Temporal Discourse Markers and the Flow of Events 
Frank Schilder 
Computer Science Department 
University of Hamburg 
Vogt-K~511n-Str. 30 
D-22527 Hamburg 
Germany 
schilder@informatik, uni-hamburg, de. 
1 Introduction 
• Temporal discourse markers such as after, before or while 
are commonly described as triggers for discourse rela- 
tions expressing a temporal relation (Mann and Thomp- 
son, 1987; Knott, 1996). However, only little research 
has been done regarding the interaction of such discourse 
markers with the context within a multi-sentence dis- 
course. 
Lascarides & Oberlander (1993), for instance, note 
that sentences containing temporal connectives express- 
ing the same temporal structure may not describe a co- 
herent discourse in certain contexts. Along these lines 
a similar theory was developed by de Swart (1994) 
wrt. temporal adverbials and also applied to temporal 
clauses. She points out that the topic/focus structure has 
to be taken into account, especially when apreposed syn- 
tactic structure as in (1) is read: 
(1) John had always been rather shy. But after he met 
H61%ne, his behavior changed quickly. He was very 
self-assured now. 
According to her proposal, the after-clause only adds fur- 
ther temporal information to the event structure of the 
discourse, since it is in the topic position of the sentence 
providing 'old' information. A sentence with a so-called 
postponed after-clause as in (2), on the other hand, re- 
quires a connection to the preceding context via a rhetor- 
ical relation (i.e. background): 
(2) Ty and John were good friends, and they often went 
to the movies together. But John's behavior changed 
quickly after he met H%l~ne. Ty said she bewitched 
him. 
Previous work seems do indicate that the syntactic 
variations of after-clauses play an important rule for the 
organisation of discourse. The purpose of this paper is to 
shed some light on the usage of the German translation 
of after (i.e. nachdem). First, an analysis is undertaken 
with respect to this tempora! connective in a German cor- 
pus. Differences regarding the syntactic variations as 
compared to English are pointed out. Secondly, the ques- 
tion is raised what kind of rhetorical relation is expressed 
by this discourse marker and how this discourse marker 
interacts with the preceding discourse. 
One important outcome of these investigations is that 
the discourse marker nachdem does not only indicate a 
rhetorical relation (i.e. narration). I will suggest that the 
nachdem-clause serves a crucial function regarding dis- 
course organisati0n. Generally speaking, the temporal 
information conveyed by this clause provides the clue to 
return to a previously mentioned thread in the discourse. 
This function of a temporal clause is not discussed by 
Lascarides & Oberlander or by de Swart. 
The remaining part of the paper is organised as fol- 
lows: Section 2 contains the corpora analysis regarding 
syntactic variations.Section 3 provides a closer look at 
the data from a semantic angle. Example sequences are 
analysed regarding their discourse structure. The last 
section summarises the results and concludes. 
2 Nachdem in written text 
2.1 Corpora 
German The online corpora maintained by the Insti- 
tutfiJir deutsche Sprache (IDS) in Mannheim were used 
to collect the German data. I The corpus chosen for this 
study (i.e. Mannheimer Korpus t) contains novels as well 
as newspaper articles. After a selection process, 80 sen- 
tences were analysed in more detail. 
English A first investigation of the British National 
Corpus (BNC) 2 was carried out to compare to the ob- 
servation obtained for German. Two simple searches for 
the English discourse marker after were initiated. Spo- 
ken language data were excluded from the list of samples 
to be considered. The number of example sentences left 
was 30. 
2.1.1 Analysis 
The selected sentences (N=80; N=30) were read and 
analysed regarding the following criteria: 
Syntax Five different structures were distinguished: 
postponed The main clause occurs before the comple- 
ment clause (e.g. Seidler selbst machte auf, nachdem 
ich geklingelt hatte. \[BT, p. 396\] or he was removed from 
office (...) after he had openly burned his certificate of 
membership (...) \[B2W 694\]). 
tThe URL for the COSMAS-system is ht~p://www, ids- 
mannheim, de/ldv/cosmas/,~intro, html. 
2The URL for this corpus is ht tp : / / in fo. ox. ac. uk/bnc/. 
