See discussions, stats, and author profiles for this publication at: https://www.researchgate.

net/publication/247497478

Subordinating conjunctions as devices for unifying sentences in memory

Article  in  European Journal of Cognitive Psychology · July 2002

DOI: 10.1080/09541440143000104

CITATIONS READS

6 782

2 authors, including:

Ralf Rummer
Universität Kassel
66 PUBLICATIONS   791 CITATIONS   

SEE PROFILE

All content following this page was uploaded by Ralf Rummer on 05 January 2015.

The user has requested enhancement of the downloaded file.

 EUROPEAN JOURNAL OF COGNITIVE PSYCHOLOGY, 2002, 14 (3), 353–369

Subordinating conjunctions as devices for unifying

sentences in memory
Johannes Engelkamp and Ralf Rummer
Saarland University, SaarbruÈ cken, Germany

In two experiments, we provide evidence that the subordinating conjunctions

‘‘nachdem’’ [after] and ‘‘weil’’ [because] unify sentences in immediate memory
better than the coordinating conjunctions ‘‘erst–dann’’ [first–then] and ‘‘deshalb’’
[therefore] are able to do. Immediate recall of the two clauses of subordinate
sentences was equally good, whereas recall of coordinate sentences was better for
the first clause than for the second in both experiments. In Experiment 2, it was
additionally shown that this recall pattern was not due to prosodic information.

Most languages offer the possibility to encode two related events either in one or
in two sentences. It is, for instance, possible to encode the fact that event 1 (E1)
occurred before event 2 (E2) either as in (1) or as in (2).

(1) After Peter had taken a walk (E1), he wrote a letter (E2).
(2) First Peter had taken a walk (E1). Then he wrote a letter (E2).

We will call (1) subordinate structure and (2) coordinate structure. The question
concerning the psychological function of these structures has rarely been dis-
cussed in the literature. The main hypothesis pursued here is that subordinating
conjunctions, as in (1), facilitate the unification of the meaning of E1 and E2 in a
higher-order chunk more so than the coordinating conjunctions in sentence (2).
Most research on sentence processing stems from the field of psycho-
linguistics and has mainly focused on subordinate structures, while paying little
attention to coordinate structures: Moreover, syntactic features of sentences have
mostly been studied in experiments of sentence comprehension (e.g., Bever &

Requests for reprints should be addressed to J. Engelkamp, Saarland University, Psychology, D-

66123 SaarbruÈ cken, Germany. Email: engelkamp@rz.uni-sb.d e
This research was supported by the German Research Foundation (DFG) within the collaborative
research centre ‘‘Resource Adaptive Cognitive Processes’’ (SFB 378). We thank Stephanie
Neubauer for running the experiment s and analysing the data. For helpful comment s on earlier
versions of this paper we thank Gilbert Mohr, Mary C. Potter, and an anonymou s reviewer. Stylistic
corrections were made by Jane Gross.

# 2002 Psychology Press Ltd

http://www.tandf.co.uk/journals/pp/09541446.html DOI:10.1080/09541440143000104
354 ENGELKAMP AND RUMMER

Townsend, 1979; Garnham, 1985; Harley, 1995; Mitchell, 1994). With the
exception of the early years of psycholinguistics, syntactic features have rarely
been studied in experiments of memory. Examples are the studies of the role of
sentence depth in sentence recall (e.g., Johnson, 1965, 1968; Martin & Roberts,
1967) and of transformational complexity (Glucksberg & Danks, 1969; Mehler,
1963; Savin & Perchonock, 1965). These studies were motivated by Chomsky’s
theory of generative syntax in which semantic factors were not the focus
(Chomsky, 1957, 1965). This situation changed when the concept of proposition
emerged in the field (e.g., Engelkamp, 1974; Kintsch, 1974). However, with this
change interest in syntactic factors was lost.
A new perspective on sentence recall, which was less psycholinguistically
and more memory oriented, was developed in the work of Potter and Lombardi
(e.g., 1990). They formulated the conceptual regeneration hypothesis. According
to this hypothesis, high accuracy in immediate memory for surface information
is not due to a (phonological) surface representation of a string of words, but to
regeneration from the conceptual representation of the sentence. In subsequent
studies, the authors additionally demonstrated that syntactic structure does, to a
certain degree, contribute to short-term sentence recall (Lombardi & Potter,
1992; Potter & Lombardi, 1998). They demonstrated that the syntactic structure
of a preceding sentence facilitated recall of the syntactic structure of a following
sentence, if it corresponded to the preceding one.
Hence, whereas the role of syntactic factors receives much attention in
studies of sentence comprehension, this role has hardly been studied in memory
experiments. It still seems to be widely accepted that syntactic factors serve to
transport meaning in order to be forgotten after this function is served. This
position is particularly evident where a series of sentences are studied as in texts
for example. What is almost completely lacking is the comparative study of
subordinate and coordinate sentence structures with regard to memory. We will
fill the gap for temporal and causal sentences in this study. To be clear, we do
not question the outstanding role of meaning in sentence or text comprehension
and memory. However, we want to underline the fact that syntactic features are
functional for building semantic structures in memory. We assume that con-
junction words influence recall via their meaning as higher-order predicates,
which binds two clauses together. In this regard, we follow Potter and Lombardi
(1990) who assumed that sentences are comprehended by a process that includes
the activation of words on a lexical and conceptual level and their functional
relations as described by propositions. These conceptual representations of
propositions form the basis for memory for later processes of remembering.
They call this retrieval process conceptual regeneration. Although Potter and
Lombardi (1998) suggested that syntactic structure contributes to a certain
degree to this regeneration process, they do not specify this contribution in
detail. Essentially they demonstrate that syntactic structure remains active for a
short period and primes the next sentence if it has the same syntactic structure.
 UNIFYING SENTENCES IN MEMORY 355

