LINKING PROPOSrFIONS 
Do S. Brae & R.A. Smit 
Rotterdam School of Management, Erasmus University, P.O.B. 1738, NL-3000 DR Rotterdam, The Netherlands. 
This research was partially funded by the Netherlands Organization for the Advancement of Pure Research. 
KEYWORDS: Semantics, Subordinating conjm\]ctions, 
Language processing, Language universals. 
ABSTRACT 
The function words of a language provide explicit 
information about how propositions are to be related. We 
have examined a subset of these function words, namely 
the subordinating conjunctions which link propositions 
within a sentence, using sentences taken from corpora 
stored on magnetic tape. On the hasis of this analysis, a 
computer program for Dutch language generation and 
comprehensien has been extended to deal with the 
subordinating,; conjunctions. We present an overview of 
the underlying dimensions that were used in describing 
the semantics and pragmatics of the Dutch subordinating 
conjunctions. We propose a Universal set of Linking 
Dimensions, sufficient to speeify the subordinating 
conjunction.,; in any language. This ULD is a first 
proposal for the representation required for a computer 
program to understand or translate the subordinating 
conjunctions of any natural language. 
I. Introduction 
Languages provide speakers with the means to express 
propositions and to link these propositions. Propositions 
are expressed in language by means of clauses. These 
can form sentences in themselves, or they may be linked 
together within one sentence, either at tim same level 
or embedded one within the other. It is this last 
category of linking, by means of subordinate 
conjunctions, in which we are interested. Tile questions 
that we ask are: 
Do languages provide a systematic way of expressing 
the possible subordinating links between propositions? 
If they do, which dimensions cat\] be used to capture 
the means that are provided in all languages for 
these links? 
What is this Universal set of Linking Dimensions 
{ULD)? 
We have atl:empted to provide a systematic description 
of the subordinating conjunctions (SCs) in Dutch. It is 
this description that we will use to propose a UI.D. At 
the top level we have divided the SCs into just four 
types: inferential, temporal, causal and manner SCs. This 
is fewer than the dozen or so types found in the 
traditional and modern grammars, which give a 'flat' 
taxonomic tree, making all tile obvious distinctions at 
the first level. At each branch in the taxonomic tree, 
we have tried to mal(e as few divisions as possible, in 
order to make tile motivation for each split clear. 
These four categories were chosen because they enable 
quite different kinds of relationship to be set up 
between a main and a subordinate proposition. They each 
indicate a different function that the subordinate 
event/state has in relation to the main predication. The 
most abstract relationship is that of inference, in which 
the speaker uses the sub proposition to give the grounds 
for his belief concerning the truth status of the main 
proposition. The other three types indicate more than 
simply a relationship between beliefs {propositions); they 
convey the speakerVs beliefs about relationships that 
exist qn reality ~. Two of these are quite specific: time 
and cause. Tile fourth category of SC, manner, also 
serves to indicate that there is a relationship 'in reality' 
between the sub and main events/states; however, this 
relationship is one that depends highly on the schema 
that the observer uses to codify what he sees. 
We will now describe the sub trees for each of these 
four types of linkage. The meanings of the Dutch SCs, 
taken from Van Wijk and Kempen {1980), have been 
determined using: 
sentences taken from a Dutch corpus (Uit den 
Boogaard, 1975; shown as e.g. 1.2345); 
the authoritative Algemene Nederlandse Spraakkunst - 
ANS (Geerts et al., 1984}. 
2. Inferential linkage 
An inference relationship exists between two propositions 
when the truth value of one can be deduced from the 
other. The grounds for the deduction are left open. They 
may be based on some causal model of reality: 
If metal is heated it expands. 
But they can also be purely definitional: 
If two angles of a triangle are equal, their opposite 
sides are equal. 
The types of inferential linkage depend in the first 
instance on tile truth statas of the main proposition. 
This may be either true, probably true but with the 
possibility of an escape, hypothetically true or 
counterfactual. A false main proposition is not indicated 
by an SC but by the use of the past tense and/or a 
modal auxiliary verb. 
True. If the main proposition is true, then the inference 
relation from the sub proposition may be used, denied or 
deemed h'relewmt. 
The tm-e of the inference is indicated in Dutch by the 
SC aangezien (since): 
De rector had besloten de school te sluiten aangezien 
her verbod was overtreden. (ANS, p.655) 
(The principal decided to close the school sinc.e the 
ban had been contravened.) 
It has been, and still is, customary to classify aangezien 
as a causal SC (ANS, p.655). This is incorrect. Causal 
SCs can be topicalized, inferential SCs cannot. The 
reason for this distinction is that causal SCs say 
something about reality, whereas inferential SCs are 
used to make an inference. This making cannot be 
topicalized. Aangezlen, however, like non-temporal since, 
cannot be topicalized, so it is not causal: 
*llet is aangezien het verbod was overtreden, dat de 
rector besloot de school te sluiten. 
An inference relationship \]nay be denied using the SC 
hoewel {although). Then the normal inference is from the 
sub proposition to the falsity of tile main proposition: 
