago Linguistic
Society Bulletin 11: 383389.
Pustejovsky, James (1995): The Generative Lexicon. Cambridge, MA: The MIT Press. Silverstein,
Michael (1976): Hierarchy of features and ergativity. In: Robert Malcolm Ward Dixon
(ed.) Grammatical Categories in Australian Languages, Camberra: Australian Institute of
Aboriginal Studies, volume 22 of Linguistic Series. 112171.
Trask, Robert Lawrence (1979): On the origins of ergativity. In: Frans Planck (ed.) Ergativity:
Towards a Theory of Grammatical Relations, London: Academic Press. 385406.
chapter 2
Grammatical metaphors
and there-insertionin Danish
Anne Bjerre* and Tavs Bjerre**

*University of Southern Denmark **Aarhus University
Kolding/Denmark Aarhus/Denmark
Transitive verbs are in general not allowed in there-constructions in Danish,

but a small subset is, and this paper gives an explanation. We suggest two
constraints on there-insertion, a subject with the semantic role of theme and an
empty object position, and show how the subset of verbs meets these constraints
while other transitives do not. The analysis is formalized within the framework
of Linearization-based HPSG.
1. Introduction
In Danish,1 as in other languages, a there-construction is used for discourse pur-

poses. It is used when we want to present a focus into the discourse, hence its
alternative name the presentational focus construction. (1) gives a typical example
of a Danish there-construction.
(1) Der ankommer et ungt par med to smbrn
There arrives a young couple with two kids
The expletive occurs in subject position and the logical subject occurs in the direct
object position. It is generally assumed that the construction is an unaccussativity
test and that only unaccussative verbs, i.e., a subset of intransitive verbs, appear in
the construction, as in (1).
With this characterization of the there-construction it is not a surprise that
most transitive verbs do not allow there-insertion in Danish. In constructions with
1. Mller & rsnes (2013), a HPSG grammar of Danish, has appeared after the submission of
this article.
90 Anne Bjerre and Tavs Bjerre
transitive verbs the direct object position of the verb is already filled and cannot be
occupied by the logical subject. So an example like (2) is not possible.
(2) *Der kyssede prinsessen en smuk prins
There kissed princess-the a beautiful prince
However, in Danish we find exceptions to this rule as has also been noted in the
literature by many others, e.g., Platzack (1983) and Vikner (1995). In this paper
we want to focus on a small set of transitive verbs which in certain contexts con-
stitutes an exception to the rule that Danish transitives do not appear in there-
constructions, cf. also Bjerre & Bjerre (2008b). (3a) and (3b) are typical examples
of such verbs.
(3) a. Da jeg en dag konfronterede ham og udredte hans spind af lgne,
When I one day confronted him and unravelled his web of lies,
ramte der mig en ubeskrivelig indvendig smerte i hjertet
hit there me an indescribable inner pain in heart-the
b. For da hun trdte igennem hovedindgangen, mdte der hende
Because as she stepped through main entrance-the, met there her
en strk duft af appelsiner
a strong scent of oranges
The verbs are typically used to convey the meaning of perception or sensation.
They appear in constructions which Halliday (1994) has termed grammatical
metaphors. In this paper we will show that the there-constructions involving such
grammatical metaphors do not constitute an exception to the general constraints
on there-insertion in Danish.
2. Grammatical metaphors
The term grammatical metaphor was introduced by Halliday (1994). A gram-

matical metaphor involves the encoding of a meaning by a different grammatical
construction than the one typically used. An example from Halliday (1994:344)
is given in (4).
(4) a. Mary saw something wonderful.
b. Mary came upon a wonderful sight.
c. A wonderful sight met Marys eyes.
If we want to encode an experience, we usually construct a sentence with an expe-

rience verb, an experiencer argument and a theme (experiencee) argument, the
Chapter 2. Grammatical metaphors 91
experiencer being linked to the subject and the theme linked to the object. (4a)
shows this congruent way of expressing the meaning that somebody perceives
something. Mary is an experiencer subject, saw an experience verb of perception,
and something wonderful is a theme object. Alternatively, we may choose a gram-
matical metaphor, as in (4b) and (4c). In (4b) we do not have a perception verb,
but instead the experience is represented as the theme argument, a wonderful
sight. In (4c), the experience is represented as the agent argument.
The effect of using a grammatical metaphor, apart from its poetic effect which
we will not have anything to say about, is a change of aktionsart of the construc-
tion. The example in (4a) does not mean exactly the same as (4b) and (4c). See is a
state verb. Come upon and meet, on the other hand, are achievement verbs and the
extra meaning they bring to the construction is an emphasis on a change of state,
rather than just a state. So we can paraphrase (4b) and (4c) as (5).
(5) Mary began to see a wonderful sight.
3. Unaccusativity
We mentioned earlier that the there-construction is treated as an unaccussa-

tive test. This goes back to Burzio (1986) who claimed that the verbs that allow
there-insertion are unaccusative verbs. The idea that intransitive verbs split into
two classes, i.e., unaccusatives vs. unergatives, is the core of the Unaccusative
Hypothesis, (Perlmutter 1978). The classification is based on the different under-
lying structures of the two classes of verbs. Unaccusatives have an initial 2 but no
initial 1, (Perlmutter 1978:160). This means that unaccusatives have an underly-
ing object but no subject. Unergatives, on the other hand, have an underlying
subject. Although the classification is syntactic, Perlmutter notes that semantic
factors determine the syntactic classes, i.e., unaccusatives take a patient argument
whereas unergatives typically describe an activity, cf. (Perlmutter 1978:162163).
The semantics of the two classes has been given more prominence by, e.g., Zaenen
(1993) who, based on Dowty (1991), proposes that the argument of unaccussatives
has more patient properties than agent properties, and the argument of unerga-
tives has more agent properties than patient properties. Levin & Hovav (1995)
argue that the division corresponds to a distinction between verbs which are exter-
nally caused and internally caused (Levin & Hovav 1995:98).
In Danish and other Scandinavian languages it is not obvious that the there-
construction is an unaccusative test. In these languages there-constructions with
unergative verbs occur freely as in (6).
(6) a. Hber ikke der lber sm brn rundt hos naboen,

Hope not there run little children about at neighbour-the,
som kan vre uheldige at falde i poolen
who can be unlucky to fall in pool-the
b. Der gik en flok ovre p den anden side, da de s os
There walked a flock on the other side, when they saw us
begyndte de at lbe vk
began they to run away
Examples like these have led some researchers, e.g., Sveen (1996) and Ldrup
(2000) to reject that the there-construction is an unaccusative test in Scandinavian.
The verbs appearing in there-constructions are not necessarily a homogeneous
class. Oxenvad (1934) and Brjars & Vincent (2005) explain the appearance of
unergatives in there-constructions in terms of the discourse function of the con-
struction, saying that the construction weakens or bleaches the meaning of certain
verbs, and hence verbs allowing this may appear.
Others, e.g., Hoekstra & Mulder (1990), have posited two entries for these
verbs, one consistent with an unacussative verb and one consistent with an uner-
gative verb. The meaning consistent with the unacussative representation may then
be considered to be a deagentivized version of the unergative, cf. also Kirsner
(1973), Maling (1987) and Ekberg (1990). In this way the there-construction as an
unaccussative test may be maintained.
In Section6 we will present a constraint on there-insertion which is not based
on unaccusativity, but will allow us to treat unaccussatives and a subset of unerga-
tives as a homogeneous class. The constraint will be formulated as a constraint on
semantic arguments. In Section8 we will show that the generalization naturally
extends to account for the transitive verbs allowing there-insertion in (3).
4. The position of the logical subject
When we talk about transitive verbs and there-insertion, the direct object position
is important. We mentioned earlier that in a there-construction, the logical subject
appears in the direct object position which is why transitive verbs have tradition-
ally been excluded from the group of verbs allowing there-insertion.
Platzack (1983) suggests that the difference between languages allowing tran-
sitive verbs in there-constructions and languages not allowing transitive verbs
in there-constructions is determined by the grammatical function of the exple-
tive pronoun. As shown in (7) the expletive pronouns have the same position as
ordinary subjects in questions in English and Swedish, but not in German and
Icelandic.
(7) a. Was there a man in the drainpipe?
b. Satt det en fgel p taket?
Was-sitting there a bird on roof-the
c. Sind (*es) in diesem Zimmer zwei Sthle?
Are there in this room two chairs
d. Eru (*a) ms n bakerinu?
Are there mice in bathtub
In English and Swedish the expletive functions as the subject and appears in sub-
ject position, whereas in German and Icelandic, the expletive pronoun appears
in topic position and does not function as the subject. Platzack (1983) proposes
that this distinction determines whether transitive verbs are allowed in there-
constructions. Languages in which the expletive functions as subject do not allow
transitives. The assumption behind this is that in English and Swedish the logical
subject occupies the NP position inside the VP, i.e., [NP, VP], and there is only
one NP position in VP. Platzack (1983:89) assumes the D-structure in (9) for the
Swedish there-construction with an intransitive in (8).
(8) Det satt en fgel p taket
There was-sitting a bird on roof-the
2(9) S
CONFL2 NP VP
V NP PP
+PAST e satt en fgel p taket
Transitive verbs are then excluded as the [NP, VP] position is occupied by the
direct object of the transitive verbs, as shown in (10) and (11), Platzack (1983).
(10) Sven kbte verkligen boken igr

Sven bought really book-the yesterday
2. CONFL is a position where both complementizers and and tense markers are generated
(Platzack 1983:82).
(11) S
CONFL NP ADVP VP
V NP PP
+PAST Sven verkligen kbte boken i gr
Platzack (1983:92) recognizes that there are exceptions to this rule and puts for-
ward the examples in (12). He does, however, not propose an explanation as to
why these verbs allow there-insertion.
(12) a. Det hnde honom ngot konstigt igr

There happened him something strange yesterday
b. Det vntade mig en verklig verraskning nr jag kom hem
There waited me a real surprise when I came home
Askedal (1986) and Ldrup (2000) also attribute the non-occurrence of transitive
verbs in there-constructions to the functional status of the logical subject as that
of direct object. They assume that in Norwegian counterparts of (12), the objects
are indirect objects, and consequently the direct object position is still vacant for
the logical subject to fill. They point out that the logical subject in there-construc-
tions appears in a VP construction which is structurally identical to a regular VP
construction with an indirect object preceding a direct object. To illustrate this,
Askedal (1986) gives the Norwegian examples in (13) and Ldrup (2000) gives the
examples in (14) which are also Norwegian.
(13) a. Da hadde det tilsttt ham noe alvorlig
Then had there happened him something serious
b. Da hadde hun gitt ham en ny bok
Then had she given him a new book
(14) a. Det blev overrakt barna en liten gevinst
There was awarded the-kids a small prize
b. Hun har overrakt barna en liten gevinst
She has awarded the-kids a small prize
An objection to the explanation regarding the direct object position is put forward
by e.g. Maling (1988) and Brjars & Vincent (2005). They refer to verbs that are
optionally transitive. Brjars & Vincent (2005:60) refers to the example in (15),
originally from Ldrup (2000).
(15) *Det spiste en mann en kake

expl eat.pst a man a cake
There was a man eating a cake
The intransitive variant of such verbs are not allowed in there-constructions, even
though the direct object position is not filled by a direct object preventing the
logical subject to fill the position, as the example in (16) from Brjars & Vincent
(2005) shows.
(16) *Det t ett litet barn i kket
expl eat.pst a.nt small.nt.sg child(nt) in kitchen.def
A small child was eating in the kitchen
These examples show that the constraint that an empty direct object position is
required, is not a sufficient constraint. Maling (1988:175) suggests instead that
verbs associated with an agent role cannot occur in there-constructions, as agents
must be assigned to subject position and not direct object position. This explains
the exceptions noted by Platzack (1983), as the verbs in (12) have no agent role,
and also the examples from Maling (1988) in (17).
(17) a. Det ndde Tomas ett brev hemifrn
There reached Tomas a letter from-home
b. Det slog mig ngot interessant igr
There struck me something interesting yesterday
c. Det slog mig en kollega
There struck me a colleague
Maling adds the further restriction that the verbs associated with an exeriencer
role cannot occur either. This addition is motivated by the observation that psych-
verbs do not occur in there-constructions, neither the subject experiencer nor the
object experiencer variants. The examples in (18) are from Maling (1988:176).
(18) a. *Det skrmmer mig sdana tanker
There scare me such thoughts
b. *Det behagede honom lite musik
There pleased him a little music
In this paper we will maintain the empty direct object position as a constraint.
The empty direct object position constraint, however, cannot stand alone, and a
further semantic constraint is required, as also suggested by Maling. However, as
will be shown in Section6, the semantic constraint is not based on the exclusion
of an agent and experiencer.
5. Aktionsart and event structure
Before we can present our constraint on there-insertion, we need to introduce

our analysis of event and argument structure, cf. also Bjerre & Bjerre (2008a),
Bjerre & Bjerre (2007) and Bjerre (2003). Our analysis is based on a complex event
structure. Regarding events as having internal structure goes back at least to Lakoff
(1965) and McCawley (1968) and is employed in combination with the Vendlerian
classification (Vendler 1967) in Dowty (1979), Pustejovsky (1991), Pustejovsky
(1995) and Levin & Hovav (1995) among many others.
Vendler (1967) classifies verbs into four classes, distinguishing four types of
aktionsart. He distinguishes between verbs describing states, activities, achieve-
ments and accomplishments. Three parameters are used, occurrence, duration and
telicity, as shown in (19).
(19) occurrence duration telicity

state +
activity + +
achievement + +
accomplishment + + +
We will also take this classification as our starting point, and as, e.g., Pustejovsky
(1991) assume that states and activities are basic and that the complex event struc-
tures achievements and accomplishments can be constructed by combining these
two event types into complex events.
We start by distinguishing between simple and complex events, as shown in (20).
(20) situation
sit-struc list-of-event-rels
simple-situation complex-situation
sit-struc event-rel temp-rel temp-rel
sit-struc event-rel, event-rel
The situation structure of a simple situation is represented as a singleton list of

event relations. A complex situation, on the other hand, is represented as a list
of two event relations. In addition it has the feature TEMP-REL which is used to
indicate the temporal relationship between the two subevents.
We also need to show what types of event relations we assume. We distinguish
between process relations and state relations as shown in (21).
(21) relation
event-rel ...
e-ind e-ind
process-rel state-rel
Semantic arguments are introduced as features on relations. We are inspired by

Davis (2001), but many details differ. In (22) our subtyping of the process-rel type
is shown and we also show how an actor and an undergoer argument are intro-
duced and inherited in these subtypes.
(22) process-rel
actor-rel und-rel unspec-rel

actor ref undergoer ref
act-only-rel act-und-rel unspec-act-rel unspec-und-rel fully-unspec-rel
Similarly, in (23) we show the subtypes of the state-rel type. We distinguish

between three types of states.
(23) state-rel
theme ref
theme-only-rel experiencer-theme-rel loc-rel

experiencer ref ground ref
In a theme-only-rel a property is ascribed to a theme argument. In an experiencer-

theme-rel an experiencer argument is further introduced and the theme argument
corresponds to the experiencee. In a loc-rel a ground argument is introduced and
the theme is located wrt. the ground.
The last type we need to describe is the TEMP-REL feature. The feature is used
to describe the temporal relation between the subevents in a complex event. As
such it plays a crucial part in determining the lexical aspect of an event. For now
we only distinguish two types, as shown in (24).
(24) temp-rel
precede-cause-rel included-rel
The precede-cause-rel implies that a subevent precedes and causes another subev-
ent. The included-rel implies that a subevent temporally includes another subevent.
Based on this set of types, we can formally define Vendlers aktionsarts. A state
verb will contain the specifications in (25) as part of its CONTENT value.
(25) word
CATEGORY | HEAD verb
simple-situation
SYNSEM | LOCAL
CONTENT state-rel
SIT-STRUC
THEME i
An activity is also a simple situation, but it will have a different type of event rela-
tion specified. In (26) we show part of the representation of an activity verb.
(26) word
simple-situation
SYNSEM | LOCAL
CONTENT actor-rel
SIT-STRUC
ACTOR i
An accomplishment is a complex event and what characterizes it is that a specified

process precedes and causes a state. A typical accomplishment verb will be partly
specified as in (27).
(27)
word
complex-situation
TEMP-REL precede-cause-rel
SYNSEM | LOCAL act-und-rel
CONTENT
SIT-STRUC ACTOR i , state-rel
UNDERGOER j THEME j
Finally, we want to show a typical achievement verb. We distinguish achievements

from accomplishments by the nature of the process in the first subevent. The pro-
cess leading to the state in achievements is not specified by the verbal content. In
the case of the typical achievement verb reach, we do not know what process took
place prior to the state of having reached something. We formalize this as in (28),
where the event relation in the first subevent is of type unspec-rel.
(28) word
complex-situation
TEMP-REL precede-cause-rel
SYNSEM | LOCAL
CONTENT state-rel
SIT-STRUC unspec-rel, THEME i
6. The there-insertion constraint
Before we present our semantically based there-insertion constraint, we want to

present some typical examples of Danish there-constructions.
In (29) we show examples of verbs of existence and appearance.
(29) a. Der er ikke en mulig verden, hvor 2+2=5 eller hvor der
There is not a possible world, where 2+2=5 or where there
eksisterer gifte ungkarle
exist married bachelors
b. Men fra den id, s opstod der 1000 andre
But from this idea, then arose there 1000 others
The examples in (30) show verbs of change of state.

(30) a. Efter den sidste istid for 10.000 r siden smeltede der s meget is,
After the last ice age for 10.000 years ago melted there so much ice,
at havene steg
that oceans-the rose
b. I bde 2001 og i 2002 bnede der nye afdelinger
In both 2001 and in 2002 opened there new aliations
We find examples of directed motion verbs, inherently directed as in (31) or in

combination with a directional PP as in (32).
(31) a. Den samme nat ankommer der ogs en mystisk fremmed
The same night arrives there also a mysterious stranger
b. Der afgr et tog hvert tyvende minut
There departs a train every twenty minutes
(32) a. Da han kom til byens port, lb der en stor flok hunde
When he came to town-thes gate, ran there a big pack dogs
imod ham
toward him
b. P et tidspunkt gik der en hest ud i sen
At a time walked there a horse out in lake-the
We also find verbs of spatial configuration in Danish there-constructions, in the

simple position sense and verbs in the assume position sense. Examples are
given in (33).
(33) a. Der str et billede af dig p mit bord
There stands a picture of you on my table
b. Lige da han sagde ordene, satte der sig en flue p ham
Right when he said words-the sat there itself a fly on him
Finally, we often find manner of motion verbs with locational PPs, as in (34).
(34) a. Har lige hrt i radioen, at der lber en tosse rundt

Have just heard in radio-the, that there runs a fool about
p motorvejen
on motorway-the
b. Tilfldigvis gik der en enlig ko p marken
Incidentally walked there a single cow on field-the
The manner of motion verbs with a directional PP, the verbs of spatial configura-
tion and the manner of motion verbs with a locative PP obligatorily co-occur with
a PP in there-constructions. This means that we do not find examples of there-
constructions as in (35).
(35) a. *P et tidspunkt gik der en hest
At a time walked there a horse
b. *Der str et billede af dig
There stands a picture of you
c. *Tilfldigvis gik der en enlig ko
Incidentally walked there a single cow
The verbs mentioned here are not all unaccusatives. This can be seen from their
selection of auxiliary verb which is a test for unaccusativity in many languages,
including Danish. The verbs of spatial configuration and the verbs of manner
of motion with location PPs all select have (have) as their auxiliary, as shown
in (36).
(36) a. Der har stet et billede af dig p mit bord

There has stood a picture of you on my table
b. Lige da han sagde ordene, havde der sat sig en flue p ham
Right when he said words-the had there sat itself a fly on him
c. Har lige hrt i radioen, at der har lbet en tosse rundt
Have just heard in radio-the, that there has run a fool about
p motorvejen
on motorway-the
d. Tilfldigvis har der get en enlig ko p marken
Incidentally has there walked a single cow on field-the
We propose that the verbs in this apparently heterogeneous group of verbs indeed
form a homogeneous class in that they all have a theme argument of a state rela-
tion linked to a subject. We refer to Bjerre & Bjerre (2008a) for a detailed analysis
of many of these types of verbs. Here we will just provide example analyses of the
verbs that traditionally have not been given an unaccusative analysis.
In (37) we show the partial representation of the spatial configuration verb st
(stand) obligatorily combining with a locative PP. A st-rel relation is a subtype
of the loc-rel relation.
(37) st,stand
word
HEAD verb
CATEGORY
ARG-ST NPi, PPj
SYNSEM | LOCAL simple-sit st-rel

CONTENT THEME i
SIT-STRUC
GRND j
And in (38) we show our analysis of a manner of motion verb with a locative PP.
(38) g,walk
word
HEAD verb
CATEGORY
ARG-ST NPi, PPj
complex-sit
TEMP-REL included-rel
g-rel
SYNSEM | LOCAL ACT i
CONTENT
SIT-STRUC loc-rel
THEME i
GRND j

Our analysis assumes that the locative PP occurring with a manner of motion
verb is obligatory and is represented as part of the event and argument structure
of the manner of motion verb, cf. e.g. Maienborn (1991) for a similar analysis. The
state subevent is temporally related to the activity event by the included-rel which
means that the activity event is temporally included in the state event.
Importantly, even though these verbs are not unaccusatives, they meet the
constraint that we propose for there-insertion in Danish, that they contain a state
relation with a theme argument linked to the subject. In (39) we show the con-
straint formally as a lexical rule, cf. also Bjerre & Bjerre (2008a).
(39) there-insertion-lexical-rule
word
HEAD verb
CATEGORY SUBJ 1 NP indef : 2 i
COMPS 3
IN SYNSEM | LOCAL
situation
CONT state-rel
SIT-STRUC list THEME i list
INFO-STRUC | TOPIC 2
word
SUBJ der
OUT SYNSEM | LOCAL | CATEGORY
COMPS 3 1
INFO-STRUC | TOPIC
The lexical rule produces a verb which will appear in a clause that has no topic.
The input to the rule is a verb that has an indefinite subject, the subject position is
coded as topic. The referent of the subject is semantically located in a state relation,
i.e., it is the theme argument of a state relation, i.e., the supertype of theme-only-
rel, experiencer-theme-only-rel and loc-rel which are the relations involved in verbs
allowing there-insertions. Der, there, is inserted on the SUBJ list.
7. Danish transitive verbs in there-constructions
In this section we will introduce a group of transitive verbs allowing there-inser-

tion. The verbs are verbs that occur in grammatical metaphors as already shown
in (3). (40) gives more examples of these verbs used in grammatical metaphors.
(40) a. Et frygteligt syn af desperat krokodille mdte os

A terrible sight of desperate crocodile met us
b. En duft af nymalet korn slog os i mde
A smell of freshly ground grain hit us toward
A smell of freshly ground grain came toward us
c. Terrorfrygten griber igen New York
Fear of terror grabs again New York
d. Hans ord ramte hende som et piskesmld
His words hit her like a crack of a whip
e. Kun en svag mumlen nede ham med mellemrum
Only a faint murmur reached him with intervals
Examples of there-constructions with the verbs in (40) are shown in (41).
(41) a. Da jeg mdte p arbejde mandag den 1. oktober

When I met at work Monday the 1. October
(efter en dejlig ferie p Samos), mdte der mig et herligt syn af flag
(after a lovely holiday on Samos), met there me a lovely sight of flag
og blomster, som brnene havde malet
and flowers, which children-the had painted
b. Da jeg kom ind slog der mig en stank af tobaksrg i mde
When I entered hit there me a stench of tobacco smoke toward
When I entered there came toward me a stench of tobacco smoke
c. Der greb mig en frygt for dden
There grabbed her a fear of death-the
d. Da jeg en dag konfronterede ham og udredte hans spind af lgne,
When I one day confronted him and unravelled his web of lies,
ramte der mig en ubeskrivelig indvendig smerte i hjertet
hit there me an indescribable inner pain in heart-the
e. Mens han flyttede om i Europa for sine studiers skyld, nede
while he moved about in Europe for his studies sake, reached
der ham gang p gang nye rygter om forestende forbindelser
there him time after time new rumors about coming liaisons
If we look at these verbs in constructions not involving grammatical metaphors,

we find that we do not get corresponding there-constructions, as shown in (42)
and (43).
(42) a. En golfbold ramte ham i panden
A golf ball hit him in the forehead-the
b. En 18 mneder gammel dreng nede kysten
A 18 months old boy reached coast-the
(43) a. *Der ramte ham en golfbold i panden

There hit him a golf ball in forehead-the
b. *Der nede kysten en 18 mneder gammel dreng
There reached coast-the a 18 months old boy
In Section8 we will show an analysis which explains why the examples in (41) are
well-formed, whereas the examples in (43) are not.
8. Formal analysis of transitive verbs
First we will characterize the verbs that enter into these grammatical metaphors.
This small group of verbs have the semantics in common that they all denote an
event where something comes into contact with something else in some way.
Jackendoff (1990) describes a group of verbs of impact. The verbs in this group
have the meaning: motion of the Theme culminating in contact with the refer-
ence object. (Jackendoff 1990:109). Jackendoff (1990) gives the verbs the struc-
ture in (44).
(44) [Event INCH[State BEc([ ]i, [Place ATc[ ] j])]]
The verbs of impact are inchoatives and they involve an event where an argument
comes to be at a location. The c subscribts indicate contact.
Using our formal apparatus, such verbs will be subtypes of the partial repre-
sentation in (45).
(45) word
HEAD verb
CATEGORY
ARG-ST NPi, NPj
complex-sit
SYNSEM | LOCAL TEMP-REL precede-cause-rel
CONTENT contact-rel
GROUND j
The verbs describe events where an unspecified process precedes and results in
a state where a theme argument is in contact with a ground argument. In (46)
and (47) we show the partial formalizations of the verbs ramme (hit) and n
(reach).
(46) word
ORTH ramme
HEAD verb
CATEGORY
ARG-ST NPi, NPj
complex-situation
CONTENT hit-rel
GROUND j
(47) word
ORTH n
HEAD verb
CATEGORY
ARG-ST NPi, NPj
complex-situation
CONTENT reached-rel
GROUND j
The hit-rel and reached-rel relations are subtypes of the contact-rel type.
The verbs of contact meet our there-insertion constraint in (39). However,
as we have seen, only when they appear in grammatical metaphors do they allow
there-insertion. And this is where the second constraint on there-insertion comes
in, as we are dealing with transitive verbs, there has to be an empty direct object
position.
The verbs that we are interested in in this paper are only found in there-con-
structions when they are used in grammatical metaphors. We may say that what
is going on in our grammatical metaphors is that achievement events involving
physical contact are used as metaphors for state experiences for some intended
effect, including that of emphasizing the transition from one state to another. Thus
in certain contexts, when we have an (indefinite) abstract argument linked to a
subject that refers to an emotion or perception, these verbs are used in gram-
matical metaphors. We will use the lexical rule in (48) to derive the metaphorical
meaning.
(48)3
contact-to-experience
word
HEAD verb
CATEGORY SUBJ NPi
COMPS NPj
IN complex-sit
SYNSEM | LOCAL TEMP-REL 1 precede-cause-rel
CONTENT contact-rel
SIT-STRUC 2 unspec-rel, THEME i
GROUND j
word
subj NP: 3
CATEGORY
comps NPj
OUT complex-sit
SYNSEM | LOCAL TEMP-REL 1
CONTENT experience-rel
SIT-STRUC 2 ,3 THEME k3
EXPERIENCER j
Applying the lexical rule in (48) to the partial lexical entries in (46) and (47) will
give an output that is admissible as input to the lexical rule inserting der shown
in (39). The ultimate result is the entries in (49) and (50) that are constrained wrt.
the definiteness and relational identity of the argument linked to the subject. We
do not provide an analysis of definiteness nor an hierarchy of nominal relations
here, though.
(49) word
ORTH ramme
HEAD verb
SUBJ der
CATEGORY
experience-rel
COMPS NPj, NP: 1 THEME i
SYNSEM | LOCAL EXPERIENCER j
complex-situation
CONTENT TEMP-REL precede-cause-rel
SIT-STRUC unspec-rel, 1
3. The index k should be structure-shared with the potential complement of the subject, but
we have not yet decided on the exact formalization of this.
(50) word
ORTH n
HEAD verb
SUBJ der
CATEGORY
experience-rel
COMPS NPj, NP: 1 THEME i
SYNSEM | LOCAL EXPERIENCER j
complex-situation
CONTENT TEMP-REL precede-cause-rel
SIT-STRUC unspec-rel, 1
So far we have not excluded the entries in (46) and (47) from undergoing there-
insertion. To rule out the latter entries, we need to take a closer look at the indirect
object position constraint and the linking properties of the semantic arguments
involved.
9. Arguments and topology
There is a long tradition within Danish linguistics for describing word order (topol-
ogy, linear structure) in terms of what is usually referred to as a stningsskema
(sentence scheme).
The sentence scheme consists of slots for the various sentence elements. The
original proposal had different schemes for main clauses, subordinate clauses,
non-finite verb phrases etc.
(51) shows the scheme for main clauses.
(51) Diderichsen (1957:186)
Hovedstning
main clause
Forb.-felt Fund.-felt Neksusfelt Indholdsfelt
connector field foundation field nexus field content field
ks F v s a V S A
og saa kunde han sikkert faa hende i tide
ikke sagt besked
alligevel
and then could he probably get her in time

not said notice
anyway
ks is the position for coordinating conjunctions, F is for almost any kind of con-
stituent, v is for the finite verb, s is for the subject if F is occupied by some other
constituent, a is for (some) adverbials, V is for non-finite verbs, N is for objects
and nominal and adjectival predicates and A is for adverbials.
The idea of describing word order in terms of a sentence scheme instead of as
the yield of a tree is to allow for a separation of word order and phrase structure
(hierarchical structure) and, e.g., be able to describe discontinuous constituents
without the use of movements. In German linguistics there is a similar tradition
taken up within HPSG by, e.g., Reape (1994) and Kathol (2000).
To model linear order as an independent level, the type sign is given a new
feature, DOM.
(52) sign
PHON ...
SYNSEM synsem
DOM [dom-obj], ...
The elements on the DOM list reflecting linear order are of type dom(ain)-obj(ect).
Both dom-obj and sign have the features PHON and SYNSEM, but only sign has
the feature DOM.
(53) sign-or-dom-obj
PHON list
SYNSEM synsem
sign dom-obj
DOM list
There are two ways in which the DOM list of the mother may be constructed from
the DOM lists of the daughters. Either the two lists are shuffled or the elements
on one of the lists are compacted to one element and then inserted with the shuffle
relation. The shuffle relation, notated , merges two lists allowing any order of ele-
ments on the output list as long as the relative order of elements on the two lists is
preserved. To give an example, if we shuffle the two lists a,b and c,d orderings
where a precedes b and c precedes d are allowed:
(54) a,b c,d
a,b,c,d,
a,c,b,d,
a,c,d,b,
c,a,b,d,
c,a,d,b,
c,d,a,b
Compaction corresponds to the traditional notion af constituent. In most cases a

non-head-daughter is compacted before it is inserted in the DOM list of the head.
Compaction takes a sign with potentially more elements on its DOM list turning
it into one domain object with a PHON list consisting of the concatenation of the
PHON lists on the DOM list and a synsem value structure shared with the synsem
value of the sign. This is shown in functional notation in (55).
(55) sign
PHON 1 ( 2 3 n ) dom-obj
compaction dom-obj dom-obj dom-obj
DOM , ,, , PHON n
PHON 2 PHON 3 PHON n SS 4
SS 4
Compaction of a sign has two purposes. It ensures that the resulting domain
object can be inserted into the DOM list of the mother as one element filling
only one slot. And it eliminates information on the internal structure of the sign
(information on its daughters) which is not needed and in accordance with the
Locality Principle should not be available in the further derivation.4 The order
of elements on the DOM list of any sign is restricted by the following (informal)
constraint:
(56) sign DOM Fvslioldoa1*VIODOPa2*H]
Except for the adverbial dom-objs (marked *) the DOM list may contain at most
one element of each type. These elements correspond to the slots in the sentence
model.
The constraint on head-comps-phr shown below says that a head can combine
with one or two complements the synsem values of which must correspond to the
synsem elements on the heads COMPS list. Each of the complement daughters
are compacted to domain objects of type obj. The DOM list of the phrase is the
product of shuffling the compaction of the non-head daughter(s) with the DOM
list of the head daughter.
4. In some cases it may be useful to distinguish between full compaction and partial compac-
tion, e.g., in the case of extraposition of postnominal modifiers (Kathol & Pollard 1995). In (1)
the constituent einen Hund, der Hunger hat is compacted into two elements that are inserted
separately:
(1) einen Hund fttern, der Hunger hat
a dog feed that hunger has
feed a dog that is hungry
(57) head-comps-phr
SUBJ 1
SS | LOC | CAT
COMPS
SUBJ 1
SS | LOC | CAT COMPS 5 ,( 6 )
DOM 2 3 ( 4 )HEAD-DTR
dom 2
NON-HEAD-DTRS 7 SS 5 , 8 SS 6

compaction( 7 , 3 obj )
(
compaction( 8 , 4 obj ) )
Some of the information in (57) originates from constraints on supertypes.