We go beyond this proposal and assume that syntactic conjunctions function as

key words for activating temporal and causal schemata. These schemata open
the slots in which the propositions of complex sentences are filled in, and a
higher-order memory unit is formed (cf., Bransford & Johnson, 1972, p. 722;
Schank, 1978).
The assumption that propositions form basic units of memory is not new
(e.g., Anderson & Bower, 1973; Engelkamp, 1980a, b; Kintsch & Keenan,
1973). What we have added here is the assumption that conjunctions form
higher-order memory units that help to integrate the two-part propositions of a
complex sentence into a larger memory chunk, and furthermore that sub-
ordinating conjunctions are better key words than are coordinating conjunctions
for activating such a frame or schema.
Our experiments were motivated by an early study from Lesgold (1972). He
explored some circumstances under which the syntactic form of a sentence
determines the extent to which underlying propositions of a complex sentence
are unified into one schema or separately stored in memory (cf., Caplan, 1972).
He focused on the unifying role of pronominalisation and showed that sentences
using pronouns were recalled better than sentences that repeated the noun of an
earlier sentence. In his study, recall was tested for groups of five to eight
sentences, with a content word of the sentence serving as probe. However, the
type of the probe word did not influence the recall pattern.
He attributed this finding to the unifying function of a pronominal reference.
This finding holds true despite the fact that both sentence types had the same
number of shared lexical items and the same number of underlying propositions.
According to Lesgold (1972), propositions must not only have items in common
in order for them to be connected in memory, but they must also have those
commonalties flagged in the surface structure of the sentence. He also notes that
there may be other ‘‘flags’’ for propositional connection besides pronominali-
sation. The flagging of temporal sequencing is one candidate, and the flagging of
cause–effect relationship is another. Flagging of these two types of relationships
by coordinating conjunctions as in (3) and (5) and by subordinating conjunctions
as in (4) and (6) were studied here.

(3) First, he had climbed on the chair. Then he switched off the light.
(4) After he had climbed on the chair, he switched off the light.
(5) He opened the window. Therefore the candle was blown out.
(6) Because he opened the window, the candle was blown out.

We assume that subordinating conjunctions such as ‘‘after’’ or ‘‘because’’,

rather than coordinating conjunctions, provide a clearer signal for the recipient
that a temporal and a cause–effect relationship, respectively, will be described.
This difference should hold true for the temporal conjunctions ‘‘first–then’’
which occur at the beginning of the first and the second clause as well as for the
356 ENGELKAMP AND RUMMER

causal conjunction ‘‘therefore’’, which only occurs at the beginning of the

second clause. Generally, subordinating conjunctions should draw listeners’
attention more strongly to the specific temporal and causal relationship than do
the coordinating conjunctions. Therefore, both clauses of sentences such as in
(4) and (6) will be unified more strongly into one memory chunk than the
clauses of coordinate sentences as in (3) and (5). This advantage of subordinate
sentences should be independent of whether the conjunction in subordinate
sentences occurs already in the first sentence, as in (3), or only in the second
one, as in (5). With the comparison of subordinate structures and coordinate
structures, we go beyond Lesgold’s (1972) study of the unifying role of pro-
nominalisations in coordinate structures.
We used the subordinate clause at the first position in order to make the order
of mentioning congruent with the order of events (event1–event2 with temporal
sentences and cause event–effect event with causal sentences). Therefore, for
temporal sentences, we used the subordinating conjunction ‘‘nachdem’’ [after].
One should be aware that in complex temporal sentences, two factors influence
processing difficulty: order of mention and position of main clause. When the
order of mention is held constant, as in our study, the position of the main clause
is decisive. It has been shown for children as well as adults that the structure
‘‘main clause—first’’ is easier to process than the structure ‘‘subordinate
clause—first’’ (e.g., Amidon & Carey, 1972; Engelkamp & Mohr, 1984; Flores
d’Arcais & Joustra, 1979; Hatch, 1971). Because we used only subordinate
sentences with the ‘‘subordinate clause—first’’, we used the more difficult
structure of subordinate sentences. Since in coordinate sentences there are only
main clauses, these sentences always start with a main clause. In this way, the
difference in processing difficulty between subordinate and coordinate sentences
should be maximised. Moreover, children describe two temporally related
events, first in coordinate structures (two main clauses), and only thereafter in
subordinate structures (Clark, 1970; Ferreiro & Sinclair, 1971), and they also
understand coordinate temporal sentences better than subordinate sentences
(e.g., Amidon & Carey, 1972; Engelkamp & Mohr, 1984; Hatch, 1971). This
observation is in line with the widely accepted assumption that this holds true
for adults (e.g., Clark & Cark, 1968; Ferreiro & Sinclair, 1971). From here it is
only a small step to assume that coordinate sentences are also easier to
remember than subordinate sentences. This assumption would clearly contrast
with our assumption that subordinating conjunctions are the better device than
coordinating conjunctions for unifying sentences in memory.
In order to test the assumption that subordinate sentences are better recalled
than coordinate sentences, we manipulated the syntactic structure (subordinate
vs coordinate). We assume that the syntactic structure of our sentences directly
determines conceptual information processing, namely the unification of two
propositions into one conceptual chunk. This assumption leads to the prediction
of an interaction effect: Recall performance for clauses should differ between
 UNIFYING SENTENCES IN MEMORY 357