Hoewel her verbod was overtreden, besloot de rector 
de school niet te sluiten. 
(Although the ban had been contravened, the principal 
decided not to close the school.) 
The irrelevance of any inference relationship is indicated 
by ongeacht (whether ... or not}: 
Ongeacht of het verbod was overtreden, zou de 
rector hebben besloten de school te sluiten. 
(Whether the ban had been contravened or not, the 
principal would have decided to close the school.) 
177 
Escape linkages. If the speaker wishes to indicate that 
the main proposition is not certainly, hut only probably, 
true, then an SC can be used to indicate the 
circumstances under which the main proposition may 
indeed by false. Such SCs provide an ESCAPE from the 
speech act being made in the main clause. This speech 
act may be an assertion, in which case the escape is 
from the truth of the main proposition; but any other 
type of speech act, such as a promise, may also be 
escaped from. The escape may be either when the sub 
proposition is true, with tenzlj (unless), or when it is 
false, using mtts (provided that): 
De oecumene zal niet slagen tenzij het gesprek met 
Israel ... wordt gevoerd. (2.3040) 
(Ecumenism will not succeed unless there is a 
dialogue with Israel ...) 
Jongeren kennen een normale behoefte aan gezag en 
normen, mits zij er de zin van weten te 
ontdekken. (5.3341) 
(Youngsters have a normal need for authority and 
norms, provided they can discover their sense.) 
Hypothetical linkage. If the truths of the main and sub 
propositions are unknown, an inference relationship from 
the sub to the main proposition can be shown by using 
either als or lndlen (if). Of these two SCs, als is the 
more common, but it is ambiguous between several uses 
(see below); indlen is more formal and emphatic: 
Als/Indien het verbod is overtreden, zal de rector de 
school slutten. 
(If the ban has been contravened, the principal will 
close the school.) 
Figure 1. INFERENTIAL SCs 
Truth of the Main Proposition 
I 1 probably I true unkown or false true 
l Escape fr°m Sub P', fits Use °f the inference I or opposite, ,,, to the Main Proposition 
Sub P. opposite HYPOTHETICAL 
ESCAPE - ESCAPE + DENIED IRRELEVANT USED 
m/ts tenz/j a/s/)nd/en hoewe/ oncjeacht aangez/bn 
provided unless if though whether or not since 
Counterfactuals. Just as with the true inferential 
linkages, the counterfactual inference may be simply 
used or denied. The inference, in this false world, may 
be used to infer a main proposition which is true in this 
false world but false in the actual world: 
If Eve hadn't given Adam the apple, he wouldn't 
have eaten it. 
The inference may also be denied to give a main 
proposition which is true in both worlds: 
Even if Eve hadntt given Adam the apple, he still 
would have eaten it. 
This is a semifactual. The same SC, als (if), is used 
both for hypotheticals and counter/semi-factuals. This is 
not the case in all languages, e.g. Polish, Japanese. 
In order to indicate that the sub clause does not 
correspond with the truth status of the sub preposition, 
the tense of sub clause verb is placed one step further 
into the past than would normally be the case. That is 
to say: if the tense would normally be past, past-perfect 
or present-perfect then it is set to past-perfect; 
otherwise it is set to past, With the true counterfactuals 
(as opposed to the semifactuals) the fact that the main 
178 
clause also does not correspond with the truth status of 
the main proposition is indicated by using the past- 
future, i.e. using the past form of the verb zullen as 
the auxiliary finite verb form: 
Als Ik geld had, zou lk op rels gaan. 
(If I money had, would I travelling go.) 
Moreover the simple (or perfect) past can also be used 
to indicate connterfactuality: 
Als lk geld had, glng tk op reis. 
(If I had money, I went travelling.) 
Als lk geld gehad had, had ik meer gereisd. 
(If I had had money, I had more travelled.) 
The ANS (p.468) provides no rule for choosing between 
using the simple (or perfect) past and using the modal 
auxiliary zullen. There are, however, systematic 
differences (Nteuwlnt, 1984). More problematic is that 
the simple past tense may indicate one of two things: 
the Time Of Reference (TOR) is in the past or we are 
dealing with a counterfactual. This ambiguity can be 
resolved by the context: if the TOR is already in the 
present, then the past tense indicates counterfactuality. 
Summarizing, we find four types of inferential linkage, 
depending on the truth value of the main proposition: 
true: the inference from a true sub proposition may 
be used, denied or deemed irrelevant; 
probably true: this truth can be escaped, either from 
the sub proposition or from its negation; 
hypothetical inference from a sub proposition whose 
truth is uncertain; 
counterfactuals: an inference from a sub proposition 
that is known to be false, to either a false or a true 
main proposition. 
3. Temporal linkage 
The temporal SCs specify the time of the main 
predication in relation to the time of the event 
indicated in the sub clause. The system we have used to 
represent the different possible temporal linkages is 
based on two dimensions: the relative temporal order of 
the main and sub events, and the place of the main 
event within this restricted time range. 
Relative temporal order. A sub clause introduced by a 
temporal SC is used to restrict the time during which 
the main proposition is true: the time of the main event 
may be at a time that is either earlier than, or later 
than, the time of the sub event, or it may be 
coincidental with the time of the the sub event. 
The position within the range. The place of the main 
event within this restricted time range is the second 