The type obj which is a subtype of dom-obj is further subtyped into light and
heavy, and direct and indirect objects as shown in (58).
(58) dom-obj
obj
light heavy indirect direct
lio ldo IO DO
Some further constraints that we will not show here ensure that non-stressed pro-
nominal objects are of type light while the rest are of type heavy.
The type head-comps-phr is further constrained as shown in (59).
(59) head-comps-phr
direct
DOM
1
SS | LOC | CONT | SIT-STRUC THEME/GROUND 1

indirect
DOM
2
SS | LOC | CONT | SIT-STRUC EXP 2
This constraint ensures that a theme or a ground is linked to a direct object posi-
tion while an experiencer is linked to an indirect object position.
The lexical rule shown in (39) moves the so-called logical subject to the
COMPS list, but since this must be a theme, if there is another object there with
the semantic role of theme or ground the constraint in (59) and the prohibition
against more than one dom-obj element of the same type on any dom-obj list will
disallow the structure. This explains why the verbs in (46) and (47) cannot appear
in there-construction.
10. Passives
In this section we will show that our analysis extends to passive transitive verbs.
Just as it has commonly been assumed that active transitives do not allow there-
insertion, it has been assumed that passive transitives do, cf., e.g., Vikner (1995).
In passives the direct object of the active verb is promoted to subject thus meeting
the constraint that the object position should be empty.
(60) a. Og et kmpebjerg af smkager og spandevis af kae blev spist
And a huge mountain of cookies and buckets of coee were eaten
og drukket
and drunk
b. Mange bde blev bygget omkring rhundredskiftet
Many boats were built around turn of century-the
They also meet our constraint on there-insertion in (39) that a theme argument
be linked to the subject so there-insertion is possible.
(61) a. Og der blev spist et kmpebjerg af smkager og drukket
And there was eaten a huge mountain of cookies and drunk
spandevis af kae
buckets of coee
b. Der blev bygget mange bde omkring rhundredskiftet
There were built many boats around turn of century-the
However, the passives of our metaphorically used transitive verbs do not allow
there-insertion.
(62) a. Det er et r siden at jeg pludselig blev ramt af en lammende angst
It is a year ago that I suddenly was hit by a paralyzing anxiety
b. Hun blev grebet af en frygtelig flelse af hblshed
She was grabbed by a terrible fear of hopelessness
(63) a. *Det er et r siden at der pludselig blev ramt en mand af

It is a year ago that there suddenly was hit a man by
en lammende angst
a paralyzing anxiety
b. *Der blev grebet en kvinde af en frygtelig flelse af hblshed
There was grabbed a woman by a terrible feeling of anxiety
They do meet the empty direct object position constraint, but they do not meet
our there-insertion constraint, as the subjects in the passive counterparts are not
themes, but experiencer arguments.
11. Conclusion
While transitive verbs in general are not allowed in there-constructions in Danish,

a small subset of metaphorically used transitive verbs are. We have suggested
two constraints on there-insertion: The subject must have the semantic role of
theme and the object position must be empty. We show that the metaphorically
used verbs meet both of these constraints while other transitive verbs do not.
Furthermore, we have shown that the analysis without further stipulation extends
to passive there-constructions.
Bibliography
Askedal, John Ole (1986): On ergativity in modern Norwegian. Nordic Journal of Linguistics
9: 2545.
Bjerre, Anne & Bjerre, Tavs (2007): Perfect and periphrastic passive constructions in Danish.
Nordic Journal of Linguistics 30: 553.
Bjerre, Anne & Bjerre, Tavs (2008a): Danish there-constructions with intransitive verbs. In:
Philippe de Groote (ed.) Proceedings of the 13th Conference on Formal Grammar. Stanford
CA: Center for the Study of Language and Information, 722.
Bjerre, Anne & Bjerre, Tavs (2008b): Danish there-constructions with transitive verbs. In:
Stefan Mller (ed.) Proceedings of the 15th International Conference on Head-Driven Phrase
Structure Grammar. Stanford CA: Center for the Study of Language and Information,
4666.
Bjerre, Tavs (2003): Syntactically and semantically complex predicates. Ph.D. thesis, Univer-
sity of Southern Denmark.
Brjars, Kersti & Vincent, Nigel (2005): Position vs. function in Scandinavian presentational
constructions. In: Miriam Butt & Tracy Holloway King (eds.) Proceedings of the LFG05
Conference. Stanford CA: Center for the Study of Language and Information, 5472. URL
http://cslipublications.stanford.edu/LFG/10/lfg05.html.
Burzio, Luigi (1986): Italian Syntax: A Government-Binding Approach. Dordrecht: Reidel.
Davis, Anthony (2001): Linking by Types in the Hierarchical Lexicon. Stanford CA: Center for
the Study of Language and Information.
Diderichsen, Paul (1957): Elementr Dansk Grammatik. Copenhagen: Gyldendal.
Dowty, David (1979): Word Meaning and Montague Grammar. Dordrecht: Reidel.
Dowty, David (1991): Thematic proto-roles and argument selection. Language 67: 547619.
Ekberg, Lena (1990): Theta role tiers and the locative PP in existential constructions. Working
Papers in Scandinavian Syntax 46: 3745.
Halliday, Michael Alexander Kirkwood (1994): An Introduction to Functional Grammar.
London: Arnold, 2nd edition.
Hoekstra, Teun & Mulder, Rene (1990): Unergatives as copular verbs: locational and existen-
tial predication. The Linguistic Review 7: 179.
Jackendo, Ray (1990): Semantic Structures. Cambridge MA: MIT Press.
Kathol, Andreas (2000): Linear Syntax. Oxford University Press.
Kathol, Andreas & Pollard, Carl (1995): Extraposition via Complex Domain Formation. In:
Proceedings of the 33rd Annual Meeting of the ACL. Cambridge MA, 17480.
Kirsner, Robert S. (1973): Natural focus and agentive interpretation: On semantics of dutch
expletive er. Stanford Occasional Paper in Linguistics: 101114.
Lako, George (1965): On the nature of syntactic irregularity. Ph.D. thesis, Indiana University.
Published as: Irregularity in Syntax. New York: Holt, Rinehard and Winston. 1970.
Levin, Beth & Hovav, Malka Rappaport (1995): Unaccusativity. At the Syntax-Lexical Semantics
Interface. Cambridge MA: MIT Press.
Ldrup, Helge (2000): Linking and optimality in the norwegian presentational focus construc-
tion. Nordic Journal of Linguistics 22: 205230.
Maienborn, Claudia (1991): Verbs of motion and positions: On the optionality of the local
argument. In: Claus-Rainer Rollinger & Otthein Herzog (eds.) Textunderstanding in
LILOG: Integrating Computational Linguistics and Artificial Intelligence, Berlin and New
York: Springer, volume 546. 621631.
Maling, Joan (1987): Existential sentences in swedish and icelandic: reference ti thematic roles.
University of Lund. Working Papers in Scandinavian Syntax. No. 28.
Maling, Joan (1988): Variation on a theme: Existential sentences in Swedish and Icelandic.
McGill Working Papers in Linguistics: 168191. Special issue on comparative Germanic
syntax.
McCawley, James D. (1968): Lexical insertion in a transformational grammar without deep
structure. CLS (Papers from the Regional Meetings of the Chicago Linguistic Society) 4:
7180.
Mller, Stefan & rsnes (2013): Danish in head-driven phrase structure grammar. URL http:
//hpsg.fu-berlin.de/~stefan/Pub/danish.pdf.
Oxenvad, Erik (1934): Om nogle upersonlige konstruktioner i dansk. In: Poul Andersen, Aage
Hansen & Peter Skautrup (eds.) Studier tilegnede Verner Dahlerup paa Femoghalvfjerdsaars-
dagen den 31. Oktober 1934, Copenhagen: Reitzel. 135140.
Perlmutter, David M. (1978): Impersonal passives and the unaccusative hypothesis. BLS (Pro-
ceedings of the Berkeley Linguistics Society) 4: 157189.
Platzack, Christer (1983): Existential sentences in English, Swedish, German and Icelandic.
In: Fred Karlsson (ed.) Papers from the Seventh Scandinavian Conference on Linguistics.
Helsinki: University of Helsinki, 80100.
Pustejovsky, James (1991): The syntax of event structure. Cognition 41: 4781.
Pustejovsky, James (1995): The Generative Lexicon. Cambridge, Massachusetts: MIT Press.
Reape, Mike (1994): Domain union and word order in German. In: John Nerbonne, Klaus
Netter & Carl J. Pollard (eds.) German in Head-Driven Phrase Structure Grammar, Stanford
CA: Center for the Study of Language and Information. 151197.
Sveen, Andreas (1996): Norwegian personal actives and the unaccusative hypotehsis. Ph.D.
thesis, University of Oslo.
Vendler, Zeno (1967): Linguistics in Philosophy, Ithaca NY: Cornell University Press, chapter
Verbs and Times. 97121.
Vikner, Sten (1995): Verb Movement and Expletive Subjects. Oxford: Oxford University Press.
Zaenen, Annie (1993): Unaccusativity in Dutch: Integrating syntax and lexical semantics. In:
James Pustejovsky (ed.) Sematics and the Lexicon, Kluwer Academic Publishers, volume
49. 129161.
chapter 3
Abstract objects of verbs
Klaus Robering
University of Southern Denmark
Kolding/Denmark
Verbs often take arguments of quite dierent types. In an orthodox type-

theoretic framework this results in an extreme polysemy of many verbs. In
this article, it is shown that this unwanted consequence can be avoided when
a theory of abstract objects is adopted according to which these objects
represent non-objectual entities in contexts from which they are excluded by
type restrictions. Thus these objects are abstract in a functional rather than in
an ontological sense: they function as representatives of other entities but they
are otherwise quite normal objects. Three examples of such a representation are
considered: the denotations of that-phrases are objects representing proposi-
tions, generic noun phrases denote objects standing for sorts, and infinitivals
are viewed as denoting objects representing attributes, i.e., the ordinary mean-
ings of verb phrases.
1. Introduction
1.1 Verbs and their arguments
Classifying linguistic items into categories has always been one of the major tasks
of the linguist. Since, furthermore, verbs have been considered to occupy a distin-
guished place among the partes orationes, the classification of verbs has always
enjoyed the linguists special interest. A classification, of course, can be based on
quite different criteria. Given, however, the view that the verb is the center of
the sentence, it is natural to attempt at a classification of verbs according to the
syntactic and semantic relationships which they bear to the items with which they
co-occur in the sentence. This idea is implemented in otherwise quite different
approaches to grammar as, for example, generative grammar, dependency gram-
mar, and categorial grammar. The technical apparatus used in these approaches
(as, e.g., subcategorization frames and -grids, category indices and cancellation
116 Klaus Robering
rules, valency frames and dependency trees) are quite different but the basic idea
is always the same, namely to classify verbs into different grammatical categories
according to the number and type of their arguments and complements.
Thus verbs and the classification of verbs are central to grammar. However,
verbs are bewildering, too. What appears to be the same verb may co-occur with
arguments of quite different types. As an example consider the sentences of (1).1
(1) a. Mary likes John
b. Mary likes taco parties
c. Mary likes the way that I walk
d. Mary likes to dance
e. Mary likes that hes gotten the young folk interested
in an endangered dance style
f. Miss Mary likes what I am playing
g. Mary likes how Keia ended things
These examples agree in the subject argument Mary of the verb (form) like. But the
object argument is either a proper noun, a single common noun, a noun phrase
(or determiner phrase, if you prefer), an infinitive phrase, a subordinate clause,
or an embedded interrogative. Obviously, this list does not exempt the range of
possibilities.2
The intuitive point of view is surely that, albeit the variety of argument types,
there is only one single relation of liking involved in all the examples of (1). Liking
is just the relation of bearing a positive attitude toward something or somebody.
One may have this relationship to entities of quite different kinds: to persons
(like John), to a special sort of social events (like taco parties), to an activity (like
dancing), to a specific mode of carrying out a certain activity (such as a way of
walking), or to a proposition (like the proposition that a certain person has got a
certain group of young people to be interested in an endangered dancing style),
or to still other entities. Intuitive as this view may be, both the ontologist and the
logician will insist that the different cases should be cleanly separated from each
other. Both will explain that there are differences in logical type between the enti-
ties said to be liked by Mary in the examples of (1) and that these differences may
not be neglected.
Splitting up the general relation of liking into a number of subrelations differ-
ing with respect to the types of objects of their codomains does of course mean
nothing else than adopting an extreme polysemy for the verb like. By admitting
1. All these examples have been found (thanks to Google) on the web.
2. As is attested, for instance, by the example Mary likes dancing suggested by (1d) above.
Chapter 3. Abstract objects of verbs 117
that the relationships distinguished from each other are nevertheless related, one
adopts the demand for some set of meaning postulates describing the interconnec-
tions between the different submeanings. The meaning postulates have to describe
in which respects liking John resembles liking taco parties and liking to dance.
Otherwise one would miss a generalization which speakers of English make when
carving up the field of possible sympathetic attitudes one can bear to entities of
different type. Obviously English speakers collect some of these relations under
the common umbrella label of liking. This label, however, so are we told by the
adherent of a rigid type theory, has no meaning in itself.
The present article is an attempt to circumvent the hypothesis of a logical
polysemy for verbs taking arguments denoting entities of different types. Instead
it is tried to to retain a simple denotation for a simple verb. The way to attain this
goal is not to give up all type restrictions but rather to admit representations of
entities belonging to one type by entities belonging to another. This strategy is
by no means unknown in formal semantics. In the theory of generalized quanti-
fiers, for instance, objects are represented by second order attributes applying
to just those first order attributes that in turn apply to the object at issue. This
representation renders possible a unified semantic treatment of proper nouns
and quantified noun phrases. Since it is most suggestive to conceive of verbs as
denoting relations between objects, the representations required for a mono
semic semantic treatments of verbs such as like would have to map non-object
arguments (like sorts, propositions, and attributes) to objects which function as
placeholders for them. Here, such placeholders in the type of objects are called
abstract objects. This does not imply that these placeholders have an extraordi-
nary ontological status. The label abstract is often reserved for entities not located
within the ontological framework of time and space; cf. p. 129 below. This
notion of abstractness may be (ontologically or linguistically) useful or not; in
any case, it is not adopted in the present article. Nor is there any claim that some
special individuals are inherently abstract. In the framework of the theory to be
sketched, to be an abstract object just means to fulfill a special semantic function,
namely to represent a non-object in a context from which it is normally barred by
type barriers. Thus the theory proposed here may be called a functional theory
of abstract objects. Spatio-temporal individuals are often taken as the prototype
examples of concrete objects. According to the present theory, nothing hinders
such a spatio-temporal thing to function as an abstract object by representing
some other entity.
118 Klaus Robering
1.2 Classifying linguistic signs
We started above by stating that classification is one of the main tasks of the lin-
guist and that verbs pose a complex problem for this grammatical task. The basic
questions to be answered in connection with each classification are: (i) What
exactly are the items to be classified? And, (ii), what are the categories available
for these items. The most natural answer to the first question and, indeed, the
most adequate one would be: linguistic signs, i.e., items consisting of an expres-
sion and its denotation.3 However, since nothing will be said here about syntactic
matters4, the expressions belonging to linguistic signs will be completely ignored
here. Hence the topic of the present article could be described more precisely as
the classification of the denotation of verbs (and other linguistic signs) rather than
the classification of verbs themselves as having both a form and a meaning.
The categories which are available for the classification of denotations are
called types here. We shall conceive here of a type as a structured collection of lin-
guistic items. This means that a type comprises (i) a type predicate which exactly
applies to the members of the type and (ii) a series of further predicates, relations,
and operations which may relate the members of the type to each other and pos-
sibly also to members of other types. We shall return to this issue in Section3.1
below; our main concern here, however, is the type predicate. Though we distin-
guish between types and the collection of their members; the metalanguage labels
for the type predicates are also used to name the types themselves. These type
labels are called type indices. Note that in the present framework type indices are
predicate constants of the metalanguage used for semantic description. As usual in
type theory and categorial grammar, basic types are distinguished from complex
one. For the purposes of the present article it is (more than) sufficient to assume
four basic types: the types of propositions, o of objects, of sorts, and of times.5
As these examples illustrate, Roman lower case Greek letters are used as con-
stants for specific (basic) types. As variables for types the italic letter either
in its bare form or ornated with number subscripts (thus 0, 1, ) is used.
3. This position is implemented in quite different ways in the categorial grammars of, for
instance, Morrill (1994), Carpenter (1997), and Villadsen (2010). As regards terminology, I
use the expression denotation as a label for the semantic component of a linguistic sign. By this,
I do not want to commit myself to any special doctrine of denotations as opposed to (real)
meanings.
4. Cf. Robering (2000) for a presentation of my ideas on categorial syntax.
5. A time as understood here need not to be instantaneous but may also be extended. Thus
moments and time spans (intervals) are both times.
Since is just a predicate of our metalanguage, we could express the fact that
a is of type by using the ordinary notation (a). However, in order to conform
more closely to standard type theories, we write a: instead. This expression is
an example of a type assignment. Some concrete examples of type assignments
are given in (2).
(2) a. Mary : o
b. John : o
c. TacoParty :
A bold face expression starting with a capital letter names the denotation of the
corresponding items of the object language under analysis. Thus, for instance,
Mary is the denotation of Mary. The type assignment (2a) says that Mary is an
object. The same is said of John by (2b). The assignment (2c) states that TacoParty
is a sort. The type assignments of (2) say that certain predicates (namely, o and )
apply to certain items. These items, thus Mary, John, and TacoParty, are thus indi-
viduals in the semantic framework of our metalanguage. Note that we distinguish
between objects and individuals. Everything belonging to a basic type is, of course,
an individual. But not all individuals are objects. TacoParty is an individual, how-
ever it is a sort and not an object.6
Complex types are recursively built up from other types (starting with the
basic ones, of course) by means of type constructors.7 Here we need only one fam-
ily of mutually related type constructors. For each m1 there is the (m+1)-con-
structor (...) which takes a basic type 0 and m types j(1jm) in order to
build up the complex type 0 (1 m). This is the type of functions which take
m arguments from the respective types 1 m and map them to values of type
0. The level of a type is the number of pairs of parentheses in its index. Thus the
level of basic types is 0 whereas the level of, e.g., (oo) is 1 and that of ((o)) is
2. The type assignment (3), for instance, says that Dance is a function mapping
objects to propositions.
6. At least as long as we do not have a further type assignment which says that TacoParty is
also an object. In Section2 below the occurrence of the expression taco parties in (1b) is ana-
lyzed as referring to an object systematically related to, though not identical with the non-object
TacoParty.
7. Modern type theories as they are used in the analysis of programming languages are often
more powerful than those applied in natural language semantics in that they have special addi-
tional type constructors (for example, for pairs, lists, and records). Many of these additional
constructors would also be useful in a linguistic context. Cf., e.g., Pierce (2002) for an introduc-
tion to type theory and its application to programming language analysis.
120 Klaus Robering
(3) Dance : (o)
Note that (o) (as well as other complex types) applies to functions; hence it is a
higher-order predicate of our metalanguage. Consequently, Dance is not an indi-
vidual (and thus, a fortiori, certainly neither an object); as already said, it is a
function. The combinatorial potential of this function may be described by the
conditional type assignment scheme (4).
(4) a : o Dance(a) :
The scheme (4) says that the application of the function Dance to any object yields
a proposition. The operation of application is notationally expressed by writing the
name of the function to be applied in front of a pair of parentheses containing a list
of the respective arguments. In the case of (4) this is a list with a single item only.
The arrow denotes the conditional of the logic used in our metalanguage. Note
that this does not need to be classical logic. Therefore it may be the case that not
all inferential transformations which are valid for the classical material implica-
tion may be applied to the conditional statement (4). We shall return to this issue
several times (pp. 122,148,152). The letter a is a free metalinguistic variable
which may be replaced by any term.
Of course, in our semantic theory the conditionals of the form (4) should be
derivable from the type assignment (3). This inferential relationship between (3)
and (4) is quite independent of the special function Dance and holds true for every
member of type (o) as well. What is at issue here is the meaning of the monadic
type constructor (...) which ensures that the following inference rule (5a) is valid.
This rule is furthermore readily generalizable to the m-adic case (5b).
F : 0(1)
(5) a. ((. . .)1)
a : 1 F(a) : 0
F : 0 (1 ... m )
b. ((. . .)m)
a1 : 1 ... am : m F(a1 ... am) : 0
The variable F in (5) ranges over functions (of arity 1 and m, respectively). The
rule ((...)m) is used for semantic composition of complex denotations out of
simpler ones. In order to achieve this, however, it has to be supplemented by rules
from the underlying logical framework: namely, by modus ponens and by the rule
(Intr) of conjunction introduction which allows to infer the conjunction of two
already derived statements. The sample derivation (6), for instance, illustrates how
the proposition expressed by (1a) is built up.
(6) a. Likes : (oo) axiom

b. Mary : o John : o Likes(Mary John) (a), ((. . .)2)
c. Mary : o axiom
d. John : o axiom
e. Mary : o John : o (c), (d), (Intr)
f. Likes(Mary John) : (b), (e), modus ponens
The rule ((...)m) in combination with the logical rules (Intr) and modus ponens
provides for meaning composition by functional application and thus corresponds
to Ajdukiewiczs Cancellation Rule of traditional categorial grammar. Modern ver-
sions of categorial grammar typically possess further rules besides Ajdukiewiczs
Cancellation Rule. These rules give rise to further modes of semantic composi-
tion. Here, too, we supplement ((...)m) by a second rule which, in some sense,
is the inverse of the first rule. Usually, a free variable as the variable a occurring
in scheme (4) is interpreted as implicitly bound by an initial universal quantifiers
governing the whole rule. Under this interpretation, (4) says that the function
Dance applied to an arbitrary member of type o yields a proposition. This, how-
ever, amounts exactly to that what is also expressed by the type assignment (3).
Hence it is an obvious move here to adopt an inverse ((...)m()) of the general rule
((...)m()) which allows to infer the premiss of the latter rule from its conclusion
if the subject terms of the type assignments in the conclusions hypothesis are all
variables. We shall do almost this, but go one step further by allowing the variables
being spread in some way over the whole expression (rather than occurring one
after the other in a list enclosed within brackets) and by prefixing the subject of
the new conclusion type assignment by the corresponding -term.8
(7) a1 : 1 ... am : m F [a1, ... , am] : 0
((. . .)m)
(x1 ... xm.F[x1, ... , xm]) : 0(1 ... m)
It is important to see what the new rule ((. . .)m) really does: it provides new
functions previously not available. The type system has been restricted here in
such a way that the value types in complex types are always basic ones. Hence no
new functions besides those given in axiomatic type assignments e.g., (3) or the
first line in (6) can be constructed merely by functional application. However,
new functions are easily constructed by the ((. . .)m) as the following derivation
illustrates.
8. F [a1, , am] is an expression in which the variables a1, , am occur. An expression of the
form F (a1 am) is, of course, also an expression of the first form. The converse, however, is not
true. There are expression of the form F [a1, , am] which do not have the form F (a1 am).
122 Klaus Robering
(8) a. Likes : (oo) axiom

b. a : o Mary : o Likes(a Mary) : (a), ((. . .)2)
c. Mary : o axiom
d. a : o Likes(a Mary) : (b), (c), propositional logic
e. (x.Likes(x Mary)) : (o) (d), ((. . . )1)

f. Every : ((o)) axiom
g. Boy : (x.Likes(x Mary)) : (o) (f), ((. . .)2)
Every(Boy (x.Likes(x Mary))) :
h. Boy : axiom
i. Boy : (x.Likes(x Mary)) : (o) (h), (e), (Intr)
j. Every(Boy (x.Likes(x Mary))) : (i), (g), modus ponens
Two remarks have to beadded here. First, it is said in the rightmost column of (8)
that the transition from lines (b) and (c) to line (d) is by propositional logic. The
logic underlying our semantic meta-theory has not yet been fixed, however. Thus
the exact justification of the inferential step at issue might still be a delicate matter.
An obvious procedure for deriving (d) from (b) and (c) would be, for instance, to
first commute the conjuncts in the hypothesis of (b) by applying the law of com-
mutativity of conjunction, then to extrapose the new left conjunct by applying
rule (9) below, and then, finally, to apply modus ponens with line (c) of derivation
(8) as the minor premiss.
(9) ABC
A [B C]
Now (9), which is known as the law of extraposition, is not a valid principle in
all propositional logics. It does not hold, e.g., in relevance logic.9 Therefore the
justification just given for the step from (b) and (c) to (d) would not be avail-
able in logics whichlike relevance logic refute the law of extraposition. If the
transition at issue is still regarded as valid, another justification has to be found
instead. The second remark to be added is that it is only reasonable to adopt the
rules ((...)m) if the logical framework of the metalanguage contains principles
for handling the -operator.
9. Cp. Anderson & Belnap (1975:261f) on what relevance logicians call the fallacies of
exportation.
2. Explaining the theory
2.1 The traditional analysis
As has already been briefly explained at the end of the introductory Section 1.1,
the gist of the theory suggested here is to analyze all the phrases which occur in
object position in the sample sentences (1) as denoting items of type o. These
object expressions are repeated here in list (10).
(10) a. John
b. taco parties
c. the way that I walk
d. to dance
e. that hes gotten the young folk interested in an endangered dance style
f. what I am playing
g. how Keia ended things
This position differs markedly from that what is else common in formal semantics.
Before explaining the analysis suggested in the present article, I will in the present
subsection give a brief sketch how these examples (or rather just the first five of
them) would be treated in a more common framework. In such a framework, the
phrases just listed in (10) would be assigned to different types. The proper name
John (10a) denotes a special object, namely the person John; hence John:o.10 The
expression taco party is often called a common (count) noun and is taken to stand
for a monadic attribute.11 Intransitive verbs like dance receive the same type of
interpretation as common noun phrases if we ignore tense for the moment. Hence
we would have both TacoParty:(o) and Dance:(o). Turning now to (10c), let
us call the denotation of (10c) just Way. The definite article in (10c) functions as
a description operator. According to a Fregean analysis, this operator picks out
a certain one from the realm of all objects. Hence: Way:o. Russell, on the other
hand, in his famous analysis of definite descriptions analyzed them as quantifiers
and many semanticist would like to follow him in this today. Hence: Way:((o)).
10. As is well known, John could alternatively be analyzed as belonging to type ((o)). If this is
done, John is identified with the second order attribute which applies to all first order attributes
which the person John possesses. For the notion of an attribute cf. the following footnote.
11. By an attribute I mean a function mapping objects (monadic attributes) or, respectively,
n-tuples (n-adic attributes) to propositions. Propositions in Montagues framework are map-
pings from indices (i.e., world-time-pairs) to truth values. Here, however, they are conceived as
individuals of a basic type.
124 Klaus Robering
It is quite common to conceive of the clause (10e) as denoting a proposition.12 Call

this proposition ThatGotten; then: ThatGotten:. I will not go into the discussion
of the last two examples of (10) in this article. They involve embedded interroga-
tive clauses and I just refer here to Groenendijk & Stokhof (1984) for a well-known
treatment of such construction in a mildly modified Montagovian framework.
Though I consider it possible, I will not try to develop an alternative analysis of
these constructions along the lines suggested for the other examples.
Given the type assignments of the last paragraph, we receive the following
analyses.
(11) a. Likes ( Mary John )
(oo) o o
b. Likes ( Mary TacoParty )
(o(o)) o (o)
c. Likes ( Mary Way )
(o((o))) o ((o))
d. Likes ( Mary Dance )
(o(o)) o (o)
e. Likes ( Mary ThatGotten )
(o) o
The analyses given in (11) involve four different typings for Likes. However, the
typings of (11a) and (11c) could be collapsed by Montagues well-known uniform
analysis of noun phrases as denoting quantifiers. This would reduce the num-
bers of typings for Likes to three. If one, on the other hand, would follow Gupta
(1980) (and other authors viewing the matter in a similar way) and would retain
a type difference between the denotations of common count nouns and those of
intransitive verbs, one would be forced to adopt a further typing. Gupta (1980)
argues that common nouns denote sorts which differ semantically from monadic
attributes. The index has already above been adopted for sorts; hence we have
TacoParty: but Dance:(o). As said, this would lead up to yet another typing
for Likes, cf. 2.1.
12. In any case, this is Montagues analysis in Universal Grammar; cf. Montague (1970:237ff).
There (p. 237) that-clauses are called proposition-denoting expressions. Their denotations have
the Montagovian type s,t, which means that they are functions from indices to truth-value,
thus possible world propositions. Verbal constructions with that-complements are differently
parsed, however, in The Proper Treatment of Quantification in Ordinary English. In
the latter article, Montague (1973:250ff) conceives of the complementizer that as belonging to
the verb rather than to its complement. Syntactically, the complex verbs believe that and assert
that require a t-expression as an argument. The translation rule T7, however, intensionalizes
the semantic input delivered by the t-argument; cf. Montague (1973:262).
(12) Likes ( Mary TacoParty )

(o) o
Given this, we would have five typings for Likes or four, in case that we adopt the
Montagovian analysis of proper nouns.
2.2 Types and polymorphism
As everybody accustomed to the fields of logic, mathematics, and programming

knows, types are extremely useful by providing transparent notations (where
untyped languages often use awkward circumlocutions) and by making the appli-
cation of meaningless operations impossible by simple syntactic prohibitions.
Therefore [a]ttention paid to logical type is, as Dana Scott (1971:350) remarks,
attention well spent. Types have, however, also disadvantages as is attested by
both the existence of type-free systems of logic and untyped or only loosely typed
programming languages13 and by the somewhat excessive use of types in (11)
above. Often one finds interesting and important relations which do not care for
boundaries between types. The relation of liking is, in spite of what (11) might
suggest, an instance of this. One may like persons, certain kinds of social arrange-
ments, modes of walking, activities, propositions, and probably all other kinds of
entities as well. Strict adherence to the principles of typing would, as (11) exempli-
fies, force one to distinguish between several relations of liking thus giving rise to
an extreme polysemy of the verb like.14 The problem is, of course, not restricted
to natural language but arises also for the type systems of the formal languages of
logic, mathematics, and programming. A famous example are the natural num-
bers. One can count the members of any type15. But does this really mean that
there are different types of natural numbers corresponding to the types of elements
13. JavaScript is an example of a loosely typed language, Smalltalk is an untyped language.