/58 
topicalised A temporal adverbial or a clause describing 
a salient situation is topicalised (e.g. Schon am Abend 
jenes ersten Tages. nachdem ich den Sonnenuntergang 
gefilmt hatte, spielten wir Pingpong \[HF, p.89\]). 
noun-first A Noun (or Noun Phrase) is followed by the 
complement clause and the verb phrase. This construc- 
tion, however, can only be found within another comple- 
ment clause (e.g ..... wo die Gitarrensanger, nachdera 
sie vor den Touristen-Restaurants gebettelt haben, ihre 
Pizza essen und Chianti per Glas trinken; \[HF, p. 150\]). 
verb-first The complement clause interjects the main 
clause between the verb and the objects or complements 
(e.g. Matzerath war im Laden und dekorierte, nachdera 
er das Geschirr vom Mittagessen abgewaschen hatte, das 
Schaufenster. \[BT, 245\]). 
preposed The complement clause succeeds the main 
clause (e.g. Nachdem er in diesem Ort die Grundschule 
hinter sich gebracht hatte, ging er zuru'ck in die Grofl- 
stadt zu seinem Vater (...) \[DBA, p. 172\] or After I made 
the cut l folded back the turf(... ) \[AOR 2942\]). 
Aspeetual Classes The main and complement clauses 
were categorised according to the four aspectual classes 
by Vendler (1967): 
State to love, to know, to cost 
Activity to run, to walk, to laugh 
Accomplishment to destroy, to create 
Achievement to notice, to win 
The classification of a situation regarding theses classes 
was tested using linguistic tests (e.g. John was happy for 
3 hours vs. *John was happy in 3 hours). 
2.2 Statistical results 
Most naturally one would expect that the syntactic sur- 
face reflects the actual ordering between the two situa- 
tions such as in (3): 
(3) After Peter had left, he went to the cinema. 
Lever (1989, p. 380), for instance, claims exactly this 
by saying that information to be expressed should be 
arranged according to the "natural ordering of its con- 
tent." But the corpus investigation did not confirm this 
claim. Instead, a variation of different syntactic struc- 
tures were found. Surprisingly enough, although the pre- 
posed structure was the most common one, it was defi- 
nitely not as dominating as one would expect. This struc- 
ture was found in only 33.75% of the studied cases. 
The verb-first structure, where the complement clause 
is inserted immediately after the verb of the main 
clause, occurred almost as often as the preposed struc- 
ture (i.e. 32.5 %). And even the postponed construction 
was found in 21.25% of the cases investigated. The top- 
icalised structure was found in 11.25% of the sentences 
(s. Figure 1). 
Assuming that the verb denotes the situation, we find 
in more than fifty percent of the cases a syntactic varia- 
tion that does not reflect the natural order of the described 
situations (i.e. the postponed and verb-first structure rep- 
resent 53.75% of all cases). 
The next section investigates in more detail how the 
syntactical structure correlates with the aspectual classes 
of the complement clause. 
v ~b- fTst : 31.2~ 
: 21.2~ 
Figure 1: The syntactical variants in German 
2.3 Comparison with after 
2.3.1 Differences in number 
The assumption that the preposed structure would occur 
most often in the corpus was not confirmed. In fact, an 
even higher distribution of the reverse syntactic structure 
was found compared to German, namely 60% (s. Figure 
2). 
Cases of the verb-first and noun-first structure did not 
occur at all in the English corpus which is not surprising, 
since these syntactic structures are not grammatical in 
English. 
toOie..,al'is~: 6.6"~ 
Figure 2: The syntactical variants in English 
2.3.2 Differences in structure 
At first sight the high number of the postponed syntac- 
tic structure would seem to be unexpected, although a 
high number of cases were found for the German data. 
However, bear in mind that more than half of the sen- 
tence in German were postponed or verb-first structures 
59 
(i.e. 53.75%) which are constellations not reflecting the 
actual ordering of the described situations. 3 
2.4 Intermediate results 
The corpora analysis showed that a variety of syntac- 
tic structures can be found in written text. The data 
disproved Levelt's claim that information should be ex- 
pressed according to a natural order. 
A brief investigation of an English corpus showed sim- 
ilar results regarding the syntactic variation. Differences 
between the languages could be explained by the differ- 
ent word order regarding temporal adverbials. 
The next section focuses on the discourse structure and 
what kind of effects can be observed between the nach- 
dem-clause and the preceding discourse. 
3 Nachdem in written discourse 
The following section gives a brief overview of how tem- 
poral connectives such as after are treated in discourse 
grammars. With de Swart's approach in mind, the pre- 
ceding context of a nachdem-clause in the 80 sentences 
of the German corpus investigation were studied wrt. the 
criteria concerning the aspectual classes. First of all, the 
aspectual class of the immediately preceding sentence 
was determined. Inspired by these data the discourse 
structure for several example discourse sequences is pre- 
sented and discussed. 
