Aspectual Requirements of Temporal Connectives: 
Evidence For A Two-Level Approach to Semantics 
Michael Herweg 
Universitiit Hamburg 
FB lnformatik 
AB Wissens- und Sprachverarbeitung 
Bodenstedtstr. 16 
D-2000 Hamburg 50 
Germany 
e-mail: herweg@rz, inf ormatik, uni-hamburg, dbp. de 
Abstract 
This paper argues for a two-level theory of semantics as opposed to a one-level 
theory, based on the example of the system of temporal and durationM connectives. 
Instead of identifying the semantics of lexical items and larger constituents with 
conceptual structures, the two-level theory assumes that there is an intra-linguistic 
level of abstract semantic representations distinct from the extra-linguistic level of 
conceptual representations. It is argued that the two-level theory can give a more 
appropriate treatment of polysemy as well as a better account of the conditions under 
which connectives may (more or less) be felicitously combined with complement 
expressions which do not respect tile aspectual requirements of the connectives. 
1 Semantic Representations and Conceptual 
Representations 
This paper addresses a specific question in the theory of natural language semantics with 
the example of a specific lexical domain. The question is whether one should opt for 
a one-level or a two-level theory of semantic interpretation. The lexical domain is the 
system of temporal and durational connectives such as after, before, while, when, as soon 
as, since, until and as long as (or rather their German equivalents). 
The one-level theory, which is the predominant one in cognitive and computational 
linguistics, considers semantic representations to be part of the conceptual system. Syn- 
tactic representations are directly mapped onto conceptual structures, with no separate, 
intra-linguistic semantic level mediating between the two kinds of representation (see, 
e.g., \[12\] and many others). As a consequence, there is no distinction in principle between 
genuine linguistic knowledge about meanings and extra-linguistic, conceptually encoded 
knowledge. 
The two-level theory of semantic interpretation (see, e.g., \[3, 8, 13\]) claims that the 
interpretation of a linguistic expression in a particular context of use involves the con- 
struction of a conceptual representation constrained by the semantic representation of the 
expression. Semantic and conceptual representations are distinct levels of mental rep- 
resentation which are governed by the principles of two separate cognitive modules, the 
systems of grammatical and conceptual knowledge, respectively. Semantic representations 
are structured configurations of semantic units which, on the one hand, are determined by 
the grammatical system of the language in question and, on the other hand, are grounded 
152 
in - or motivated by - the conceptual system. Semantic representations are abstract repre- 
sentations of meaning in that they are not identical with specific conceptual interpretations 
which an expression may have in a l~articular context of use. Instead, they constrain the 
range of possible conceptual interpretations which may be contextually assigned to an 
expression. They do so by fixing general, abstract conditions for admissible conceptual 
interpretations. Semantic representations are the unique level on which the principles of 
compositionality of meaning are operative. 
Conceptual representations are structured configurations of conceptual units, which 
are mental representations of certain aspects of the external world. In the semantic inter- 
pretation of a linguistic expression in a given context of use, they serve as contextually 
specified representations of meaning. Being subject to the principles of the conceptual 
system, which mediates between various cognitive systems (visual, auditive, motoric, mo- 
tivational, linguistic, ... ) and in itself consists of several subsystems, conceptual represen- 
tations provide the level of integration of extra-linguistic, conceptually encoded knowledge 
in the course of the semantic interpretation of an expression. Thus, conceptual represen- 
tations are the level on which non-compositional processes of interpretation take place. 
The crucial question for a decision between the two competing views is whether or not 
there are generalizations and regularities concerning semantic structure which cannot be 
traced back to conditions of syntactic or conceptual structure. And indeed, experimental 
research in psycholinguistics suggests that, for example, delays in the production of the 
marked item as opposed to the production of the unmarked item of a pair of gradable 
adjectives (small vs. big, short vs. long) in describing relations between objects differing 
in size are due to features of the lexical items themselves and not to features of the 
conceptualization of the scene \[14\]. In particular, the markedness effect appears only in 
language production tasks but not in non-linguistic tasks, as opposed to a congruency 
effect (the retrieval of a particular concept may be inhibited by a competing activated 
concept), which could be observed both in linguistic and non-linguistic tasks. These 
findings argue for the need to distinguish between the complexity of the abstract semantic 
representations assigned to the lexical items in question and the complexity of the more 
concrete conceptual representation of a situation. Bierwisch (in \[3\]) proposes a general 
format for semantic representations of gradable adjectives within the two-level theory 
which meets these demands. 