14. Strictly speaking this is not true for Montague (1973). Montague would as his treatment
of the verbs believe, assert, try, and wish shows (cp. footnote 12) parse the examples (1d) and
(1e) in such a way that the complementizer that and the particle to belong to the verb rather than
to its complement. Obviously, this analysis can be extended to (1f) and (1g) giving rise to two
further complex verbs like what and like how. This, of course, does not lead up to a proliferation
of readings of verbs. Rather it enlarges the stock of verbs itself, which is as unwelcome. In any
case, this is not the way natural language works. There is just one single verb like. Furthermore
this procedure is surely not very attractive from a syntactical point of view.
15. Provided there is a well-defined notion of identity associated with the type which makes it
possible to distinguish its members. In this article, the notion of a type is understood in such a
way that this is always the case.
126 Klaus Robering
counted? As is well known, exactly this is what happens in the logicist arithme-
tic developed by Whitehead & Russell (19101913) in their seminal Principia
Mathematica. There are numbers for counting objects besides numbers for
counting sets of objects and besides numbers for counting sets of sets of objects,
etc. In order to reduce the complexity of the system with its infinitely many types
of numbers and to make it more easy to use, the notational convention of typical
ambiguity is adopted by Russell and Whitehead. This convention allows to ignore
type distinctions whenever this is possible. Of course, it only hides (but does not
eliminate) the multiplication of structurally completely similar entities and does
not really eliminate it.16
A step toward such an elimination is the admittance of polymorphism. This
means that one allows for polymorphic entities belonging to different types.17 A
typical example from combinatory logic is the analysis of the identity combina-
tor I. The combinator I returns for each argument entity this same entity as the
corresponding value. Obviously, this makes sense for every type to which an
argument might belong. Consequently, it makes good sense to say that I belongs
to every functional type of the form (). Perhaps something similar can be said
for the case of liking. The Likes-function returns for each person p and each object
x of an arbitrary type the proposition that p has a positive attitude toward x and
prefers it to a large number of other members of . Actually, it has been already
admitted above that any entity whatsoever might be the object of such a positive
attitude. Assume that is the type of meanings of spatial prepositions. There is no
problem in imagining a linguist specialized in research on prepositions and hav-
ing a somewhat exotic affection toward the relation of being in something. True
as this may be, we do not formulate ourselves in the most direct way opened for
us by polymorphism. We do not, for example, just say that the imagined linguist
likes in. Instead, we employ circumlocutions such as the above used the relation
of being in something or, perhaps, the Heidegger-like sounding the relation of in-
ness or even just in-ness. Thus we speak of the In-relation as if it were an object
using a formulation which contains the preposition either adorned with further
somewhat technical terminology like relation or in morphologically modified
form, e.g., augmented by the suffix -ness.
16. Typical ambiguity is explained in Whitehead & Russell (19101913: Vol. I, 41ff). Cardinal
arithmetic is treated in Part III of the Principia Mathematica; this is the first part of the
second volume. The first section of this part is mainly devoted by studying the the relation of
a cardinal number in one type to the same or an associated cardinal number in another type;
Whitehead & Russell (19101913: Vol. II, p. 5).
17. Cf. Hindley & Seldin (1986: Chapter16) or Pierce (2002: Chapter V) for an introduction to
polymorphic typing.
2.3 Representation by abstract objects
The theory proposed here agrees with the polymorphism hypothesis in assuming
only one single relation Likes. In contrast to the polymorphism hypothesis and
in agreement with more traditional type theories, it assumes that this relation
belongs to a certain determinate type, namely (oo). Hence we retain the type
assignment Likes:(oo) used as an axiom in the sample derivations (6) and (8)
without supplementing it by further type assignments. This single Likes-relation
from (oo), however, can also be used to describe situations where a non-object
x is liked. In order to do this, one has just to employ modes of expression that
represent x from the non-object type (x:) as an object. This means that one
has to specify a representative or placeholder r(x) of x in type o (r(x):o). This is
schematically illustrated in Figure1 for the cases of propositions and sorts.
Individuals
(Propositions)
x
o (Objects)
(x)
(y)
(Sorts)
Figure 1. Representation of propositions and sorts by objects
In order to apply the idea to our examples from (1), remember the type assign-
ment TacoParty: from the paragraph preceding (11). Furthermore, the proposi-
tion expressed by (13).
(13) Hes gotten the young folk interested in an endangered dance style
will be called Gotten here. Since Gotten: and TacoParty:, Mary (of type
o) cannot bear the relation likes:(oo) to these two individuals. In the theory
proposed here there are, however, the operations and which assign objec-
tual placeholders from type o to sorts () and propositions (), respectively:
(TacoParty),(Gotten). By the mediation of these placeholders, the items rep-
resented by them become available as intentional (with a t) objects for Marys
positive attitude; cf. (14).
128 Klaus Robering
(14) a. Likes ( Mary ( TacoParty ) )

(oo) o o()
b. Likes ( Mary ( Gotten ) )
(oo) o o()
The complementizer that by which (10e) differs from (13) may be analyzed as
the syntactic reflex of the -operation. This view is already implicit in Currys
work on categorial grammar and in his formalist approach to logic. [C]ertain
subordinating conjunctions such as that are called subnectors by Curry &
Feys (1958:275) and Curry (1963:33) describes these as functors which form
nouns out of sentences.18 The denotation of a subnector is called a subnexus in
Curry (1963:33). Using Currys terminology, we could say that that is a subnector
denoting the subnexus . Similarly, we may conceive of the plural marking of
taco parties as indicating a functor whose application to the argument TacoParty
from type renders accessible the second argument place of the Likes-relation
by providing a representing object which can fill its second argument place. Be
aware that I do not say that this is the (only) function of plural but only that a
plural marking can function in this way. The existence of type shifting functions
as and whose application are recognizable either on the syntactical (sub-
nector/complementizer) or morphological (plural) level is a counterargument
against theories explaining the type shifts at issue here by polymorphism. If both
propositions and sorts were also objects besides being what they in fact are,
namely propositions and sorts , i.e., if and were just subtypes19 of o, then
we would expect that both sentences and common count nouns could readily
occupy the object position of the verb like. This, however, is not the case; in order
to occupy this position, a sentence needs a complementizer and a common count
noun has to be in plural form although these modifications would be semanti-
cally void from the viewpoint of the polymorphism hypothesis. An additional
hypothesis would thus be needed under this hypothesis in order to explain these
modifications.
The theory developed in the present article is that propositions, sorts, and
attributes, though they make up their own types, are nevertheless sometimes
represented as if they belonged to the type of objects. Now, what an object
(like John) distinguishes from a proposition (like: Gotten) is that John by his
body is an individual existing in space and time whereas the proposition is
not. Though it makes sense to ask of John where he is now or when he will be at
a certain place, this obviously does not make sense for a proposition. It makes
18. In Currys system of categories they belong to the category Fsn.

19. For subtyping see, for instance, Pierce (2002: Chapter III).
(presumably20) good sense to ask when and where (at which times and at which
places) a proposition is true; but, in contrast to this, it does not make sense to
ask where a certain proposition is or when it has been at a certain place. Similar
things apply to sorts (and attributes). John belongs to the sort of human beings.
But while we can ask of John at each moment of the span of his life where he is
at that moment, it does not really make sense to ask where the sort of human
beings is. Thus propositions and sorts are entities which relate to the categories
of time and space in ways quite different from that of such things as stones, cars,
and houses (and even persons) which have determined positions within space
and time. The latter entities are called concrete whereas the former are said to
be abstract.21 This notion of abstractness may be (ontologically or linguistically)
useful or not; in any case, it is not adopted in the present article. Nor is any claim
accepted which assumes some special individuals which are inherently abstract.
However, as the title of the present article indicates, I shall nevertheless talk
about abstract objects.
According to the position taken in the present article propositions, sorts, and
attributes are no objects (i.e., they do not belong to type o) neither abstract nor
concrete ones. They are entities of their own type, but they can be represented by
objects. These representing object, then, may be called abstract. Hence abstract
objects, as understood here, are just ordinary objects which are used to represent
non-objects such as sorts (like the sort TacoParty), activities (Dance), proposi-
tions (like Gotten), etc. as if they were objects. Thus, to dance refers to an abstract
object, the activity of dancing. This object is not called abstract here because it
has (or lacks) a certain ontological characteristic. Rather it is called abstract since
it fulfills a certain semiotic function, namely: to represent an activity. This is, so
to speak, an instance of double codification: the expression to dance stands for an
object and this object stands in turn for an activity. For the information conveyed
20. This restrictive modification is added in order to take care of such absolute conceptions of
propositions as that of Frege. According to Frege, a proposition may have or lack a truth value.
But if it has a truth value it does never change it. An expression which is sometimes true and
sometimes false as, for example, The weather is fine here today thus does not itself express a
proposition (Gedanke) for him. The hearer of an utterance of such an expression has to supply
the information given by the words by the special circumstances of the utterance made; cf. Frege
(1918/19:348f).
21. Since this is not a essay in ontology, the presentation has to be somewhat superficial in not
taking account of some subtle philosophical problems and distinctions. However, the charac-
terization of the distinction at issue given in the main text above is sufficient for the purposes
of the present article. Lewis (1986:8186) discusses different approaches toward the distinction
between abstract and concrete entities. One of these is the Negative Way: abstract entities have
no spatio-temporal location; they do not enter into causal interaction; they are never indiscern-
ible one from another. This matches the explanation given above.
130 Klaus Robering
by a sentence such as (1d), which contains the expression at issue, the activity of
dancing is essential and not the object denoted by the object phrase. The latter
merely has a mediating function. The same holds true for the examples (10b)
and (10e) as they are used in the respective sentences of (1). They refer to objects
which in turn stand for a sort and a proposition, respectively.
The advantage of a theory which admits for objectual placeholders of entities
from other types is the reduction in complexity. Looking back to the traditional
analyses of (11), we recognize that the - and the -function make it possible to
skip the type assignments of Likes used there in (11b) and (11e) and to use the
simple assignment of (11a) instead. If, furthermore, a Fregean analysis of definite
descriptions is adopted, the same simple assignment may also be used in the case
of (11c). The only remaining case is thus (11d), where the Likes-function has
to be combined with an entity from type (o). Following the same idea, whose
application for arguments from the types and has been explained above, we
would postulate here, too, a type shifting function converting monadic first-order
attributes (type: (o)) into objects (type: o), say :o((o)). Using this type shifting
function, we would have (15).
(15) Likes ( Mary ( Dance ) )
(oo) o o((o)) (o)
What distinguishes this cases from that of and is that the latter functions are
defined on basic types whereas is applied to functions. Furthermore, the values
of are objects and these are just the entities which the argument functions of
require as their arguments. Assume, for instance, that F:(o). Then we would
have F((F)) : o. The proposition F((F)) says that the attribute F by applying to the
object (F) representing the attribute in some sense applies to itself. Now, since the
days of Russell and Whitehead type theories have been developed to avoid such
cases of circularity and self-reference, which in the foundations of mathematics
have given rise to the well-known paradoxes of class theory. This fact should be
seen as a warning that representing attributes by objects is quite a dangerous issue
which requires special care. We will return to it in Section4 below.
Before we take up that problem however, we turn to a more basic question.
According to our theory non-objects are sometimes represented by (abstract)
objects. This presupposes that we can distinguish objects from non-objects and
that we when classifying semantic units have principled reasons for putting
some entities into the type o of objects whereas other entities are assigned to other
types. The occurrences of our sample expressions (10) in the examples (1), for
instance, have been analyzed as referring to objects. But up to now no justification
for this classification has been given. Why, for example, should to dance in (1d)
stand for an object (type: o) rather than for an attribute (type: (o))? This problem
will be treated in the following subsection.
2.4 Identifying objects
The notion of an object as it is common in modern formal semantics derives

from the semantic theories of Frege. For Frege, objects are what referential expres-
sions refer to. Thus the distinction between objects and non-objects are correlative
to the linguistic distinction between referential expressions and expressions of
other kinds. Unfortunately, Frege was not very definitive on this distinction22 a
shortcoming, which several of his interpreters, among them Peter Geach (1950),
Michael Dummett (1973: Chapter 4), and Bob Hale (2001) have attempted to
improve. These authors tried to set up a battery of inference tests in order to distin-
guish referential expressions from non-referential ones in terms of their inferential
behavior. The precise formulation of these rule would require a careful consider-
ations of intricate syntactic issues. We shall not go into these matters here, how-
ever, but rest content in just stating the test rules in a somewhat loose manner.
The first rule of the test battery is an English version of the logicians existential
generalization.
(16)
There is something/someone such that pro
The Greek letter stands for an expression which is to be tested for referentiality.
As is easily seen, all the expressions of (10) as used in the examples (1) pass this
test since
(17) a. Mary likes someone
b. Mary likes something
are correctly deducible by (16) from all the sample sentences in (1). The semantic
motivation of this test is obvious. Clearly, if really is a referential expression, then
there should be an object to which it refers.23
The two next rules are designed to exclude quantified noun phrases. These
rules are displayed in (18) below.
22. Frege nowhere discusses the quest for criteria of referentiality in a systematic manner.
However, he mentions several criteria in his writings: (a) starting with the definite article or a
demonstrative pronoun (Frege 1884, cf. 51, 61, 68 fn., Frege 1892:195); (b) lacking a plural form
(Frege 1884, cf. 51, 68 fn.); (c) being unable to occupy the predicate position in a sentence (Frege
1892:193) .
23. I deliberately ignore here the problem of designationless terms. Consider, for instance, the
following argument, due to Anderson (1984:357): Tut worshiped Ptah There is (or was) some-
thing which Tut worshiped. Clearly, there is a reading of the conclusion under which the argu-
ment is invalid.
132 Klaus Robering
(18) a.
( )

and pro

b. or

( )

or pro

Both rules in (18) are double direction rules; this is indicated by the in the rule

names following the two rule schemes. A double direction rule may be used in
both direction: bottom up and top down. The rule ( ) catches the following two

criteria: (i) Suppose the expression to be tested occurs in two sentences. Then
we may conjoin these sentences by and and replace the test expressions occur-
rence in the second conjunct of the resulting complex sentence by some suitable
pro-element (pronoun, trace, or whatsoever) referring back to the test expressions
occurrence in the first conjunct. If really refers to an individual, then the complex
sentence may be validly inferred as a conclusion from the two original sentences
as premisses. (ii) Conversely, from a conjunction containing an occurrence of the
test expression in its first conjunct and a corresponding pro-form in the second,
we may, if the test expression really is referential, infer both the first conjunct and
the expression resulting from the second when the pro-form by is replaced by its
antecedent. As one easily recognizes, taco parties passes the test, cf. (19), while, for
instance, the indefinite noun phrase a poodle in (20) does not; the subrule ()
fails in this case though its two converses () are valid.
(19) Mary likes taco parties Tom detests taco parties

Mary likes taco parties and Tom detests them

Mary likes taco parties

Tom detests taco parties
(20) Mary likes a poodle Tom detests a poodle
/
Mary likes a poodle and Tom detests it

Mary likes a poodle

Tom detests a poodle
The reader will furthermore easily verify that taco parties also passes the ( )-test

whereas a universally quantified noun phrase like every poodle does not. This time
it is subrule () which fails. If every poodle is liked by Mary or detested by Tom, it

does not follow that Mary likes every poodle, nor does it follow that Tom detests every.
Not only taco parties but also the infinitive phrase to dance passes the two
tests. In order to save space, we just present a sample instance of ( ) and leave

it to the reader to check ( ).

(21) Mary likes to dance or she has not learned to dance

Mary likes to dance or she has not learned it
Mary likes to dance or she has not learned it

Mary likes to dance or she has not learned to dance
In this case, universally quantified noun phrases may serve as expressions which
do not pass the test; cf. the invalid second inference in (22).
(22) Mary did every dance or she disliked every dance

Mary did every dance or she disliked it
Mary did every dance or she disliked it
/
Mary did every dance or she disliked every dance
The semantic rationale behind ( ) and ( ) is that the Boolean operations

and (denoted by and and or) cannot only be used in the type of propositions
but also, as is well known, in all functional types of the form (0 m) (where
0 m are arbitrary argument types) in order to form more complex functions of
that type. In the present framework this can be explained as follows. Let G,H:(o);
then we proceed in the following way.
(23) a. G : (o) axiom

b. H : (o) axiom
c. a : o G(x) : (a), (( )1)
d. a : o H(a) : (b), (()1)
e. a : o G(a) : H(x) : (c), (d), propositional logic
f. : () axiom
g. G(a) : H(a) : [G(a) H(a)] : (f), (()2)
h. a : o [G(a) H(a)] : (e), (g), propositional logic
i. x.[G(x) H(x)] : (o) (h), (()1)
A completely analogous reasoning goes through for disjunction . Let us
therefore use the signs and not only for propositional conjunction
and disjunction but also for the corresponding operations on the type (o) of
monadic attributes which assign to G,H:(o) the complex attributes GH def
=
x.[G(x)H(x)] ({,}) whose construction is provided by (23) and the
analogous derivation for .
(o) axiom (b), (( )1(i) ) (a), x.[G(x) (( ) 1 ) axiom: (o)
H(x)] (h), (( )1 )
(o) axiom (a), (c), ((A (d), )1 )propositional
completely (b), (( logic
) ) axiom
analogous1reasoning goes through for disjunction . Let us therefore use the s
oH(x) G(x)
axiom (b),
: : axiom
(( )1and ) not (c),only (d), for (a), (( )logic
propositional
propositional 1 )conjunction and disjunction but also for the correspo
oH(a)] H(a) (a), ((: (f),
: (c), (d),
ing )1(( propositional
) ) ) axiom logic (b), (( )1 )
134 Klausoperations Robering on the type (o) of monadic attributes which assign to G, H : (o) the com
2
o: G(a)
(b), axiom
[G(a) ((
: (e),
)1H(a)]H(x)
(g),
attributes ) propositional :: G
(f), H = x.[G(x) (( logic
)2 ) (c), (d), propositional
H(x)] ( {, }) whose logic construction is provided by (23) and
)] H(a)]
() : (c), (d),(h),
:(f), ((propositional
(( )2 ) )1 ) (e), logic def(g), propositional axiom logic
analogous derivation for .
] ) :
:
gh for disjunction
(o) axiom
H(a)
(e), (g), : propositional
. Let[G(a) (h), H(a)]logic (( : )
) (f),
the(( )2 )
letletusbe therefore use signs
1
Now, Now, bethe theexpression
expression which
whichisistotobe betested
testedfor forreferentiality
referentiality and let Q be its denotat
o:ction
ning goes (f),
[G(a) (h),(( ((
through H(a)])2 ) )1for ): disjunction Let (e), (g), propositional
us therefore useHattributes
the logic
and disjunction
Letbe its furthermore but also G
denotation. forLetand.the H correspond-
furthermore be the monadic G and be signs
the monadic denotedattributes by the expressions
denoted resulting f
[G(x)
or tributes
ositional H(x)]
(e),
disjunction (g), propositional
which
conjunction .: assign
(o)Let us to
and G,
therefore logic
H
disjunction : use
(o) the
the
but (h),
signs
complex
also ((for )the
1 ) correspond-
theby the unconjoined
expressions sentences
resultingentering from thethe tests whensentences
unconjoined the test expression entering the tests is excised from t
(h), (( )1 )goes
attributes
})logous
n) of andwhose reasoning
monadic
disjunction
construction
(thus,
when but thealso
for through
is which
instance,
test for
provided
expression forMary
assign
the disjunction
correspond-
by (23) to isG,excised
likes H.
and :the Let
(o) or Tom
from usthem
the therefore
complex
detests(thus, use forthe in signs
the examples
instance, Mary likes given above). Wh
onlyH(x)]for
disjunction
utes which {, assign
(propositional
. at Let
})
orissue us
to
whose
Tom detests G, H
therefore
conjunction
is the truth :
construction
(o)
in theuse the
of thethe
and signs
complex
disjunction
is
examples provided
type assignment but
by
given above). also
(23) for
and
Q :What the
o. The correspond-
is at issue
tests ()is theandtruth ()ofinvolve
the the inferen
and
whose the disjunction
ested type construction
for (o) referentiality ofbut
type
relation monadic
isalsoprovided
assignment forattributes
and
between theQ
let bycorrespond-
be(23)
:o.
propositions its
Thewhichand
denotation.
tests assign
the ( up
built ) to and
byG,(QHand : )(o) the the
involve simple complex
the attributes G
inferential and H on the one h
relation

=sibutes whichisassign
x.[G(x) H(x)] toby G,theQH
( :referentiality
(o) the the whose complex construction isthe
provided by(23) and theother.
efwhich denoted tobe tested
by for expressions and
resulting let be G its denotation.
andbetween propositions
{,
and }) complex built up byfrom
attributes Q and H and G
simple attributes H on Gthe and H on All the the
one attributes at i
ose
ion
hed for
when construction
forreferentiality
monadicthe. testattributesarehand isofand
expression provided
and
type let
denoted
by by
Q isand
(o). beIn (23)
itsthe
by
excised
order and
denotation.
the from
complex
for the
expressionsQ to them
attributes
build resultingGHfrom
propositions and G withHthese on the other. AllQthe
attributes, has thus to be ei
etests
es
ering expression
denoted
the tests inbywhich
of the the
when
attributes
type isothe
examples toor
expressions atbe test tested
ofgiven
issue type resulting
expression
are for
above).
of type
((o)). referentiality
from
What
(o).is
The excised
Inis andfrom
order
purpose letofQthe
for bebuild
them
to its
tests denotation.
propositions
is, of course,with these the latter c
to exclude
Gnor
o.and theThe H
referentiality
test
or Tom
beexpression
tests the detests
()
Howattributes,and
monadic dolet
and ()
theyQ be
isattributes
excised
in
has its
the
involve
attain thus denotation.
examples
to
thisdenoted
from
thebegoal? them
either by
given
inferential Inofthe expressions
above).
type
order of typeresulting
What
otoorunderstand is((o)). this, from The
we have purpose of the
to remember that the
sentences
nd
denoted
assignment
the simplein by the the
entering
Q examples
tests
: of expressions
o.
attributes theoftests
is,
The
propositions G and
given
course,
tests resulting
when
() Hisexclude
above).
to thethe
and
onnot from
test
What
()
just the
one aexpression
iscollection
latter
hand case.
involve ofis special
How
theinferential excised
do they from
attainthem
entities this
but thatgoal?itInhas ordera certain structu
the
The , Mary
built
nd Gtest likes
tests
up expression
Hby()onQin toand
and
the which or
understand Tom
()
the
other. issimple
theinvolve
All detests
excised this,
the
operations from
the
attributes we inferential
attributes them
have Gand
in the
to
and H
examples
at remember
(onon)the
issue given
that
onethe
participate handabove).
type andto What
ofwhich is also a consequence
propositions is not relatio
hitions
he ibutes inKlaus
ofsimple
the Gthe
withtype examples
HRobering
attributes
these just
assignment
andattributes,
belongs. Ggiven
aGcollection
and HQon
Most Habove). ofThe
:Qproponents
o. on
the
has special
theother.
thus What
tests
one toAll isformal
entities
()
hand
be
of the
eitherandbutsemantics
that involve
()
attributes it athas issuea the
assume certain structure
inferential
that is a Boolean 24 in which Algebra. There
G ethe
propositions
to tests
H
build
tests
Robering 62 ()
on the andbuilt
propositions
is, of the
other.
however, () operations
course,up involve
by
All withtothe
Q and
these the and
the
attributes
exclude
alternative
Klaus Robering
inferential
simple
attributes,
the
logics (onat )
issue
latter participate
attributes
has
case.
(many-valued,
Q thus G toand be H
to which
eitheron
intuitionistic, the also
one a consequence
hand
relevance, etc.) rela-
and the question w
ns simple
derstand
The complex with purpose attributes
theseattributes
this, tion
attributes,
of we the G
have G belongs.
and
tests to H has
is, on
and
remember
of Mostthe
thusG
course, one
to H
proponents
be
thathand
on
toeitherthe
the
exclude
kind of logic is to be found in natural language (or a certain natural language) seems not to h
Q of
other.
type formal
the All semantics
the
latter attributes
case. assume at that
issue is a Boolean
pecialn3) InHorder
tests onis,
order(a)
axiom
the
foroftoG
entities other.
Qcourse,:Algebra.
to
understand
been but build
(o) All
that
addressed There
the
topropositions
itthis,
exclude attributes
has are,
awe the however,
certain
seriously have at issue
with
latter to
up these alternative
remember
case.
structure
till 24
attributes,
now. that
However, axiom
logicsQthe has(many-valued,
thusrealistic
type
most to be either intuitionistic,
candidates forrel- the logic of nat
G with
tand
pe : ((o)).
collection
ipate (o) (b)
axiom
these
this,
and we
toH
attributes,
ofThe
which (23)
:have
evance,
special
language(o) to
purpose also
Q (a)
etc.) hasa and
remember
entities of
assume Gthe
thus
consequence : (o)
the
but to question
that
tests
some be either
that the
is,
kind type
of
itrelation
has axiom
what
ofcourse, kind
to
alattice
certain of axiom
logic
exclude
structure
structure is to 24
thebe found
induced latter axiom
by in
,natural
case. , andlanguage on . Correspondin
H
in
stsal sassume :this
is,
entities
(on
(o) (c)
(a),
goal?
of
)course, a :the
((
but
participate
that In othat
(or
)1 isto aitG(x)
a)certain
order
relationexclude
has
and (b)
to :
Boolean ato H
natural
understand
on the
certain
which :, language)
latter
Algebra.
(o) we also this,
case.
structure
dene axiom
we 24 havenot
seems
a consequence
There now are,antoanalogue to(a),have
remember
relation (( been )1that
relation )addressed
theaxiom
(denotedtype seriously
by theup till sign) on (o
same
aeormal dis:and
uitionistic, o
this,
not (d)
(b),
to G(x)
we
just which
semantics ahave
(( :now.
a :(24). o)also
collection
relevance, 1to
25
assumeH(a)
)However,
remember(c)of:that
a etc.)
consequence amost
special
and :that
othe
is theG(x)
realistic
aentities
relation
Boolean
questiontype : but
(a),
candidates
(( itfor
that
Algebra.
what )1has(b),
)There ((
thealogic )1of)natural
certain
are, structure 24
(a), language
(( )1 ) assume
eaations
entities
sume
any-valued,: o a
(or (e)
(c),
but
that H(a)
certainandathat
(d), is::someopropositional
a(on
intuitionistic,
natural itBooleanG(a)
kind
has (d)
participate
)language) a
: Algebra.
lattice
a ofcertain
relevance,
: H(x)
logic oseems
structure
structure
and H(a)
: etc.)
Thereto 24
notwhich:are,
(b),
induced
and
to
have ((
also
the by )a, (c),
question
1 ) , (d),
and
consequence what
propositional
on . (b), ((
Corresponding
relation )1 ) to the
logic
(24) G H (x : o).G(x) H(x)
a,nistic,
and
oponents
natural : oto
most (f)
axiomG(a)
which offormal
relevance,
language
realistic :relation
also aetc.)
(or
candidates
() a
H(x) (e)
consequence
semantics and
certain: for
on athe
, :def
we oquestion
assume
natural
the define
relation
logic G(a) that now
language)of :
what (c), is
natural H(x)
(d),
an analogue
aseems
Boolean axiom
propositional
: not relation
to have(denoted
Algebra. logic There (c),are, (d),thepropositional
by same sign) logic
me
ve
lre :
a:now. that
logics
certain
induced
() (g)
(f),
However, is G(a)
((
byaG
(many-valued,
natural on, )2:most
Boolean (o)
language)
)
,H and
byH(a)
(f) (24).
Algebra.
realistic
means
on :seems
intuitionistic,
25
.:candidates
that There
()
[G(a)
not is ato
Corresponding
G are,
relevance,
have
forH(a)]
axiom
subattribute : etc.)
thetologic of(f), of
H. ((the
andnatural
It )question
is easily
2 ) seen axiomwhat
then that the type (o) with the
G(a)
stic,
ost
ue fbelattice :(h)
(e),
relevance,
found
realistic
relation in H(a)
acandidates
(g),
structure :dened
natural
(denoted :induced
propositional
oetc.) [G(a)
and (g)
language
by [G(a)
forthe
relation the
the
by G(a)
H(a)]
logic
same ,H(a)]
question
(or
logic
and :asign)
, :what
:certain
of
and
the
H(a)
natural
on on
above (f),:.
natural
(o) ((
by [G(a)
language) )2 )
Corresponding
explained (e), (g),H(a)]
seems
operations to : not
propositional
and (f),have
to (( )2 )the lattice structure o
logic
inherits
(24) G H (x : o).G(x) H(x)
ariously
nducedcertain
now :o (i)
(h),
an by[G(a)
natural
up ((
till
analogue
, type
x.[G(x)
,
def 1H(a)]
)and
language)
now. )oHowever,
on
relation
induced
H(x)]
(h)
: . seems aCorresponding
(denotedon :(o)
:most othis
not
realistic [G(a)
to
by
type have
theby (e),H(a)] (g),
candidates
sameto
the sign)
relation:propositional
(h),
for
on and ((the
the
(o) logic
by 1logic
) of (e),
)operations natural (g),
denoted propositional
by the same logic
symbols
elation realistic
x.[G(x) some
ompletely kind
(denoted
disjunctionanalogous H(x)]
candidates
of lattice
by :
24 Let usreasoning for
the
(o) (i)
the
structure
same logic sign)
x.[G(x) induced of on natural
H(x)]
by (h),
by : ((
(o)and ) on
) Corresponding (h), ((
to ) )
Htherefore use
goes Gthe through
asigns for disjunction Let usseen
therefore use thethe type
signs
(o) , , 1 . 1
.G We shall means
be a littlethat bit more is subattribute
denite about H. It .
thisofstructure isineasily
subsection then
3.1 below. that
uced
analogous we by dene now
reasoningAand an
completely on analogue
goes Corresponding
through
analogous relation for to
(denoted
disjunction
reasoning by
goes the
Let same
through
us sign)
therefore
for on
disjunction
use the by
signs Let us
and anddisjunction
not (o) only but for also propositional
forthus thetheoretic
correspond- conjunction and disjunction but also for theand andon the use
correspond- therefore the s
G(x), ,
H(x) , 25
We use .
the lattice symbols such
. as and explained for (o)
operations . relations semantic
with the defined relation and the
, , above operations
tion
tot es is
operations (denoted
easily
which seen
forassign on by then the
denoted same
that
G, the
by
H(o) sign)
type
expressions on ofwithour by the
object thuslanguage. The normal logical
theG, signs
Htype such as, for instance, and
only the
propositional toand type : not
conjunction ofonly monadic
thestructurecomplex
for and attributes
propositional
disjunction which
conjunction
but assign
also for to and : (o)
correspond-
disjunction the complex
butrelation
also for the correspo
(o) (o) (o)
inherits the lattice of the type o induced on this by the
logical constants of the metalanguage.
attribute
hose rations
on (x
ibutes the :of
Gtype
construction HH.(o)
and
o).G(x) ing= It inherits
is
26 x.[G(x)
and is
the
easily
operations
of monadic
provided
H(x) seen
the
operations H(x)] lattice
onby then
attributes
the
(23)
denoted
that
(structure
type
and {,the
which
the
(o)
by })type
ofwhose
theof
the
assign
monadic
(o) with
to G, the Hthus
attributes
construction 26: (o) is which
the complex
provided assign
by (23) to G, and H the
: (o) the com
26same symbols.
defThis kind of transfer of logico-semantic principles from one type to another is extensively used in Keenan
easily
dH ef the explainedseen
operations thenoperations that
H(x)]
attributes denoted the ( type
by
Gnow and the
HBoolean same with
inheritswhose the
symbols.the thus lattice
H(x)] structure of the
ourconstruction (Isitis{, provided
}) whose by a(23)
construction
testand the is provided
with by (23) and
(o)
logous = x.[G(x)
def
derivation for us
Let
Faltzs (1985)
. {, })x.[G(x)
=
return to
Semantics question:
for Natural possible
Language. 26 that expression
hat
ons the G isand
relationa subattribute
inherits
and the theof H.
operationslattice It isdef easily seen
structure
denoted of
by then
the
the that the
same type (o)
symbols. with the thus
Now, vation
re letfor
forin referentiality be
subsection . the analogous
denotation
3.1 and letderivation
expression
below. Q be which
passes forto.
its denotation.
is
our be tested
tests though foritreferentiality
is not the case andthat let Q:o? be its denotation.
Assume that
ily and seen
operationsthe then
above that
denoted explainedtheand type
by the operations
(o)same with symbols.the thus26
and
semantic inherits the lattice structure of the
eseand the
furthermore
denoted
about thisfor
expression by operations
structure Gthe
this Now,
which
and Hletis
expressions
is subsection
in indeed relations
be tobethe be3.1
the
possible. on
tested
resulting
monadic
below.the
expression Weforfromattributes
had items denoted
referentiality
which
already isseen
toand bethatlet
tested
by the
Q befor itsreferentiality
denotation.
expressions
:((o)) in26thisresulting
case. and Nowlet
fromQ be its denotat
slsnormal and
this type
such logical
as3.1inherits
by the
signs and the
relation such
lattice and
as,
for for structure
the
operations operations
instance, and of thedenoted
and
relations on arethe by the same
semantic items symbols.
the G test
eunconjoined
subsection H beLet
and expression , below.
, thefurthermore
sentences monadic
isenteringexcised G and
attributes the
from H denoted
tests be when
them the monadic
bythe thetest expressions
attributes
expression denoted
resulting
is excised by from the from
expressions
them resulting f
perations
ject language. denoted
forininstance,
operations The normal and bythis the
relations same
logical on signs symbols.
the such 26
semantic as,items
for instance, and are
us, bit
uage.
d sentences
for
more the
denite entering
the
examples about Maryunconjoined
from one type to another is extensively used in Keenan and likes
the
given
structuretests above).sentences
inwhen or
subsectionTom the
What detests
entering
test
3.1is below.
expressionthe tests
in the
when
is excised
examples the test
from
given expression
them
above). What
is excised
is from t
mal heoretic logicalsymbols signs 24. such such
We as,asfor
shall be instance, and
aassignment
little bit
and
for
more are about
operations
definite and relations
this structure onand the semantic
in above).
subsection items
3.1examples
below.
he nce,
ssue
bsection testsMary
is the
()
3.1 likes
truth
below. (thus,
and of () for
the or Tom
instance,
involve
type
,
emantic principles from one type to another is extensively used in Keenan and
guage. , detests
the Mary
inferential likes
Q :in o. theThe examples
or Tom
tests ()detests
given
() in the
involveWhat is
the given
inferential above). Wh
sions for of our object
operations and language.
relations The
on normal
the semantic logical items signs such as, for instance, and are
sruth
ation
eof simple
one of
Natural
the type
for the
between to
metalanguage.
type
attributes at
Language.
another issue
propositions
25. WeisGextensively
assignment is the
and
use theHlattice
truth
built
on
Qused : the
o.ofinThe
up
theoretic the
one
by
Keenan Q type
tests
hand and
symbols and assignment
()
the
such andas()
simple ,:involve
o. The
,Qattributes
and for Gthe
tests
and
operations H on
inferential
() and
and ()
one involve
therelations hand
on the inferen
logical signs such as, for instance, and are
en
e. by H
propositions
Q on and the the relation
other.the built
complex All
semantic upbetween
theby
attributes
items attributes
Q
fer of logico-semantic principles from one type to another is extensively used in Keenan and and
denoted G
propositions
the at
by H
simple
issue and
expressionsbuiltG
attributes
up
of H by
ouron G the
object
Q and H
the
other.
language. on simple
the
All The one
the attributes
hand
attributes
normal logicalG atand H
issue
signs on the one h
sthe of with complex
type these
Semantics
(o). such
attributes
and
attributes,
Inisorder as,
by Qfor
Natural for QG instance,
and has
Language.
Qused H
the
to
thus and
complex
build and
to Gbe
propositions
either are
Hand logical
attributes
on the other. constants
withG these H of
Alland the
the metalanguage.
Gattributes
attributes, H on Q has the
at issueother.
thus to All the attributes at i
be either
neolean type to another for extensively in Keenan
ests ).ype Inis, oorder
orof of fortype
course, are Q to
26. of
This tobuild
type
((o)). exclude
kind propositions
of The
(o). transfer Inpurpose
the order
latter with
for ofQthese
case.
of logico-semantic theto build attributes,
tests propositions
is, of course,
principles Qfromhasone thus
with
totype tothese
tobe
exclude either
attributes,
another theislatter has thus to be ei
Qcase.
extensively
w nd dothis,
type they weattain
((o)). have
of used The
typethis
to remember
in Keenanogoal?
purpose or of and In type order
that
ofFaltzsthe((o)). the
to understand
tests
(1985) type
is, The
Booleanof course,
purpose this,towe
Semantics of have
exclude
the
for tests to
theremember
Natural is,latter
of course,
case.
Language. thattothe type the latter c
exclude
attain
lf entities thisbut
propositions goal? How
that is In itdo
not order
has they
just a to attain
acertainunderstand
collection this
structure goal?
ofthis, 24 In
special weorder have
entities totounderstand
remember
but that it has this,
that athe we type
certainhavestructure
to remember 24 that the
ons 24
which andistonot thewhich justalso
operations aofcollectionpropositions
and of
a consequence special
(on is not relation
) entities
just a collection
participate butand thattoitof has
which special
a alsocertainentities structure
a consequence but that it has a certain
relation structu
umeperations
ongs. thatMost and in
isproponents which
a Boolean (on the ) of participate
operations
Algebra.
formal semantics There and
and to which
are, (on )
assume also participate
that a consequence
is a Boolean and torelation which
Algebra. also
There a consequence
are, relatio
wever,istic,proponents relevance,
alternative of belongs. formal
etc.)
logics semantics
Most
and theproponents
(many-valued, question assumewhat ofthat formal
intuitionistic, is asemanticsBoolean
relevance, assume
Algebra. etc.) that There
and the isare,
aquestion
Booleanwhat Algebra. There
anative
dcertain of logic logics is to(many-valued,
natural however, foundalternative
belanguage) in seems intuitionistic,
natural logics
notlanguage
to have (many-valued,
relevance,
(or a certain etc.) intuitionistic,
and the
natural question
language) relevance, what etc.)
seems not toand have the question w
: oa : o [G(a)
(i) (e), (23)
H(a)
x.[G(x) (g), (c)H(a)]
(h), : a H(x)]
propositional
((
axiom
(a) : :o) G1 ): :(o) G(x)(o)logic : (e),(b), (g), (( propositional
(h),
)1 ) (( )(a), 1 )logic axiom
(( )1 )
axiom axiom axiom
x.[G(x)
mpletely ugh axiom
a :for (h),
o analogous
H(x)]
G(a) (( (d):
: ) axiom
1(b)
(o)a.:H(x)
) H: us
oLet H(a)
:(o) : through (h), for(c), (( )
(d),1 propositional
) (b),usaxiom (( )1 ) use the signs
logic
: : disjunction :reasoning therefore
goes use the signs
disjunction . Let therefore
H(a)
H(a)] [G(a) H(a)]