In section 3.3 the outcome of these investigations is 
used to gain a better view of how the discourse struc- 
ture was constructed with the help of temporal discourse 
markers. 
I argue that a nachdem-clause is used to provide di- 
rections through the discourse structure. The temporal 
information expressed can be seen as a reference back 
in time and discourse structure. These clues have to be 
given from the writer so that the reader will not lose the 
thread of the story. 
3.1 Temporal connectives 
De Swart proposes an analysis of temporal adverbials 
and temporal clauses within Segmented DRT (SDRT) 
(Asher, 1993). As mentioned in the introduction, a dif- 
ference in discourse structure can be observed for the two 
discourses ((1)=(5a), (2)=(5b)): 
(5) a. John had always been rather shy. But after he 
met H~l~ne, his behavior changed quickly. He 
was very self-assured now. 
3Furthermore. an analysis of a German sentence according 
to the to~uology approach to word order yields the following 
Vorfeld link¢ Kl. Mittelfeld rechte Kl. schema: Peter u'ank heute zwei Bier. Temporal 
adverbials can be most often found in the Mittelfeld according to this 
analysis. In English, however, temporal information is normally pre- 
sented at the end of a sentence. 
(4) Peter drank two beers today. 
Hence I will consider nachdem occurring in this construction as a tern- 
poral adverbial and not as a discourse marker. 
b. Ty and John were good friends, and they often 
went to the movies together. But John's behav- 
ior changed quickly after he met Hrl~ne. Ty 
said she bewitched him. 
The discourse in (5a) contains a preposed temporal 
clause that according to de Swart's account only adds fur- 
ther temporal information to the existing event structure. 
No further rhetorical information is supposedly derived. 
The main clause in (5b), she argues on the other hand, 
is connected with the preceding discourse via the rhetor- 
ical relation background. 
One question comes to mind after reading this expla- 
nation for the two discourses: why does de Swart not use 
the discourse marker bur in (5a) to derive the rhetorical 
relation contrast? Note that the discourse marker but is 
actually required to establish a coherent discourse. It is 
also necessary to derive this rhetorical relation, because 
otherwise the temporal relation between John's shyness 
and his change of behavior cannot be derived. Bear in 
mind that temporal relations a.re by-products of rhetori- 
cal relations within the SDRT framework. 
Lascarides & Oberlander (t994) propose an account 
that takes more into account the interaction between 
world knowledge and the underlying rhetorical relations. 
However, they do not consider the effect different syntac- 
tic variants may have, but provide a sound formal expla- 
nation for the example sentences discussed above. Their 
starting-point is a discourse that contains a state as the 
first sentence fo!~owed by a sentence containing a after- 
(or before-) clause. A similar example discourse they 
discuss is presented in the following: 
(6) Mary was cross with John, 
a. She was pacified after John gave her the tickets 
for the concert. 
b. ?John gave her the tickets for the concert before 
she was pacified. 
They show how the after-clause in (6a) interacts with 
the state described in (6). A rhetorical relation is derived 
between the situations described by these two clauses 
(i.e. Background). In addition, a Result relation between 
the after-clause and the main clause in (6a) can be estab- 
lished due to our world knowledge. 
The contrast between (6a) and (6b) is explained by 
Lascarides & Oberlander as follows. They point out that 
temporal connectives are presuppositional. The situation 
described by the temporal clause is assumed to have oc- 
curred for the entire sentence to have a truth-value (cf. 
(Heinam~iki, 1972)). If this is not the case, the situa- 
tion has to be accommodated into the reader's current 
discourse model. But this can't be satisfactorily done for 
the situation described by the before-clause in (6b). 
The example discourses discussed show that several 
factors have to be considered. Rhetorical relations hold- 
ing between the situation described by the after-sentence 
60 
and the situations described by the preceding context 
have to be derived. 
In order to get a better understanding of what kind of 
relations between an after-clause and the context can be 
derived, I examined in more detail what context such a 
clause is normally embedded in in the following section. 
3.2 The preceding context 
The data extracted for the corpus investigation in section 
2.2 were analysed a second time. The preceding sentence 
was classified according to its aspectual class. Interest- 
ingly enough, a high number of homogeneous aspectual 
classes were found (i.e. 65%). Consider (7): 4 
(7) (a) Peter schol~ seinen dreizehnten Spatz an diesem 
Nachmittag. (b) Er war sehr zufrieden mit seiner Aus- 
beute. (c) Nachdem er den dreizehnten zu den an- 
deren Spatzen gelegt hatte, (d) ginger nach Hause. 
(Peter shot his thirteenth sparrow at this afternoon. 