coordinate and subordinate sentence structures. Recall of the second clause

should be better for subordinate than coordinate structures. In addition we varied
word length. We did so because one might assume that word length affects
immediate recall on the phonological level. This assumption is at least widely
accepted for recall of unrelated word lists (e.g., Baddeley, 1986, 1997, for
reviews). If this assumption also holds true for immediate sentence recall, then
one would expect independent effects of syntactic structure and word length in
our study. Such an effect would underline that the effect of syntactic structure is
not a phonological effect.1
To summarise, in Experiment 1 we studied immediate recall of coordinate
and subordinate sentences containing short (one syllable) or long (three syl-
lables) nouns in the clauses. One should be aware that the conjunctions we use
(causal and temporal) serve as examples. The assumptions should hold true also
for other, functionally similar conjunctions. Therefore, we do not expect any
differences between causal and temporal sentences, and in particular no inter-
actions of these material variations with effects of syntactic structure or word
length.
In order to avoid ceiling effects, we presented our sentences in pairs and then
requested recall. That is, participants listened to two sentences before they
started to recall them. Because our predictions refer to the relative recall level
for clause 1 and clause 2, and recall could also differ between the two sentences,
we included clause and sentence position as factors in our statistical analyses.

EXPERIMENT 1
Experiment 1 aimed at testing the effect of syntactic structure and word length in
temporal and causal sentences on immediate serial recall of acoustically pre-
sented sentences. The experiment aimed at testing the assumptions of the uni-
fication hypothesis. According to this hypothesis, it is expected that subordinate
sentences are better recalled than coordinate sentences, and that this effect of
subordination is due to better recall performance of the second clause of the two
presented sentences.

Method
Participants
Thirty-two undergraduate students from Saarland University, all of whom
were German native speakers, participated in the experiment. They were paid for
participation.

1
We did not control the short and long words for frequency, familiarity, and imagery, because the
factor word length essentially served as a control factor. However, Konieczny and Rummer (1997)
showed that word length manipulations affected recall performance in a substantial way, even if the
words were matched according to their frequencies of occurrence.
358 ENGELKAMP AND RUMMER

Materials
Sixty-four sentences—half temporal and half causal—were constructed. Each
sentence consisted of two clauses which consisted of a subject, an object, and a
verb, respectively. The subject of the clauses was a personal pronoun, whereas the
object was a (monosyllabic or three-syllabic) noun. For each of the sentences, a
subordinate and a coordinate version was constructed. The first clause of the
subordinate sentences started with the temporal conjunction ‘‘nachdem’’ [after]
or with the causal conjunction ‘‘weil’’ [because]. In the coordinate sentences, the
temporal conjunction ‘‘erst–dann’’ [first–then] and the causal conjunction
‘‘deshalb’’ [therefore] were used. In subordinate and coordinate constructions,
the order of mention corresponded to the order of events. Orthogonal to this
syntactic variation, the length of the nouns in the clauses (monosyllabic vs three-
syllabic) was manipulated. Thus, for each of the 64 basic sentences four versions
were constructed (subordinate–monosyllabic, subordinate–three-syllabic, coor-
dinate–monosyllabic, coordinate–three-syllabic). In every trial two sentences
were presented. Both sentences were assigned to the same condition. The fol-
lowing examples illustrate the construction of the material. The translation of the
sentences into English is given in brackets. Notice that the first clause of the
German sentences was always in pluperfect and that the number of syllables per
sentence was identical for subordinate and coordinate constructions.

Temporal sentences
Coordinate:

Erst hatte er Bier/Apfelsaft getrunken; dann bemerkte er den Freund/Bekannten.

[First, he drank beer/apple-juice, then he noticed his friend.]
Erst hatte er die Post/BehoÈ rde informiert; dann eilte er zur/m Bar/Restaurant.
[First, he informed the post office/authority, then he hurried to the bar/restaurant.]

Subordinate:

Nachdem er (Bier/Apfelsaft) getrunken hatte; bemerkte er den Freund/Bekannten.

[After he had drunk beer/apple-juice, he noticed his friend.]
Nachdem er die (Post/BehoÈ rde) informiert hatte, eilte er zur/m Bar/Restaurant.
[After he had informed the post-office/authority, he hurried to the bar/restaurant.]

Causal sentences
Coordinate:

Er war Hans/Johannes begegnet, deshalb rief er Ralf/Sabine an.