[G(a)
(f), (( H(a)]
)2(f), ) :(( )(f),
2 ) (( )2 )
nalogous (23)
axiom : axiom (a) (e) G :
(a),
(c)a : ((o a : o
) G(a) G(x) : : H(x) : axiom axiom (c), (a),
(d), ((
use propositional )1 )correspond- logic
unction nd]r :disjunction
reasoning
not
and :only
disjunction Let
forgoes us(e),
propositionaltherefore
through
but also for
use
for disjunction
thepropositional
conjunction
the signs
correspond- and. Let disjunction us therefore but also the
for thesigns
() (o) 1 )

a) H(a)] . (g),
(e), propositional (g), propositional
(e), logic
(g), logic logic
perations G(a)
(a),
attributes axiom
(( : on(b)
)
which H(a)
(f)
1 the type H :
(b),
(d)
:
assign : (( a :[G(a)
to for o
)G,
of H
1 monadic H(a) H(a)] : : (f), (( ) axiom axiom (b),
G,correspond-
H (( )
tno) only
and :disjunction
for propositional but )also conjunction the )1:) correspond-
and attributes
the
disjunction complex which but2 assign
also for tothe : (o)1 the complex

) (o)

(o)() )
(o) ) )
H(x)] (o) (h), (( (h),
1 ) (( (h), (( )1 ) Chapter 3. Abstract objects of verbs 135
utesutes n})
a
(b), : G
thewhich o
whose(a),
((
H assign
type
(c)
((
)
[G(a) (g) )
construction
1
(o)
=
) a1
x.[G(x)of
):
(c),
(e)
to G(a)
o
H(a)]
monadicG, (d), a G(x)
H: :
is propositional
H(x)] o
:
provided

(o): H(a)
attributes
G(a)
( the : :
complex
by

{, logic
which
(23)

}) H(x)
[G(a) (e),
whose
andassign : (g),
the
H(a)] to
construction G, :
propositionalH(a),
: ((
(f),
is
(o) (c),
((
)
logic
the
provided1 )
(d),
)
complex
2 ) propositional
by (23) and the logic
goesgh asoning for through
disjunction
goes axiom
forthrough disjunction axiom
Let for usdisjunction
therefore axiom
Let us use
therefore
Let the ussignstherefore
use the signs
use the signs
whose (c),
x.[G(x) (b),
(d),
construction (d)
((
def H(x)]
propositional
(h)
H(x)]
)a :
axiom
1 is (
) (f)
.a
o : : o
provided
(o) H(a)
:
logic [G(a)
(). : by whose H(a)]
(23) and
. :
construction
the
(h), (( )
is provided
1 ) (b), ((
(e), axiom
(g),
by (23)
) 1 propositional
) and the logic
gous
propositional
nction nal
= x.[G(x)
def
derivation
conjunction
and(d), 62 axiom
disjunction
conjunctionfor
and . Klaus axiom
disjunction
but
{,
and Robering
also
}) axiom
disjunction
for but the also correspond-
for
but thealso correspond-
forthe correspond-
(a) axiom H(a)]
(c), (e) : (i)
a
propositional
:
(f),
(g)o
x.[G(x) (( G(a)
G(a) ) logic
:
)
: H(x)]
H(x)
H(a) : (o) : : [G(a) H(a)] (c), : (d),
(h),
propositional
(f),
(( ((
) ) )signs
)the logic
eyation
w,
monadic )(o)
ttributes
analogous
for
tested
: let be
of for .the
monadic
which
attributes
reasoning
referentiality
(a), expression
determines
((:attributes
assign which )(a), 1to
goes
)and the
G, ((through
which
assign
2
H let
set
which )(a),
1Q is
:[G(a) ) of
to be
to for
((
G,
assign be
its Hdisjunction
attributes
the tested
)denotation.
:1complex
to) G,:for FH (i.e.,
the : complex
Let us therefore
referentiality
.F:(o))
(o) the such
complex andthat let use 1
Q(F) be 2is itsadenotation.
true propo-
(f), A axiom
(( axiom

completely(f)
) ) (e),
(h) ()
analogous (g), a :propositional
o (o) H(a)]
logic (o) axiom (e), (g), propositional logic
):not only forG propositional reasoning
conjunction goes
and through
disjunction fortests,
butdisjunction
letalso for the Let
correspond-us therefore ofuse the the signs
2 be
dttributes
urthermore
]x) the (for
:
expression
H(x)]
whose referentiality
denoted
( (b),
construction and
whose which
sition. (( by Hand the
)be
Since is
(b),
construction
1 let
whose
is
) to
the Q be
expressions
((
provided be tested
,monadic
by )its
(b), assumption,
construction
1:) is denotation.
by for
((
provided(23)referentiality
resulting
attributes
[G(a) )1 ) is
and passes from
provided
by denoted
the(23) our and
andby bythe
(23) the
Qand aits
isexpressions
very
the denotation.
. special resultingsubset from
(e), (f),
(g), ((
(g)
propositional
(23) ) G(a) (h),
(i) (a) : (( logic Gfor H(a)
) 1: (o) H(x)] : H(a)] : (f), (( (h),) (( )
2axiom
}) {,axiom }) {, ) }) x.[G(x)
) (o) ) )
ns ses )G:when
nconjoined
(x) on :
denoted
and the H
the
H(x) type
and be by :the
sentences
test
(c),
2 not
the
lattice
(o) monadic
expression
(d), of
expressions
of only monadic
entering
monadic
propositional
(c), (d), attributes
is attributes
propositional
resulting
the excised
attributes.
propositional
(c), tests
logic
(d), denoted from
when which
Passing
propositional logic theassign
conjunction
bythem the
( testexpressionstoand G,disjunction
expression
) means
logic H that : resulting
(o) is the butcomplex
excised fromalso
1 for the
from them correspond-

(e), (g),
(h) a
propositional: o [G(a) logic H(a)] : (e), (g), propositional logic
.through (h),
Hthe= ing ((
(a),
for ) ((
Aexpression )
completely
operations disjunction )
H(x)] on (b) H : axiom G,andHus
detests
nfor sentences test
instance,
x.[G(x) 1
inMary
entering
axiom
the 1 ) likes
examplesthe analogous
axiomthe (
is
.
tests Let
type
excised
{, (o)us
when
given or(o)
axiom therefore
}) reasoning
Tomfrom ofdetests
whose
the
above). monadic
test
them usegoes
What theattributes
construction
expression through
signs
is in theisfor whichdisjunction
isprovided
excised
examples assignby from
given (23)
. Let
toabove).
them : therefore
the (o) the
What use the signs
is complex
sion
tested honjunction
is def to
which(h),
for
be
(b), (i)
((
referentiality
tested
is
(( to )x.[G(x)
(25) G
and be
)for) tested
referentiality
(c)
and afor H H(x)]
:let
o be
referentiality :and
G(x)
(o) its
Glet Q( and
denotation.
:
be H its
let Q be
conjunction
denotation.
. whose (h),
its denotation. ((disjunction
)(a),
1 ) (( )
unction
sue ce,
erivation
H(a) H(a)] : isMary
o.
[G(a) attributes
:the :in
The .

the
truth
for Let
likes
tests
H(a)]and
[G(a)
(f),
.
Gus

examples
of ()
(( 1
:the
)H
therefore
1disjunction
and
H(a)] or
type
)def= not
2(f),)Tom
given
() only use
assignment
x.[G(x) but
:((detests
above).
Q
the
for
involve
)(f), also
2 ) ((
H(x)]
signs
propositional
for
Q What the
:)2in
the o. )the correspond-
is
Theexamples
inferential
{, tests ()given
}) and
construction
and () above). involve Whattheis inferential correspond-
is but
provided
1 )also for
by the
(23) and the
ributes
onadic
be the
disjunction
dic A completely monadic
attributes
denoted
(c),
attributes ing (d),but Let byattributes
analogous
operations
whichalso denoted
propositional usthe(d)for expressions
therefore
assign on
the a by
denoted
:
reasoning
the o
to the
logic
correspond-G,
use
type expressions
H(a)
H by
resulting
the
goes
: the
:
signs expressions
through
of the from
resulting
monadic complex for resulting
from which
disjunction
attributes from (b),
Let us
assign (( therefore
to) 1 G,
) H use: the signs
the complex
The uth
on
a) ]and be of
tests
:between
the the the
analogous
H(a)] ()type
simple
expression .
:(e),The andassignment
propositionsattributes
derivation
(g), ()
which
right propositional
(e), involve
tobuiltGfor
is(g), Qtoand
left :. up
be o.the
direction H The
by
tested
propositional
(e), logicinferential
onQtests
(o)
(g), the
(o)andof one
for
propositional ()
the hand
referentiality
(25)logic and that
simple
says ()
logic and
theinvolve
attributes let QG the
set . isand
be its
closed H
inferential onunder
denotation. the one the hand (o)
oper-
when
the
entering tests
the
axiom the
when
test tests
expression
G, the H G when
(e)test H expression
athe : is
o test excised expression
G(a) is
:from excised them
H(x)
is from
excised
: them from them (c), (d), propositional logic
hich
yand assign
disjunction
and whose
attributes to not but only
construction :)monadic
also for for the
propositional
the
is )complex
correspond-
provided H(x)]
byconjunction
)(23)to( beand theand disjunction
whose construction but also Gfor
is the
provided correspond- by (23) and the
at
he
no)
H(x)] ore propositions
and
{,
QsimpleGand
})G: Now,
and
(o) theH
attributes
H on
let
(h), built
complex
be
ation the
(( be
the .Gup
other.the and
(h), by
1attributes
(o)
As
) expression
= H
All
regards
((
Q and
on the
x.[G(x)
1G the
the
attributes
(h), ) which
the ((Hopposite
one
simple
attributes and hand
1 is
)denoted G
attributes
at issueHby
direction on
tested
{, the Gthe
}) ofand
for other.H on All
referentiality
expressions
(25), assume theresulting
the one
that and hand
attributes
let
from Q be
and issue
itsthat denotation.
etests
es ] kes
or:
ing
construction Tom
which
(f),detests
operations axiom
or
assigninTom
(( is the
)
onto
provided
2 detests
examples
) G,
the(f) in H the
type :
by(o)
def :examples
given
(23) () inthe and the
ofabove). examples
complex given
monadic
the Whatabove). given
is What
attributes above).
which is What assign isaxiom
to G, H : the complex
Gositions
goes
gh ned etype
complex
asoning
HLet onwith
sentences
(o). theattributes
analogous
furthermore other.
IntheseGorder H.
entering All
attributes,
for
derivationGfor
Since Q the
and
to H(o) attributes
Hand
build
(o)
testsQfor ishas
be G the
when
a. at
propositions
thus
lattice, H
monadic
the issue
on
to Gbe H
the
test eitherother.
with
attributes
expression
is just these
the the
Alldenoted the
attribute
the isattributes
attributes, by
excised G,Qthe has
and at issue
thus
expressions
fromsince to
thebe
(o)
them either
latter resulting
(by from
ype
ose o.for
nment
attributesthrough
The disjunction
assignment
(e),
construction goes
Qtests o.for
:axiom
(g),
G through
()
The disjunction
propositional
H Qis. tests
and
:provided
o.
(g) Let The
()
() usdisjunction
G(a) therefore
tests
involve
.and
logic
by :Let ()
H(x)]
(23)() us theuse
H(a)
.
and

therefore
and
involve
( Let the
inferential
the :()us signs
therefore
theuse
involveinferential
[G(a)
whose signs
the use
H(a)]inferential
construction signs
:thus
(f), is ((either
provided )2 ) by (23) and the
ns
pe
of o In
tance,with
the
o
be order
or
thetestedthese
tests
of
Mary for
type
unconjoinedis,
Now,
for attributes,
likes
Q ofto
assumption)
((o)). build
course,
referentiality
let = x.[G(x)
disjunction be propositions
sentences
Qorand has
The
to
the belongs thus
exclude
purpose
expression
and
Tomdisjunction
let
detests
entering
toto Qwith
be
the
, of
be
the either
which
for these
latter
the
its {,
left tests attributes,
case.
denotation.
to is
})
inalso the to
right is,be
when
examplesof course,
tested
direction theQ has
for
test to to
exclude
referentiality
expression
given above). What is
of (25) be
requires the
and
that islatterlet
H
excised Qcase.
, be
too. its
from denotation.
them
propositional
nction
ons nal
nd up the byconjunction
built and
Qsimpleand
up disjunction
(a), byconjunction
the (( )and
attributes def
)1(h)
1simple
and but
theG aalso
attributes
simple
and for
Hbut the
Galso
attributes
on and
thecorrespond- but
H
one H(a)]G the
on hand
and correspond-
the for
H one the
on hand
thecorrespond-
one hand(e), (g), propositional logic
is,(h), (( Q In:of oorder [G(a) : this,
62 The ) ) Klaus
exclude Robering
type
monadice icdonderstand
tests truth
tributes
they
analogous
eferentiality
attributes
(thus, of
((o)).
of
attain
of
Let the
monadic
which
this,
course,
for
attributesderivation
and
denoted this
furthermorewe
instance,
Hence
type assign let goal?
have
purpose
to
assignment
Q by for
be
is Mary to
theG
its
also
G, . remember
H
theH
denotation.
and likes
expressions
closed
Q : o. to
tests
latter

be
under
The
G,H(x)]
understand
that
the is,case. of
resulting
or
Hthe:complextests the
Tom
monadic type
course,
superattributes.
G, ()
H detests
from towe
attributes
and () have
exclude
Together to
denoted
in
involve the
the remember
with latter
theby
examples the
closure case.
inferential that
expressions
given
under
theabove).type this resulting
What isfrom
rsfor nd GG
attributes
(o) HH G
(b),
and on. G ((
the H attributes
which
H
other.
and )us(i) )to
on G the assign
All H which
other.
the on : (o)to assign
attributes
the All the
other. to at
attributes
: All
(o) issue the
the : complex
attributes
attoissue
(o) the complex at issue (h), (( type )1 structure
itx.[G(x) (o)
tain and disjunction
this
this, goal?
we
have InLet order
to 1that remember to
therefore by understand use
that the signs 24 ) 24
ween propositions
noted special
tests Now,
referentiality
at by
when entities
issue
propositions let
the
the is be not
expressions
the
unconjoined
is means test
the butthe
and just
truth
built expression
let
expression upaof collection
has
be
resulting
sentences
is
the a its awhich
filter
type and certain
denotation.
is from
in ofthe
excised
entering
assignment
the this,
isspecialtype
structure
to
lattice
simple bewe
from
the have
entities
tested
(o). tests
: them
attributeso. for but
when
Passing
The remember
G that
referentiality
tests the
also
and it(
H has
test
() that
on aandthe
certain
expression
)andrenders
the ()let
one be
involve
hand its
is
furthermore denotation.
excised
the from them
inferential

x) H(x)] Q Q Q Q
nuild
})
ns alor(
sitions Qwhose
and
propositions
entities
is to {, with
build
disjunction
not
construction
})
just
(these
(c),
A
but
whose

a
(d),
propositions
{,
completely
that with
but
collection
})
it
construction
propositional
attributes, also
has
whose
theseis providedwith
analogous
a for
of
construction
attributes,
Qspecial
certainthe hasthese isthus
logic byprovided(23)
attributes,
Qtohas
reasoning
correspond-
structure
entities be is
and24 provided
bythe
thus
either
but Q(23)
goes to
has
that beand
throughit
by
thus
either
has
the
(23)
tofor
a be and either the
disjunction
certain structure Let
24 us therefore use the sign
dom icipate
ich
test Let
denoted the
thedetests
relation and
complexoperations
furthermore
expression (thus, to
bybetween the
to which
for in
be
attributes G
expressions
a
the and
isinstance, and
veryalso
excised H
examples
propositionsG a
special
be(on
consequence
Mary resulting
Hthe
from and
) given
filter. participate
likes
monadic
them
builtG from
above).
Passing
up relation
H by on and
attributes
orQthe
this Tom
Whatand to
test
other. which
detests
denoted
is
means
the simple also
Allthe a
byattributes
in consequence
thethe
attributes expressions Gatand
examples
. issuerelation
resulting
Hgiven on the above).
from
one What is
hand
purpose )).
.ics
tes the Thetests
which ofpurposetheof(23)
axiom
is,
assign
tests
andcourse,
of
to is,
the
G,
(a) of
H tests
to
not G
:course,
exclude
only : (o)
is, of
the
for tocourse,
the exclude
complex latter
propositional tocase.
the
exclude latter
conjunctionthe
case. latterand case. axiom but also for the correspond
disjunction
eerations
nt gs.
(o).the and
the
in Qassume
Most
test
the
and
In to
o.order
which
:unconjoined
at
byThe and
proponents
expression
examples that
issue
Qfor tests
andalso
Qis(on is
sentences
given
()
the
the
to aabuildof
) consequence
Boolean
is participate
formal
excised
above).
truth
and
complex entering
()
of
propositions
(o) Algebra.
semantics
the What from
involve
attributes relation
typeand the
isthem toThere
with G
assignment
the which
assume
tests theseH are,
when
inferential also
andthatQthe G:ao.
attributes, consequence
is
test
H
The aQ Boolean
expression
on tests
hasthe () relation
Algebra.
other.
thus toand
be is ()
All the
either There
excised involve are, the
from
attributes them
at inferential
issue
oal?
sion hrder(a)]
derstand
ested is to :to
In
whichfor
be (f),
understand
order
this,
referentiality
tested ((
towe
isisrelevance,
to
(26) G be
for is )tested
understand
have this, )(b)
referentiality to
we
and for H H
remember
lethave
this,
:by be
referentiality
Q(o) we
to its
and remember
G have
that
let denotation. and
the to

be Hremember
type
its that the
denotation.
let be
. its type
that the type
denotation. axiom
yhose
proponents
sume
ntuitionistic,
ver,
sts of(thus,
Q() and
typeareconstruction
that
alternative
in for
the
and
ofthe ing
relation of
instance,
type
((o)). examples
()
simple aoperations
formal Boolean
logics
(o). involve
between
The 2provided
Mary semantics
attributes
In (many-valued,
given etc.)
order
purpose on
Algebra.
the likes the
and
propositionsG
above).
inferential
for of
andtype
(23)
assume Qthe
the There
Hor
What
to and
question
on
built
build
tests
(o) Qare,
that
intuitionistic,
Tom the
is of
the up monadic
is
what
detests
one by
propositions
is, of 24
QBoolean
acourse,
Qrelevance,
hand
and attributes
inAlgebra.
the
with
to24 the etc.)
simple
these
exclude which
examples and There assign
thegiven
attributes
attributes,
24 the latter are,
question
G tohas
case.
Q G,thus
above).
and Hwhat
H :on What
to
(o) be the
the onecomple
is
either hand
ributes
onadic pecial
be st
ction athe collection
ofentities
monadic (e),
special
attributes
denoted attributes(g),but
ofby propositional
entities
special
that
attributes
denotedthe G (c) it but
has
expressions
H by a
entities
denoted :
that
thea o logic
certain
itbut G(x)
has
expressions
by that
resulting the :
structure
a certain
H(x)]it has
expressions
from
resulting
( astructure
certain resulting
from structure
whose from (a), (( )is1 )
construction provided by (23)
yative
of
nistic,
ge eat
ple Hlogic
(or
tests
attainissue logics
arelevance,
attributes
and
of iscertain
()
typeG isto
and
this (many-valued,
theobe
and
by
goal?G
H or
Since,found
natural
truth
Q ()
and
on etc.)
of andthe
In in ofHorder
type in and
language)
involve
the
on
athe natural
other.
lattice, intuitionistic,
thethe
typetothe
complex
((o)). = All question
one language
understand
x.[G(x)
we seems
assignment
inferential
hand
the
The
have attributes
attributeswhat
relevance,
not
purpose

both (or GtoaG
this,
Q G:have
certain
o.
at
we H
issue
of H
The etc.)
have
the {,and toand
natural
tests
})
and GH G
()the
language)
remember
is, Hofand question
H
on the
course, ()
that
and seems
other.
towhat
involve
theexclude
since not
(25)the
All
type toalready
the have
inferential
the attributes
latter case. at issueth
and
cipate nwhen andthe
entering
) tests (on(h),
participate
and
the to
)which
when
the
test ((
participate
tests and
expression
the )when also
1 to
test )(d) which aand
expression
the adefis :test
consequenceto oexcised
also which a
expression H(a)
consequence
is also
from : athem
relation
excised consequence
is relation
from
excised them relation
from
them (b), (( )124 )
daddressed er,
tionsoaon be
certain
simplemost
relation
ropositions
for the
Howfound are
not
referentiality natural
realistic
attributes
is other. between
do in
seriously
with
ofthey
just
analogous natural
All
type
implies language)
acandidates
G
these
attain up
propositions
the
collection
and and
(o). language
tillthis
attributesH
attributes,
derivation
that let In
Q now. seems
isfor
on
beorder
goal?
of (or
the
However,
the
built
its
closed at
special
for not
alogic
one
issue
for certain
has
Inup
Qdenotation.
under
. Qto hand
by
thus
to
order have
entitiesmost
of Qnatural
build natural
totoand realistic
be the language)
propositions
but
superattributes, either
understand that simplecandidates
it thehas seems
attributes
with
this, a we
right for
these
certain not
tohave the
left todirection
to have
logic
Gattributes,
and
structure H on
remember ofQ natural
ofthe
has
(26) one
thus
that is hand
theto betype either
cs
etests lkes
of
or semantics
forassume
formal
Tom detests
disjunction that
or semantics
inassume
Tom the isdetestsathat
examples
Let in(e)
Boolean
assume usthe aexamples
isgiven
therefore :athatoin
Algebra.
Boolean
G(a)
the
above).isexamples
use athe
given : What
There
Algebra.
Boolean are,
above).
signs H(x)isAlgebra.
given There : above).
What are, isThere What are, is . (c), (d), propositional logic 24
ose
es seriously
ture
age
hoperationsst andHrealistic
these
denotedinduced
ofon assume
bythe the
Q up
attributes,
of candidates
and
by till
by
some
other.
ofpropositions
teststype Now,and
the now.
the
already .
, All
orkind
complex
is,oQexpressions
of let
, has However,
andfor
the
course,
(onof
implied of
isthus
be the
type
not
) lattice
the on
attributeslogic
attributes
tobyto
just
participate most
be exclude
expression
resulting
(25).
.
((o)). aeitherof
structurerealistic
Corresponding
at G
collection
The natural
issue
The
and the
fromwhich
left H candidates
induced and
purpose
tolatter
to of
which
right toG
case.
isspecial byof beH
also
todirection,
,theforon the
a tests
entities
tested
, and
the logic
consequence
for
however, other.
is,
but onofthat of Allnatural
Corresponding
course,
referentiality it the
requires hasto
relation attributes
a exclude
and
the let Q at
certain
filter to
theissue
be
to latter
structure
its case.
denotatio
valued, uitionistic,
nment
ype (many-valued,
o. The intuitionistic,
assignment tests:, o.relevance,
()
The intuitionistic,
tests
and
: o. (f)
relevance,
etc.)
()
The
() and
tests : relevance,
involve
and ()the
etc.)
()
() question
the and
involve etc.)
inferentialthewhat
() question
theand
involve the question
inferential whatinferential
the what axiom
ntion and disjunction but also for the correspond-
me duced some Q
by kind of and Q
lattice on structure Corresponding induced by to
guelation
to with
s,
stare
the of relation
of
in
proponentsthese
understand type
course,
test
whichon
How Let (denoted
attributes,
expression
,,
(o). todo
the
beofwe
this, In
exclude
an dene
they we
operations
formal
furthermore order
ultrafilter,
Q by ishas
attain
have now
the
the for
semantics
. excised Gthussame
to
this
latter
an
and
i.e.,
Q and toto
remember analogue
sign)
build
goal? be
case.
H
afromassume
filterbe either
(on Inon
the
them
which ) relation
propositions
that
order
that by
the
participate
monadic
(o)
, does
, toand (denoted
istype awith
understand
not onthese
and
Boolean
attributes
possess byCorresponding
.to the
attributes,
athis,
which
Algebra.
denoted
proper same
we also havesign)
There
by athe
refinement.
Q hastoon thus
remember
consequence
are,
expressions
(o) to by berelation
either
that thetype
resulting fro
al
ons
nd eup ibutesin(or
5lationlanguage
natural
the
by builtaQsimple
certain
which uplanguage
and (or
bythe natural
aQsimple
attributes
assign certain
and (or
to (g)
language)
theG,Ganatural
attributes G(a)
certain
simple
Hand : : natural
Hlanguage)
seems Gon
attributes
on
and
thethe H(a)
not language)
Hseems
one
complex toG: hand
on have
and not
the [G(a)
Honeseems
to
on
24 have
handnot
the H(a)]
one to hand : (f), (( )2 )
have 24
nnhis,
estsernative ofof we
we
is,
type
special(denoted
belongs.
in thethe
, of dene
have
o course,
logics
or
of toof
entities
Most
examplesnow by
remember
type
propositions to
(many-valued,
unconjoined the
Summing an exclude
but
proponents analogue
same
that
givensentences
((o)). that
is up sign)
itthe
not the
The
has
of relation
latter
intuitionistic,
(o)
above). What
our just type
a purpose
formal certain
results, a case.
entering
(o) (denoted
collection by
semantics
weis of the
structure
relevance,
can by
sayof tests the
special
assume
thus:etc.)same
is, If of
that
and
an sign)
course,
entities the is
expression
on
but
a to
question exclude
that
Boolean
when the test expression is excised from the
(o) byit
what
passes has the
Algebra. a
our latter
certain
tests case.
There structure
are,
r,s)pnd . attributes
However,
most
Gtill
whose G now.
H realistic
H and However,
G
on most
construction G the candidates
H
realistic
H
other.
and most
onis(h) G the
All
provided
a
candidates
realistic
forH :
other.
the o
the
on logic
attributes
by theAll [G(a)
candidates
for
other.
(23) the
24 type the
of
and at H(a)]
natural
logic
for
attributes
All
issue
the the the:
of logic
natural
attributes
at issue of atnatural
issue (e), (g), propositional logic
nd he How
ties
participate
is this,
to
testsbutbe do
however, we
inthey
that
found
H() have
which
and
(thus, itand attain
has
tofor
alternative
for in tothe
naturala:remember
which
referentiality
() this
certain
operations
instance,involve goal?
also
logics
language structure
aand
Marythat
the In
consequence order
(many-valued,
does (orthe
and
likes
inferential to
a nevertheless
certain
(onunderstand
relation
)or participate
intuitionistic,
naturalTom not this,
language)
detests
denote we
and anhave
to
relevance,
seems in to
which
object, the remember
etc.)
not alsotoand
examples
then ita denotes
have the that
consequence the
question
given type
anabove). relation
what What
or ice
nd re G
itions
ild Q induced
of lattice
structure
propositions
to with
build by structure
induced
these (x
propositions
, withattributes,
, and o).G(x)
(i)
by induced
these
, on
with
attributes,
x.[G(x)
,
Q and
has
. by H(x)
these Corresponding
thus
, on H(x)]
and
attributes,
, Q to has
. be 24 :
Corresponding
thus on
either Qto
(o) tohas
. beCorresponding
thus
either to be either to (h), (( )1 ) 24
mantics e to
ed entities
of
whichkind
seriously
simple propositions
but
assume also
belongs.
of def
at
attributes
:is, that
logic
up
o).G(x) a
issue that
till
ultrafilter it
consequence
Most
is is
has
now.
G
is to
not
the
and
H(x) is
be a
in certain
just
proponents
a found
However,
truthH
theBoolean on a
relation
lattice
of collection
in
the structure
the Algebra.
of
natural
most
onetype formal
realistic
attributes.
hand of
language
assignment special
There
semantics
candidates
Being are,
(or entities
Q ana :assume
certain
o. for but
ultrafilter,
The the that
that
natural
tests logic it
however,
is
() has a
language)
of anda
Boolean certain
natural
leaves () seems
only structure
Algebra.
involve two not to
the There
have
inferentiare,
ue
ne an
)).urpose
the relation
analogue
now
The (xof
tests an
purpose (denoted
the analogue
relation
of tests is, by(denoted
relation
thecourse, same is, (denoted
byofsign) the same onby (o) the
sign) same
by on (o) sign) byon for by
Gcourse,of the ofto
tests exclude to the
course,
exclude latter to case.
theexclude latter the
case. latter case.
A completely analogous reasoning goes through (o)
d,
sted and
that H
It
sume
def
is
H meanseasily
inintuitionistic,
which
to
been
for
iswhich
some
on athe that
theseen
Boolean
however,
addressed
kind
referentiality
relation also then
operations
possibilities
other. isalternative
of aAlgebra.
abetween
relevance, that
subattribute
consequence
lattice
seriously
All and the the
let
for and
etc.)type
structure There
logics
attributes
Q up
. be and
till
Either
propositions of
its
(o) H.
relation
(on are,
now.
induced with
It
(many-valued,
the
denotation.
atit) is
issue ais the
easily
participate
question
However,
built byup
principal thus
, seen
bywhat and
most and then
,intuitionistic,
ultrafilter
Q and to that
which
realistic
on the or. itdisjunction
the alsotype
relevance,
candidates
simpleCorresponding
is a free a(o)consequencewith
etc.)
for
.
ultrafilter.
attributes Let the
toand
the
G us
logic
and therefore
A thus
therelation
prin-H question use
of natural
on the one thehan
what si
oal?rder
derstand to
G In understand
order
this, towe understand
have
and this, to we H. remember
not have
this, only to
we for remember
that
have the
propositional to remember
type that the type
conjunctionthat the type
and disjunction but also for the correspo
ed
butes
sasily
perations that
me onrelation
belongs.
relevance,
guage
with that seeniswe
(or
language
denoted
, these athen
kind subattribute
Most
is
a
andandand
deneaof
etc.)
certainby
cipalthat
Boolean
assume
by
attributes, the
proponents
and
logic
thenow
Q the
inherits above
natural
ultrafilter
and the
isQ type
of Algebra.
some
an
expressions the the
question
tohas be of
analogue Itlattice
explained
language)
kind
isthus
(o)complex isfilter
aformal
found witheasily
There
of
to what the
structure
lattice
in
relation
resulting
bewhich seen
operations
semantics
seems are,
natural
attributes
either thus then
not of the
language
structure
(denoted
from
contains to
G