Once one accepts the need for distinguishing two levels of representation, i.e. a level 
of semantic representation subject to genuine linguistic knowledge and a level of concep- 
tual representation subject to extra-linguistic knowledge, one may assume a theoretical 
perspective different from that adopted in various versions of the one-level theory with 
regard to a number of well-known problems in lexical semantics. Let me briefly indicate 
this new perspective with two examples: firstly, the interpretation of expressions which are 
semantically deviant from a purely compositional point of view without neccessarily being 
strictly unacceptable, and secondly, the treatment of polysemy. Each of these problems 
will be addressed in more detail below with regard to temporal connectives. 
2 The Two-Level Perspective on Semantics 
First of all, the two-level approach can shed some light on a problem for which, in compet- 
ing approaches, no theoretically satisfactory account has been given so far. The problem is 
the following. In general, we have rather clear criteria by which the compositional seman- 
153 
tics of an expression can be determined. However, actual discourse occasionally disregards 
compositional semantics to a certain extent: There may be combinations of expressions 
which are deviant from a semantic point of view without being strictly unacceptable. In 
a number of cases, these constructions can more or less easily be given a special inter- 
pretation shaped by the context. As a matter of fact, their semantic irregularity may 
often reveal itself only through a systematic study of the whole paradigm. Consider the 
following examples from the realm of the semantics of aspect, i.e. that part of semantics 
that primarily deals with the basic distinction between state expressions and event-type 
expressions. In each of these examples, there is a clash between the demands of the count 
adverbial and the aspect of the sentence the adverbial is applied to. Example (1.a) is 
taken from \[1\]. 
(1) a. ?John slept three times last night. 
b. ??John was healthy three times last night. 
c. *Three times, John did not write a paper on semantics last night. 
In each of these sentences, the count adverbial is applied to a sentence overtly describing 
a state, i.e. a situation which homogeneously holds for a certain period of time. Since 
state sentences, in contrast to event-type sentences such as John fell asleep, John regained 
his health and John wrote a paper on semantics, do not carry an internal criterion of 
individuation for the situations they describe, these cannot be counted straightforwardly. 
If a state holds homogeneously over a certain period of time, it also holds at all parts of 
that period; so on what grounds could we claim that a certain state held exactly three 
times last night? The presence of a count adverbial may, however, force one to implicitly 
supply an external criterion of individuation. From a logical point of view, the sentence in 
question is thereby treated on a par with event-type expressions. The count construction 
is then understood as specifying the numbcr of distinct maximum quantities of the state 
(cf. \[10, 11\]). The examples in (1) show that some sentences allow for such an adjustment 
of their interpretation to the requirements of their context more easily than others. 
One popular strategy for treating examples such as (1) is to note that these phenom- 
ena exist, but, due to a lack of theoretical means to cope with them, to not further dwell 
on the subject. Another strategy is to assign less restrictive meanings to the expressions 
in question, so that the critical constructions turn out to be regular. The latter strategy 
has a number of drawbacks. In the first place, there is the borderline problem: It is hard 
- if not impossible - to give a theoretically sound explanation of why specific cases such 
as (1.c) are really unacceptable. In the second place, reduced acceptability often comes 
in degrees, as is shown by (1.a) and (1.b). Ill the third place, the interpretation of the 
critical combinations always involve some amount of inferential effort, which can easily 
be acknowledged. Neither the second nor the third problem can be dealt with simply 
by allowing more combinations to be regular from a semantic point of view. As I will 
argue below, both call for an explanation in terms of extra-linguistic knowledge of and 
assumptions about features of types of situations. The two-level theory of interpretation 
claims that semantically deviant, though more or less interpretable, constructions always 
require a non-compositional reinterpretation of some expression. The need for such a rein- 
terpretation follows directly from the semantic representations assigned to the respective 
expressions. However, the conditions under which certain reinterpretations can or cannot 
be performed are spelled out on the extra-linguistic, conceptual level. Consider the ex- 
amples in (1). The difference between (1.a) and (1.b) is primarily based on a difference 
regarding what we are willing to assume about the lenght of periods of the respective 
154 
kinds of states. (1.a) does and (1.b) does not match our standard assumptions in this 
respect. When we try to assign an interpretation to (1.b), we have to perform quite a lot 
of additional reasoning about the person's state of health. However, even such effort can 
hardly help us to find a sensible interpretation for (1.c). Our knowledge about states such 
as the one described there gives us hardly any clue about what the situation which the 
sentence purports to describe to us might look like. 
By providing us with a distinction between linguistic and extra-linguistic determinants 
in the interpretation of a linguistic expression in a particular context of use, the two-level 
theory yields a differentiated, modular theory of acceptability judgements with a clear 
distinction between semantically regular and irregular constructions on the one hand and 
conceptually interpretable (acceptable) and uninterpretable (unacceptable) constructions 
on the other hand. The requirements of temporal connectives regarding the aspect of the 
clauses they can combine with, which is a phenomenon in line with the one exemplified 
in (1), form a convincing case in favour of the two-level approach to semantics. This will 
be the subject of §3. 