[He met Hans/Johannes, therefore he phoned Ralf/Sabine.]
Er wollte den Berg/das Gebirge besteigen; deshalb erwarb er ein Seil/einen
Eispickel.
 UNIFYING SENTENCES IN MEMORY 359

[He wanted to climb the mountain/the mountain-range, therefore he acquired a

rope/an ice axe.]

Subordinate:

Weil er Hans/Johannes begegnet war, rief er Ralf/Sabine an.

[Because he met Hans/Johannes, he called Ralf/Sabine.]
Weil er den Berg/das Gebirge besteigen wollte; erwarb er ein Seil/einen Eispickel.
[Because he wanted to climb the mountain/the mountain-range, he acquired a rope/
an ice axe.]

Design and procedure

The factors syntactic structure (subordinate vs coordinate) and word length
(monosyllabic vs three-syllabic nouns) were orthogonally varied. All partici-
pants were presented half the trials with temporal and half the trials with causal
sentences. Further factors were the position of the clauses and the sentences. In
each trial, two sentences with the same syntactic structures each consisting of
two clauses were presented. Thus, besides syntactic structure and word length,
the statistical analyses also included clause (clause 1 vs clause 2) and sentence
(first sentence vs second sentence). All factors were varied within subjects.
Participants were presented with 64 sentences grouped into 32 pairs (each one
containing two sentences of the same syntactic structure and the same word
length condition). The pairs of two sentences were digitised and subsequently
stored acoustically on the hard disk of a personal computer. They were presented
in a randomised sequential order. The presentation was realised by means of
external loud speakers. Each participant was presented with only one variant of a
basic sentence, that is, a basic sentence was either presented in the subordinate
short noun, the subordinate long noun, the coordinate short noun, or the coor-
dinate long noun version of the respective sentence. Immediately after the last
word of a sentence pair was presented, participants had to orally recall the four
clauses as accurately as possible and in the correct sequential order. The recall
was tape-recorded.

Results
As dependent variable for each clause, it was recorded whether it was accurately
(verbatim) recalled or not. Thus, performance for each pair was scored
according to the number of correctly recalled clauses. Because the sentences
were repeatedly presented with the same structures, the order of the clauses was
consistently recalled correctly.
Because the use of causal vs temporal sentences served to realise two types of
conjunctions considered as theoretically equivalent, and no interaction with
syntactic structure and clause was expected, this assumption was tested first by
360 ENGELKAMP AND RUMMER

overall ANOVAs. We calculated ANOVAs over participants (F1 analyses) and

over items (F2 analyses). Because these overall analyses did not show any of the
previously discussed interactions, we averaged over causal and temporal sen-
tences. Therefore, the subsequent ANOVAs were based on the factors syntactic
structure, word-length, sentence position, and clause position.
All four main effects were significant. The main effect for syntactic structure
was F1(1, 31) = 4.27, MSe = .0446, p < .05; F2(1, 31) = 3.56, MSe = .0687, p <
.07, and for clause it was, F1(1, 31) = 19.82, MSe = .0196, p < .001; F2(1, 31) =
6.48, MSe = .830, p < .05. Subordinate sentences were better recalled than
coordinate sentences and clause 1 was better recalled than clause 2. However,
these main effects were modified as expected by a significant interaction
between syntactic structure and clause position, F1(1, 31) = 9.84, MSe = .0246,
p < .01; F2(1,31) = 13.68, MSe = .0321, p < .001. This interaction is depicted in
Figure 1.
This interaction made it clear that only in coordinate sentences, recall for the
second clause was worse than for the first clause (LSD test p < .001). The LSD
test refers here and subsequently to the F1 analyses. For subordinate sentences,
however, the difference between clause 1 and clause 2 was not significant. A
comparison between subordinate and coordinate sentences showed corre-
spondingly no difference in recall for clause 1, however, it showed a recall
advantage of the subordinate structure for clause 2 (LSD test p < .001).
The two other main effects were also significant: Sentences with mono-
syllabic nouns (54%) were better recalled than sentences with three-syllabic

Figure 1. Recall performance as a function of the syntactic structure of the sentences and the
position of the clauses within the sentences in Experiment 1.
 UNIFYING SENTENCES IN MEMORY 361

nouns (46%), F1(1, 31) = 45.16, MSe = .0306, p < .0001; F2(1, 31) = 24.49, MSe
= .0718, p < .0001. Moreover, this effect did not interact with syntactic structure
and clause. The other main effect concerns the sentence factor. Recall perfor-
mance for the first sentence (63%) was better than performance for the second
sentence (49%), F1(1, 31) = 51.65, MSe = .0487, p < .0001; F2(1, 31) = 19.80,
MSe = .1516, p < .001. This effect did not interact with syntactic structure and
clause.