24
that
and
assume
have Hthe
ainduced by (or
minimal
and type
inherits
that
the
24 aGsame
certain
by
the
is,
H
element
(o)
with
asign)
Boolean
on
,
24 ofthe
lattice
natural
and
the on the thus
structure
Algebra.
language)
other.
on
underlying
(o) .by
All ofthe the
There
seems
Corresponding lat- are,
not totoathave
attributes issu
ons
ond pecial
ction
st athe
H(x)
o).G(x)and
induced collection
ofentities
and
the special
operationson H(x)
above ing
inherits
this but
ofexplained
entities
special
typethat
operations
denoted the by itlattice
but
has
the entities
by on that
operationsastructure
relation
the certain
the itbut
same has
type and thatstructure
aof
and
symbols. certain
the it has
the of 26 astructure
monadic
inherits
operations certain the structure
attributes
lattice
denoted which
structure
by the same assign
of the
symbols. to G, 26 H : the comp
wever, stic,
ain
hen
however, the
ests is, of are
relevance,
natural mostbeen
relation
test alternative
language)
realistic
expression
course, tice etc.)
addressed
of to exclude
on
together and
logics
seems
candidates
(o). In the
, the
seriously
we is
with (many-valued,
not
dene question
excised all
order to
for up have
the
now
larger
latter till
from
for Q what
logicnow.
an
(o)
elements.
case. intuitionistic,
of
However,
analogue
them natural
In the most
relation
case relevance,
of realistic
(denoted
our
to build propositions with these attributes, Q has thus to be eith etc.)
lattice candidates
by (o) and
the the
same
of for question
the
sign)
attributes, logic
ona what
(o)
(o) of natural
by
on nand ipate
)
operations
this participate
and(on25to
type ) which
denoted
by participate
attributes
the andrelation also
to G
by which
the aand consequence
Hand
same toalsothe which aoperations
symbols. consequence
also
relation
26 a consequence
H(x)] denoted( relation by relation
the whose
same construction
symbols. 26 isexclude
provided bylatter
(23) and
listic
ests
and tructure certain
kind
this, of natural
candidates
(24). logic
induced
language
in
we of thehave is
type language)
to
for
by
examples
minimal to be
o the
assume
, found
logic
attribute
remember
or , of and seems
given some
type in = of Fnatural
on not
natural
above).kind
x.[G(x)
would
that
((o)). . to
the language
have
be lattice
Corresponding
of What
type
The an attribute
purpose(or awhich
isstructure certain
to{, induced
})
of below. natural
applies
tests by to
is, language)
exactly
,of, and
course, one seemson
object,
to .not thus to the
Corresponding have to
cas
shall
ute tlsH ture
subattribute
is of
ofsemantics
assume in
be
easily
formal
subsection
a
H. little
(xIt
seen
thatis :
bit
of
semantics
assume o).G(x)
more
3.1
H.
easily
then is below.
It a denite
that
seen
is
thatBoolean
assumeeasily
the
H(x)
then about
is type aseen
defthat
that this
Algebra.
Boolean
(o) the
thenstructure
is with
type
a that
There
Algebra.
Boolean in
the
the
(o) subsection
thus
type
are,with Algebra.
There the
(o) 3.1 thus
with
are, There the thus
are,
dnalogue
l,.use
le
realistic
been
by
The
entities
and
subsection
bit
def
the,
more
addressed
tests
(24),
lattice candidates
and
relation
the but
denite relation
()
How analogous
fortheoretic
G an
that

operations seriously
on
(denoted
attribute
and
do it .
Hsymbols for
they ()
has
the
Corresponding
on
derivation
a toby
attain
andstructure up logic
involve
certain
,
relations be
such till
we
the this now.
dene
same
identical
inon
:relevance,
o).G(x) of
for
the
as natural
goal?
structure
the toHowever,
sign)
nowwith
inferential
. semantic
and
24
In
H(x) an
on
some
order most
analogue
(o)
items to by
certain realisticrelation
understand a:o. candidates
Then, (denoted
this,
for operations and relations on the semantic items of we for
course, by
have the the logic
there
to same would
remember of
sign) natural on
that the
(o) by
typ
plained
above
erations
uitionistic,
alued,
(many-valued, 3.1
explained
operations
and
intuitionistic, below.
relevance,aboutoperations
inherits
intuitionistic,
and this
relevance,
etc.) the
(x inherits
andlattice
and the
etc.) subsection
the , ,
structure
inherits
question lattice
and etc.)
the 3.1
the
ofwhat below.
structure
the
lattice
question
and thesigns structure
ofwhat the of
question what the
dnoted
he euced
ans language
(denoted
the
andnormalthat
theoretic
by
for by
simple G (24).
tooperations
which by
expressions
logical
, symbols assume
is, aof25and
the
subattribute
Now,
hardly
attributes
also
signs same
propositions
and
of some
aour
such on
be let
def
relations sign)
Gobject
aas
consequence
such . kind
and
,be
as,Corresponding
of
difference on , on
H.
the
Hthe
is
language.
for ofinstance,
and on
not(o) lattice
Itsemantic
is
expression
between
the
justby
easily
relation
The for astructure
one to
normal
and
items seen
which
saying
hand
collection
operations
26 then
logicalinduced
are is to
that
and that
of be by
the
denotes
special
relations
26 testedtype a,for
onentities
such
, and
as,
the (o)
26 and
for
semanticwith
but onitems
referentiality
toinstance,
say the.thus
that Corresponding
it
andand
denotes
has let
a are Q. be its
certain tostructure
denotat
edeelation
al ical by
the
in
(or
languagethe operations
natural
a
constants and
relation
certain the
language
(or
of operations
natural
thea denoted
and
certain
metalanguage. the
(or language)operations
by
natural
a denoted
certainthe same
language)
seems bydenoted
natural symbols.
the not same
language)
seems
to by have the
symbols.
not same
seems
to symbols.
have not to have
ation
mal dion x) ume the
G
essions
logical
relation
G (denoted
and
H
that on
ofsigns
our
H
thein
(24) is Let
the
by
means
above
Thus
which
objecta on other.
Boolean
such
G the
furthermore
our
language.
, the
as, we
that
explainedsame
His question
All
for dene
G
operations
Algebra. sign)
the
instance,
The is a
G now
operations
andon
subattribute
whether
attributes
normal Thereand
an
H and
logical
(o)
:candidates
o).G(x)
beanalogue
at
by
are,
theof
and
non-referential
issue
(on
signs
are H.
monadic
H(x)

) relation
It
such is
inherits easily
participate
as, for(denoted
the
attributes
expressions seen
lattice
and
instance, then by
can
to andthe
that
structure
denoted pass
which thesame
by
our type
of
the
also
are sign)
the
tests
aexpressionson
reduces
consequence
(o) with (o) the by thus
resulting
relation fr
es .ps,tkind However,
most
till
offrom the now.
of25 realistic
onetransferHowever,
type
metalanguage. most tocandidates
of realistic
another most
logico-semantic candidates
realistic
for
extensively theprinciples
(x logic for inthe
used of from natural
Keenanlogic
onefor the
and
typeof logicnatural
to another of is natural
extensively used in26Keenan and
ure ed ions
enite
nistic, tzs
(24).
this on
in dened
withthis
subsection
(1985)
structure
about
relevance, type
these
this
Boolean
relation
the
in
belongs. to
3.1 by
attributes,
structure
subsection the
unconjoined
thebelow.
etc.)
Semantics
problem
Most
inrelation
andsubsection
3.1 def Q the has
below.
proponents above
and
sentences
whether thus
question
Natural
3.1 the explained
to
below. operations
there
of
Language.be entering
whatformal could
either operations
denoted
the
be
semantics tests
expressions bywhenand
the
assume same
the inherits
denoting
that testsymbols.
is the
expression
freea lattice
ultrafilters
Boolean structure
is
Algebra.excised
in of the
from
There th
ar
ymbols ice
nd
anguage. re
nsfer
uch one
and induced
of
structure
astype lattice
of
such
for by
logico-semantic
to another
and structure
induced
,
operations
as , is for and
andand by
extensively induced
principles
operations ,
relations
on
for
forused
, . and
by
from Corresponding
operations
and on in
,one on
Keenan
,
relations
the and
type
.
semantic
andon Corresponding
andthe onsemantic
torelations
another
items to.on Corresponding
is extensively
the items to used
semantic items to
in Keenan and 26
he
Boolean
f
a .
H. tests
certain
(24)
It
type , is G
is,
,easily
G o
natural
Semantics
induced
of (thus,
however,
the
H Hseen
course,
, means
for
lattice
language)
, onthen
Natural tothis
alternative of that
instance, type
monadic
exclude: seems Gthe
o).G(x)
Language.
is bytype
Mary
logics a
the
not subattribute
the
attributes.
latter
to relation
likes
(many-valued,
(o)
H(x) have with
We
case. the
ofhave
and H.thus
the
orbelow.Tom It
to isadmit
easily
operations
detests
intuitionistic, that seen
denoted
there then
in could
relevance, the that
byexamplesthe
the
be type
same
such
etc.) and ultra-
(o) thewith
symbols.
given above).
question the thusWha
wh
aur
ue ne
nguage. little
an relation
normalanalogue
object bit Themore
now an(denoted
logical
language.
normaldenite
analogue
relation
signs
The
for logical
about by
normal
such (denoted
relation
(x thethis
signs
as, logical
for
structure
same (denoted
by
such
instance,signssign)
the
as, in
forsubsection
same
on by
such
and
instance, the
(o)sign)
as,
3.1
same
forby
are
on
instance,
and sign)
(o) by
on
are and (o)
by
are
edseen
erstand
stattice then
operations
realistic this,
theoretic that
dened kind at
we
candidatesthe and
filters
issue
have
of type
relation
def logicand
is toinherits
for the
(o) hence
remember
isthe towith
and
truth be the
logic thethe
expressions
of
found thus
lattice
above
the
that
of type
in structure
the
natural explained
whose
natural assignment
type of
language operations
denotation the Q :(or fix
o. a such
The
and
certain an
tests inherits
ultrafilter
()
natural and the (in
()
language) lattice
the way
involve structure
seems the
not of the
inferen
to hav
language. 24
We shallsymbols be a little such bit more as , denite, andabout this for operations
structure26inand relations3.1
subsection on below.
the semantic items
26
on sily and G and seen H then
inherits
the type means
operationsthat
o the induced
explained
relation that
the
lattice G
type
denoted
between is
on
in a
structure
this subattribute
this by with
type
subsection).
propositions theof the
the
by
same thus
of
the H.
symbols.
relation
Free
built 24
It is easily
and
ultrafilters,
upKeenan by seen
the and then
operations
though, the that
have
simple the
denoted
a type
special
attributes by the
status G withsame
since
and the thus
symbols.
Hlogic
on the
ecial
duced
ic expressions
ico-semantic
from
principles entities
one
25 by
Wetype ,of
usebeen
from but
our
principles
,to and
theanother
one that
addressed
object
lattice type
from it
on
is tohas
language.extensively
theoretic one
. (o)
another atype
seriously certain
Corresponding
The
symbols is
toused normal up
extensively
another structure
intill is now.
logical
Keenan
such to signs
extensively
used
as , andHowever,
in
,
and such
used Q and
in most
as, for
Keenan
for realistic
instance,
operations and and candidates
and are
relations (o) on forthe the
semantic of one
items naturh
H(x)
o).G(x)
tants
sstructure
dened
and of the H(x)
metalanguage.
relation 26 exist
pate lation
Natural
antics tions
guage. denoted
for to
(denoted 24inherits
Language.

Natural
denoted and
which
language by
they
by by the
by
Language.
the
can
also and
the
expressions Q same lattice
assume the
only
and
a
same above
the
consequenceof structure
symbols.
provedsome
sign)
our complex explained
objecton to
kind(o) of
relation the
attributes
of
language. by
by operations
lattice use
Theof G
structure
normal H
aspecial
and
and induced
logical G inherits
set-theoretic
signs
H by
on the
such
, the lattice
principle,
as,
, other.
and structure
for instance,
onnamely
All . the of
Corresponding
and the
attributes
are at is
transferlogical inofsubsection
We shall be
logico-semantic 3.1a little below. bit more from
principles denite one about this
26type
structure
totoanother in subsectionused
is extensively 3.1 in below.
Keenan andthat only 26
perations
type
assume o induced
25denoted
that the are
constants
Zorns
is on
of
relation a by this
type Lemma
Boolean oftype
the the same
on
(o). metalanguage.
by
which In
Algebra.
, symbols.
wetheorder relation
isdene
equivalent
for
There now
Q and
to
are, build
an the the operations
analogue axiom
propositions of
relation denoted
choice.with We by
these
(denoted the
conclude same
attributes,
by the symbols.
same Q has thusonto(o)
sign) be eib
stion , 3.1
, and
26H.below. We for
use operations
the lattice and
theoretic relations symbols on the semantic
such as ,items , and for operations and relations on the semantic items
ute t5) subattribute
isBoolean
of
easily This Itseen isofeasily
Semantics
kind H.
then
of25
referential It that
transfer
for seen
is easily
the
Natural then of
expression type seen
that
Language.
logico-semantic(o) the
then
will with
type that
pass thethe
principles
(o) our thus
type
with
tests from the
(o)or thus
onewith type
expressions the to thus
another
whose is extensively
denotation used
have in aKeenan and
rtionistic,
ubsection
e.operations
24The normal
We explained
shall3.1
relevance,
(24).
and
be
below.
of
denoted
logical type
relations
a(1985)
little
byetc.)
signs
bit
oexpressions
onor
more
the of
and
such type
semantic
denite
the
as, offor question
our
about
instance,
((o)).
itemsobject language. The
andpurpose
what The arenormal of the tests
logical is, of
signs course,
such as, for to instance,
excludethe and latter
arec
erations
above lained operations
and
Faltzs operations
inherits and
Boolean
special the inherits
lattice
and
Semantics
the forthis
structure
inheritslattice
Natural structure
the
ofstructure
the in subsection
lattice
Language. the 3.1 of
structure
ofassumption below. the
(or cal a
25signs certain such logical
How very
naturalas, forconstants
do language)
instance,they of attainthe
set-theoretic and metalanguage.
seems this not
are constitution.
goal? to In
have order The to understand that
this, the
we latter
have possibility
to remember that the t
for
theWe operations
use 26the(24) lattice
and relations theoretic
byon symbols the semantic such items
as , , and for operations 26 and relations on the semantic items

deelation
nciples by theoperations
and
from relationonethe
This operations
type
kinddenoted
and toG the
ofanother Hoperations
transfer denoted
isthe same
extensively
of (x by :symbols.
denoted
logico-semantic o).G(x)
the used same by
in
26the H(x)
symbols.
Keenan
principles same andfrom
26
symbols.one type to another is extensively used in Keenan and
most
asily logical realistic
seen
denoted signs then by
may of
candidates
that(1985)
expressions
such be
propositions
as,the type neglected
for of for(o)
instance,our def does
object is
the Semantics logic
not
with the
and not
language.
justseem
offor
thus
area
natural to
The be
collectionnormalcompletely of
logical specialunjustified
signs entities
such and
as, but
forwill thatthus
instance, it adopted
has
anda certain
are structur
al pe Language.
tological
anotherconstants Faltzs
is extensively
here ofthe
(at the least
Boolean
used
metalanguage. ina Keenan
asbelow. working and hypothesis).
Natural Language.
Thus itItisisnot a matter
enite tnsreinduced
this
26
in and
subsection
structure
about by G
this in
inherits
in
, 3.1 which
structure H
subsection
, below.and means in the
on
lattice
3.1 operations
subsection that
. Corresponding
G
structure 3.1 is a
below.
and
subattribute
of the to (on of ) participate
H. easily and seen to of which
then arbitrary
that also thedecisions
atype
consequence (o) with relatio the thu
ymbols
uch ne and type
This
as , to
kind another
for
such andofaswhat
transfer
operations is
, extensively
for and
and
denotations ofrelations
operations logico-semantic
used
for and in relations
operations
on
are Keenan
the
included principles
semantic and and
26 on relations
the
into items from
semantic on
thesemanticsone
objectthe type
items to another
semantic
type o. items
There is extensively
are objective usedtests in Keenan for and
relation
operations , (denoted
denedbelongs.
denoted , by
relation
by Most
thethe samesame
proponents
and sign) the
symbols. on
above of (o) formal by
explained operations assume and that is a Boolean
inherits
the lattice Algebra. structure Thereof th
ur
nguage. normal
object Faltzs The
logical (1985)
language.normal signs Boolean
The logical
normal suchsigns Semantics
as,logical for
such instance, signs
for as,Natural
for such and
instance, Language.
as,for are
instance,
and areand are
language. 3.1type
however, o induced alternative on this type logicsby(many-valued, the relation and intuitionistic,
the operations relevance,
denoted etc.) by theand same question26w
the symbols.
subsection below.
ico-semantic cfrom principles
forone type from kind
principles
24to of logic isontypetothebe found inis natural language (or a certain and 3.1natural language) seems not to h
operations Weanother
one
andshall type frombeisto
relations aextensively
one
another
little bit more is to used
extensively
semanticanother in Keenan
denite items extensively
usedand
about in
thisKeenan used and
structure in in Keenan
subsection below.
Natural
antics al logical signs
guage. for Language.
Natural 25 been
Wesuch addressed
Language.
use as, thefor lattice instance, seriously
theoretic and symbolsup
are till now. However, most
such as , , and for operations and relations realistic candidates for on thethe logic
semantic of natu
item
s easily seen language thendenoted thatbythe assume type (o)
expressions some ofwith our kind theofthus
object lattice structure
language. The normal induced
logical signs by ,, such and as,for oninstance,
. Correspondin and a
one type to another the is extensively
relation on used thein we Keenan
deneand
ations and logical
inherits constants thelattice of , metalanguage.
structure ofnow the an analogue relation (denoted by the same sign) on (o)
136 Klaus Robering
membership in o. If an expression passes these tests, the most obvious semantic

procedure is to recognize it as a referring expression rather than to enrich the
type theory by mechanisms circumventing in some way such a classification of
that expression.
3. Types and representation
3.1 Types as data types
While Russellian types are just collections of sets, modern type theories provide
their types with more structure. A simple but important example for this are prod-
uct types. Type theories allowing for such types are usually also equipped with
projection operations which assign to each pair a : 1. 2 its left and right compo-
nent: p1(a) and p2(a), respectively. By such additions logical types approach what
are called data types27 in computer science (or concrete algebras28 in mathemat-
ics). A type is no more conceived as a mere collections of entities but rather as an
aggregate consisting of such a collections plus a series of important and useful
relations and operations defined on them. Thus the reader should not conceive of
the type system as something resembling the hierarchy of sets in the Principia
Mathematica but rather as a collection of interrelated algebraic systems or as
something like the class hierarchy of an object oriented programming language.
Each class comes along with its fields and methods; cp., for example, the Java class
hierarchy as documented in Javas API Specification (cf. http://docs.oracle.com/
javase/7/docs/api/).
Actually, the types of propositions and (o) of monadic attributes were
already conceived in this way in the discussion of subsection 2.4. Both types are,
as we have seen there, provided with lattice operations and as well as with a
complement operation and a lattice order . There have been more concepts
implicit in that discussion which deserve to be mentioned here. In defining the
27. There are, however, several uses of this term in computer science. In the present context, I
mean by data type any collection of values (e.g., characters, strings, truth values, integers, etc.)
plus a set of relations between them and operations defined on them.
28. The set of axioms of group theory, for instance, determine the abstract structure of a group.
Each individual model of these axioms is a group. The former general concept is what I call an
abstract algebra whereas the individual groups are concrete algebras. Similarly, that what I have
called a data type is a concrete algebra; cf. the previous footnote. The structure determined by
the signature of such a data type would be the corresponding abstract algebra. In computer sci-
ence the latter structure is sometimes also called a data type.
), (( (h),
)1 )
(( )1 )
tion
disjunction
. Let us
. therefore
Let us therefore
use the use
signsthe signs
junction
and disjunction
but alsobutforalso
the for
correspond-
the correspond-
es
h assign
which to G, Hto: G,
assign H the
(o) : (o)
complex
the complex Chapter 3. Abstract objects of verbs 137
ose
struction
construction
is provided
is provided
by (23)byand(23)
theand the
for
rentiality
referentiality
and set
let andQ for
belet
its
a Qdenotation. for instance, we made use of the notion of a true proposi-
be its denotation.
:((o)),
ddenoted
by the expressions
by thetion. expressions
Weresulting
could do resulting
from
this since from the datatype of propositions has two subtypes
the
expression
test expression is excised
(gr. is excised
from them
true) from
and them
(gr. false) for which principles such as those in
the examples
in the examplesgiven
(27) above).
hold given
true above).
Whatrelate
which isWhat theistwo subtypes and to the operations and .29
()
e testsand()()and involve
() involve
the inferential
the inferential
(27) a. p : p :
simple
attributes G and H
attributes G onandthe H oneon the hand one hand
p : p :
theH other.
on the All other.the Allattributes
the attributes
at issueat issue
b. p : p :
ese
withattributes,
these attributes,
Q has thus Q has to thusbe either
to be either
of
stscourse,
is, of course,
to exclude to exclude p : the
the latter case.
p : case.
latter
ndwethis,
havewetohave c.
remember to remember (p
that thethat q) :
type ptype
the :q:
entities
but thatbut it has
that aitcertain (p
has a certain q) :structure
structure 24 p : 24 q :
which
and to alsowhich d. (p
a consequence
also a consequence q) : p : q :
relationrelation
me isthat is a
a Boolean Boolean
Algebra. (p
Algebra. q) are,
There pare,
:There :q:
stic,
evance,relevance, e.
etc.) and etc.)
theand p
question :
the questionp q
what what q:
certain
natural natural
language) language)
Another seems not
ratherseems to have
useful notnotion
to have already used implicitly is identity. In subsection2.4,
crealistic
candidatescandidates
for the logic
for the of
logic
natural of natural
for instance, we said that F G is the same attribute as F if F G. Order and
uced
, ,byand , ,onand .Corresponding
identity onare Corresponding
. notions to
which to their appearance in (nearly) every type.30 As
make
enoted
tion (denoted
by the regards
same
by thesign) same
order, onsign)(o)
the on
basic by(o)
type byis partially ordered by the relation . For the times
in type , we have the linear order of temporal precedence; e.g., A.D.1999
A.D.2002. The sorts from are partially ordered by the subsortrelation ; e.g.,
TacoPartyParty). If we assume that the type o of objects contains plural
nily
then
seenthat
thenthethat
typethe
objects type
(o) orwith thewith
ensembles,
(o) thusthe thus
then there is also a natural partial order (being a part
sinherits
and inherits
theof)lattice
inthe structure
thislattice
type; structure
of Fidodogs
e.g., the of the where dogs is the mereological whole of all dogs.
ns
perations
denoteddenoted
by Whereas
the sameby the symbols.
same 26
symbols. 26
order and identity are concepts permeating the whole type system, there
bsection
3.1 below. 3.1 below.are other concepts such as, for example (truth) and (falsity) which are
erations
for operations special
and relations for semantic
and relations
on the a on
certain type (here:
the semantic
items items ).
igns
logical
such
signsas,
for
suchinstance,
as, for instance,
and and
are are
ne
o another
type to isanother
extensively
is extensively
used in Keenan
used in and
Keenan and
29. e
: means that entity e does not belong to type . Probably one would like to admit
propositions in which are neither true nor false and hence reject the principle p:p:. If
this is the case, there are several options to consider for possible amendments to (27c) and
(27d). One such option is the principle that (loosely speaking) a proposition has a definite truth
value only if all its components have. If this is adopted, one is forced to reject the converses of
those subtheses of (27c) and (27d) which are only conditionals (and not biconditionals). If, for
instance, p were false but q lacked a truth value, then pq would not be false but would rather
lack a truth value, too. A related issue is whether, in parallel to (27e), principles of indirect
reasoning such as, for example, q : p q p : should be adopted. Such principles
are missing from Nelsons logic of constructible falsity (cf., for instance, Nelson 1949 and
Wansing 1993: Chapters 68) and Ackermanns type-free logic (cf., e.g., Ackermann 1950,
Robering 2008).
30. This is also true for the types of many programming languages.
138 Klaus Robering
Identity is an ubiquitous notion. Indeed it could be asked whether there is only

one single notion of identity for all types or for at least all basic types.31 Compare,
however, the ways how we tell apart items from our basic types. In order to distin-
guish two objects from each other, we would look after an attribute possessed by
one of them and being lacking from the other. Two times are different from each
other when there is a moment falling within the first but not within the second.
For distinguishing two sorts, we should consider whether there possibly could
be an object belonging to one of them while being absent from the other. And
in order to distinguish two propositions from each other, we should describe a
situation in which one of them is true while the other is false. Though there are
obvious analogies relating these four cases, they clearly involve different opera-
tions and procedures and this strongly suggests the position that types come along
with their own special identity relations. We will write a=b: when a is the same
member of type as b, and ab: if they can be told apart from each other.32 The
general thesis
(28) a = b : a b :
states that the members of each type can be distinguished from each other. It
appears natural and even unavoidable if the principle tertium non datur is
accepted in our metalanguage.33 Furthermore, it would make a type (or more
exactly: the collection of members belonging to a type) rather similar to a set in
the original Cantorian sense. Cantor, as is well known, explained a set as a collec-
tion of well-distinguished objects of our thinking or intuition.34
31. Proponents of relative identity reject the idea of an absolute relation of identity generally
defined for all entities and insist that identity is relative to a criterion. Objects which are identical
according to one criterion may differ with respect to another. According to Geach (1962, 1967)
such criteria of identity are associated with common nouns, thus with sorts. What is meant by
this is nicely illustrated by the following example from Gupta (1980:23): (a) National Airlines
served at least two million passengers in 1975. (b) National Airlines served at least two million
persons in 1975. Though each passenger is a person, (b) does not follow from (a) since the same
person may be two or more passengers.
32. Martin-Lf (1984:5, 9f), in his intuitionistic type theory, distinguishes two kinds of identity
judgments: A = B, i.e., A and B are equal sets, and a = b : A, i.e., a and b are equal elements of
the set A.
33. Note, however, that we introduced non-identity as a primitive notion and did not define
it by requiring x y : as [x = y : ]. Even if the tertium non datur was accepted, these two
statements have still to be declared equivalent in order to convert (28) into an instance of that
logical principle.
34. Unter einer Menge verstehen wir jede Zusammenfassung M von bestimmten wohlunter
schiedenen Objekten m unserer Anschauung oder unseres Denkens (welche die Elemente
3.2 Uniqueness and elimination
Until now not very much has been said about the functions and mapping
respectively sorts and propositions to objects representing these items. Let us call
an object representing a proposition a state of affairs and an object representing
a sort a genus.35 The question which we approach in the present subsection is
how propositions and sort are related to the state of affairs and genera representing
them. One reasonable requirement for the kind of representing functions which
we are interested in is that they should not confuse entities distinguished in their
own type by representing them by the same objects. Hence it seems natural to
postulate the following uniqueness conditions [] and [], which make use
of the notation for identity just explained in the previous subsection.
(29) a. []: (p) = (q) : o p = q :
b. []: (a) = (b) : o a = b :
These two conditions require the mappings and to be injective which implies
that there must be both at least as many objects as propositions and at least as
many objects as sorts. The second consequence seems not to be very problematic
since one will probably assume that the objects outnumber the sorts of objects.
But the first consequence is of a somewhat nominalistic flavor which will not be
welcomed to everybody. Rejecting it means, however, to reject the theory put for-
ward here in its strongest form. A weakened form of the theory might be possible
but will not be considered here.36
vonM genannt werden) zu einem Ganzen; Cantor (1895:481). Ackermann (1965:5f) inter-
prets the adjective wohlunterschieden (well distinguished) as meaning that, for each pair of
elements a, b of the set, it must be the case that either a = b or a b. This means that the tertium
non datur with respect to identity should hold true within the set. Of course, this is only non-
trivial if that principle is not accepted universally for all propositions.
35. These two terms are of cause used both in philosophy and semantics in many very different
meanings. It is not intended here to catch any of the various meanings attached to one of these
terms. Rather, states of affairs are just the objects from the counter-domain of the function
and genera are the objects from the counter-domain of the function .
36. Here I am thinking about something similar to the hashfunctions used in different subfields
of computer science. A hashfunction represents the elements of a source set by a key from a
target set, which has at most as many members as the source set. Typically, the keys are in some
sense shorter or simpler than the elements from the source sets. For example, entire texts
may be represented by natural numbers. If the hashfunction is injective, it is called perfect.
Perfect hashfunctions are not possible when the source set exceeds the target set in cardinality.
Nevertheless, partial hash functions are considered in these cases where only some elements of
the source set are assigned keys.
140 Klaus Robering
The representation of items from other types by members of the object type
has, as we have seen, the advantage to reduce the typings necessary for such
denotations as, for example, Likes. Just one typing, instead of four or five, suf-
fices. However, this kind of economy has its prize in that the mode of expression
employing representation by abstract objects is indirect: one does not directly
talk about what is at issue but represents this by an abstract object. Amazingly,
the indirect mode of expression is often shorter and more to the point than the
direct one. Consider the following example. A polyhedron is a three-dimensional
geometrical solid with flat faces and straight edges (such as a cube, prism, pyra-
mid, etc.). Now, (30) is true.
(30) In a polyhedron the number of vertices minus the number of edges plus the
number of faces equals 2
Call the proposition denoted by (30) Euler. Then, of course, (Euler):o.