Furthermore, the two-level theory opens up a new perspective on the treatment of pol- 
ysemy in different linguistic fields. This perspective diverges considerably from what has 
been suggested within the one-level approach, most notably in prototype semantics in the 
style of the Lakoff school. Instead of assigning a setof rather specific (though systemati- 
cally connected) conceptual meanings to a polysemous expression, each of which incorpo- 
rates a considerable amount of contextual information as well as assumptions about object 
types, situation types and so on, one may confine oneself to a unique, sufficiently abstract 
semantic representation, which constrains the range of admissible conceptual interpreta- 
tions. The differentiation of particular conceptual interpretations in various contexts is 
carried out by a number of general conceptual operations, the choice of which depends on 
various kinds of extra-linguistic knowledge and assumptions as well as contextual features. 
This general idea has been applied to a number of polysemous expressions already. See, 
for example, Bierwisch \[2\] on nouns such as school and theatre, which can be used to refer 
to buildings, institutions, cultural achievements and collections of processes, depending 
on the particular context of use. Herweg \[7\] studies polysemous spatial prepositions such 
as in; Herweg \[8\] deals with polysemous temporal connectives. Lung \[3\] gives an extensive 
analysis of the semantics and conceptual interpretations of dimensional adjectives such as 
long, short, wide and narrow. Lung et al. \[13\] describe a computer model of the inter- 
pretation of dimensional adjectives implemented in the LILOG project of IBM Germany. 
The basis of this model is the two-level theory of interpretation with specific formats for 
semantic representations and for the representation of the relevant conceptual knowledge 
about spatial properties of objects. 
In short, the two-level approach to polysemy claims in contrast to its competitors that 
the different readings of a polysemous lexical item in various contexts do not give rise to 
an abundance of semantic representations assigned to that expression. Rather, the vari- 
ous readings are differentiated on the conceptual level, where extra-linguistic knowledge 
about object types, situation types and so on may be assessed in the interpretation of an 
expression. In §4, I will illustrate the treatment of polysemy within the two-level theory 
with the example of temporal and non-temporal readings of the connective be/ore. 
155 
3 Aspectual Requirements of Connectives 
By a number of tests, most temporal connectives can be shown to regularly accept only 
complement clauses of one particular aspect ill its whole range. This defines the aspectual 
requirements of the connectives as determined by their semantics. Nevertheless, one often 
encounters combinations of a connective with a clause which, though not fully regular 
from a semantic point of view, can easily be conceptually understood. However, these 
combinations vary considerably regarding their acceptability. Their conceptual interpre- 
tations always involve some amount of inferential effort in order to produce an aspectual 
reinterpretation of the complement expression. This reinterpretation serves to adjust the 
conceptual interpretation of the clause to the aspectual requirements of the connective. 
Obviously, this is essentially the same phenomenon we observed above (example (1)) with 
state sentences in the context of count adverbials. 
In the underlying semantic theory of tense and aspect, I distinguish between expres- 
sions of two basic aspects: event-type expressions and state expressions; the latter include 
expressions about ongoing processes (cf. \[5\]). A sentence based on an event-type ex- 
pression reports that a situation which is conceptually categorized as an individual event 
of the type in question occurs within a certain period of time. A sentence based on a 
state expression asserts that a situation which is conceptually categorized as a state holds 
homogeneously over a period of time. In both cases, the respective times are specified by 
the tenses and possibly the temporal adverbials of the sentences. As is well known from 
the literature (e.g. \[4\]), the aspect of a sentence can be determined by its combinatorial 
potential with respect to time span adverbials ((with)in two hours) and durational ad- 
verbials (for two hours): Event-type expressions only combine with time span adverbials 
(and count adverbials, for that matter), but not with durational adverbials; for state ex- 
pressions, it is just the other way around. By this test, which makes use of the inherent 
semantic properties of these kinds of adverbials, it can be shown that negation always 
produces a state expression. (Some putative counter-examples are discussed in \[10\]). The 
negated sentence either says that a negative state, i.e. the opposite of the mentioned 
state, holds over a certain period of time, or that there is no event of the respective type 
within that period. In either case, the sentence specifies a homogeneous property of the 
time in question. This will be relevant for the ensuing discussion (cf. (1.c) above). The 
following set of examples illustrates what has been said so far. 
(2) \[event-type expressions\] 
a. Peter drank a bottle of wine in two hours. 
b. *Peter drank a bottle of wine for two hours. 