Discussion
The most important finding of this first experiment is the interaction between the
syntactic structure of the sentences and the recall performance for the clauses.
This interaction supports the unification hypothesis of subordinate structures.
The two clauses of subordinate sentences were recalled equivalently well. This
finding indicates that there is a strong unification of both clauses into one chunk.
In coordinate sentences, on the other hand, the second clause was recalled worse
than the first one. This finding demonstrates that the chunking of both clauses
was weaker in this case.
Because we assume that chunking takes place on a conceptual level, we
consider the recall of our sentences as concept- or meaning-based as do, for
instance, Lesgold (1972) or Potter and Lombardi (1990). And we claim with
Lesgold (1972) that syntactic conjunction words indicate conceptual informa-
tion. In our experiment, the syntactically flagged information was temporal and
causal relation. However, this flagging was done either by a subordinating
conjunction (after, because) or by a coordinating conjunction (‘‘first–then’’,
‘‘therefore’’). It was assumed that the subordinating conjunction is the more
efficient flagging for integrating two clauses and their underlying propositions
than the coordinating conjunction. It could be argued that this is the case in our
experiment because the subordinating conjunction comes at the beginning of the
first clause, whereas the coordinating conjunction only at the beginning of the
second clause. However, this confounding variable holds true only for causal
sentences. Temporal sentences begin with ‘‘first’’ as the first word of the first
clause. This argument can be refuted because the results do not differ for
temporal and causal sentences. Also the fact that, in our experiment, the main
clause was always second in the subordinate sentences, does not seem to be
critical, given that Lesgold (1972) varied the position of the clauses (or
propositions) without observing changes in recall.
The interaction which we observed between syntactic structure and clause
position is clearly in line with the unification hypothesis. However, it also
clearly contrasts the purging-the-buffe r hypothesis suggested by Jarvella (1971,
1979). This hypothesis claims that listeners hold verbal information in short-
term memory until the information is no longer of use, that is, until they have
used them to form the underlying propositions. Jarvella (1970, 1971) showed in
362 ENGELKAMP AND RUMMER

several experiments that listeners purge short-term memory (which is suggested

to store the verbal surface of the sentence) typically after a sentence boundary.
That means that participants recall the last sentence better than those before it. In
his studies, participants listened to a prose passage that was interrupted several
times with the request that participants write down, word for word, as much as
they could remember of the end of the passage immediately preceding the
interruption. These expectations were confirmed in Jarvella’s studies (e.g., 1970,
1971).
How can these different findings from Jarvella (1970, 1971) and our study be
reconciled? Whereas in our experiment, recall performance declined along the
order of presentation (with better recall for the first presented material), recall
performance in Jarvella’s experiments declined in the other direction. The more
recently presented the items are, the better is the recall performance. We believe
that the participants of Jarvella’s and our experiment followed different recall
strategies. Jarvella’s participants probably started from the end of the passage.
They first recalled what was, so to speak, still in their ear, and only then
proceeded to the other part of the text. Our participants usually started from the
first sentence of a pair and proceeded then to the second sentence. The use of
these different strategies could be due to divergent procedural aspects of the
studies.
Moreover, the texts presented by Jarvella were difficult and abstract (e.g.,
‘‘The confidence of Kofach was not unfounded. To stack the meeting for
McDonald, the union had even brought in outsiders.’’). In contrast, our sen-
tences were concrete. Abstract material is more difficult to understand and can
reinforce the storage of the surface structure rather than concrete material (e.g.,
Begg & Paivio, 1969). Furthermore, writing down the last clause also induces
the forgetting of the earlier clauses, at least their surface structure, and because
the sentences are abstract, the occurrence of synonyms is enhanced if the exact
surface is forgotten. However, synonyms are considered as incorrect. This
explains the decline of recall from the last to the first clause. In our study, the
sentence structures are homogeneous and the focus is on the content of the
clauses and on their relationship within each sentence.
These considerations make clear that details of the material and the procedure
must be taken into account if the findings of Jarvella’s and our study are to be
compared and interpreted. It turns out that different recall strategies are most
likely to be responsible for the inconsistent findings of both studies.
The finding that word length of the nouns within the clauses influenced recall
level, but that word length did not change the pattern of our interaction effect
speaks for the assumption that the syntactic structure of our sentences and the
word length operate on different levels of processing. Whereas syntactic
structure via the conjunction words operates on the conceptual level and
determines the unification of clauses, word length determines word recall within
clauses and it might well be that phonological processing also contributes to
 UNIFYING SENTENCES IN MEMORY 363

word recall within clauses (cf., Martin, 1993; Rummer & Engelkamp, 2001;
Rummer & Schrobildgen, 2000).
However, there is one explanation that would extend and modify the
assumption that the conjunction words are critical for the unification effect.
One could argue that the critical aspect is not the conjunction word per se but
the prosodic structure of the acoustically presented sentences. It was repeat-
edly shown that prosodic information influences sentence understanding (e.g.,
Bader, 2000; Fodor, 1998; Schafer, Carter, Clifton, & Frazier, 1996). From
this position, it might be expected that in Experiment 1, prosodic information
supported the encoding and the remembering of the sentences in that it helped
unify subordinate clauses. If this holds true, the interaction effect should be
reduced or even abolished if sentences were presented without prosodic infor-
mation. The goal of Experiment 2 was to test this hypothesis. Because we
asked for an oral recall, it is possible that the participants were motivated to
generate prosodic structures, even if the sentences were not presented with
such. We therefore instructed our participants in Experiment 2 to repeat the
sentences in the monotonous, prosodic-free form in which they were pre-
sented.

EXPERIMENT 2
In Experiment 2, we used the same sentences and the same design as in
Experiment 1. However, in this experiment the sentences were presented in a
prosodic-free manner (i.e., without the typical intonation and rhythm), and they
had to be reproduced in this prosodic-free manner. If prosodic information is
critical to the unification of subordinate sentences, the interaction between
syntactic structure and type of clause should disappear or be reduced.