According to the theory developed in the present article, (Euler) is a (math-
ematical) state of affairs denoted by the that-clause corresponding to (30). Now,
that what is said by (30) can also be expressed by a sentence incorporating the
that-clause referring to (Euler); cf. (31).
(31) That in a polyhedron the number of vertices minus the number of edges plus the
number of faces equals 2 is true
This, of course, is even longer and less transparent than (30). However, elements
of type o like (Euler) are nameable; we may give names to them. I do not use
the term name in the present context in the technical Fregean sense (for a referring
expression) but rather in the everyday sense according to which a name is a short
label or description used to refer to something or somebody.37 Actually, Euler
has a name, it is called (or may be called) the Euler-Descartes-conjecture. If we use
EDConj for this, we have thus (32).
(32) EDConj = (Euler) : o
Thus, instead of (31) we may just say:
(33) The Euler-Descartes-conjecture is true
which is shorter than both (30) and (31). This example illustrates that representing
entities by abstract objects may offer the possibility to express oneself in a rather
brief but nevertheless succinct way.
Note that the adjective true cannot be the object language counterpart of our
subtype of since the subtype contains propositions while the adjective applies to
37. Every name in the latter sense is also a name in the first sense, but not conversely.
state of affairs, i.e., to objects which represent proposition. At a first look this seems
queer38 but this corresponds exactly to the fact that we cannot attach the adjective
(or the predicate is true) directly to an affirmative sentence. In order to do this, we
have to transform the sentence either to a that-clause or we have to use some other
syntactic device.39 Let us use Obtains for the attribute denoted by the predicate is
true occurring in (31) and in (33); clearly, Obtains:(o). The propositions denoted
by (31) and (33) are thus, respectively, Obtains((Euler)) and Obtains(EDConj).
In view of (32), one would of course say that this is the same proposition and, fur-
thermore, Obtains((Euler)) actually was equated with Euler, cf. the statement just
before (31). Hence, we have the three identifications made in (34).
(34) a. Obtains((Euler)) = Euler :
b. Obtains(EDConj) = Euler :
c. Obtains(EDConj) = Obtains((Euler)) :
Of these equalities, (34a) is of special interest because it shows how one can, at
least in certain contexts, get rid of the star operation and the abstract object intro-
duced by it. Instead of saying that something is true, we can just as well utter that
what is asserted to be true. In this way, we dispense with both the star operation
and states of affairs. Representation by abstract objects offers the possibility of
brief and succinct formulations as, for instance, (33) but they are often unnec-
essary. Each representing function, such as the star operation, thus comes along
with a set of elimination procedures which specify how the representation in ques-
tion can be avoided (at least in certain cases). In the case of the star operation, for
instance, we have the following generalization [] of (34a).40
(35) []: p = Obtains((p)) :
38. This impression is re-enforced by the fact that we normally say of a state of affairs that it
obtains rather than that it is true. Remember, however, that the choice of the term state of
affairs was somehow arbitrary. States of affairs are just the members of the co-domain of the star
operation. It was not intended to catch any intuition connected with the colloquial use of the
term state of affairs or with its use in any theory of states of affairs.
39. We could, for instance, add to the bare sentence the comment That is true; e.g., London is
the capital of the Empire. That is true. The word That in the second sentence of this example
functions like a demonstrative pronoun but does not refer back to a noun it does not stand
pro nomine but rather to the whole preceding sentence; it thus stands pro sententia. Thus
it can justly be called a prosentential; cf. Grover etal. (1974).
40. [] captures the basic tenet of so-called deflationary theory of truths according to which
one does not state anything more when one says that a certain proposition is true than one
does when one simply asserts that same proposition. A locus classicus for this position is Frege
(1918/19:345): So scheint denn dem Gedanken dadurch nichts hinzugefgt zu werden, da ich
ihm die Eigenschaft der Wahrheit beilege. Another classical reference is Ramsey (1927:44f).
142 Klaus Robering
The case for sorts and genera is similar to that of propositions and states of affairs.
Sorts are used to state rough generalization which play a distinguished role in
everyday explanations using (more or less) common knowledge. There is, for
instance, often something relevant to be stated which normally holds true for all
or no humans, fish, or vegetables; cf., e.g., (36).
(36) Normally, no fish of the coral reef will attack a cleaner wrasse
In a first analysis, the proposition denoted by this sentence may be represented

as in (37).
(37) No ( ReefFish x.( A ( CleanerWrasse y. Attacks(xy))))

() (((o)) (((o)) (oo)
Using a genus object for representing the sort CleanerWrasse, this can be simpli-
fied to (38).
(38) No ( ReefFish x. Attacks(x ( CleanerWrasse )))

() (((o)) (oo) o()
In plain English, this would correspond to a re-formulation of (36) by means of

the generic definite article.
(39) No fish of the coral reef will attack the cleaner wrasse
As a (very) tentative formulation of an elimination procedure, it may be suggested

that a proposition within the scope of a normality-operator which results from
the application of an indefinite quantifier to a sort S and a predicate F may be
replaced by the proposition F((S)).
(40) []: A(K F) = F((K)) :
The original example (1b) differs from (36) in that plural morphology is used
instead of the generic article in order to express the representation of a sort by a
genus. For (1b), [] yields the following rephrasing of it.
(41) Normally, Mary likes a taco party
This is not too bad. However, the sketchy analysis given surely needs a more care-
ful elaboration of the details.
4. Comprehension
4.1 Problems with functions
Up till now we have only considered the representation of certain individuals

(namely, sorts and propositions) by abstract objects (namely, genera and states of
affairs). In the present subsection, we shall turn to that what, generalizing the case
of set theory, may be called comprehension: namely, the representation of func-
tions (entities of higher types) by individuals (entities of basic types).41 As we shall
see in the present subsection, comprehension is the main obstacle for a theory of
representation as that envisaged here. The first question arising for comprehension
is that for a reasonable uniqueness condition. The idea of [] and [] is that
identical individuals (propositions or sorts) should be represented by the same
object (functioning as state of affairs or as genus, respectively). If one wants, as
seems reasonable, apply the same idea in the present case, one has to explain when
functions are identical. An obvious idea is to declare functions to be identical if
they make the same assignments, i.e., if they assign the same values to the same
arguments. For unary functions, we thus receive (42).
(42) f = g : 0(1) x : 1. f (x) = g(x) : 0
One might argue that (42) is inadequate since, by this principle, functions are
identified with their graphs, i.e., the collection of the argument-value-pairings
induced by them, and this will lead up to extensionality. Though the first is cor-
rect, the second does not follow. Let us consider the case of attributes; this is the
only case considered in the following. Attributes are functions of type (o) (thus
0 = and 1 = o) and the extension of an attribute would be the collection (set)
of objects to which the attribute applies. What (42) requires for this special case
is that attributes are identical if they assign the same proposition to the same
objects. Obviously, this does not reduce attributes to sets. Sets are identical if they
have exactly the same members. But remember that is the type of propositions
rather than that of truth values (which is missing from the type system intro-
duced here). Coincidence of truth values () could be explained in the present
framework as in (43).
41. If is a functional type, I will call the representation of functions from by objects
-comprehension. What is of interest in the present article is (o)-comprehension or attribute
comprehension. In contrast to the iterative conception of set (cf., e.g., Boolos 1998a:1622),
which conceives of sets as collections, the naive or logical conception of set (cp. pp. 1416
of Boolos article) views sets as objects representing functions from objects to truth values. Let
be the type of truth values, then set comprehension is the special case of (o)-comprehension.
144 Klaus Robering
(43) p q p, q : p, q :
def
Given this explanation, one can, following Myhill (1963:305), also state when
attributes determine the same set; cf. (44), where F and G are used as variable for
attributes.
(44) F G x : o.F(x) G(x)
def
As one easily recognizes, PQ is a weaker requirement than P=Q; cf. Myhill

(1963).
Using the notation already introduced in (15) for the representation of attri-
butes by abstract objects, we can now formulate a uniqueness condition for this
kind of representation which is in line with the formerly given ones [] and [].
(45) []: F = G : o F = G : (o)
In order to have a name for objects representing attributes, I will call them proper-
ties. As was the case for genera (in connection with the -operation) and states of
affairs (in connection with the -operation), it is not maintained here that proper-
ties make up a special kind of entities ontologically inherently different from other
kinds of entities.42 Properties, according to the theory proposed here, are just nor-
mal objects fulfilling a special functions: namely, that of representing attributes.
Nothing hinders an object to fulfill simultaneously different representation tasks,
i.e., to be both a concept, a genus, and a state of affairs.
[] turns out to be a severe restriction on the comprehension of attributes. In
order to see this, have a look on the diagram of Figure2. Individuals are situated
on the lower plane of Figure2, functions on the upper one. Assume that there are
just two objects a and b and two propositions p and q. Two attributes are
shown on the upper plane: F and G. The functional pairings induced by them are
represented on the lower level by dashed (F) and dotted (G) arrows, respectively.
Each arrow is connected by a line of the same kind to the point representing
the attribute function to which it belongs. F and G agree on the argument a to
which both assign the value p. They differ, however, in that F assigns the same
value to the other object b while G assigns the proposition q to that object. Hence:
F(a)=p=G(a), but F(b)=qp=G(b). Since the attributes F and G thus differ,
the principle [] requires them to be represented by different objects. This is
actually the case in Figure2: F =ba=.
42. Note, however, that there indeed is a fundamental difference between attributes and proper-
ties: whereas properties are individuals and even objects, attributes are functions.
Functions
Attributes (o)
F
Propositions
a p
b
q
Objects o
Individuals
Figure 2. Representation of attributes by objects fulfilling []
Thus, at a first glance, everything seems to be all right in the miniature model
of attribute comprehension depicted in Figure2. A second glance, however, will
reveal the problem that F and G are only two of four attributes which are combi-
natorially possible. Obviously, it is not possible to represent these four attributes by
just two objects in accordance with principle []. And this is not an artifact of the
miniature model. Quite generally, if M and N are two sets, the class of functions
from M to N exceeds the set M in number of elements so that it is not possible to
assign to each such function an element of M in accordance with principle [].
Georg Cantor, the founder of set theory, gave a proof of this for the special case
that N contains just two elements (the characters m and n in Cantors original
presentation); cf. Cantor (1890/91).43 In this case, the functions correspond in a
one-to-one manner to the subsets of the set M. Thus what Cantor proved is that
the power set of a given set always contains more elements than the set itself. This
is known as Cantors Theorem. In the present context this means that there can be
no representation of attributes by objects if
43. Actually, Cantor proves the assertion for the even more special case that M is an interval of
the real numbers. However, he remarks that the idea of his proof can be used in order to prove
the more general result that there is no upper bound for the number of elements of sets (Cantor
1890/91:279), and he formulates that what nowadays is called Cantors Theorem in a letter to
Dedekind from 31st August, 1899; cf. Cantor (1932:448) and Meschkowski (1967:85).
146 Klaus Robering
1. one wants to keep the uniqueness condition [], and

2. one models types and functions within the framework of classical logic and
standard set theory, and
3. one insists on unrestricted comprehension, i.e., on the requirement that all
attributes shall determine a property, thus have a representing object.
4.2 More problems
The solution which I want to suggest for the problem of the previous subsection is
to give up both 1. and 2. above, but to keep 3., thus unrestricted comprehension.
In the context of the present article, attribute comprehension is interesting since
it makes it possible to render sentences like (1d) containing a finite verb which
takes an infinitival argument by formulas like (15), which is repeated here as (46).
(46) Likes ( Mary ( Dance ) )

(oo) o o((o)) (o)
No special, extraordinary typing is required for Likes as it occurs in (46). The same
garden variety of that relation occurs there as in the mundane (11a). Keeping this
analysis of infinitival complements of verbs but adopting simultaneously some
restricted form of attribute comprehension (as it may be inspired by some sort
of axiomatic set theory with set comprehension restricted in some way or other)
would mean to rule out some infinitival clause as illegitimate for semantic reasons.
But there does not seem to be any such restriction for the formation of infinitival
clauses. Therefore I conclude that there are only two meaningful options here:
either one entirely gives up the idea that infinitivals stand for abstract objects or
one admits unrestricted attribute comprehension.
Unrestricted attribute comprehension, however, leads up to problems with
Cantors Theorem as we have seen in the previous subsection. Assuming (some
kind of) standard set theory enforces a rather strong interpretation of the phrase
all attributes in item 3. of the list given above. The problem is that set theory
delivers too many attributes to represent. But perhaps this problem is just an arti-
fact of set theory. An attribute is a mapping from (o) and according to set theory
mapping here means any arbitrary association of objects and propositions in
which each object is assigned just one unique proposition. This is a relatively
recent understanding of mapping. Before the rise of set theory, a mapping was
considered to be associated with some kind of rule which states which value (here:
proposition) has to be assigned to which argument (here: object). Such a rule has
to be formulated in some kind of language. Given that a language only has a finite
vocabulary and that a rule is a finite string of items from this vocabulary, there
would be at most denumerably (countably) many rules and thus only denumer-
ably many functions (here: attributes). If then the object type o contained infinitely
many elements (e.g., by containing the natural numbers as a subtype), there
should be no real problem with matching each attribute with an object.
It is true, the supply of objects would be large enough under the circumstances
just described; still we would have a problem, however. And this demonstrates that
the problem at issue is not just one of the cardinalities of the types involved. As
and , attribute comprehension, too, comes along with certain concepts which
make it possible to eliminate the representing objects (properties in this case, gen-
era and states of affairs in the former ones). Consider the sentences in (47).44
(47) a. Amber Guardian has the property to suit every taste and every budget
b. Fire has the property to open substances according to the alchemical sense
c. This vitamin has the property to delay the change of the cellular structure
which leads to natural decadence
It is natural to analyze these examples as stating that the attributes represented

by the infinitivals adjoined to the phrase the property apply to the denotations of
the subject phrases. Thus I take the sentences in (47) to denote exactly the same
propositions as the sentences in (48) below.
(48) a. Amber Guardian suits every taste and every budget
b. Fire opens substances according to the alchemical sense
c. This vitamin delays the change of cellular structure which leads to natural
decadence
Thus I assume a function Htp: (oo) (has the property) with the following elimi-
nation procedure (49).
(49) []: Htp(a F ) = F(a) :
Now assume that we have not uncountable many but only denumerably attributes
Fj (0 j) each of which is represented by a property aj. We want to apply the idea
Cantor used to prove his theorem about the size of the power set to the present
situation. The method Cantor used is his celebrated diagonalization procedure.
Adopting it to the present case, we have to set up a quadratic matrix M = (pi j)i, j0
of propositions: pi j is the proposition Htp(ai Fj). Note that, by our initial assump-
tion, it holds true that Fi = ai : o; hence Htp(aiFi) = pii = Htp(ai ai) : o. Let for each
i 0 Di be the set of propositions different from pii. Since there are surely more
than two propositions, each Di is non-empty. Hence, by the axiom of choice, there
44. These are again sentences found on the web by the help of Google. (47b) and (47c) have
been slightly abbreviated.
148 Klaus Robering
is a function sel:(o) which choses for each argument ai one proposition sel(ai)
from the set Di and which takes arbitrary values for an argument not occurring
among the aj (j 0). By our construction sel(ai) pii :o for each i0; hence sel
must differ from each Fi. But according to our assumption all attributes are among
the Fis. In the classical set theoretic framework one concludes that this assump-
tion must have been wrong. But in the present context we rejected the concept of a
function inherent in that framework for the sake of the argument. So, by what rea-
sons else could we reject the argument presented? Its crucial point is, of course, the
construction of the choice function sel. This construction relied on the axiom of
choice. So we might reject the axiom of choice.45 There is, however, a less dramatic
alternative for rejecting the argument. The notational system in which the rules
determining the mappings Fi(i 0) has not been spelled out. Therefore it remains
uncertain whether that system is really sufficient for constructing the function
sel. Remember what was said concerning example(8) above: If some classical
principle are missing from the grammar of our notational system, then functions
present in a classical framework will not be available. And, of course, attributes
not available cannot be represented by properties. The attribute at issue in (8) was
x.Likes(xMary). If the exportation law did not hold for the conditionals stating
our semantical rules (and if the typification at issue there could not be achieved by
other principles which in fact were valid), then there would be no such attribute.
In the present case, we cannot be sure whether there really is the critical function
sel before the rules of the system are precisely stated which allow the construction
of new attributes out of those initially given.
A variant of the argument above is possible which avoids reliance on the axiom
of choice. Just use x.Htp(x x):(o) instead of x.sel(x). Call the first attribute R;
then we have R(ai)=Fi(ai):.46 The attribute R will assign to each ai the negation
of the proposition which is assigned to this argument by Fi. If we generally have
p:p: (remember that was the subtype of true, that of false proposition),
then (27a) implies that R(ai)Fi(ai): and that R differs from each Fi. However,
nothing forces one to adopt the principle of bivalence according to which each
proposition is either true or false; we may admit indefinite propositions which
are neither. If we then insist that there must be an i such that R=Fi:(o), then
45. Though most set theorist would not like this option very much, there is at least one set
theory which refutes the axiom of choice, namely Quines NF; cf. Specker (1953). The intuition
underlying Quines system differs from the usual cumulative conception of sets which underlies
Zermelos set theory.
46. This argument is essentially an adaptation of Russells Paradox. Russells paradox may be
viewed as a special application of Cantors diagonalization method; cf. Robering (1994:913),
Boolos (1998b:135138).
we have for that i: Fi(ai)=Fi(ai):. Thus the proposition Fi(ai) is its own negation.
Of course, this is only possible for an indefinite proposition. But what is else to
complain about this?
Summing up the above considerations, it may be said that attribute compre-
hension seems to be possible if we revise (in a way still to be described more
precisely) the classical assumptions about sets and functions and if we admit
indefinite propositions. More issues are to be taken into account, however. In (27)
we considered the operations of negating propositions and of conjoining them
by conjunction and disjunction. Conditions which govern these operations have
been formulated there which differ from the classical ones only by giving up the
principle of bivalence. Our object language, however, contains besides the seman-
tic operations mentioned also that what Curry (1963:139f) calls a ply-operation
(ply < imply), i.e., an operation corresponding to the conditional connective (if-
then). I will use for this operation. Some principles governing this operation
are listed in (50).
(50) a. p p :
b. p (p q) q
c. p q r q p r
(50a) may be called the principle of self implication, (50b) is just modus ponens
and (50c) is the so-called deduction theorem. Since these principles are rather
weak and hold true already in minimal and intuitionistic logic, they seem to be
unproblematic. However they give rise to a problem in connection with attribute
comprehension.47 Before describing this problem, I want to point out that (50a)
and (50c) (in conjunction with the lattice laws and (27e)) imply that pq: if
pq; cf. (51).
(51) a. pq hypothesis
b. p (r r) p lattice law
c. p (r r) q (a), (b), transitivity of
d. rrpq (c), (50c)
e. rr: (50a)
f. pq: (e), (d), (27e)
Now, the problem alluded to above is that the principles (50) in the presence
of attribute comprehension imply that coincides with its subtype , i.e., each
proposition is true: p for each p. In order to see this, let p be an arbitrary
proposition and consider the attribute x.Htp(xx)p. Let, furthermore, cp be
47. This problem is known as Currys Paradox; cf. Curry (1942).

150 Klaus Robering
the property x.Htp(xx)p where, quite generally, x.F def

=x.F. By [], then, we
have (52).
(52) Htp(cp,cp) = Htp(cp,cp) p :
Now we argue as in (53).
(53) a. Htp(c p c p) (Htp(c p c p) p) p (50b)

b. Htp(cp cp) Htp(cp cp) p (a), (52)
c. Htp(cp c p) p (b), lattice law
d. Htp(c p c p) p : (c), (51)
e. (Htp(c p c p) p) (Htp(c p c p) p) p (a), (52)
f. Htp(c p c p) p p (e), lattice law
g. p: (d), (f), (27e)
4.3 A way out (?)
The last problem discussed is decisive for the project of analyzing infinitival com-
plements of verbs by means of unrestricted comprehension. If the conditional con-
nective of a natural language has the properties listed in (50), then such an analysis
is untenable. Whether these conditions are fulfilled is, of course, an empirical
question about the semantics of the conditional in natural languages. Main stream
formal semantics often somehow naively assumes that the conjunctions of natural
languages (and, or, if) match the connectives of classical logic. There are, however,
authors outside the main stream which display a more critical attitude toward this
problem; cf. Cooper (1978: Chapter8). In any case, the following sketchy remarks
base on the assumption that natural language conditionals do not have all the
properties listed in (50).
Then there is at least a chance for working out the hypothesis that infinitival
complements stand for properties representing attributes as objects. One natural
framework for such an analysis is a type-free logic with completely unrestricted
comprehension axiom for sets as it has been developed by Wilhelm Ackermann in
a series of publications; see, for instance, his articles (1950), (1965) and the appen-
dix below (Section6).48 What I would suggest, then, is to develop a type theory for
48. Actually, Ackermann has developed several systems of the kind described above; cf.
Robering (2008:28f) for a brief discussion. A useful overview over type-free systems is provided
by Feferman (1984). A typical trait of type-free logic is the lack of an axiom of extensionality.
This is the technical reason to skip []. Feferman (1975:78) plays down the importance of such
an axiom when he writes the mathematical role of extensionality [] is much less important
than ordinarily thought.
the denotations of natural language expressions within such a type-free framework

and to use the unrestricted set comprehension of the framework in order to model
unrestricted attribute comprehension for the analysis of infinitival complements
of verbs. Something similar to this has already been done by Ackermann (1953)
in a logico-mathematical context. In that article, Ackermann defines a ramified49
theory of types for a language of arithmetic. Unrestricted comprehension is used
in his system in order to represent numbers by specific terms.
How might such a type system look like? Well, here is a brief sketch. We start
from the following symbols:
1. the variables x, y, z, x0, y0,...;
2. the constants for the elementary entities in which we are interested; these
include
a. the descriptive constants: Mary, likes, taco party, ,
b. the semantic constants: , , ,
c. markers for abstract objects: , , ;
3. the type indices: , , , o, (o), etc., (true propositions), (false proposi-
tions), ;
4. judgment forming operators; these include (besides the membership predicate
listed below under 5d)
a. the type-theoretic signs: the colon (:) for type membership and the com-
plex sign = : for identity between entities of a certain type
b. special signs coming from the individual (data) types; thus, for instance,
(implication between propositions), (temporal precedence),
5. the logical symbols; namely,
a. the connectives: (a truth constant), , , , ,
b. the quantifiers and ,
c. the abstraction operator ,
d. and the class-theoretic signs: (judgment forming) and {|};
6. technical signs (dots and square brackets) for disambiguating formulas, paren-
theses to denote the application of functions.
The formation rules are obvious from the examples previously discussed.
The hallmark of Ackermanns system is that comprehension is not restricted in
it; every condition C[a1, a2,...an] on n individuals gives rise to a relational indi-
vidual {x1 x2...xn | C[x1, x2,...xn]}, cf. the axioms stated in subsection 6.4 of the
appendix. One-place relational individuals are, of course, classes. The distinction
49. Ramified type theory as a framework for natural language semantics has been recom-
mended by Thomason (1989).
152 Klaus Robering
between attributes on the one hand and objects on the other is spurious in a sys-
tem such as Ackermanns. Instead of saying that a has attribute F, one may just
as well say that a belongs to class F.50 Hence the only attributes really necessary
are the membership relations (plural! ). For each arity we have a corresponding
attribute. In the appendix, we denote all these membership relations by the single
symbol .
Class formation, in Ackermanns system, fulfills two different tasks: (1st) it
builds attributes from formulas containing free variables; simultaneously (2nd),
it objectifies these attributes, i.e., it turns them into objects. In the system we
envisage here, we want to keep these two functions separate from each other,
however. We employ the lambda-operator for the first task and the objectifying
cap-operator for the second. The most natural framework to express our basic
principle that each attribute is represented by an abstract object would be a second
order logic which would allow us to quantify over attributes. In such a framework,
we could express unrestricted comprehension (of monadic attributes) by (54).
(54) F : (o).x : o.[x = F :o]
It is important to note that the domain of quantification for the second order
quantifier has been restricted by adopting Ackermanns more restrictive laws for
the conditional. Conditional laws govern, as explained in the first part of this
article, the construction of attributes. Thus, for instance, the construction of the
complex attribute x.Likes(x Mary) as in (8) is no longer available within the more
restricted propositional principles given in subsection 6.1 below. Fortunately,
there is another way to construct this attribute which is still viable in the more
restrictive framework. However, such an alternative is not available in every case.
Employing now the resources of Ackermanns system, one may chose the abstract
object required to exist by (54) to be the graph of the attribute F, i.e., the class
{uv|F(u)=v:}. What Ackermanns system guarantees without yielding a con-
tradiction is that this class actually is an individual, i.e, within the domain of
first-order quantification. What is required, however, is more: this individual
should belong to type o, it should be not only an individual, but also an object.
The problem whether this farther going requirement can also be met must be left
open in this article.
50. In Ackermanns system a primitive relation symbol such as = functions as both a term and a
relation symbol. Thus, for instance, the verum is defined by Ackermann by the formula ===
stating that the identity object (first =) bears the identity-relation (second =) to itself (third =).
5. Historical remarks
The idea to represent functions by objects comes, of course, from Frege; cf., espe-
cially, Frege (1893/1903). Frege uses this idea in order to avoid ascent to types
higher than those of second order and to define numbers as objects. Freges theory
of classes as value courses (Wertverlufe) of what he calls properties namely,
functions mapping objects to truth values is extensively studied by Cochiarella
(1986); Cocchiarella (1987). Cocchiarella, in the works cited, also relates Freges
conception of classes to that implemented in Quines non-standard set-theories
NF (Quine 1953) and ML (Quine 1940). Freges general strategy of definition
by abstraction is investigated within a more comprehensive framework by Fine
(2002). Quines NF is used for a type-free semantics of a fragment of English in
Chierchia (1982, 1988). As is done in the present article, infinitival complements
are treated as object-denoting expressions by Chierchia. However, his semantics
does not admit for the representation of each attribute by an abstract object since
Quines NF allows comprehension only for conditions which fulfill a certain con-
dition, namely stratification.
The idea to represent attributes rather than extensional functions from objects
to truth values by abstract objects is the common core of the different property
theories which have been developed since the appearance of Bealers book (1982)
on this topic. Seen from the perspective of the present article, Bealer develops a
framework for different property theories but does not propose a certain theory
which avoids the troubles connected with the antinomies which might arise in
a full-fledged property theory in analogy to the well-known antinomies of set-
theory. For his logic of the predication relation, Bealer (1982:94100) consid-
ers the set-theoretic strategies for avoiding antinomies used by Zermelo and
Fraenkel as well as those suggested by von Neumann, Bernays and Gdel without
however reaching at a definite proposal for property theory. Any solution of this
kind would suffer from the ad-hoc-character which these strategies take on when
applied in the framework of a logical conception of classes as opposed to the itera-
tive conception.
Supecki (1971) seems to be the first author dealing with a device similar
to the -operator of the present article. Actually, the notation used here derives
from Supeckis At which, in his generalization of modal logic stands for the
proposition A states t (A a formula, a a term). Hi (1984) explains how a theory
of the representation of propositions by abstract objects results from a Fregean
class theory if one concentrates on the case that the attribute to be represented is
0-ary.
154 Klaus Robering
The semantic literature on the topic of genericity is vast, of course. A discus-

sion from a perspective similar to that backing up the -approach sketched in the
present article can be found in Bacon (1973).
6. Appendix: Ackermanns original system
In this appendix we specify Ackermanns original system as it is presented in Ackermanns last

publication on this topic; cf. Ackermann (1965:79). The notation is adopted to that used in the
present article. is defined by and, similarly, t1 t2 and t1tn t0 stand, respectively, for
t1 = t2 and t1tn t0.
6.1 Axioms of propositional logic
1. AA 2. [ A] [B A B]
3a. AB A 3b. AB B
4a. A AB 4b. B AB
5. A [B C] B [A C]
6. [A B] [A C] [A B C]
7. [A C] [B C] [A B C]
8. [A B] [B C] [A C]
9. [ A] [B A]
10. [ A B] [ A] [ B]
11. [ A] A 12. A A B
13a. A A 13b. A A
14a. A B [A B] 14b. [A B] A B
15a. A B [A B] 15b. [A B] A B
16. [ A] [B ] [A B]
17. [A B] [ A] [B ]
6.2 Axioms of predicate logic
1. x.A[x] A[a] 2. A[a] x.A[x]

3. x.[A B[x]] [A x.B[x]]
4. x.[A[x] B] [x.A[x] B]
5. x.[A B[x]] A x.B[x]
6. A B[x] x.[A B[x]]
7. [ x.B[x]] x.[ B[x]]
8a. x.A[x] x.A[x] 8b. x.A[x] x.A[x]
9a. x.A[x] x.A[x] 9b. x.A[x] x.A[x]
Conditions: The variable x does not occur free in A in axioms of the types 3, 5, and 6, nor does
it occur free in B in axioms of type 4.
6.3 Axioms for identity
1. a = a 2. a = b A[a] A[b] 3. A[a] A[b] a b
6.4 Axioms of class theory
1. A[a1,, An] a1an {x1xn | A[x1,, xn]}