dimension. It may be either: 
at some time within the proximity of the sub eventl 
- either immediately adjoining the sub event; 
- or in the vicinity of the sub event time. 
at some time within an interval. The way the bounds 
of this time interval are Indicated depends on the 
relationship between the TOR and the interval itself. 
If the TOR is to fall within the interval, then the 
duration of the sub event determines the interval. 
Otherwise the TOR marks one bound of the interval, 
the other bound being set by the time of the sub 
event. Again this category has two alternatives: 
either the main event occurs at some moment 
within the interval; 
or it occurs for the whole of the interval, in 
which case the event must be able to have a 
duration or be repeatable. 
These four different places within the time range, 
together with the three ways of specifying the range, 
give twelve different possibilities for indicating the time 
relationship between the main and sub events. 
It is not to be expected that any language will have SCs 
to distinguish between all these twelve possible temporal 
linkages. In Dutch three of the relationships cannot be 
expressed using an SC. Moreover, the sub-distinctions 
made in the second dimension are not always made. 
Figure 2. TEIVIPORAL S~RDINATING CONJUNCTIONS 
Relation Main event in Main event in interval 
of main to proximity of sub: of TOR & sub event: 
sub event. Next to Nearby Sometime Durative 
Earlier than: voor totdat 
(before) (until) 
toen/als 
Coincidental: wanneer/nu terwljl zolang 
(when) (while) (as long as) 
Later than: zodra nadat Ands sinds 
(as soon as) (after) (since) (since} 
The difference between the four SCs als, toen, wanneer 
and nu requires an explanation: 
teen is used to set the TOR to some point before 
the time of utterance, and so only occurs when the 
TOR is in the past (ef. the use of when as discussed 
by Kamp, 1981). The TOR Is set to the time of the 
sub event; 
nu is used when the TOR has already been fixed, and 
an event, the sub event, which happens to be 
coincidental with the TOR, is a cause or reason for 
the main event; 
als and wanneer are used: 
for a temporal coincidence after the TOR, 
without bringing the TOR forward; 
to indicate a repeated or repeatable temporal 
coincidence. 
Wanneer (when), which hardly ever occurs in spoken 
Dutch except as an interrogative, is temporal. Als (if) is 
not confined to a temporal role, being used also for 
manner and inferential linkages. So its use puts the 
burden of interpretation onto the addressee. 
We have seen that the temporal linkage is specified on 
two principle dimensions: 
the time range to which the main event is restricted 
may be before, after or coincidental with the time 
of the sub event; 
the time; of the main event may be related either 
directly to the time of the sub event, or it may fall 
within an interval. In the former case the proximity 
may be indicated. In the latter case the main event 
may be considered to occur once in the interval or 
during the whole of the interval. The interval itself 
is bound between the sub ewmt and the TOR, unless 
the TOR falls within the time period of the sub 
event, In which case the interval is equivalent to the 
duration of the sub event itself. 
4. Causal linkage 
A temporal linkage is not the only relationship that can 
be indicated in reality between the sub and main 
propositions. A causal linkage can also be made from 
the sub event or state, to the main event or state. 
There are two main types of causal linkage: teleological 
and ateleological. 
An ateleological cause Is a purely physical link, i.e. 
mechanistic, in the sense that no will is posited. The 
mechanism operates inevitably, e.g. gravitation that 
controls the motion of the planets. E.g. 
De peehdag voor de NS ward glsteren nog 
gecompleteerd, doordat op bet centraal station in 
Utrecht twee machinisten van aanslultende 
posttreinen biJ bet wlsselen van treln allebei 
precies in de verkeerde stapten. (1.5847) 
(The day of troubles on the railways yesterday was 
even more complete, because at the central 
station in Utrecht two drivers of connecting post 
trains, when changing trains, each stepped into 
the incorrect train.) 
An ateleologleal link may also be proportional: the more 
there is of some sub property the more there will be of 
the main property, as in: 
De dagen lengen naarmate de nachten korten. 
(The days lengthen as the nights shorten.) 
A teleological link, on the other hand requires that a 
will be present. They are volitional. The being that 
exerts tlds will has two components of interest: a 
perception of his own state and an awareness of his own 
goal. There are, correspondingly, two types of 
teleological cause: reason and motive. Reason is 
primarily state controlled, e.g. 
ttet kwam hem voor dat hij, jutst omdat hlj zo 
gewoon mogelijk wilde doen, zich zo ongewoon 
voelde. (4.1610) 
(He realized that he, just because he wanted to 
behave as normally as possible, felt himself to be 
so abnormal.) 
Motive is primarily goal controlled, e.g. 
De regering heeft hat bedrag van de steun verhoogd 
opdat de armsten geen honger zullen lijden. 
(Donaldson, 1984, p. 195) 
(The government has increased the amount of the 
support so that the poorest people will not suffer 
from hunger.) 
Note that the description of the goal state is not true, 
as it has yet to be achieved. The standard way of 
indicating this is to use the conditional auxiliary (zullen 
will) in the subordinate clause. As an illustration of 
the contrast: 'feeling ill' is a reason for going to bed, 
'to get better' is a motiw. ~ for going to bed. 
Figure 3. CAUSAL SC s 
no l_ 
I Pr°p°rti°nal i 
i= • 