(3) \[state expressions\] 
a. *Peter waited for a wine dclivery in two years. 
b. Peter waited for a wine delivery for two years. 
(4) \[negated event-type expressions\] 
a. *In two years, Peter didn't drink a bottle of wine. 
b. For two years, Peter didn't drink a bottle of wine. 
(5) \[negated state expressions\] 
a. *In two years, Peter didn't wait for a wine delivery. 
b. For two years, Peter didn't wait for a wine delivery. 
In addition to the two basic aspects, there is a third, derived aspect, namely the perfect 
aspect. The perfect aspect is the result of applying an operator to event-type expressions 
156 
which associates to an event of the type in question a potentially bounded consequent 
state proximal to that event. For further details about this theory of the aspects and its 
formal specification, the reader is referred to \[8, 9, 10\]. 
In the following, I will illustrate ~the aspectual requirements of temporal connectives, 
the criteria used to determine these properties of the lexical semantics of the connectives, 
the operations of reinterpretation and the conceptual conditions for these operations with 
some examples taken from the system of German connectives. German connectives show 
very interesting aspectual requirements, which in many cases are more restrictive than 
those shown by their English counterparts. For example, nachdem ('after') does not 
merely require an event-type expression in its complement clause (see (6.a)), but more 
specifically a clause in the perfect aspect, i.e. an expression representing an event of a 
particular type plus a consequent state proximal to that event (see (6.b)). The relevant 
time of the event or state reported in the main clause (either the definite time of an 
event or some arbitrary time at which the state in question holds) is located within this 
consequent state. Note that the potentially bounded proximal consequent state plays 
a crucial role, i.e. more than a simple relation of temporal ordering is needed in the 
semantics of +nachdem (and, for that matter, in the semantics of English after). (6.b) 
could not describe a situation in which Peter sits down in his chair, gets up, moves to 
another place and only then reaches for the newspaper. 
(6) a. ??Nachdem Peter sich in seinen Sessei setzte, griff er nach der Zeitung. 
\['after Peter sat down in his chair, he reached for the newspaper'\] 
b. Nachdem Peter sich in seinen Sessel gesetzt hatte, griff er nach der Zeitung. 
\['after Peter had sat down in his chair, he reached for the newspaper'\] 
Occasionally, nachdern is combined with a complement clause which, if taken on its own, 
has to be classified semantically as a state expression (see (7.a)). This combination must, 
however, be considered as semantically deviant, since nachdem does not combine freely 
with negated complement clauses (see (7.b), (7.c)). Since negation, irrespective of whether 
it is applied to a state expression (7.b) or an event-type expression (7.c), always produces 
a state expression, there should be no such restriction if the combination of nachdem with 
a state expression were fully regular from the point of view of compositional semantics. 
(7) a. Nachdem Peter in seinem Sessel sal3, griff er nach der Zeitung. 
\['after Peter sat in his chair, he took tile newspaper'\] 
b. *Nachdem Peter nicht in scinem Sessel sa6, griff er nach der Zeitung. 
\['after Peter did not sit ill his chair, he took tile newspaper'\] 
c. *Nachdem Peter sich nicht in seinen Sessel setzte, griff er nach der Zeitung. 
\['after Peter did not sit down in his chair, he took the newspaper'\] 
In assigning a conceptual interpretation to (7.a), the complement clause cannot be un- 
derstood as reporting a state, but calls for a non-compositional reinterpretation which 
supplies a change-of-state event as the first argument of tile relation expressed by nach- 
dem/after. The appropriate reinterpretation of (7.a) is the ingressive reinterpretation, 
which assigns to the state of Peter sitting in his chair the event of the state's ingression. 
Thereby, the conceptual interpretation of the state expression is adjusted to the arrange- 
ment of events and states characteristic for the perfect aspect: The state that holds when 
Peter sits in his chair is considered to be the consequent state of its own ingression, i.e. 
the state which results from a change-of state event of Peter sitting down and eventually 
comes to an end when Peter gets up again. 
157 
With very few exceptions, the change-of-state interpretation is difficult to obtain for 
negated complement clauses, since the fact that something is not the case usually does 
not provide us with any significant temporal demarcation to which we can relate other 
events or states. It is therefore in general neccessary to assert explicitly that a change-of- 
state event has occurred. This can be done by means of mehr \['any more'\]: 
(8) a. Nachdem Peter nicht mehr in seinem Sessel saS, griff er nach der Zeitung. 
\['after Peter did not sit in his chair any more, he reached for 
the newspaper'\] 
b. ??Nachdem Peter sich nicht mehr in seinen Sessel setzte, griff er nach der 
Zeitung. 
\['after Peter did not sit down in his chair any more, he reached for 
the newspaper'\] 
But even then, (8.b) is odd. The reason is that the corresponding positive clause does 
not describe a state, but an event. Thus, some additional reasoning is neccessary in order 
to obtain a state which held at some time in the past and then changed into the state 
that there is no event of Peter sitting down in his chair. If there is an interpretation for 
(8.b) at all, it is a habitual interpretation (which of course is a special kind of a state 
interpretation): Peter took the newspaper after he quit his habit of regularly sitting down 
in a particular chair of his - whatever this is supposed to tell us. 
