CONTENT AND EPISTEMIC RELATIONS: A DEVELOPMENTAL STUDY OF RECALL

* Sara Verbrugge and Aline Sevenants, Laboratory of Experimental Psychology, KU Leuven. We want to thank the anonymous referee for his or her helpful comments on an earlier version of this paper. The remarks have significantly improved both the argumentation and the readability of the text. Needless to say, all remaining errors are our own responsibility. Correspondence concerning this article should be addressed to Sara Verbrugge, Laboratory of Experimental Psychology, KU Leuven, Tiensestraat 102 bus 3711, B-3000 Leuven. E-mail: Sara.Verbrugge@ppw.kuleuven.be CONTENT AND EPISTEMIC RELATIONS: A DEVELOPMENTAL STUDY OF RECALL

(1) I dropped a vase. So it broke (2) The lights are out. So the neighbours are not home. These sentences have the same outward appearance: two sentences between which a relation is expressed by means of the connective so. However, research in both linguistics and psychology has shown that these relations carry different characteristics. They are called content (or semantic) and epistemic (or pragmatic) coherence relations. Roughly speaking, sentences with a content relation express relations between events in reality, whereas epistemic causal sentences express relations between states of thinking (Dancygier, 1998;Dancygier & Sweetser, 2005;Sanders, 1997;Sweetser, 1990). Sentences with a content relation, such as (1), express relations between events in reality: real-world causality connects the two events described in the two clauses. The first part is often a cause or condition for the second part. Coherence exists because the world that is described is perceived as coherent (Sanders, Spooren, & Noordman, 1992). Content relations hold between the propositional content of the sentences of a text.
(3) He had been working late yesterday. So he was tired.
(4) He was yawning. So he was tired.
In (3) a relation is expressed between events in the outside world: being tired is a result in the real world of working late. In epistemic sentences, such as (4), relations between states of thinking are expressed: the speaker arrives at a particular conclusion and gives evidence how he arrived at that conclusion (premise-conclusion relation). Coherence exists because of the writer's goal-oriented communicative acts (Sanders et al., 1992). Epistemic relations hold between the utterances of the text or the beliefs the text is based on (Knott, 2001), e.g., (4). A relation is expressed between beliefs: the speaker infers on the basis of seeing someone yawning that that person must also be tired. The relation comes into being solely by means of the speaker's inferential system. That is why epistemic relations are also often called inferential relations.
Content and epistemic relations are often used to venture hypotheticalities about possible outcomes of events and thus occupy a very valuable position in everyday thinking and reasoning about both the future and the past. Overall, epistemic relations are said to be more complex as regards reasoning and epistemic relations are later acquired.
The relations between the two parts of a coherence relation can, but need not be made explicit by means of connectives (e.g., so, therefore, because). Connectives give instructions how to integrate the second segment with the first one.
Below we will go deeper into the difference between content and epistemic relations and the importance of connectives for these relations.