2. a1an {x1xn | A[x1,, xn]} A[a1,, An]
3. A[a1,, An] a1an {x1xn | A[x1,, xn]}
4. a1an {x1xn | A[x1,, xn]} A[a1,, An]
6.5 Rules
1. Modus Ponens.
2. From A infer A.
3. From A[a] infer x.A[x].
4. Bound variables may be re-named.
5. If t is a term, infer A[t] from A[a].
Bibliography
Ackermann, Wilhelm (1950): Widerspruchsfreier Aufbau der Logik, I. The Journal of Symbolic
Logic 15: 3357.
Ackermann, Wilhelm (1953): Widerspruchsfreier Aufbau einer typenfreien Logik, II. Mathe-
matische Zeitschrift 57: 155166.
Ackermann, Wilhelm (1965): Der Aufbau einer hheren Logik. Archiv fr mathematische
Logik und Grundlagenforschung 7: 522.
Anderson, Alan Ross & Belnap, Nuel D. (1975): The Logic of Relevance and Necessity. Vol. 1.
Princeton and London: Princeton University Press.
Anderson, C. Anthony (1984): General intensional logic. In: Dov Gabbay & Franz Guenthner
(eds.) The Handbook of Philosophical Logic, Vol. II, Dordrecht: Reidel. 355385.
Bacon, John (1973): Do generic description denote? Mind 82: 331347.
Bealer, George (1982): Quality and Concept. Oxford: Clarendon Press.
Boolos, George (1998a): The iterative conception of set. In: Boolos (1998b: p. 1329). Origi-
nally published in The Journal of Philosophy 68 (1971). 215232.
Boolos, George (1998b): Logic, Logic, and Logic. Cambridge MA: Harvard University Press.
Cantor, Georg (1890/91): ber eine elementare Frage der Mannigfaltigkeitslehre. Jahresbericht
der Deutschen Mathematiker-Vereinigung 1: 7578. Reprinted: Cantor (1932: p.278281).
Cantor, Georg (1895): Beitrge zur Begrndung der transfiniten Mengenlehre. Mathematische
Annalen 46: 481512. Reprinted (together with the continuation of this article): Cantor
(1932: p. 282351).
Cantor, Georg (1932): Abhandlungen mathematischen und philosophischen Inhalts. Edited by
Ernst Zermelo. Berlin: Springer. Reprinted: Hildesheim and New York: Olms 1966.
Carpenter, Robert L. (1997): Type-logical Semantics. Cambridge MA: MIT Press.
156 Klaus Robering
Chierchia, Gennaro (1982): Nominalization and Montague grammar: A semantics without

types for natural languages. Linguistics and Philosophy 5: 303354.
Chierchia, Gennaro (1988): Topics in the Syntax and Semantics of Infinitives and Gerunds. New
York and London: Garland.
Cocchiarella, Nino B. (1987): Logical Studies in Early Analytic Philosophy. Columbus OH: Ohio
State University Press.
Cochiarella, Nino B. (1986): Logical Investigations of Predication Theory and the Problem of
Universals. Naples: Bibliopolis.
Cooper, William S. (1978): Foundations of Logico-Linguistics. Dordrecht: Reidel.
Curry, Haskel B. (1942): The inconsistency of certain formal logics. The Journal of Symbolic
Logic 7: 115117.
Curry, Haskell B. (1963): Foundations of Mathematical Logic. New York etc.: MacGraw-Hill.
Curry, Haskell B. & Feys, Robert (1958): Combinatory Logic. Vol. 1. Amsterdam: North-Holland
Publishing Company.
Dummett, Michael (1973): Frege. Philosophy of Language. London: Duckworth. 2nd edition
1981.
Feferman, Solomon (1975): Non-extensional type-free theories of partial operaions and classi-
fications, I. In: Justus Diller & Gert Heinz Mller (eds.) |= ISILC Proof Theory Symposium.
Dedicated to Kurt Schtte on the Occasion of His 65th Birthday, Berlin, Heidelberg, and
New York: Springer. 73118.
Feferman, Solomon (1984): Toward useful type-free theories, I. In: Robert L. Martin (ed.)
Recent Essays on Truth and the Liar Paradox, Oxford: Clarendon Press. 237306. Also in:
The Journal of Symbolic Logic. 49 (1984). 75111.
Fine, Kit (2002): The Limits of Abstraction. Oxford: Clarendon Press.
Frege, Gottlob (1884): Die Grundlagen der Arithmetik. Eine logisch-mathematische Untersuchung
ber den Begri der Zahl. Breslau: Koebner. Centenary edition. Hamburg: Meiner 1986.
Frege, Gottlob (1892): ber Begri und Gegenstand. Vierteljahresschrift fr wissenschaftliche
Philosophie 16: 192205. Reprinted: Frege (1990: p. 167178).
Frege, Gottlob (1893/1903): Grundgesetze der Arithmetik. 2 Vols. Jena: Pohle. Reprinted:
Hildesheim and New York: Olms 1966.
Frege, Gottlob (1918/19): Der Gedanke. Beitrge zur Philosophie des deutschen Idealismus 1:
5877. Reprinted: Frege (1990: p. 342362). English translation: The thought: A Logical
Inquiry. Mind 65 (1956). 289311.
Frege, Gottlob (1990): Kleine Schriften. Edited by Ignacio Angelelli. Hildesheim and New York:
Olms, 2nd edition. 1st edition 1967.
Geach, Peter (1950): Subject and predicate. Mind 59: 127162.
Geach, Peter (1962): Reference and Generality. An Examination of Some Medieval and Modern
Theories. Ithaca NY and London: Cornell University Press. 3rd edition 1980.
Geach, Peter (1967): Identity. Review of Metaphysics 21. Reprinted: Geach, Thomas: Logic Mat-
ters. Oxford: Blackwell 1972. 2nd, corrected edition 1981. p. 238247.
Groenendijk, Jeroen & Stokhof, Martin (1984): Studies in the semantics of questions and the
pragmatics of answers. Ph.D. thesis, University of Amsterdam.
Grover, Dorothy L., Camp, Joseph L. & Belnap, Nuel D. (1974): A prosentential theory of truth.
Philosophical Studies 27: 73125.
Gupta, Anil (1980): The Logic of Common Nouns. An Investigation in Quantified Modal Logic.
New Haven and London: Yale University Press.
Hale, Bob (2001): Singular terms (1), (2). In: Hale & Wright (2001: p. 3147 (1), 4871 (2)).
Hale, Bob & Wright, Crispin (eds.) (2001): The Reasons Proper Study. Essays towards a Neo-
Fregean Philosophy of Mathematics. Oxford: Oxford University Press.
Hindley, J. Roger & Seldin, Jonathan P. (1986): Introduction to Combinators and -Calculus.
Cambridge GB: Cambridge University Press.
Hiz, Henry (1984): Frege, Leniewski and information semantics on the resolution of the
antinomies. Synthese 60 : 5172.
Keenan, Edward L. & Faltz, Leonard M. (1985): Boolean Semantics for Natural Language.
Dordrecht: Reidel.
Lewis, David (1986): On the Plurality of Worlds. Oxford: Blackwell.
Martin-Lf, Per (1984): Intuitionistic Type Theory. Naples: Bibliopolis.
Meschkowski, Herbert (1967): Probleme des Unendlichen. Werk und Leben Georg Cantors.
Braunschweig: Vieweg.
Montague, Richard (1970): Universal grammar. Theoria 36: 373398. Reprinted: Thomason
(1974: p. 222246).
Montague, Richard (1973): The proper treatment of quantification in ordinary English. In:
Jaakko Hintikka, Julius Moravcsik & Patrick Suppes (eds.) Approaches to Natural Lan-
guage: Proceedings of the 1970 Stanford Workshop, Dordrecht: Reidel. 221242. Reprinted:
Thomason (1974: pp. 247270).
Morrill, Glyn V. (1994): Type Logical Grammar. Categorial Logic of Signs. Dordrecht, Boston,
London: Kluwer.
Myhill, John (1963): An alternative to the method of extension and intension. In: Paul Arthur
Schilpp (ed.) The Philosophy of Rudolf Carnap, La Salle IL: Open Court. 299310.
Nelson, David (1949): Constructible falsity. The Journal of Symbolic Logic 14: 1626.
Pierce, Benjamin (2002): Types and Programming Languages. Cambridge MA: MIT Press.
Quine, Willard Van Orman (1940): Mathematical Logic. Cambridge MA: Harvard University
Press. Revised Edition 1951.
Quine, Willard Van Orman (1953): New foundations for mathematical logic. In: From a Logi-
cal Point of View. 9 Logico-Philosophical Essays. Cambridge MA: Harvard University Press.
80101.
Ramsey, Frank P. (1927): Facts and propositions. Aristotelian Society Supplementary Volume
VII : 153170. Reprinted: Ramsey, Frank P.: Foundations. Essays in Philosophy, Logic, Math-
ematics and Economics. Ed. by D.H. Mellor. London: Routledge and Kegan Paul 1978.
4057.
Robering, Klaus (1994): Stufen, Typen, Sorten. In: Klaus Robering (ed.) Sorten, Typen, Typen-
freiheit. Probleme der Klassifikation semantischer Einheiten, Berlin: Technische Universitt
Berlin. 555.
Robering, Klaus (2000): Categorial graph grammar: A direct approach to functor-argumentor
structure. Theoretical Linguistics 26: 3173.
Robering, Klaus (2008): Ackermanns class theory. In: Klaus Robering (ed.) New Approaches
to Classes and Concepts, London: College Publications. 2356.
Scott, Dana (1971): The lattice of flow diagrams. In: Erwin Engeler (ed.) Symposium on Seman-
tics of Algorithmic Languages, Berlin, Heidelberg, New York: Springer. 311372.
Supecki, Jerzy (1971): A generalization of modal logic. Studia Logica 28: 713.
Specker, Ernst P. (1953): The axiom of choice in Quines New Foundations for Mathematical
Logic. Proceedings of the National Academy of Sciences 39.
Thomason, Richmond (ed.) (1974): Formal Philosophy. Selected Papers by Richard Montague.
New Haven: Yale University Press.
158 Klaus Robering
Thomason, Richmond (1989): Motivating ramified type theory. In: Gennaro Chierchia,
Barbara H. Partee & Raymond Turner (eds.) Properties, Types and Meaning. Volume I:
Foundational Issues, Dordrecht: Reidel. 4762.
Villadsen, Jrgen (2010): . A Linguistic System Based on Type Theory. Berlin: LIT Verlag.
Wansing, Heinrich (1993): The Logic of Information Structures. Berlin and New York: Springer.
Whitehead, Alfred N. & Russell, Bertrand (19101913): Principia Mathematica. 3 Vols.
Cambridge GB: Cambridge University Press.
chapter 4
Object-orientation and the semantics of verbs
Andrea C. Schalley
Griffith University
Brisbane/Australia
The paper introduces a decompositional approach to verbal semantics that is

object-oriented in nature and based on the Unified Modeling Language (UML),
the de-facto standard formalism for object-oriented software design and analysis.
In addition to laying out the very rich semantic structure of verbal meaning and
its object-oriented modeling, the paper focuses on the advantages a semantic
representation along the lines of the UML provides, in particular looking at the
multi-layered architecture of UML and its implications for verbal meaning. Other
semantic phenomena and problems (such as metaphors, questions of classifica-
tions of verbal meaning, or selectional restrictions) will be touched on and dis-
cussed in the light of an object-oriented, UML-based approach to verbal meaning.
1. Introduction
Verbal meaning and how to best represent it is still one of the hotly debated
issues in lexical semantics, partly due to the centrality and complexity of the
semantics of these linguistic expressions (cf. Section 4 for a brief overview).
Rigorous approaches to the representation of verbal semantics have been pre-
dominantly of a logical (cf., e.g., Montague 1973, Montague 1974, Portner &
Partee 2002, Gutirrez-Rexach 2003) or functional (cf., e.g., Jackendo 1990,
Jackendo 2002) nature.
This paper will look at verbal meaning from a theoretical linguistics point
of view, while being inspired by computer science. Logic, functional and object-
oriented programming can be seen as three dominant programming paradigms in
computer science. As indicated, both the logical and the functional approach have
received appropriate attention in linguistics and the treatment of verbal meaning.
This is not the case for the remaining paradigm, object-orientation. Therefore, the
question poses itself what an object-oriented approach might be able to oer to a
successful representation of verbal meaning. In this paper, I will give an overview
160 Andrea C. Schalley
of an object-oriented approach to verbal meaning, one which adapts the Unified

Modeling Language (UML, cf. OMG 19972010), the de-facto standard formalism
for object-oriented software analysis and design.
The paper is structured as follows: Section 2 outlines object-orientation as
a computational paradigm and argues for its relevance for semantics in general.
Section 3 concisely introduces the computational modeling language on which the
approach to verbal meaning advocated here is based. Section 4 provides a brief
introduction to verbal meaning and what aspects have to be captured when rep-
resenting verbal semantics. The remaining sections give an overview of the basics
of this approach (Section 5), the architecture of the modeling framework and its
implications (Section 6), and the issue of verb classifications (Section 7), which is
addressed in-depth in a number of companion chapters in this volume.
2. Object-orientation and its relevance to semantics
Before heading into the discussion of object-orientation and its relevance to lin-
guistic semantics, the first question is what motivation there is to look at concepts
from software development at all. What do software models and semantic represen-
tations have in common? Human perception filters reality and creates structured
mental representations, models that are reduced in their complexity to relevant
aspects of a situation. Meanings are such models: relevant recurring aspects of spe-
cific situations are selected and categorized (in particular as far as lexical meaning
is concerned, as in the case of verbal meaning). This results in semantic representa-
tions of a conceptual nature. Similarly, software models represent relevant aspects
in a structured way: Modeling languages from computer science (languages on a
higher abstraction level than programming languages) are targeted at formulating
models of reality as well, at reducing reality to the relevant aspects the software sys-
tem needs to be able to handle. Resulting models are structured representations of
conceptualizations of the domain in question. Both software models and semantic
representations thus aim at representing conceptual structures, and aim at doing
this as eciently, rigorously and intuitively as possible. There has been extensive
research over decades in computer science on how to best approach this (cf., e.g.,
Section 3), which is why it should be helpful to look at software modeling languages
in order to develop a convincing semantic representation framework.
The next question, following on from the discussion in Section 1, is: Why
object-orientation? Due to the limitations of this paper, I will not be able to explic-
itly compare the object-oriented approach advocated here with existing logical
and functional approaches, but I will outline why object-orientation appears to
be well-suited for the task at hand.
Chapter 4. Object-orientation and the semantics of verbs 161
What is the essence of object-orientation? As the name indicates, the cen-

tral, core elements of this approach are objects. An object represents a concrete
entity. In particular, an object has an identity and is thus distinguished from other
objects (even if two objects looked identical because of what is known about them,
they would still be dierent). The representation of an object comprises a descrip-
tion of the entitys (i) characteristics (attributes), (ii) relations to other objects,
(iii) behavior (roughly, actions, states, and changes of state), and (iv) interactions
(communication with other entities through message passing) all these pieces of
information are encapsulated in the computational object definition. This encap-
sulation parallels what happens in cognitive systems, where the entity is conceptu-
alized as central element, together with its (i) characteristics, (ii) relations to other
entities, (iii) behavior, and (iv) interactions with other entities. For example, the
table I am sitting at while writing this paper is (i) brown and wooden, (ii) enter-
tains a supporting relation to the laptop which is sitting on it, (iii) is stationary
(does not move), and (iv) if being pushed by me will move somewhat away from
me. A cognitively oriented verbal semantics will strive to represent (inter)acting
entities, and hence there is a direct correspondence between object-orientation
and semantic modeling.
Objects are instances in the software system, they are instances of classes,
in which the structure and behavior of a set of similar objects is described in a
more abstract way. I.e., it is defined in a class what kind of characteristics (e.g.,
animacy) a member object has (but generally not the exact value, such as inani-
mate this is specified at the object level), which relations it entertains (e.g.,
part-whole relation), and which behavior and which interactions it shows under
which circumstances (e.g., if x happens, the objects of the class all react in the
same way to x). From a computational point of view, an object is an instance that
exists at execution time and behaves according to its class definition, while the
corresponding class is a salient concept of shared characteristics and behavior
of similar objects. This works similarly in linguistic semantics and cognition: an
object is a discourse referent and context dependent (i.e., a concrete entity), while
it is a member of a context independent, more abstract category, the concept that
underlies a lexical element. For example, the table I am sitting at while writing
this paper is an object, and an instance of the concept TABLE, which is generally
coded by English table.1
1. Concepts are indicated by capital letters (such as the concept TABLE), while linguistic expres-
sions are italicized (such as the noun table). Since meaning is conceived here as being of a con-
ceptual nature (cf. Section 2), capitals are also used when referring to semantic representations.
Model elements are set in typewriter font (such as Dog).
At the same time, the table I am sitting at while writing this paper is an
instance of the class DESK, which is a more specific class than TABLE. This takes
us to one feature object-orientation is hailed for, and which is useful for linguistic
semantics: inheritance. Inheritance is a taxonomic relation between a more gen-
eral element and a more specific element, with the latter being fully consistent with
the former (i.e., inheriting the formers characteristics and behavior), while adding
additional information. In particular, classes can inherit from one or several other
classes, thus allowing the establishment of a taxonomic or hierarchical order (e.g.,
Dog being subordinate to both Animal and Pet, while at the same time being
superordinate to German Shepherd and Poodle). Inheritance readily allows for
complexity management (through the emerging structuring of the domain and its
resulting modularity) as well as expandability and hence evolution of the system.
Inheritance as a concept is highly relevant to linguistic semantics, as shown not
only by hyponymic relations (between, e.g., table and its hypnonym desk) and the
general taxonomic structuring of the lexicon (as amply demonstrated by WordNet,
cf. Fellbaum 1998), but also by the general expandability of speakers lexicons
(lexical acquisition of both adults and children, and phenomena such as over-
and under-extension) and ontological knowledge. The latter is relevant for, for
instance, anaphora resolution cases such as (1), where the reptile is anaphorically
related to a lizard, based on LIZARD being subordinated to REPTILE:
(1) When you try to catch a lizard, the reptile may drop the tail and escape
(Nickles et al. 2007:38)
Another feature of object-orientation is polymorphism (the in principle same

behavior can be adjusted to the aected entity), the relevance of which to lin-
guistic semantics is extensively discussed in Benz (2014). Together, both features
are important for an adequate representation of linguistic semantics, given that
language is a general purpose and self-organizing system and changes over time
need to be replicable by our representation framework.
3. Modeling object-oriented: Unified Modeling Language (UML)
How can classes and objects and their behavior and interactions actually be cap-
tured? This section introduces the computational modeling language on which the
semantic representation framework advocated here is based, the Unified Modeling
Language (UML).
UML is a general-purpose modeling language in computer science. It is a
language for specifying, visualizing, constructing, and documenting the artifacts
of software systems, as well as for business modeling and other non-software
systems (OMG 2001:1-1). Extending this, a UML-based semantic representation

framework has been developed in order to make UML utilizable for verbal seman-
tics (cf., among others, Schalley 2004). UML has been and is being standardized by
the Object Management Group (OMG, OMG 19972014), and is a de facto indus-
try standard.2 The result of a synthesization process, UML fuses the concepts of
mainly three, over many years developed predecessing notations Booch Method,
Object-Modeling Technique (OMT), and Object-Oriented Software Engineering
(OOSE) into a single widely usable modeling language (OMG 2001:1-6). The
UML formalism includes a syntax (notation) and a semantics for its modeling
concepts. Syntax and semantics can be found in the specification, together with
well-formedness rules for the modeling concepts. Eorts are still under way to
develop a formal semantics for UML.
UML comprises an extensive number of graphical notation techniques it is
a mixture of graphical, two-dimensional elements and linear textual constructs
(expressed in either natural or logical language). Its representational form is dia-
grammatic. UML provides for dierent views onto a system, with dierent dia-
gram types for both structural (i.e., static) and behavioral (i.e., dynamic) aspects.
The diagrams are essentially graphs containing nodes connected by paths. The
information is mostly in the topology, not in the size or placement of the symbols.
Three kinds of visual relationships are important (cf. OMG 2001:3-6): (i) connec-
tions (usually of lines to two-dimensional shapes); (ii) containment (of symbols
by two-dimensional shapes with boundaries); and (iii) visual attachment (one
symbol being near another in a diagram). Furthermore, there are four kinds of
graphical constructs that are used (cf. OMG 2001:3-63-7: (a) icons (graphical
figures of a fixed size that cannot hold contents); (b) two-dimensional symbols
(which have a variable height and width and can expand to hold other things);
(c) strings (which present various kinds of information in an unparsed form);
and (d) paths (sequences of line segments whose endpoints are attached to each
other, a two-dimensional symbol, or an icon). Paths may have terminators, i.e.,
icons that appear on the end of the path and that qualify the meaning of the path
symbol. An example notation of inheritance relations between dierent classes is
shown in Figure 1.
The rectangles represent classes, holding in this example only their respec-
tive name string which identifies the class. The paths connecting the classes, with
the hollow triangle as terminators, indicate inheritance (or, as called in UML,
generalization) relationships. The end of the path with the hollow triangle meets
the more general, i.e., superordinate element. That is, Sailboat is more specific
2. The semantic representation framework has originally been based on specification version 1.4
(OMG 2001), while the current specification version is 2.4 (OMG 2011).
Vehicle
power venue
power venue
WindPowered MotorPowered
WaterVehicle LandVehicle
Vehicle Vehicle
Sailboat Truck
Figure 1. Inheritance relations as represented in UML
and inherits from both WindPoweredVehicle and WaterVehicle, which

both in turn inherit from Vehicle, the most general class in the Figure 1 exam-
ple. The strings attached to the paths with Vehicle as more general element are
discriminators, which declare that the more specific elements (such as the class
WindPoweredVehicle) are in a given partition, in this example between how
a vehicle is powered (power) and where it operates (venue).
4. Verbal meaning
In order to be able to discuss and find an appropriate meaning representation for

verbs, it needs to be clarified (i) what a verb is, and what its role in supra-lexical
expressions (such as utterances) is, and (ii) which meaning elements or aspects are
required for a comprehensive representation of verbal meaning. In the following,
I will address at (i) and (ii) in turn.
What is a verb? In primary school, we learn that verbs are do-words, in con-
trast to, e.g., nouns, which are thing-words. This is, of course, a crude simplifica-
tion; yet, it points us in the right direction. First, verbs are linguistic expressions
that are members of a particular lexical class. The delimitation of this lexical class
is not trivial. It can best be made on the basis of functional considerations (cf. also
Schalley 2004:11), in describing verbs as elements which typically function as lin-
guistic signs for events and similar entities (called eventities, following Zaeerer
2002), entities that are characterized by their dynamism and hence that run within
time. Secondly, verbs generally are considered to be the core or central words or
word combinations of supra-lexical expressions (in analogy to syntax analyzing
the verb phrase as the most prominent phrase in a sentence, with the verb as head
of the verb phrase and thus its most prominent element); their semantics provide
the scaolding for the semantic interpretation of the supra-lexical expressions.
In other words, there are two dierent viewpoints on verbs one being lexical,
where a verb is seen as an abstract unit (or type), the other being discourse rep-
resentational, looking at the concrete usage of these abstract units in supra-lexical
expressions, i.e., their tokenization or instantiation.
For question (ii), I am adopting the view that meaning is decomposable, i.e.,
that meanings can be deconstructed into meaning components and their struc-
tural relations in line with, e.g., Dowty (1979), Jackendo (1983), Jackendo
(1991), Pustejovsky (1995), and in contrast to, e.g., Fodors holistic approach
(Fodor 1975, Fodor 1983). So, which components (together with their relations)
require inclusion in the representation of verbal meaning? As posited above, verbs
are linguistic signs for eventities, for conceptual entities that run within time.
Thus, the core meaning elements are dynamic in nature, representing what is hap-
pening. In addition, it is important to include who is doing something or to whom
something is happening, i.e., who or what can participate in the eventity in ques-
tion, i.e., is a potential participants. This means that the representation of verbal
meaning has to be able to capture at least:
The overall eventity coded by the verb, as conceptual unit, such as ARRIVE.
The dynamics inherent in the eventity.
What is happening as part of the eventity (what behavior3 do the involved
entities [participants of the eventity, cf. below] show)? E.g., in the case of
ARRIVE the entering of the arrival location.
How is the eventity dynamics substructured? (Which behavior follows
which behavior?) E.g., in the case of ARRIVE there is an instantaneous
transition into a state of being at the arrival location.
Which part of the overall dynamics can be attributed to which partici-
pant? E.g., can the arriving be attributed to a specific participant?
The potential participants of the eventity (who is it that is carrying out or
undergoing the dynamics?).4

Which participants are conceivable? What ontological types do the par-
ticipants represent? Are they, for instance, individuals, or substances, or
3. The term behavior is used here as a very broad notion, encompassing not only the actions
of an entity/participant, but also the states and changes of state (transitions) they undergo.
4. On a lexical level (the type level), there is generally information on potential participants
available, while the specification of actual participants only happens at the discourse level (the
instantiation level). I focus here on the lexical level, as I will primarily be looking at verbs as
lexical items, irrespective of their actual embedding in specific supra-lexical expressions.
eventities themselves? E.g., the eventity coded by the English intransitive

lift as in x lifts does not allow individuals as participant, but rather sub-
stances such as MIST (cf. the mist began to lift).

Which characteristics or selectional restrictions do apply to the partici-
pants? E.g., a stone cannot be woken up, that is, the WAKE_UP eventity
requires an animate undergoing participant with a stone not generally
being conceptualized as animate.

Which roles do the participants have within the eventity are they agent,
experiencer, instrument, patient, location, or goal?5 E.g., the arrival loca-
tion in the ARRIVE eventity holds the goal role.

Which relations between participants are important for a participant con-
figuration to be possible? E.g., in PUT the agent has to have location and
movement control of the entity that is being put (Schalley 2004:282285).
Other aspects a representation system for verbal semantics should be able to
systematically capture through abstraction mechanisms are, for instance, under-
specification and verbal classification (cf., e.g., Levin 1993, or the extensive dis-
cussion on aktionsarten, as also sampled in this volume). Here, I understand
underspecification as a phenomenon in which certain pieces of information (in
many cases structurally requested) are not specified in semantic representations.
For example, most if not all changes of state (transitions) that occur in verbal
semantics have either an unspecified source or target state. This is exemplified by
the eventity WAKE_UP as coded by the English intransitive wake up (for more
details, cf. Section 5), in which it is unspecified if the undergoer is in a source
state of sleep, unconsciousness, day dreaming or similar. Underspecification is
important to allow for verbs to be flexibly usable in many dierent contexts
only if this is provided for does it make sense for a language to include an item
in its lexicon. For instance, a verb coding the eventity WAKE_UP_FROM_
UNCONSCIOUSNESS is presumably much less likely to evolve than a verb
encoding the more general WAKE_UP eventity. Only if the eventity is general
enough and applicable in many situations, it is likely that a lexical item for the
coding of this eventity is stable.
5. For a discussion of such role notions, cf. Schalley (2004:1722), or Van Valin & LaPolla.
(1997).
5. Object-oriented representation of verbal meaning: The basics
So far the discussion has remained somewhat in the abstract I have argued for an
object-oriented approach to meaning in general (Section 2), provided background
information on the computational modeling language the proposed representa-
tion framework is based on (Section 3), and addressed the question of what a verb
is and which meaning components need to be captured when representing verbal
semantics (Section 4). This section introduces the basics of the object-oriented
representation framework for verbal semantics. The framework is called Unified
Eventity Representation (UER), because it is an adaptation of UML while at the
same time focusing on representing the concepts underlying verbs, eventities.
Earlier, objects have been named core elements of an object-oriented
approach, in whose specification their behavior is also defined (in accordance to
their class definition). In other words, the focus of an object-oriented approach
appears to lie in the specification of non-eventity entities, on structural aspects
of the system. Verbal semantics is focusing on the opposite on dynamic, pri-
marily non-structural aspects, on eventities as the concepts underlying verbal
meaning. It thus seems that an object-oriented approach and UML might not be
usable in a straightforward way. This is, however, counteracted by the dierent
views on the system (i.e., modeled domain) and dierent diagram types that
UML supports, some of which are targeted at dynamic aspects (namely, the
dierent behavior and interactions diagrams, for more information, cf. OMG
(2001) or OMG (2011)). The UER as a UML adaptation has combined some
of these diagram types in order to provide a unified representation suitable for
eventities. While a specification of behavior within class definitions has its indis-
putable benefits,6 a representation of eventities in their own right is necessary
for a comprehensive coverage of the linguistically relevant conceptual system.
For a representation of verbal semantics, eventities are modeled on the type (or
model) layer, as this is the layer on which salient concepts, such as the mean-
ings of lexical items, are represented,7 as general and underspecified as possible.
Essentially, representations of verbal meaning are models of behavior that can
be realized by classes, potentially adjusted according to the class in question
(polymorphism, cf. Benz 2014).
6. Cf. Benz (2014) for a discussion of its relevance.

7. The model layer is one of the architectural layers of the UML (and UER) specification. It is
the layer on which classes (and hence salient concepts) are defined. For a more in-depth discus-
sion of the architecture and relevance of the dierent architectural layers, cf. Section 6.
In order to show what a semantic representation with the UER looks like, I
will discuss the prototypical reading of English wake up as instantiated in (2).8
(2) a. Conrad woke Vincent up.
b. The loud music woke up the little boy.
c. The blowing of the wind woke Savannah up.
The corresponding WAKE_UP-eventity model is depicted in Figure 2.9
1
13
14 wake up
10
/ Agent / Effector
3 11
[[y]] / Patient : Individual 12

[[x]] / Instigator : Entity intrinsic ani:Animacy = animate
9
do undergo
8
x y
4
6
cause
cause Awake
2
7
Figure 2. Example: the WAKE_UP-eventity
8. Much of the remainder of this section is adapted from Schalley (2004:7680). A precise speci-
fication of the modeling elements referred to in the following can also be found in Schalley (2004).
9. In addition to the modeling of the eventity, Figure 2 has fourteen notes with numbers
attached (such as 1 ), which are merely additions to the representation itself. They are used
for referencing specific modeling elements in the course of the examples discussion.
First and foremost, the UER has a graphical container representing eventities:
octagons which are called eventity frames (No. 1). The character string wake
up in the upper left corner indicates the name of the eventity frame within the
model. The eventitys components are encapsulated within the eventity frame. As
discussed in Section 4, both dynamic aspects and related participant information
are part of verbal meaning and thus eventity concepts and require modeling. The
dynamic aspects are shown in the dynamic core, a dashed rectangle with rounded
corners (No. 2). Related participant information is depicted around the dynamic
core, but within the eventity frame. Participants x and y of the eventity (the vari-
ables stand for potential participants names) are represented by rectangles (No.3),
called participant classes (cf. below for further discussion).
The dynamic core is separated into two compartments (swimlanes), each of
which displays the behavior of one participant, namely of the participant that is
referenced in the upper left corner of the respective swimlane. Elements crossing
the vertical swimlane border depict interactions between participants. In the left
swimlane, the behavior of participant x is modeled, showing that x is performing
some action. In the UER, active simple states, containers with straight top and
bottom and convex arcs on the sides (No. 4), represent actions. The underscore in
the body of the active simple state illustrates that it is not conceptualized what x is
doing the action is unspecified. Yet, an interaction between x and y is provoked
by it and causes a change of state of y. Causation is expressed in the UER by the
cause-signal (No. 5), consisting of the two pentagons (the convex one in xs swim-
lane indicates the sending of the signal, the concave one in ys swimlane the receipt
of the signal) and the connecting arrow. ys change of state is represented by a
transition shown in ys swimlane. A transition comprises a solid connecting arrow
(No. 6) bipartite in the example because the signal receipt triggers the transi-
tion and source and target state. The target state in the WAKE_UP-eventity is a
passive simple state, depicted by a rectangle with rounded corners, which is speci-
fied as Awake (No. 7). That is, y is in the (passive) state of being awake as a result
of the transition. Nothing is known about the source state (the state in which y is
before the change occurs) it is displayed as an unspecified source state (No.8).
This unspecification does not only extend to the states detailed specification (e.g.,
whether y is asleep or day dreaming), but also to its structural characteristic (i.e.,
it is unclear whether it is an active or a passive state). Yet, as a transition and hence
change of state is taking place, what is known is that the source state does indeed
dier from the target state.
Dashed connectors between participant classes and the dynamic core, the par-
ticipate associations (No. 9), indicate participant status as such. The attached ste-
reotypes do and undergo mark actor and undergoer of the eventity, singling
out the prominent participants of the eventity.10 Explicit marking of prominent

participants is important: their behavior is explicitly conceptualized, i.e., each
prominent participant owns a swimlane in the representation. The definition of
participants (not only of prominent participants, although in this example all par-
ticipants are also prominent participants) is given in the participant classes (No.3),
in the static periphery of the eventity frame. The basic participant class syntax is:
(3) representative-variable / role-expression : type-expression
The representative-variable is, as mentioned above, a placeholder for potential

participants names, such as y for the undergoing participant. Participant classes
function as filters: only those entities are potential participants which meet the
requirements specified in the respective participant class definition. The role-
expression, given as Patient (No. 10) for y, names the semantic role that best
captures the role of the participant in the eventity.11 The type-expression, in the
example modeling Individual (No. 11), lists the ontological category of the
participant. The ontological categorization is merely a top-level categorization,
in accordance with top-level distinctions as reflected in natural languages gram-
mars. The modeling so far would thus still allow a stone to be woken up, which is,
however, not supported by the prototypical reading under discussion and exem-
plified in (2). In order to address this, selectional restrictions are displayed. In
the example, the undergoer is hence additionally specified as animate entity by
way of an attribute (No. 12) in the undergoers participant class. In the actor xs
participant class specification, the role-expression Instigator is italicized, indi-
cating that it is an abstract characterization that cannot be directly instantiated.
Actors are instead either Agents or Effectors (No. 13), as expressed by the
generalization or inheritance relationship (No. 14). They inherit all the proper-
ties of Instigator but add the additional information of whether the actor
acts volitionally (and is an agent) or non-volitionally (and is an eector). (2a) on
p.168 is an instantiation where the actor is interpreted per default as agent (but
could also be an eector), whereas (2b) and (2c) are instantiations where the actor
is an eector. Note in particular that the actor participant of (2c) is an eventity
itself (coded by a noun phrase with a nominalized verb as head), which is why its
type expression has not been restricted in any way but shows the top ontological
category of Entity.
10. Prominence of participants largely corresponds to the notions of proto-agent and proto-
patient in Dowtys (1991) terms or to actor and undergoer in Van Valin and LaPollas (1997)
terms. For further discussion, cf. also Schalley (2004:1622, 225226).
11. For a more extensive discussion of role notions in UER, including how to specify (standard
and non-standard) semantic roles, cf. Schalley (2004:200206).
6. Multi-layered architecture
Important from a system and theory-building point of view is not only the intro-
duction of adequate modeling elements (cf. the previous section), but also the
explicit specification of their semantics. A prime example of this is the spec-
ification of logical languages, the semantics of which are rigorously defined.
For the UML and UER, this is achieved through its multi-layered architecture,
which consists of four distinct modeling layers. These are: (i) instance layer
(called user objects layer in the standard), (ii) model layer, (iii) metamodel layer,
(iv)meta-metamodel layer. While (i) and (ii) are part of the actual modeling
and representation language, (iii) and (iv) are meta-level structures, with (iii),
the metamodel, being most relevant for the semantic specification of the UML
and UER modeling languages. I will discuss layers (i) and (ii) first, followed by
a discussion of (iii) and (iv).
The distinction between (i) and (ii) has already been touched on above, as it
corresponds to the distinction between instance and type. Objects are elements
on the instance layer, classes are elements on the model layer. Accordingly, on the
model layer conceptual categorizations are represented, salient concepts of shared
characteristics, relations and behavior of similar objects, while on the instance
layer contextualized instances of such concepts are modeled. Instances are used to
represent specific tokens or entities. If I want to represent the conceptual category
TABLE, I will do so via a class on the model layer; if I want to represent the table
at which I am sitting while writing this paper, I will do so via an object representa-
tion on the instance layer.
Similarly for the verbal domain: Figure 3 on p.172 depicts the representa-
tion of SIT and hence of a reading of sit as well as an instantiation of this eventity
as coded by English James sits on the sofa, where James and the sofa are actual
participants. Hence, the latter is a contextualized instance of the eventity SIT
coded by English sit. Instead of representative-variables, the participants of the
contextualized eventity (James and the sofa) are listed in the participant object
specification on the instance layer. The same applies to the swimlane attribu-
tion, which names the participant James. Position is an association class on
the model layer, and as such represents a relationship between the theme and
ground participants depicted by participant classes, while also defining a set of
features (via attributes) that belong to the relationship itself and not to any of the
connected classes.12 On the model layer, these features are only defined, without
12. The representation of relationships and their features and characteristics is unique to the
UER. No other semantic representation system I know of provides such an elaborate system for
modeling dierent relations and relational types.
sit
[[x]] / Theme : Ineventity