Can we, however, conclude from examples such as (7.a) that at least all positive state 
expressions are accepted by nachdem, even though they must get a special interpreta- 
tion? This is not so, because not all positive state expressions allow for an ingressive 
reinterpretation. A state can be assigned an event representing its ingression only if there 
exists a previous phase of the opposite state. As an example, see (9), which cannot mean 
something like 'after becoming a young person (= at the time when he was young), Peter 
already played the piano like a virtuoso'. 
(9) ??Nachdem Peter jung war, spielte er schon wie ein Virtuose Klavier. 
\['after Peter was young, he already played the piano like a virtuoso'\] 
Although the state of a given person being young is of course a temporary state, it does 
not have a beginning in the sense of an event marking the ingression of the state (see also 
*he became young, *he is already young). There is no preceding state in which the person 
is not young. Before being young, the person simply did not exist, which means that the 
negative state expression cannot be applied to times prior to the person's lifetime. (By 
the same token, the state of a given person being old does not have an end in the sense 
of an event consisting in the state's egression.) This contrasts with (7.a): Before Peter 
sat down in his chair, he did not sit in his chair, i.e. the opposite of the state of Peter 
sitting in his chair obtained. Therefore, the complement clause in (7.a) does and the 
complement clause in (9) does not allow for all ingressive reinterpretation. This difference 
in acceptability cannot be traced to any difference between the state expressions Peter in 
seinem Sessel sitz- \['Peter sit- in his chair'\] and Peterjung sei. \['Peter be- young'\] regarding 
their semantic properties crucial in the context of temporal and aspectual interpretation; 
in other respects, these expressions show exactly the same relevant properties. 
It is part of our conceptual knowledge about different kinds of states whether a state 
fits the sequences of negative and positive phases that characterize the ingressive event 
type (as well as other kinds of event types such as the egressive event type). A formal the- 
ory of the relevant properties of the conceptual domains of times, event types, events and 
158 
states (including processes), in which the conditions of different kinds of reinterpretations 
of state expressions can directly be expressed as conditions concerning the arrangement 
of phases of the states in question, has been developed in \[8, 9, 10\]). The theory contains 
two basic sorts of predicates and individuals: event-type predicates, which express hetero- 
geneous properties of individual events, and state predicates, which express homogeneous 
properties of periods of time. In addition to standard propositional negation, there is a 
predicate negation restricted to state predicates. This allows one to distinguish between 
a proposition asserting that a positive state predicate does not hold at a given time, and 
the application of a negative state predicate to a time, where the latter is obtained by 
negating the corresponding positive predicate. 
A system of operators establishes various kinds of connections between the two do- 
mains. There is, for example, an operator relating an event to a state consisting of the 
set of times which are temporally included in the time of occurrence of the event. This 
operator is intended to capture the (basic) meaning of the progressive aspect. On the 
other hand, there is an ingressive and an egressive operator. These operators map states 
onto the corresponding ingressive and egressive event types, respectively. Both operators 
can only be applied to a state if phases of the state are preceded or followed by phases of 
the corresponding negative state. Another operator worth mentioning maps states onto 
the so called 'pofective' event type. This event type characterizes events which consist in 
the occurrence of a maximum period at which the state obtains (as in Peter was at the 
beach twice this week; see also (1) above). In contrast to the ingressive and the egressive 
operator, the pofeetive operator does not require that a phase of the state is flanked by 
phases of the negative state. All that is required is that there are earlier and later times at 
which the state does not hold (in the sense of simple propositional negation). With these 
means, the formal system can capture tile difference between a state such as Peter being 
young, which, though being a temporary state, cannot be said to have a beginning, and 
states such as Peter sitting in Iris chair, which actually have a beginning. This is exactly 
what is needed in order to cope with the difference between (7.a) and (9). 
Let us now turn to constructions involving a direction of reinterpretation opposite to 
that discussed so far. Take the following examples. 