He was quite happy with this result. After he had 
laid the thirteenth beside the other sparrows, he went 
home.) 
This example sequence shows that the situation de- 
scribed by the after-.clause (i.e.(Tc)) refers back to a sit- 
uation mentioned in (7a). This example also shows that 
nachdem does not only add temporal relation to the event 
structure. The previous discourse structure has to be 
taken into account as well. 
3.3 Discourse structure and discourse guidance 
I agree with de Swart's observation that apreposed struc- 
ture triggers a break in the narration. But the connection 
has to be made wrt. the discourse structure derived to that 
point. Only adding a temporal constraint is not enough to 
cover the meaning and function of a nachdem-sentence. 
I suggest that nachdem's function in a written dis- 
course is to connect discourse segments. The difference 
in the syntactic position can explain how this connection 
is to be made. A preposed structure triggers a search 
in the discourse structure for a position where the situa- 
tion described by the main clause is to be connected to. 
A postponed structure, on the other hand, adds temporal 
information to the main clause situation and incorporates 
it with the discourse structure derived before-hand. But 
bear in mind that a rhetorical relation has to be estab- 
lished between the main clause and the preceding con- 
text. 
Consider a slightly altered version of (7): 
(8) Peter schol3 seinen dreizehnten Spatz an diesem Nach- 
mittag. (b) Er war sehr zufneden mit seiner Aus- 
beute. (d) ?Er ging nach Hause, (c) nachdem er den 
dreizehnten zu den anderen Spatzen gelegt hatte. 
The main clause can be connected to the shooting 
via narration. But, the nachdem-sentence cannot ini- 
tiate a search for a connecting situation then, since the 
"~This is a modified discourse found in \[BT, p. 448\]. 
narration relation has closed off this discourse segment. 
Hence the discourse structure has to be reorganised and 
(8c) is inserted in the already generated structure. 
4 Conclusions 
I have shown how a corpus analysis ofnachdem can shed 
some light on the usage of the temporal connective nach- 
dem in real text. The claim by Levelt that information is 
expressed according to a natural order was disproved by 
the data presented. 
Moreover, I have discussed the different discourse 
functions of nachdem wrt. the postponed and preposed 
structure. It was shown that de Swart's account of tem- 
poral clauses was not sufficient to explain the data dis- 
cussed. Lascarides & Oberlander's theory, on the other 
hand, is limited to a specific construction shown in (6). 
Discourse phenomena such as returning to an earlier 
thread in the discourse are not covered by their approach. 
Ongoing research is being carried out with the focus 
on the reading time of the two different syntactic vari- 
ants. Psychological experiments will provide more evi- 
dence that there is a difference between the variations. 
A Work of Fiction 
\[AC\] Ansichten eines Clowns ('The Clown') by Hein- 
rich BSII, \[BT\] Die Blechtrommel ('The Tin Drum') by 
Gtinter Grass, \[HF\] Homo Faber ('Homo Faber') by Max 
Frisch 

References 
Nicholas Asher. 1993. Reference to abstract Objects in 
Discourse, volume 50 of Studies in Linguistics and 
Philosophy. Kluwer Academic Publishers, Dordrecht. 
Henri6tte de Swart. 1994. Time adverbials in sentence 
and discourse. In Peter Bosch and Rob vander Sandt, 
editors, Focus and Natural Language Processing, vol- 
ume 2, pages 415-424. IBM Deutschland Informa- 
tionssysteme GmbH. 
Orvokki Heintimtiki. 1972. "before". In Papers from the 
Eighth Regional Meeting of the Chicago Linguistics 
Society, pages 139-151, Chicago, Illinois, USA. 
Alistair Knott. 1996. A Data-Driven Methodology for 
Motivating a Set of Coherence Relations. Ph.d. the- 
sis, Department of Artificial Intelligence, University 
of Edinburgh. 
Alex Lascarides and Jon Oberlander. 1994. Temporal 
connectives in a discourse context. In Proc. of the 7 th 
EACL, pages 260-268, Dublin, Ireland. Association 
for Computational Linguistics. 
W.J.M. Levelt. 1989. Speaking: From Intention to Ar- 
ticulation. MIT Press, Cambridge, Mass. 
William Mann and Sandra Thompson. 1987. Rhetorical 
structure theory: A theory of text organisation. Tech- 
nical Report ISI\]RS-87-190, USC/ISI, June. 
Zeno Vendler, 1967. Linguistics in Philosophy, chap- 
ter 4, pages 97-121. Cornell University Press, Ithaca 
NY. 