Method
Participants
Thirty-two undergraduate students from Saarland University, all of whom
were German native speakers, participated in the experiment. They were paid or
received credits for participating in the experiment.

Materials and procedure

The same sentences as in Experiment 1 were presented to the participants.
However, in this second study sentences were presented without prosody. In
order to realise prosody-fre e versions of the sentences, all morphological forms
of the words appearing in the presented sentences were separately recorded. The
sentences, then, were generated out of this lexicon. Thus, the material lacked
sentence-specific prosodic information but still included stress information
concerning the single words of the sentences. However, presentation time per
364 ENGELKAMP AND RUMMER

sentence was longer for the prosody-fre e sentences of Experiment 2, than for
sentences presented with natural prosody in Experiment 1.
Otherwise, the procedure and analysis were identical to Experiment 1 except
for the recall instructions. Participants were instructed to recall the sentences in
the same way they were presented, that is, they had to orally reproduce them in a
prosody-fre e way.

Design
Just as Experiment 1, this second experiment is based on a 2 £ 2 £ 2 £ 2
within-subjects design with the factors syntax, word length, sentence position,
and clause position.

Results
As in Experiment 1, there were main effects for syntactic structure, F1(1, 31) =
4.72, MSe = .0445, p < .05; F2(1, 31) = 5.87, MSe = .0367, p < .05, and for
clauses, F1(1, 31) = 10.59, MSe = .0422, p < .01; F2(1, 31) = 4.07, MSe = .1115,
p < .05. These main effects were, as in Experiment 1, modified by a marginally
significant interaction, F1 (1, 31) = 3.18, MSe = .0291, p < .09; F2(1, 31) = 3.60,
MSe = .0195, p < .07. This interaction is illustrated in Figure 2.
Planned comparisons with the LSD tests of F1 analyses showed that the recall
levels of clause 1 and clause 2 did not significantly differ for subordinate sen-

Figure 2. Recall performance as a function of the syntactic structure of sentences and the position
of the clauses within the sentences in Experiment 2.
 UNIFYING SENTENCES IN MEMORY 365

tences, but they did so for coordinate sentences (p < .01). Recall performance of
both sentence structures was equal for clause 1 but different for clause 2 (p <
.01).
In addition, there was a main effect of word length, F1(1, 31) = 13.84, MSe =
.0367, p < .001; F2(1, 31) = 11.40, MSe = .0438, p < .01. Sentences with
monosyllabic words were recalled better (48%) than those with three-syllable
words (42%). Also, recall differed for sentence 1 and 2, F1(1, 31) = 43.33, MSe
= .0556, p < .000; F2(1, 31) = 20.15, MSe = .1188, p < .0001. Sentence 1 (52%)
was recalled better than sentence 2 (38%). However, these effects were
modulated by a triple interaction of the factors syntactic structure, word length,
and sentence position, F1(1, 31) = 9.90, MSe = .0212, p < .01; F2(1, 31) = 4.27,
MSe = .0457, p < .05. This interaction is illustrated in Figure 3. This pattern of
data essentially shows that there were no longer any word-length effects for the
first sentence if it was coordinate and for the second sentence if it was sub-
ordinate.
A common ANOVA over both experiments, including ‘‘experiment’’ as an
additional between-subjects factor, was carried out in order to show more
directly that the triple interaction is due to the lack of prosodic information. This
analysis showed that the triple interaction was modified by the factor
‘‘experiment’’. The fourfold interaction was clearly significant, F1(1, 62) =
16.30, MSe = .0168, p < .001; F2(1, 62) = 6.85, MSe = .0384, p < .05. In other
words, the triple interaction in Experiment 2 was different from the findings in
Experiment 1. Moreover, this analysis revealed that the recall level of Experi-

Figure 3. Recall performance as a function of syntactic structure, number of syllables of the nouns,
and the position of the sentence.
366 ENGELKAMP AND RUMMER

ment 1 was higher than that of Experiment 2, F1(1, 62) = 12.10, MSe = .2739,
p < .001; F2(1, 62) = 19.40, MSe = .1738, p < .0001.

Discussion
The findings of Experiment 2 can be summarised as follows:

. Lack of prosodic information decreased recall performance.

. Lack of prosodic information did not abolish the unification effect of sub-
ordinate structures; it attenuated it.
. Lack of prosodic information modified the effect of word length. It abolished
the effect in the first sentence if it was subordinate and in the second if it was
coordinate.

The main effect of lacking prosodic information is plausible. Prosodic infor-

mation is obviously helpful to encode and retrieve sentence information. This
positive effect of prosodic information is known from on-line studies (cf., Bader,
2000; see also the special issue of Language and Cognitive Processes on
‘‘Prosody and Parsing’’ edited by Warren, 1996). It could also be shown here for
the performance of sentence recall. However, the monotonous manner of recall
could also be responsible for the decline of recall.
However, the finding that the interaction of syntactic structure and clause is
only slightly attenuated and still there in Experiment 2, makes it clear that an
important contribution to this interaction effect stems from the conjunction word
and not from the prosodic structure of the sentence. The conjunction word of a
subordinate structure induces the unification effect. It is also interesting to notice
that this unification effect is independent of the position of the sentences and of
word length. This finding underlines that the unifying effect of subordination is
due to the conjunction word and not to the prosodic structure which relates to the
structure of the whole sentence. It is obviously the lack of this supra-segmental
prosodic information of the whole sentence that modifies the word-length and
sentence position effects in Experiment 2.