(io) a. Solange Peter in Hamburg war, lebte er bei seinen Eltern. 
\['as long as Peter was in Hamburg, he lived with his parents'\] 
b. (?)Solange Peter Maria traf, lebte er bei seinen Eltern. 
\['as long as Peter was meeting (meet-PAST) Maria, he lived with his parents'\] 
e. *Solange Peter Maria heiratete, lebte er bei seinen Eltern. 
\['as long as Peter married (marry-PAST) Maria, he lived with his parents'\] 
Whereas the interpretation of (7.a) involves the transition from a state to a type of event 
systematically associated with that state, the interpretation of (10.b) requires that an 
event-type expression is conceptually reinterpreted as a state expression. The durational 
connective solange ('as long as') is semantically restricted to state expressions in the 
complement clause. Thus, (10.a) is grammatical. In (10.b), however, where solange is 
combined with an expression reporting the occurrence of an event (note that there is no 
Progressive form in German; (...) Peter Maria traf is an event expression), the com- 
plement clause cannot be interpreted straightforwardly but must again be reinterpreted. 
The preferred reinterpretation gives tile clause a habitual or iterative reading. By this, we 
obtain a (habitual or iterative) state, which eventually conforms with tile requirements of 
the connective. The complement clause of (10.c) does not - or at least not that easily - al- 
159 
low for this or any other kind of statal reinterpretation of the event expression. Therefore, 
(10.c) is irrevocably unacceptable. 
One major advantage of the two-level theory of interpretation is that one can precisely 
locate the source of the difference in acceptability between (10.b) and (10.c) without having 
either to weaken the aspectual requirements of the connective or to alter the semantic 
classification of the complement expressions. The semantic part of the lexical entry of 
solange determines that this connective is restricted to state expressions as complements, 
and the clauses Peter Maria traf and Peter Maria heiratete are semantically nothing 
but event-type expressions. Whenever one encounters semantically deviant combinations 
such as (10.b) and (10.c), one tries to find a reinterpretation of the complement clause 
which fulfills the semantic requirements of the connective. This involves checking which 
kind - if at all - of statal reinterpretation may be obtained for the event-type expression 
in question. While doing so, one is led to access extra-linguistic, conceptually encoded 
knowledge about the types of events described. Conceptual knowledge about the type of 
event of some person meeting some other person tells us that these events may - and in 
fact often do - occur more than once, which is a prerequisite for a habitual or iterative 
reinterpretation. Conceptual knowledge about the type of events of person x marrying 
person y tells us that, with the exception of some rather unusual cases, this type of events 
does not occur often enough in x's and y's lifetime to form a habitual state of that couple. 
But note that nothing in the meaning of the verb marry Jcan sensibly be said to prevent 
x and y from getting married on one day and getting a divorce on the next day over and 
over again, just as x and y are free to meet as often as they like - in the former case, x's 
and y's habit simply would not fit our assumptions about the world. And this is just what 
makes us accept (10.b), but leads us to reject (10.c). 
4 Temporal and Non-Temporal Readings of before 
It is a well known fact that the temporal connective before as well as its equivalents 
in other languages has a wide variety of non-temporal readings. Consider the following 
examples. In (ll.a), before has a purely temporal reading. In (ll.b-d), the connective 
has various kinds of non-temporal readings: (ll.b) shows a likelihood reading, (ll.c) a 
preference reading, and (ll.d) a conditional reading. 
(11) a. Before the Edmonton Oilers played the Minnesota North Stars, they beat the 
Los Angeles Kings. 
b. Before the Pittsburgh Penguins win the Stanley Cup, the Minnesota North 
Stars will. 
c. Before Wayne Gretzky signs with the Edmonton Oilers again, he will quit 
playing hockey. 
d. The Pittsburgh Penguins will not get their money before they have 
beaten the Boston Bruins. 
In its temporal use, before says that the relevant time of the situation reported in the 
main clause is located on the time scale prior to the time of the event introduced in the 
complement clause. Since before puts no aspectual restriction on the main clause, the 
relevant time is either the definite time of occurrence of an event or some indefinite time 
at which a state holds. In addition to this purely temporal relation, before determines 
that the time associated with the main clause situation is proximal to the complement 
160 
clause event, i.e. it is placed within what I call the proximal pre-state of the event. Note 
that the notion of proximity is again crucial to the meaning of before, just as it is in 
the case of German nachdem and English after: (ll.a) can hardly be used to convey the 
information that the Oilers played the North Stars in the spring of 1991 and beat the 
L.A. Kings a year or a decade ago (which would not be excluded if the connective were to 
express nothing but a temporal ordering relation), even if the Oilers actually happened 
to beat the L.A. Kings in 1990 and 1980 as well as in 1991. Apart from marginal cases, 
relating the events in (ll.a) by means of before most naturally indicates that both events 
occurred within the same season (or any other period which can serve as the conceptual 
basis for fleshing out the notion of proximity in a given context). 