no yes 
PHYSICAL PROPORTIONAL 
doo~at naa/mat~ 
because according as 
Teleological? ! 
? yes 
\[Sub proposition is goal i 
no yes 
? REASON MOTIVE 
(zndat omdat opdat 
because because so that 
Omdat may be used either for a teleological reason or 
for an ateleological mechanistic link. As a result doordat 
is used to emphasize an ateleological cause. In, e.g. 
Because there were several new dancers in the 
troupe, the form of the ballet (was) changed. 
because would be translated by doordat to indicate that 
the new dancers themselves caused the change; whereas 
omdat would indicate that the choreographer made the 
change to accommodate tile ballet to the new dancers. 
179 
We have found that there is a distinction between 
mechanistic and volitional causal linkages. Mechanistic 
links may also be proportional. Volitional, or teleological, 
links may be based either on reason or on motive. 
5. Manner linkage 
The fourth and last type of linkage is the least specific. 
The sub proposition indicates something about the 
manner of the main proposition. A manner SC is used to 
add a descriptor, which includes the sub event or state, 
to the main proposition. The principle distinction to be 
made is whether this sub event/state actually exists or 
is (perhaps) imaginary, i.e. whether the sub proposition is 
true or of unknown truth value. 
True sub proposition. If the sub proposition is true, then 
either the manner of the main proposition is specified as 
being restricted to the same as the manner in the sub: 
Hij speelt viool zoals hi\] piano speelt. 
(He plays violin in the same way as he plays piano.) 
or the sub proposition is an addition to the main one: 
Hij speelt viool evenals zijn vader dat heeft gedaan. 
(He plays the violin, just as his father did.) 
Non true sub proposition. If the sub proposition is false 
or of unknown truth value then alsof is used: 
Hij speelt viool alsof hii piano speelt. 
(He plays the violin as if he plays the piano.) 
Whether the sub proposition is false or merely of 
unknown truth value must be determined using the 
context. The speaker can indicate a false value by using 
the past tense, just as with conditional counterfactuals: 
Hij speelt viool alsof hi\] piano speelde. 
(He plays the violin as though he played the piano.) 
Figure 4. MANNER SUBORDINATING CONJUNCTIONS 
Sub proposition: True Non-true 
Restriction zoals (same way as) 
Addition evenals (just as) alsof (as though) 
6. Conclusion. 
In this analysis of the semantics of Dutch SCs, we have 
concentrated on the main aspects in order to distinguish 
the wood from the trees. We have ignored SCs that are 
archaic, dialectal or formal. We have also ignored 
secondary uses of certain SCs, e.g. that the temporal SC 
terwljl (while) can be used to highlight a contrast. We 
believe that this is not prejudicial to our case. 
Secondary uses are just that; they are not different 
meanings, as we have argued elsewhere for the non- 
standard uses of if (Br(e & Smit, 1985). 
The definition of the meanings of the Dutch SCs is 
specific enough to be implemented in a sentence 
generation program. We have demonstrated this using 
Kempen's Incremental Procedural Grammar (Hoenkamp, 
1983). Our extension (see Br(e, Smit & Schotel, 1984) 
allows a user to enter two or more propositions and the 
type of relationship between them (inferential, temporal, 
causal, manner). Then it asks questions corresponding to 
the semantic tree for the corresponding type of SC, in 
order to select the appropriate kind of SC (e.g. 
hypothetical, denial, etc.). The program then uses the 
selected kind to find the correct SC in Dutch. So the 
selection of the kind of SC is independent of the 
language; the actual SC is selected from a table of SCs 
built up from the semantic definition of each SC within 
one language. Our program can also take a sentence as 
180 
input and break it down into main and sub propositions, 
replacing the linking SC by its language independent 
semantic definition. The important point is that the 
definition of the kinds of SC is language independent. 
We set out with the aim of establishing a Universal set 
of Linking Dimensions (ULD) that speakers use in linking 
propositions. It is the semantic trees that provide us 
with the ULDs. There are two levels at which we could 
hypothesize universality. The first, the strong hypothesis, 
is that the trees are the same in all languages. Then 
the kinds of SCs should be the same in all languages. 
For Dutch and English this is more or less the case. The 
differences in the SCs in the two languages come about 
in the different ambiguities that arise because the same 
SC is used for more than one kind of relationship (e0g. 
als ==> if/when, since ==> slnds/aangezlen). As these 
ambiguous uses are not the same in the two languages, 
difficulties arise for translation programs, llowever, this 
does not negate the strong hypothesis. 
If it does turn out that there are languages with SCs 
that cannot be defined using these trees, then a second, 
weaker, thesis may hold, namely that the building blocks 
from which the trees are made, are the same in all 
languages. That Polish and Japanese have SCs 
specifically for counterfactual inferences, leads us to 
suspect that it is this second thesis that will be found 
to hold. It will provide tlm basis for constructing a 
means for representing the functions performed by SCs 
in all languages in linking propositions. 
In either case, why is it that people relate propositions 
using the ULD? Is it because their brains are so 
constructed, or because their minds reflect the nature of 
the environment in which they find themselves? Is the 
ULD a natural or artificial phenomenon (Simon, 1981}? 