GENERAL DISCUSSION
The comparison of subordinating and coordinating conjunctions in sentence-
processing and sentence-memory research has received little attention, although
Engelkamp (1976) and Kintsch and Keenan (1973) have discussed higher-order
predicates. Causal connection, for instance, was considered as such a higher
order predicate that integrated two propositions into a higher one. However,
within this propositional approach, it was not distinguished how the causal
connection was communicated at the surface, that is, subordinating and coor-
dinating conjunctions were considered to be functionally equivalent. Our results
show that this equivalence does not hold true. It would be worthwhile to study
 UNIFYING SENTENCES IN MEMORY 367

whether similar differences hold for other subordinating and coordinating

conjunctions such as ‘‘whereas’’ vs ‘‘however’’ or ‘‘although’’ vs ‘‘never-
theless’’, etc.
Another aspect that has become obvious in this study is that immediate
sentence recall is multidimensionally determined. It was not only shown that
conjunction words were critical for sentence recall, but also that word length and
prosodic structure influence sentence recall independently of the conjunction
words. Hence, the assumption that immediate sentence recall is primarily based
on phonological information, as the ‘‘purging-the-buffer ’’-hypothesis assumed
(e.g., Jarvella, 1971), as well as the assumption that it is more or less exclusively
based on conceptual information, as the conceptual regeneration hypothesis
suggested (e.g., Potter & Lombardi, 1990), is too constrained. It rather seems
that immediate sentence recall relies on whatever information is available. This
latter hypothesis is in agreement with the more general assumption that struc-
tural aspects of language surface are functional for sentence processing and
recall. For some aspects, this functional role becomes obvious even if the aspects
are realised in isolated sentences, as in this study.
A final note refers to the role of the dependent variable used. In our study, we
measured immediate sentence recall. It has to be recognised that measures, other
than sentence reading time or comprehension, measured as acting-out, do not
correspond in a one-to-one manner. It may be that a factor influences on-line
comprehension and off-line recall differently. This holds true, for instance, for
temporal sentences, as a comparison of this study and other studies show (e.g.,
Amidon & Carey, 1972; Engelkamp & Mohr, 1984). It may be that by making
sentences difficult to understand, readers or hearers must elaborate harder on
them, and this elaboration eventually facilitates their recall. Hence, it is desir-
able to take into account the kind of task used if one theorises about sentence
processing.

Manuscript received August 2000

Revised manuscript received January 2001

REFERENCES
Amidon, A., & Carey, P. (1972). Why five-year-olds cannot understand ‘‘before’’ and ‘‘after’’.
Journal of Verbal Learning and Verbal Behavior, 11, 417–423.
Anderson, J.R., & Bower, G.H. (1973). Human associative memory. Washington, DC: Winston &
Sons.
Baddeley, A.D. (1986). Working memory. Oxford, UK: Oxford University Press.
Baddeley, A.D. (1997). Human memory. Hove, UK: Psychology Press Ltd.
Bader, M. (2000). On reanalysis: Evidence from German. In B. Hemforth & L. Konieczny (Eds.),
German sentence processing (pp. 187–246). Dordrecht, The Netherlands: Kluwer Academic
Publishers.
Begg, J., & Paivio, A. (1969). Concreteness and imagery in sentence meaning. Journal of Verbal
Learning and Verbal Behavior, 8, 821–827.
368 ENGELKAMP AND RUMMER