As for the preference and likelihood readings, a number of relevant features, such as 
the choice of tenses, has already been brought to attention by I-Iein~im~iki \[6\]; so I will 
not unneccessarily repeat them here. There are, however, two things which Hein~im~iki 
did not make particularly clear: first, under what circumstances these readings arise, and 
second, whether or not she thinks that the temporal and non-temporal readings follow 
from different semantic representations of the connective. 
Note in the first place that tile non-temporal readings are only options we may choose 
in the interpretation of (ll.b) and (ll.c). If we consider (ll.b) to be a claim about two 
successive seasons of the National Hockey League, say tile 1991 and the 1992 season, (ll.b) 
may perfectly well be understood to tell us that in 1991 the Minnesota North Stars and in 
1992 the Pittsburgh Penguins will win the Stanley Cup. (Since in this case NHL seasons 
or years in their entirety form the basic units of time, we must count the time of the North 
Stars' triumph as proximal to the Penguins' 1992 victory.) If, however, (ll.b) is intended 
to be a claim about the very same season, as, for example, when we are discussing the 
prospects of our favourite teams for the season to come, a temporal interpretation would 
not make any sense. After all, we know that there can be only one champion per season. 
In the same way, (ll.c) can simply mean that Wayne Gretzky will first quit his career 
and then sign a contract with his former team. But this temporal interpretation of the 
sentence would contradict our assumptions about what the president of a team normally 
expects from the players he employs. 
This shows that the preference and likelihood readings of before are alternatives to 
the temporal interpretation of the connective which we choose in case the latter hardly 
makes any sense. The conditions under which these non-temporal readings car/ appear 
must be spelled out in terms of our extra-linguistic knowledge of and assumptions about 
features of types of situations, i.e. they are genuinely conceptual in nature. Tile relevant 
conditions are the following: A temporal interpretation contradicts what we think is 
plausible, and the two statements linked by the connective allow us to establish a contrast 
between types of situations. In (ll.b), the neccessary contrast is given by two possible 
outcomes of the hockey season. In (11.c), the contrast consists in a player's choice either 
to end or to continue his career, the latter being a prerequisite for being offere'd a new 
contract. Where there is no contrast, a preference or likelihood interpretation can hardly 
be achieved; compare (ll.b) and (ll.c) to (12). The two readings differ in that tile 
preference reading requires that the situations are under a subject's control; see sentence 
(13), which has a likelihood reading only. 
(12) a. Before the Pittsburgh Penguins win tile Stanley Cup, Mario Lemieux will 
score for them more than once. 
b. Before Wayne Gretzky signs with the Edmonton Oilers again, he will 
decide to continue his career. 
161 
(13) Before Wayne Gretzky signs with tile Edmonton Oilers again, the Pacific 
Ocean will dry out. 
I consider the basic meaning of before to be temporal. The preference and likelihood 
readings require no special semantic representations, but arise through a conceptual rein- 
terpretation of the ordering relation conditioned by the circumstances outlined above. Ill 
the basic temporal use, the ordering relation is interpreted as the temporal precedence 
relation between times associated with situations. In preference and likelihood readings, 
this relation is replaced by the relation of subjective prcference or likelihood, respectively, 
which ranks one type of situation superior to another type of situation. (What comes 
first on the scale is more likely or preferred than what comes later, just as what is present 
or in the past has a higher degree of certainty than what is yet to come.) In both the 
temporal and the non-temporal cases, there is a projection of parameters associated with 
situations - their times or their types - onto a scale, on which situations can be located 
with respect to that parameter. The notion of proximity in the semantics of before also 
carries over to non-temporal uses. In the cases discussed above it means that the type 
of situation described in the main clause is among the closest alternatives to the type of 
situation described in the complement clause with respect to the nature of the scale. 
Nothing has been said so far about tile conditional reading of before in (11.d), which 
is repeated here as (14): 
(14) The Pittsburgh Penguins will not get their money before they have beaten 
the Boston Bruins. 
A conditional reading can only arise if the main clause contains a negation and the sub- 
ordinate clause is in the perfect aspect. The reasoning that leads to a conditional reading 
can be elucidated by considering what the standard temporal meaning of before yields in 
connection with these types of clauses. In its temporal interpretation, (14) says that prior 
to the consequent state of the Penguins' victory there is no event of the type that the 
team is getting money. This means that tile earliest possible change from a period without 
such an event to a period including such an event is when the victory has occurred. If 
there will be such an event at all, it must occur after the victory. This reasoning leads 
us to regard a victory of the Pittsburgh Penguins over the Boston Bruins as a condition 
for the occurrence of any event of the Penguins getting their money. Here again, the two 
types of events are projected onto a non--temporal scale, which in this case is a scale on 
which events are preceded by their conditions. The relation of proximity in the meaning 
of before now means that the realization of the event type introduced in the main clause 
is among the immediate consequences of the type of event described in the subordinate 
clause. 