Content and epistemic relations
Differences between content and epistemic relations have been shown both in the purely linguistic and the psycholinguistic literature: they were shown to differ with respect to reasoning, clause integration, and acquisition.
As for adult participants, it has been shown that they react differently to content and epistemic relations in a reasoning task. Verbrugge, Dieussaert, Schaeken, Smessaert, and Van Belle (2007) have shown that participants reason differently with content and epistemic conditionals. It appears that reasoning is easiest with conditionals that most closely reflect the order of events of the real world.
It has also been shown that content and epistemic conditionals differ on the level of clause integration (Verbrugge & Smessaert, 2011). The difference between antecedent subclauses of content conditionals and those of epistemic conditionals is standardly associated with a high degree versus a lower degree of clause integration with their respective consequent main clauses. Clause integration was measured in terms of the semantic and grammatical compatibility between conditional sentences and alternative formulations by inserting an evidential adverb into the consequent main clause, replacing the subordinating conjunction or clefting the conditional subclause. Clear differences came to the surface between content and epistemic conditionals on these tests. Consider the examples (5-6). Clefting the content conditional (5b) yields a grammatical result with the same basic meaning as the original sentence, whereas this does not hold for the epistemic conditional in (6b).
(5a) If you press the button, the bell rings. Acquisition studies have shown that epistemic relations are usually acquired later than content relations (see also Givón, 2009). Evers-Vermeul (2005) and Kyratzis, Guo, and Ervin-Tripp (1990) show via different studies that the epistemic domain seems to be acquired last. Spooren and Sanders (2008) present evidence for the fact that epistemic relations are more cognitively complex. Kyratzis et al. (1990) make a distinction between three types of causal expressions: Speech Act-Level Causals, Content-Level Causals and Epistemic Causals. In Speech Act-Level Causals (e.g., Is everything all right? Because you look really sad today), "the clause with 'because' or 'so' justifies why something is said, and why the hearer should comply with what is said, rather than explaining the event that is referred to in the matrix clause" (Kyratzis et al., 1990, p. 205). The Content-Level Causals and Epistemic Causals comply with our definitions given above. The purpose of the study of Kyratzis et al. (1990) was to examine the relative frequency of these three types of causals in child discourse. For the data-analysis, children were divided in three age groups. The first group included children aged 2.4 until 3.6. The second group included children aged 3.7 until 6.6. The third group included children aged 6.7 until 12.0. Kyratzis et al. observed that epistemic causals were very infrequent for all age groups, contrary to content and speech act causals. They relate this to the fact that "epistemic meanings are more cognitively complex for young children" (Kyratzis et al., 1990, pp. 209-210).
Evers-Vermeul (2005; see also Evers-Vermeul & Sanders, 2011) conducted an experiment among Dutch children of four and five years old. One of the goals was to investigate whether young children can already produce epistemic relations. Therefore, particular contexts were manipulated to favour epistemic utterances. She shows that "children as young as four are capable of producing epistemic relations when they have the communicative goal of persuading someone, which is the case in an argumentative context" (Evers-Vermeul, 2005, p. 235). Moreover, the five year old children produced more epistemic relations than the four year old children. This shows that the ability to produce epistemic relations increases with age. In another, longitudinal study, investigating the acquisition order of the domains by observing emergence and use of connectives such as 'because' (omdat), 'because/for' (want), 'that's why' (daarom), 'so' (dus) in the different domains, Evers-Vermeul shows that "the epistemic domain seems to be acquired last" (2005, p. 243). However, she also claims that the different contexts in which the sentences are uttered plays a crucial role. Spooren, Tates, and Sanders (1996) conducted an experiment with sixseven year olds and eleven-twelve year olds. Overall, children produced more semantic than pragmatic relations. Young children produced proportionally fewer pragmatic relations than older children. Spooren et al. (1996) explain this by referring to the fact that semantic relations are cognitively less complicated than pragmatic relations. They claim that, in general, less complex relations are acquired before more complex situations.

Connectives and epistemic modal markers
As we have briefly indicated above, the two parts of a coherence relation can be integrated by means of a connective. Connectives (and often also modal markers) can influence the ease of understanding of these types of relations. Evidence has been adduced that particular contexts and settings can enhance the use and understanding of epistemic relations, which are by default the more complex and more difficult type of relation for children. Here we explore the contribution of epistemic modal elements and connectives to the straightforwardness with which epistemic relations are understood. Traxler, Sanford, Aked, and Moxey (1997) showed that the use of modal constructions can help the reader in the direction of a particular interpretation. They conducted a reading time experiment with causal (i.e., content) and diagnostic (i.e., epistemic) sentences involving because (e.g., causal sentence: Tina had to walk five miles because her engine stalled on the motorway; diagnostic sentence: Tina ran out of gas because her engine stalled on the motorway). Overall, the processing of the second clauses of diagnostic sentences took longer than the processing of the second clauses of causal sentences. Yet, in their second experiment they report that when a particular modal construction (e.g., perhaps, maybe, might) was added, reading times decreased considerably for the second clauses of the diagnostic sentences. They were then as easily processed as causal sentences. Diagnostic interpretations become more readily available and are consequently more easily processed.
In Verbrugge (2007) two elicitation tasks with children and adults showed that adults produce more epistemic conditionals spontaneously than children. Moreover, the use of epistemic markers (i.e., markers indicating the speaker's stance towards the truth or falsehood of an utterance; e.g., 'may well' (zal wel), 'probably' (waarschijnlijk), 'then it means that' (dan betekent dat)) helps to boost the production of epistemic relations, both for children and adults. It has been stated in the literature (Bloom, Lahey, Hood, Lifter, & Fiess, 1980;Evers-Vermeul, 2005;Kyratzis et al., 1990) that content relations are acquired before epistemic relations. Verbrugge (2007) has added that even in the later phases of language acquisition (9 until 12 years old) children still have a clear preference for content conditionals. They spontaneously use fewer epistemic conditionals than adults. The two experiments also revealed that, when participants are asked to insert particular epistemic markers into the consequent of their conditionals, the number of epistemic conditionals increases both for adults and children. The fact that the insertion of epistemic markers can speed up the processing of epistemic sentences has been repeatedly shown in reception studies involving adult participants (Sanders & Noordman, 2000;Traxler, Sanford et al., 1997). However, Verbrugge (2007) has contributed that the effect of epistemic markers is also prominent in production studies and extends to children.