REFERENCES 

Brae, D.S. & R.A. Smit 1985, Non-standard uses of if. In 
The proceedings of the 2nd Conference of the 
European Chapter of the Association for 
Computational Linguistics. Geneva: 218-225. 

Brae, D.S., R.A. Smit & H.P. Schotel 1984. Generation 
and comprehension of Dutch subordinating 
conjunctions by computer. In O'Shea, T., ed., 
Proceedings of the 6th European Conference on 
Artificial IntelIlgence. Elsevler, Amsterdam: 205-208. 

Dik, S.C. 1981. Functional grammar. Foris, Dordrecht. 

Donaldson, B.C. 1984. Dutch reference grammar. Nijhoff, 
Leiden. 

Geerts, G., W. Haeseeryn, J. de Rooij, & M.C. van den 
Toorn, eds, 1984. Algemene nederlandse spraakkunst. 
Wolters-Noordhoff, Groningen, Holland. 

Hoenkamp, E.C.M. 1983. Een computermodel van de 
spreker: psychologlsche en llngulstlsche aspecten. 
Ph.D. thesis. Katholieke Universitelt van Nijmegen. 

Kamp, H. 1981. A Theory of truth and semantic 
representation. In Groenendijk, J.A.G., T.M.V. 

Janssen, & M.B.J. Stokhof, eds, Formal methods in 
the study of language, vol. I. Mathematisch Centrum, 
Amsterdam: 277-322. 

Nieuwint, P.J.G.M. 1984 Werkwoordstijden in nederlandse 
"counterfactuals". De Nleuwe Taalgids, 77(6): 542-555. 

Simon, H.A. 1981 Sciences of the artificial. 2nd ed. MIT 
press, Cambridge. 

Ult den Boogart, P.C. 1975. Woordfrequentles van 
geschreven en gesproken Nederlands. Oosthoek, 
Scheltema en Holkema, Utrecht. 

Wijk, C. van & G. Kempen 1980. Funktlewoorden: een 
inventarisatie voor het Nederlands. Review of Applied 
Linguistics, 47, 53-68. 