5 Conclusion 
Let me sum up my position. There is evidence from experimental psycholinguistics that 
we have to take into account an intra-linguistic level of semantic representations distinct 
from the extra-linguistic level of conceptual representations. In theoretical linguistics, the 
two-level theory can, for example, give an explanation of the differences in acceptability 
among expressions which are not fully correct from the point of view of compositional 
semantics. The theory claims that semantically deviant, though more or less acceptable, 
constructions require a non-compositional reinterpretation of some expression. Thereby, 
162 
the theory reflects the fact that their interpretation always involves a certain amount of 
reasoning. The need for an reinterpretation can be predicted from the semantic repre- 
sentations assigned to the respective i expressions. However, the conditions under which 
certain reinterpretations can or cannot be carried out are spelled out on the conceptual 
level. Due to extra-linguistic knowledge about types of situations (or, in cases other than 
those discussed in the present paper, tyi:)es of objects or other kinds of entities), some 
expressions allow for an appropriate reinterpretation and others do not. In addition, the 
two-level theory suggests a different perspective on the treatment of polysemous expres- 
sions. Rather than assigning a load of highly specialized semantic representations to the 
expressions in question, each of which incorporates a considerable amount of contextual 
information and whose number might easily get out of control in various linguistic fields, 
the theory highlights that there is a division of labour between the semantic and the con- 
ceptual level. Each level of representation makes its own contribution to what eventually 
will be the interpretation of a linguistic expression in a particular context. 
Acknowledgements: 
I wish to thank Elke Jiiger for her assistance in the preparation of the manuscript and 
Geoff Simmons for correcting nay English. 

References 
\[1\] Bach, Emmon (1986): "The algebra of events". Linguistics and Philosophy 9, 5-16. 
\[2\] Bierwisch, Manfred (1983): "Semantische und konzeptuelle Repr~entationen 
lexikalischer Einheiten". in: R. Ruzicka & W. Motsch (eds.): Untersuchungen zur 
Semantik. Berlin: Akademie-Verlag, 61-99 (studia grammatica 22). 
\[3\] Bierwisch, Manfred & Lang, Ewald (eds., 1989): Dimensional Adjectives. Grammat- 
ical Structure and Conceptual Interpretation. Berlin: Springer. 
\[4\] Dowty, David (1979): Word Meaning and montague Grammar. The Semantics of 
Verbs and Times in Generative Semantics and Montague's PTQ. Dordrecht: Reidel. 
\[5\] Galton, Antony (1984): The Logic of Aspect. An Axiomatic Approach. Oxford: 
Clarendon Press. 
Hein~m~iki, Orvokki (1978): Semantics of English Temporal Connectives. Blooming- 
ton/Indiana: Indiana University Linguistics Club. 
Herweg, Michael (1989): "Ans~tze zu einer semantischen Beschreibung topologischer 
Pr~.positionen". in: C. Babel et al. (eds.): Raumkonzepte in Verstehensprozessen. 
Interdisziplindre Beitriige zu Sprache und Raum. Tiibingen: Niemeyer, pp. 99-127. 
Herweg, Michael (1990): Zeitaspekte. Die Bedeutung yon Tempus, Aspekt und tem- 
poralen Konjunktionen. Wiesbaden: Deutscher Universit~itsverlag. 
Herweg, Michael (to appear/a): "Temporale Konjunktionen und Aspekt. Der sprach- 
lithe Ausdruck von Zeitrelationen zwischen Situationen". to appear in Kognitionswis- 
senschafl. 
\[10\] Herweg, Michael (to appear/b): "Perfective and imperfective aspect and the theory 
of events and states", to appear in Linguistics. 
\[11\] Herweg, Michael (to appear/c): "A critical examination of two classical approaches 
to "aspect". to appear in Journal of Semantics. 
\[12\] Jackendoff, Ray (1983): Semantics and Cognition. Cambridge, Mass.: The MIT 
Press. 
\[13\] Lang, Ewald; Carstensen, Kai-Uwe; Simmons, Geoff (1991): Modelling Spatial 
Knowledge On A Linguistic Basis. Theory-Prototype-Integration. New York, Berlin, 
Heidelberg: Springer (Lecture Notes in Artificial Intelligence 481). 
\[14\] Schriefers, Heribert (1990): "Lexical and conceptual factors in the naming of rela- 
tions". Cognitive Psychology 22, 111-142. 