dispositional
disp : Posture = sitting
undergo
x Position
be-at post : Posture = sitting
Z loc : Situatedness
cont : Contact
supp : Support
[[x]] / Ground : Ineventity

axis: Axis sit
dim : Dimension
James / Theme : Ineventity
dispositional
disp : Posture = sitting
undergo
James Position
be-at post : Posture = sitting
Sofa loc : Situatedness = on
cont : Contact = yes
supp : Support = yes
Sofa / Ground : Ineventity

axis: Axis = horizontal
dim : Dimension = volume
Figure 3. Suggested modeling for the concept SIT and its instantiation as coded
byEnglish sit and James sits on the sofa, respectively (Schalley 2004:183)
providing actual values. Any instantiation of this association class (and hence
the overall eventity) will be required to specify the values of those features. In
Figure3, for instance, the Position relation will have to be specified as to
whether a contact exists between the two participants or not, via the attribute
cont which is of the enumeration type Contact its value is specified as yes at
the instance layer. This allows for a distinction between on and above, as exempli-
fied in (4a) and (4b), respectively.
(4) a. James sits on the sofa
b. James sits above the sofa
As for the Position relation, the attribute values for the ground participant
have to be specified on the instance layer. This allows for details of the situation
conceptualization to be captured, such as that the ground participant is an object

which has a prominent horizontal axis and is a volume (i.e., 3-dimensional, in
contrast to a line or plane). The importance of such feature information is well
exemplified by Talmy (2007) for the spatial domain. Note, however, that for some
attributes the values are already specified at the model layer. These are essential for
the representation of SIT on the model layer: in order for SIT to occur, a potential
undergoing participant has to have the disposition of assuming a sitting position,
and the Position relation is further characterized via a sitting posture.
Figure 3 hence exemplifies the two dierent viewpoints on verbs and their
underlying conceptualizations mentioned earlier, with the model layer represen-
tation for SIT depicting the modeling of an abstract lexical concept, while the
instance representation displays the conceptualization underlying a concrete usage
of the lexical item in the supra-lexical expression (4a). In distinguishing layers
(i) and (ii) in the UER (and UML) architecture, the semantic dierence between
lexical item and discourse item is reflected explicitly in the modeling on dierent
layers. This is a clear advantage of the object-oriented approach to semantic rep-
resentation; the UER semantic representation system is thus able to cover both
lexical semantics and the basics of discourse representation.
Layers (iii) and (iv), the metamodel and meta-metamodel layer of the UML
and UER, are meta-level structures. While representations on layers (i) and (ii) are
models of a domain of interest (in our case the representation of natural language
meaning), representations on (iii) specify how such models on layers (i) and (ii) are
to be understood. That is, the metamodel defines a language for specifying models,
and explicitly states the semantics of the modeling elements that are used on (i)
and (ii). For example, the association class Position in Figure 3 is an instance
of the metamodel element AssociationClass, which inherits from both Class and
Association (the latter defines general relationships). The general semantics of asso-
ciation classes is defined in the metamodel (e.g., that they are both a class and an
association). Through the metamodel, the semantics of the representation system is
explicitly specified: The metamodel architectural pattern is a proven infrastructure
for defining the precise semantics required by complex models that need to be reli-
ably stored, shared, manipulated, and exchanged across tools (Kobryn 1999:33).
The metamodel layer is itself an instance of the meta-metamodel layer, which
forms the foundation for the metamodeling structure and defines the language
for specifying the metamodel. It is more compact than the metamodel that it
describes. It is essentially a core static structure (classes and their characteris-
tics, types of relationships and wellformedness rules/constraints). Generally, the
meta-metamodel is a common core language that can be used in the definition
of dierent metamodels; a standardized meta-metamodel language ensures easy
comparison and cross-portability between modeling languages. An important
advantage of having a meta-metamodel and hence a four-layer system architecture
is that it provides the infrastructure for defining extensions and model-layer exten-
sion mechanisms. This has been vital in the design of the UER (which adapts the
UML and has a number of its own modeling elements). An extension of the UER
from a verbal semantics representation framework to a compositional semantics or
discourse representation system will also make extensive use of this infrastructure.
A specific characteristic about metamodeling is the ability to define languages
as being reflective, i.e., languages that can be used to define themselves (OMG
2011: Infrastructure Specification: 18). The same language that is used to repre-
sent models on the model and instance layers is used to represent the metamodel
(and the meta-metamodel) and therefore the modeling language itself. The UER
is thus a (partially) self-describing system. The ability to carry out such meta-level
discussions and reflect about the representation language itself in a rigorous way
is very important. It clearly helps to keep the representational framework concise
and precise and helps to be attentive to arising issues and necessary developments.
Noteworthy about the architecture is furthermore that each element in a
lower layer is an instance of an element in the next higher layer. For instance, a
model element is an instance of a metamodel element, which in turn is an instance
of a meta-metamodel element (e.g., model class Car being an instance of the
metamodel class Class, which in turn is an instance of the meta-metamodel class
MetaClass). Navigation between layers is built-in, achieved via the instanceOf
dependency relation which is defined as part of the UER, the instantiation relation
that is one of the backbone relations of ontologies more generally.
The distinction of dierent layers in the UERs architecture has advantages
from a conceptual point of view, too:
Specific model information is separated from general definitional meta-
information. There is a clear conceptual delineation between, for instance,
characteristics that are common to all part-whole relations (definitional char-
acteristics, e.g., the whole has an independent ontological existence which
transcends its parts) and those that only apply to specific part-whole relations
as represented on the model layer (e.g., separability: the separable part can
be removed from the whole and may exist independently of the whole, cf.
Schalley to appear and Winston et al. 1987:420).13
13. It might be surprising to find one of the other backbone relations of ontologies, the part-
whole relation or meronomy, as an example here, as it might not be seen as having relevance
to the realm of verbal semantics. Yet, it is indeed important for verbal meaning and discourse
representation: (a) If, for example, a body part of the agent constitutes the undergoer of an even-
tity (as is possible in the case in SETZEN, an eventity coded by German setzen put, cf. Schalley
2004:309311), then this part-whole relation between agent and undergoer often triggers a
special though in my eyes questionable treatment in discussions of the verbal meaning (as,
e.g., in Schindler 2001 for German legen) due to an apparent dierence in conceptualization.
The system can be understood as being based on conceptual frames in

Barsalous sense (Barsalou 1992; Barsalou & Hale 1993), which are considered
to be the basic format of concept formation in cognition. Recursion and itera-
tion through and within layers defines frames and categorizations that can be
and are recursively used for modeling and further specification. Recursively
refining semantic constructs through and within successive instantiation layers
thus results in a highly sophisticated semantic representation system. While
the system is essentially frame-based, cognitively relevant concepts and sortal
types become readily available for semantic representation at the model layer.
A separation into architectural layers is essential to make available and manage
the conceptual complexity required for verbal meaning representation.
As a result, the UER representation framework iconically distinguishes
dierent conceptual types by providing dierent modeling elements (as
defined in the metamodel) for the dierent types (e.g., state vs. transition,
class vs. relation). Important conceptual types are graphically depicted in the
system (e.g., the rectangle depicting classes, connecting paths showing rela-
tions), which highlights their conceptual and structural relevance and their
centrality within the system. And conceptual units are represented as units
within the system (together with their characteristics, relations etc.), which
renders a system close to conceptualization, as already argued earlier.
7. Eventity classes
One theme that is taken up in a number of contributions to this volume is aktion-

sarten and aspectual classes of verbs, or more generally how to semantically
classify verbs. As I will show in this section, deploying the UER leads to a new
approach toward the discussion of aktionsarten and semantic classification, one
which is generative and onomasiological in nature, and which will result in a quite
extensive departure and at the same time refinement of Vendlers (1967) quadri-
partition of state, activity, achievement, and accomplishment.14
Aktionsart is a notion deployed to classify verbs into dierent categories
according to particular properties of the temporal or logical structure of the
(b)In Example (1) above, there is another anaphora that requires resolution, namely between a
lizard and the tail. This is possible due to the part-whole relation that holds between the lizard
and its tail.
14. I will not recapitulate Vendlers classification and argumentation here, the interested reader
is referred to Vendler (1967) and, e.g., other contributions to the present volume, such as Dlling
(2014) and hl (2014).
underlying eventities (cf. Dowty 1979:51f). The categorization is based on inher-

ent properties of eventities, e.g., telicity and durativity (temporal extendedness).
However, assigning aktionsarten to verbs and hence classifying this lexical class
has proven to be a dicult task (for a discussion cf., e.g., Zybatow 2001), as not
only properties of the lexical verb (e.g., run) appear to be relevant, but also verb
phrases with their complements and other modifications (e.g., cf. He is running
vs. He is running home). Hence, such a classification should not be thought of as
classifying verbs themselves (the surface elements), but the underlying eventities
they code. Their classification will be targeted in this section, and the UER is well-
suited for this task, as it represents the underlying eventities.
As outlined in the previous section, the UER modeling language, by way of its
architecture, results in a representation system in which (i) dierent conceptual
types are represented by dierent modeling elements and (ii) structurally impor-
tant conceptual types are graphically depicted in the system. There is a graphical
container, a rectangle with rounded corners, representing passive simple states,
while the name of the state and thus its specification is given as a character string.
This interplay between graphical and textual elements allows to abstract away from
the specificities of a model element through parameterizing the textual informa-
tion, while maintaining the graphical structural information. Cf. Figure 4, which
shows a simple passive state of knowing as well as its parameterization, leaving us
in the second case with the information that we are dealing with a passive simple
state, without specifying this any further. Only structural information remains in
this case. Parameterized model elements are called templates in the UER.15
Know Z
Figure 4. Passive simple state Know and abstraction through parameterized

passivesimple state
Templates are elements on the model layer of the UER, but are not directly deploy-
able because of their parameterization: A template cannot be instantiated, as the
parameters have to be bound to actual values first. Templating is one abstraction
mechanism that is built-in in the UER, together with others such as the inheri-
tance and instantiation relations (i.e., the abstraction achieved through superor-
dination and through the dierent architectural layers, respectively).
15. More precisely, the syntax of templates also comprises a list of parameters and those param-
eters types, thus constraining the actual values the parameters may be bound with. For the
sake of graphical brevity, I will disregard this list here. For the specification of templates and a
discussion of their linguistic relevance, cf. Schalley (2004:169178, 222224).
From a conceptual point of view, this abstraction can be interpreted as gen-

erating categories (of a primarily structural nature) on the model layer, where
families of ordinary (i.e., nontemplate) model elements are defined: each mem-
ber of such a family is specified by binding all the template parameters to actual
values. For instance, the parameterized state in Figure 4 (on p.176) represents all
passive simple states, and binding its only parameter Z to Know results in the the
representation of the passive state of knowing.
In other words, templates are patterns with slots to be filled in order to yield ordi-
nary semantic representations of concepts. Accordingly, they do not represent
concepts as such, but abstractions of concepts, or, if one takes all possible bindings
into consideration, they define families of concepts. (Schalley 2004:222)
Eventity classes are exactly this: families of concepts. That is, templating results in
exactly those kinds of representations that are required for modeling eventity classes.
Therefore, templating the dynamic core of eventity frames leads to a system-
atic classification of eventities with regards to their temporal and logical structure.
One approach toward aktionsarten hence is to take the central dynamic modeling
elements of the UER (state, transition, and cause-signal in both their durative
and non-durative [= default] version) and to systematically model parameterized
dynamic core configurations deploying these modeling elements. The resulting
configurations represent the dynamic structure (without further specificities) of
eventities that are potentially coded by verbs or verb phrases. The onomasiologi-
cal approach thus presents possible conceptual configurations, and then discusses
which of these configurations are lexically coded. (The combinatorial properties
of the modeling elements restricts which configurations are possible and hence
which aktionsarten can in principle be expected.) Schalley (2004:Chapter8,
pp.251280) presents the following basic classification, subdivided into three
eventity class categories (for a specification of the classes, cf. the corresponding
figures, for example lexicalizations cf. Table 1 on p.181):
Statal eventities (Figure 5, p.178):
The most basic element of dynamic modeling is the simple state, both passive
and active. Passive simple states are what is generally called state in aktions
art discussions (they are inherently extended), active simple states represent
actions, which are generally subdivided into those actions that show a pro-
gression in time (and are hence durative), called activities, and those that
show punctiformity, called acts in the UER.
Change eventities (Figures 6 and 7, p. 179):
Change eventities take up another building block of dynamic modeling: the
transition. Accordingly, they also require a source and a target state for the tran-
sition. Change eventities only have one prominent participant who undergoes
x
Z State
Z Activity
Z Act
Figure 5. Statal eventities
a transition, and are hence coded by intransitive verbs (as are statal eventities).
In order for lexicalizations of such eventities to be suitably flexible, either the
source or the target state of the transition is unspecified. There are therefore
the following variation dimensions for change eventities: (i) source or target
state unspecified (while disregarding whether it is a passive or active state), (ii)
parameterized passive or active state, (iii) transition gradual16 (i.e., durative) or
non-durative. This leaves us with 23=8 configurations, which are: achievement,
terminative, alterative, desitive, ingressive, egressive, inchoative, and conclu-
sive. (This basic classification disregards dierences regarding the durativity of
the active state which have been taken into account for statal eventities.)
Interactional eventities (Figure 8, p.180):
Interactional eventities integrate the last building block of dynamic modeling:
the cause-signal. This results in eventities involving two interacting prominent
participants, with the agent carrying out an action (modeled via an active
simple state) that causes a transition of the undergoer. As with change eventi-
ties, the transition has one unspecified state (source or target) to maintain
the coded eventitys utilizability in dierent contexts. Variation dimensions
are: (i)whether the agents active state is unspecified or specified (and hence
parameterized), and (ii) whether the parameterized state of the undergoers
transition is the source or target state. This leaves us with 22=4 configura-
tions: causative and deducive (unspecified active state of agent), resultative
16. A gradual transition is represented by a gradual pseudostate, which is shown as a circle sur-
rounding a small hollow circle: .
x x
Z
Achievement Terminative
x x
Alterative Desitive
Figure 6. Change eventities (with passive state)
x x
Z
Ingressive Egressive
Z
x x
Inchoative Conclusive
Figure 7. Change eventities (with active state)

and expulsive (specified and hence parameterized active state of agent).

Several other variation dimensions are disregarded in this basic classifica-
tion: (a) whether the agents active state is durative or not, (b) whether the
transition of the undergoer is gradual or not, (c) whether the cause-signal is
continuous (i.e., durative) or not,17 (d) whether the parameterized state of the
transition is an active or a passive state, and (e) if the parameterized state of
the transition were an active state, whether this active state would be durative
or not. These dimensions have been disregarded by Schalley (2004) for the
basic classification, as they would increase the conceptual complexity beyond
the level of granularity desirable for a foundational approach to eventity clas-
sification.18 In addition, the variation dimensions are not orthogonal to one
another, and a more fine-grained classification would have to take dependen-
cies into account, such as the one mentioned in footnote 17.
Causative Deducive
x y x y
cause
cause
cause cause
Z
x y x y
A Z
A
cause
cause
cause cause
Z
Resultative Expulsive
Figure 8. Interactional eventities
17. Note that a continuous cause-signal requires that the agent is performing an activity and
that the undergoers transition is gradual.
18. Admittedly, there is no objective measure for how complex conceptual structures can be to
be still acceptable for a basic classification. This judgment is based on intuition only and would,
if possible at all, require the scrutiny of psychological research.
Table 1 lists all the generated eventity classes again, together with example lexical-
izations from English or German. Interestingly, eventities with a specified source
state instead of a specified target state appear to be a rather marginal phenom-
enon, possibly because of the primacy of goals in human conceptualization (cf.
Hommel et al. 2001 and Gallese & Metzinger 2003). Desitives, conclusives, dedu-
cives, and expulsives are way less prominent than their goal-oriented counterparts.
Table 1. Basic eventity classification with example lexicalizations in English and German
Class category Eventity class Example lexicalization
Statal eventities State English know
Activity English run
Act English flash
Change Achievement English arrive
eventities Terminative English forget
Alterative German fllen fill up
Desitive English deflate (intrans.)
Ingressive German loslaufen start running
Egressive English cross
Inchoative German anrollen start rolling
Conclusive German ausrollen roll to a stop
Interactional Causative English kill
eventities Deducive English deflate (trans.)
Resultative German wachrtteln wake up by shaking
Expulsive German aussaugen suck out/dry, pump
This systematic approach to eventity classification, based on the foundational

dynamic modeling elements of the UER (state, transition, and cause-signal), has
generated 15 eventity classes, three of which correspond to Vendlers categories
of state, activity, and achievement. Yet, these are embedded in a much more fine-
grained classification. The fourth Vendler category, accomplishments, appears as
a more heterogeneous superordinate class in the eventity classification proposed
here. Accomplishments are generally described as telic, i.e., bounded processes,
which involve some durativity before reaching their resultant state (the target state
of the transition). This can, however, apply to either the source state of the tran-
sition (which then has to be a durative activity state) or the gradualness of the
transition (which has to be gradual in this case).19 Therefore, accomplishment is
really a superordinate term to alterative, desitive, egressive, inchoative, and con-
clusive eventities, and to corresponding interactional eventities that boost such
transitional configurations as part of their dynamic structure. That is, the notion
19. In principle it could also apply to a durative, i.e., continuous, cause-signal, but this requires
a gradual transition of the undergoer and is thus a special case of the gradual transition case.
accomplishment covers both eventities with one as well as eventities with two
prominent participants. From a conceptual point of view (and a UER modeling
point of view), it is an enormous dierence to be talking about change eventities
as in (5a) or about interactional eventities as in (5b).
(5) a. The pool filled up
b. The housekeeper filled up the pool
Even within change eventities, it can be easily demonstrated that accomplishment

is a cover term for dierent conceptual configurations:
(6) a. The pool filled up in thirty minutes
b. The boy ran home in thirty minutes
In (6a), the pool undergoes a gradual transition to being full, and the duration of
the gradual transition is specified by the time adverbial in thirty minutes. In (6b),
on the other hand, the agent carries out some activity that culminates (the boy is
home as a result),20 and the duration of the activity state is modified by the time
adverbial. The time adverbial hence modifies dierent elements in the dynamic
structure of dierent eventity conceptualizations.
The basic eventity classification outlined so far can of course be refined, for
instance by introducing the variation dimensions previously disregarded. I will
give a few examples in the following, looking at durativity in active states, transi-
tions, and cause-signals in turn:
For change eventities, I abstracted away from the dierence between durative
and non-durative active states, i.e., between activity and act (a distinction used
for the statal eventities). Reinstating this distinction results in two subclasses each
for the classes ingressive, egressive, inchoative, and conclusive. For ingressives,
for instance, they could be termed activity-ingressive (with an activity as resultant
state) and act-ingressive (with a non-durative resultant state). Example lexicaliza-
tions for this would be German loslaufen start running and aufschreien cry out,
respectively. The latter case is, however, marked:
The conceptual punctiformity of acts makes it hard to conceptualize their begin-
ning (or end). Acts are, although in principle valid target states, not prototypical
as target states in ingressives. If they occur, not the general punctiformity of the
act is highlighted, but its boundedness. (Schalley 2004:261)
In accordance with this, activity-ingressives and act-ingressives are quite clearly

dierently conceptualized: While activity-ingressives allow for a continuation with
20. Home does not code a prominent participant in (6b), but the termination condition (change
event in the UER) that triggers the agent to exit the activity state.
machte eine Stunde lang weiter continued for an hour, this is not the case for act-
ingressives. Conceptually, we are dealing with a true continuation of the running
in (7a), but not a true crying-out continuation in (7b).
(7) a. Er lief los und machte eine Stunde lang weiter
He started running and continued for an hour
b. ?Er schrie auf und machte eine Stunde lang weiter
He cried out and continued for an hour
The dierence between non-gradual (= default) and gradual transitions has pre-
viously been taken into account in change eventities, but not in interactional
eventities. Yet, similarly to the example just discussed, there is a clear conceptual
dierence between, for instance, achievement-causatives and alterative-causatives
(causatives with a default transition vs. with a gradual transition). Example lexical-
izations are the English transitives wake up and inflate, respectively. Example(8)
shows that the encoding of the achievement-causative can be modified by instan-
taneously, which does not seem to be possible for the alterative-causative.
(8) a. The boy instantaneously woke up his mother
b. ?The boy instantaneously inflated the mattress
A final example for a possible refinement of the eventity classification is the varia-
tion between a continuous (i.e., durative) and an unmarked cause-signal for inter-
actional eventities. Again, the class of causatives will be used for illustrational
purposes. Figure 9 on p.184 depicts the modeling of a continuous causative (with
the agents activity state and the undergoers gradual transition, cf. footnote 17
on p.180). Example lexicalizations are the English transitives lower and drop. A
modification with slowly appears possible for the continuous causation in (9a), but
not acceptable for the non-durative one in (9b):
(9) a. Slowly she lowered her glasses
b. ?Slowly she dropped her glasses
Of course, there are many more refinements and other eventity classifications that
due to space limitations could not be discussed here. These include, for instance,
classifications of more complex dynamic structures and dynamic substructuring,
or classifications based on the static periphery (e.g., using role notions). The UER
lends itself both to a systematic study of the classification refinements that could
only be touched on in this contribution, as well as to further classificational tasks.
This section demonstrated that an object-oriented approach, utilizing a graphical
tool such as the UER, does not only result in representation results that are close to
conceptualization, but also enables a fruitful debate on issues of verbal semantics,
such as the discussion of aktionsarten.
x y
cause
us
uo
ntin
co
cause
Z
Figure 9. Continuous causative
Bibliography
Barsalou, Lawrence W. (1992): Frames, concepts, and conceptual fields. In: Eva Feder Kittay &
Adrienne J. Lehrer (eds.) Frames, Fields, and Contrasts: New Essays in Semantic and Lexical
Organization, Hillsdale NJ: Lawrence Erlbaum. 2174.
Barsalou, Lawrence W. & Hale, Christopher R. (1993): Components of conceptual represen-
tation: From feature lists to recursive frames. In: Iven Van Mechelen, James Hampton,
Ryszard S. Michalski & Peter Theuns (eds.) Categories and Concepts: Theoretical Views and
Inductive Data Analysis, San Diego CA: Academic Press. 221242.
Benz, Anton (2014): Ergativity and the object-oriented representation of verb meaning. This
Volume, pp. 6587.
Dlling, Johannes (2014): Aspectual coercion and eventuality structure. This Volume,
pp.189225.
Dowty, David R. (1979): Word Meaning and Montague Grammar: The Semantics of Verbs and
Times in Generative Semantics and Montagues PTQ. Dordrecht: Reidel.
Dowty, David R. (1991): Thematic proto-roles and argument selection. Language 67: 547619.
Fellbaum, Christiane (ed.) (1998): WordNet: An Electronic Lexical Database. Cambridge MA:
MIT Press.
Fodor, Jerry Alan (1975): The Language of Thought. New York: Crowell.
Fodor, Jerry Alan (1983): The Modularity of Mind. An Essay on Faculty Psychology. Cambridge
MA: MIT Press.
Gallese, Vittorio & Metzinger, Thomas (2003): Motor ontology: The representational reality of
goals, actions, and selves. Philosophical Psychology 16: 365388.
Gutirrez-Rexach, Javier (ed.) (2003): Semantics. Critical Concepts in Linguistics. 6 Vols. London:
Routledge.
Hommel, Bernhard, Msseler, Jochen, Aschersleben, Gisa & Prinz, Wolfgang (2001): The the-
ory of event coding (TEC): A framework for perception and action planning. Behavioral
and Brain Sciences 24: 849937.
Jackendo, Ray (1983): Semantics and Cognition. Cambridge MA: MIT Press.
Jackendo, Ray (1990): Semantic Structures. Cambridge MA: MIT Press.
Jackendo, Ray (1991): Parts and boundaries. Cognition 41: 945.
Jackendo, Ray (2002): Foundations of Language: Brain, Meaning, Grammar, Evolution. Oxford:
Oxford University Press.
Kobryn, Cris (1999): UML 2001: A standardization odyssey. Communications of the ACM 42:
2937.
Levin, Beth (1993): English Verb Classes and Alternations. A Preliminary Investigation. Chicago
IL: University of Chicago Press.
Montague, Richard (1973): The proper treatment of quantification in ordinary English. In:
Jaakko Hintikka, Julius M.E. Moravcsik & Patrick Suppes (eds.) Approaches to Natural Lan-
guage: Proceedings of the 1970 Stanford Workshop on Grammar and Semantics, Dordrecht:
Reidel. 221242. Reprinted in: Montague (1974: 247270), Portner & Partee (2002: 1734),
Gutirrez-Rexach (2003: Vol. I, 225244).
Montague, Richard (1974): Formal Philosophy: Selected Papers of Richard Montague. Edited and
with an introduction by Richmond H. Thomason. New Haven CT: Yale University Press.
Nickles, Matthias, Pease, Adam, Schalley, Andrea C. & Zaeerer, Dietmar (2007): Ontologies
across disciplines. In Schalley & Zaeerer (2007: 2367).
hl, Peter (2014): Predicate classes: a study in compositional semantic. This Volume,
pp.329362.
OMG (19972010): Unified Modeling Language. UML resource page. Object Management
Group. URL http://www.uml.org. Accessed 30 January 2014.
OMG (2001): OMG Unified Modeling Language Specification. Version 1.4. Object Manage-
ment Group. URL http://www.omg.org/cgi-bin/doc?formal/01-09-67. Accessed 30 January
2014.
OMG (2011): OMG Unified Modeling Language Specification. Version 2.4.1. Object Manage-
ment Group. URL http://www.omg.org/spec/UML/2.3. Accessed 30 January 2014.
Portner, Paul & Partee, Barbara H. (eds.) (2002): Formal Semantics: The Essential Readings.
Oxford: Blackwell.
Pustejovsky, James (1995): The Generative Lexicon. Cambridge MA: MIT Press.
Schalley, Andrea C. (2004): Cognitive Modeling and Verbal Semantics. A Representational Frame-
work Based on UML. Berlin and New York: Mouton de Gruyter.
Schalley, Andrea C. (To appear): Types of part-whole relations in common-sense reason-
ing. In: Hans Burkhardt, Johanna Seibt & Guido Imaguire (eds.) Handbook of Mereology,
Mnchen: Philosophia.
Schalley, Andrea C. & Zaeerer, Dietmar (eds.) (2007): Ontolinguistics. How Ontological Status
Shapes the Linguistic Coding of Concepts. Berlin and New York: Mouton de Gruyter.
Schindler, Heidrun (2001): Das Verb legen. Eine Untersuchung seiner rumlich-konkreten
Bedeutungsvarianten. Tbingen: Niemeyer.
Talmy, Leonard (2007): The representation of spatial structure in spoken and signed language:
A neural model. In Schalley & Zaeerer (2007: 231277).
Van Valin, Robert D. & LaPolla, Randy J. (1997): Syntax: Structure, Meaning and Function.
Cambridge UK: Cambridge University Press.
Vendler, Zeno (1967): Linguistics in Philosophy, Ithaca NY: Cornell University Press, chapter
Verbs and times. 97121.
Winston, Morton E., Chan, Roger & Herrmann, Douglas J. (1987): A taxonomy of part
whole relations. Cognitive Science 11: 417444.
Zaeerer, Dietmar (2002): Polysemy, polyvalence, and linking mismatches. The concept of
rain and its codings in English, German, Italian, and Spanish. DELTA Documentao de
Estudos em Lingstica Torica e Aplicada 18: 2756.
Zybatow, Tatjana (2001): Grammatische Determinatoren von Zeit- und Sachverhaltsverlauf im
Deutschen. Ph.D. thesis, University of Leipzig, Leipzig.

4 Articles On Verb Meaning and Argument Structure

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

4 Articles On Verb Meaning and Argument Structure

Uploaded by

Copyright:

Available Formats

chapter 1

Ergativity and the object-oriented

There is an interesting parallelism between the representation of one- place

(16) a. John cut his arm / his foot

2. The object-oriented paradigm

What exactly the object-oriented paradigm makes up is a question which has no

Figure 1. A Triangle class

This named, abstract collection of attributes and operations defines a specific

necessary to define a separate drawFigure() procedure for each geometrical

Figure 2. Two classes of the same type

Figure 3. Sub-classes and super-classes in inheritance relation

3. See the collection of Briscoe et al. (1993), and Pustejovsky (1995).

Figure 4. The Newspaper class and multiple inheritance

Pustejovsky (1995) introduced dotted types in order to handle such example. If

3. Polysemy and argument roles

Polymorphism refers to the phenomenon that operations show different behavior

This is also confirmed by the following German/English examples:

(20) a. Hans / der Mhdrescher / der Ae schlgt den Mann

(20) c. Hans / der Mhdrescher / der Ae schlgt die Trommel

(24) a. The soldiers cross the river

marking a NP by affixation, by means of adpositions and particle, and by cross-

(26) a. uma banaga-nyu

c. uma yabu-gu bura-n

1st pers. 2nd pers. Demonstratives Proper Common nouns

5. Correlates of the ergative/absolutive distinction

5.1 Bondedness to the verb

(28) a. John painted a picture / the wall

Sentence (29b) will normally be understood as meaning that Rembrandt painted a

In all examples, we find some slight difference in interpretation. In the second,

8. Pragmatic enrichment here refers to pragmatic strengthening of meanings as postulated by

Another property which is common to S and O relation but not to A relation

9. See Keenan (1984: Section A.3) for references.

(dust-sucking, i.e., vacuum cleaning). Keenan uses a survey by Mardirussian (1975)

5.2 Control phenomena

As mentioned before, Keenan considers here expressions such as adjectives and

The following examples also contain a direct object:

(32) a. John saw Bill angry

connected with the object-oriented structure of the lexicon. We discussed a num-

Armstrong, Deborah J. (2006): The quarks of object-oriented development. Communications

Anne Bjerre* and Tavs Bjerre**

Transitive verbs are in general not allowed in there-constructions in Danish,

In Danish,1 as in other languages, a there-construction is used for discourse pur-

The term grammatical metaphor was introduced by Halliday (1994). A gram-

If we want to encode an experience, we usually construct a sentence with an expe-

We mentioned earlier that the there-construction is treated as an unaccussa-

(6) a. Hber ikke der lber sm brn rundt hos naboen,

4. The position of the logical subject

+PAST e satt en fgel p taket

(10) Sven kbte verkligen boken igr

+PAST Sven verkligen kbte boken i gr

(12) a. Det hnde honom ngot konstigt igr

(15) *Det spiste en mann en kake

5. Aktionsart and event structure

Before we can present our constraint on there-insertion, we need to introduce

(19) occurrence duration telicity

The situation structure of a simple situation is represented as a singleton list of

Semantic arguments are introduced as features on relations. We are inspired by

actor-rel und-rel unspec-rel

act-only-rel act-und-rel unspec-act-rel unspec-und-rel fully-unspec-rel

Similarly, in (23) we show the subtypes of the state-rel type. We distinguish

theme-only-rel experiencer-theme-rel loc-rel

In a theme-only-rel a property is ascribed to a theme argument. In an experiencer-

An accomplishment is a complex event and what characterizes it is that a specified