Bever, T.G., & Townsend, D.J. (1979). Perceptual mechanisms and formal properties of main and
subordinate clauses. In W. Cooper & E.C.T. Walker (Eds.), Sentence processing (pp. 159–226).
Hillsdale, NJ: Lawrence Erlbaum Associates Inc.
Bransford, J.D., & Johnson, M.K. (1972). Contextual prerequisites for understanding : Some inves-
tigations of comprehensio n and recall. Journal of Verbal Learning and Verbal Behavior, 11, 717–
726.
Caplan, D. (1972). Clause boundaries and recognition in latencies for words in sentences. Perception
and Psychophysics, 12, 73–76.
Chomsky, N. (1957). Syntactic structures. The Hague, The Netherlands: Mouton.
Chomsky, N. (1965). Aspects of the theory of syntax. Cambridge, MA: MIT Press.
Clark, E.V. (1970). How young children describe events in time. In G.B. Flores d’Arcais & W.J.M.
Levelt (Eds.), Advances in psycholinguistics (pp. 275–284). Amsterdam: North-Holland Publishing.
Clark, H.H., & Clark, E.V. (1968). Semantic distinctions and memory for complex sentences.
Journal of Experimental Psychology, 20, 129–138.
Engelkamp, J. (1974). Psycholinguistik. MuÈ nchen, Germany: Fink.
Engelkamp, J. (1976). Satz und Bedeutung. Stuttgart, Germany: Kohlhammer.
Engelkamp, J. (1980a). Some memory tests for instrument and beneficiary as propositional argu-
ments. In F. Klix & J. Hoffmann (Eds.), Cognition and memory (pp. 125–130). Berlin, Germany:
Deutscher Verlag der Wissenschaften.
Engelkamp, J. (1980b). Some studies on the internal structure of propositions. Psychological
Research, 41, 355–371.
Engelkamp, J., & Mohr, G. (1984). Temporalsatzverarbeitun g bei geistig behinderten Kindern und
Jugendlichen. Zeitschrift fuÈ r Entwicklungspsychologi e und PaÈ dagogisch e Psychologi e, 16, 306–
322.
Ferreiro, E., & Sinclair, H. (1971). Temporal relationships in language. International Journal of
Psychology, 6, 39–47.
Flores d’Arcais, G.B., & Joustra, J. (1979). The expression of temporal order in descriptive language.
Italian Journal of Psychology, 6, 203–223.
Fodor, J.D. (1998). Learning to parse. Journal of Psycholinguisti c Research, 27, 285–319.
Garnham, A. (1985). Psycholinguistics. London: Methuen.
Glucksberg, S., & Danks, J.H. (1969). Grammatical structure and recall: A function of the space in
immediate memory or of recall delay? Perception and Psychophysic s, 6, 113–117.
Harley, T.A. (1995). The psychology of language. Hove, UK: Psychology Press Ltd.
Hatch, E. (1971). The young child’s comprehensio n of time connectives. Child Development, 42,
2111–2113.
Jarvella, R.J. (1970). Effects of syntax on running memory span for connected discourse.
Psychonomic Science, 19, 235–236.
Jarvella, R.J. (1971). Syntactic processing of connected speech. Journal of Verbal Learning and
Verbal Behavior, 10, 409–416.
Jarvella, R.J. (1979). Immediate memory in discourse processing. In G.H. Bower (Ed.), The
psychology of learning and motivation (Vol. 13, pp. 379–421). New York: Academic Press.
Johnson, N.F. (1965). On the relationship between sentence structure and the latency in generating
the sentence. Journal of Verbal Learning and Verbal Behavior, 5, 375–380.
Johnson, N.F. (1968). Sequential verbal behavior. In T.R. Dixon & D.L. Horton (Eds.), Verbal
behavior and general behavior theory (pp. 421–450). Englewood Cliffs, NJ: Prentice-Hall.
Kintsch, W. (1974). The representation of meaning in memory. New York: John Wiley.
Kintsch, W., & Keenan, J.M. (1973). Reading rate and retention as a function of the number of
propositions in the base structure of sentences. Cognitive Psychology, 5, 257–274.
Konieczny, L., & Rummer, R. (1997). Phonological storage and syntactic complexity. Paper pre-
sented at the Architectures and Mechanisms for Language Processing conference (AMLaP-97),
Edinburgh, UK, September.
 UNIFYING SENTENCES IN MEMORY 369

Lesgold, A.M. (1972). Pronominalization: A device for unifying sentences in memory. Journal of
Verbal Learning and Verbal Behavior, 11, 316–323.
Lombardi, L., & Potter, M.C. (1992). The regeneration of syntax in short term memory. Journal of
Memory and Language, 31, 713–733.
Martin, E., & Roberts, K.H. (1967). Grammatical factors in sentence retention. Journal of Verbal
Learning and Verbal Behavior, 5, 211–218.
Martin, R.C. (1993). Short-term memory and sentence processing: Evidence from neuropsychology .
Memory and Cognition, 21, 176–183.
Mehler, J. (1963). Some effects of grammatical transformations on the recall of English sentences.
Journal of Verbal Learning and Verbal Behavior, 6, 335–338.
Mitchell, D.C. (1994). Sentence parsing. In M.A. Gernsbacher (Ed.), Handbook of psycholinguistic s
(pp. 375–409). San Diego, CA: Academic Press.
Potter, M.C., & Lombardi, L. (1990). Regeneration in the short-term recall of sentences. Journal of
Memory and Language, 29, 633–654.
Potter, M.C., & Lombardi, L. (1998). Syntactic priming in immediate recall of sentences. Journal of
Memory and Language, 38, 265–282.
Rummer, R., & Engelkamp, J. (2001). Phonological information contributes to short-term recall of
auditorily presented sentences. Journal of Memory and Language, 45, 451–467.
Rummer, R., & Schrobildgen, M. (2000). Phonologisch e Information beim kurzfristigen Behalten
von SaÈ tzen. Zeitschrift fuÈ r Experimentelle Psychologie, 47, 144–155.
Savin, H.B., & Perchonock, E. (1965). Grammatical structure and immediate recall of English
sentences. Journal of Verbal Learning and Verbal Behavior, 4, 348–353.
Schafer, A., Carter, J., Clifton, C., Jr., & Frazier, L. (1996). Focus in relative clause construal.
Language and Cognitive Processes, 11, 135–164.
Schank, R.C. (1978). Predictive understanding. In R.N. Campbell & P.T. Smith (Eds.), Recent
advances in the psychology of language : Formal and experimental approaches. New York.
Plenum Press.
Warren, P. (Ed.) (1996). Prosody and parsing. Language and Cognitive Processes, 11 (1/2), 1–215.

View publication stats