Recall of coherence relations with(out) connectives
Although the differences between content and epistemic coherence relations have been shown on different levels (as we outlined above), no clear results could be reached as regards recall of these relations and the interaction with the presence of connectives (e.g., Cain & Nash, 2011;Degand & Sanders, 2002;Kamalski, Sanders, & Lentz 2008;Millis & Just, 1994;Sanders, Land, & Mulder, 2007). Some papers have studied the influence of coherence in a text on comprehension and recall of fragments of the text. However, a conclusive picture could not be reached. Martins, Kigiel, and Jhean-Larose (2006); Millis, Graesser, and Haberlandt (1993) and Sanders and Noordman (2000) have shown that adding connectives did not improve performance in tasks participants had to perform after reading. Martins et al. (2006) conducted a study involving the connective 'that's why'. They showed that adding this connective did not improve recall of the relations (i.e., the recall of textbase-related information). The connective did not improve performance in the after-reading task (Martins et al., 2006, p. 8). Neither did Millis et al. (1993) find a positive effect of adding this connective to the recall of encyclopaedic texts, in their study on the causal connective 'because'.
Similarly, Sanders and Noordman (2000) did not find a lasting effect of signalling coherence relations by means of causal connectives. While the effect is often present during reading, it disappears afterwards. They claim that relational markers have an effect during online processing, but that the influence of connectives decreases afterwards (Sanders & Noordman, 2000, p. 37). Martins et al. (2006, p. 9) provide the following rationale for this effect: "One possible reason for this is that connective processing is made too quickly and so does not permit a positive effect on long term memory. It is perhaps possible to enhance this type of processing by inviting readers to consider more deeply the semantic causal meaning of the causal connectives".
All these studies seem to suggest that the effect of adding connectives to make a text more coherent is transient. It may help in order to better understand texts during reading, but the effect disappears quickly once the texts are not present anymore. Positive effects of connectives on long term memory could not be shown. With our study below, we take the edge of this finding and demonstrate that there is more to it than meets the eye: age, type of coherence relation and connective all play a role for the recall of fragments of texts. Because effects of connectives for the online reading process have been shown over and over again, we focus only on the recall of coherence relations with and without connectives in order to reach more transparency concerning this topic. We aim to show that there is an important developmental perspective involved in the contribution of connectives to the comprehension of content and epistemic relations. Moreover, the manipulation of the type of recall task should not be underestimated.

Hypotheses
In the experiment we investigate the comprehension of content and epistemic relations in relation to the use of connectives by means of recall questions. We test whether content relations stick better in memory than epistemic relations.
Are participants clearly better at answering recall questions about content relations than about epistemic relations? Do children and adults read these relations in a similar way, or do children encounter more difficulties with epistemic relations than adults? Is there a beneficial effect of connectives and does it interact with the other variables? We will discuss our findings in the light of the linguistic skills required to understand and produce relations in the content and the epistemic domain.
Several authors have addressed the issue that free recall might not be sensitive enough to find effects of marking local coherence relations (Kamalski, 2007;Sanders, 2001). In order to overcome this obstacle, we asked the participants specific questions about text fragments after they had read texts (see appendix).
On the basis of the relevant literature we reach the following hypotheses: 1. Content relations will be easier to recall than epistemic relations. Precisely because epistemic relations are more complex, we think people will encounter more difficulties when they have to recall epistemic relations than the straightforward content relations. 2. Relations with a connector will be easier to recall than relations without a connector. 3. We hypothesise an age effect: adult participants will recall more relations than younger participants simply because they have better working memory capacities. Moreover, we think that the variable age will interact with the presence of connectives and the type of relation. 4. We hypothesise there will be an interaction between presence/ absence of connective and type of relation. Because epistemic relations are more complex, we think participants reading epistemic relations will benefit more in recall questions when a connective was present in the text that explicated the relation.

Participants
In total 539 children and adults participated in the experiment: 169 9-10 year olds; 192 11-12 year olds; and 178 17-18 year olds. These children and students read a text comprising content and epistemic relations, either with or without connectors. Participants were asked to read texts comprising content and epistemic relations and answer comprehension questions about these relations after having read the texts.

Materials, procedure and design
The participants read the texts in Dutch, their mother tongue. All the relations studied involved the connective 'so' (dus). Every participant was given four short stories to read. These stories were based on existing children's stories (see appendix for an example), but were adapted in order to fit the standards of the experiment. Participants were instructed to read the text and explained that they would receive comprehension question afterwards. Before starting with the actual experiment, the participants were given a story and questions in order to get used to the procedure. Participants read the text. After that, the text was taken away and they had to fill in four questions per story on a piece of paper. The same stories were given to all four groups, however two groups were given the stories with connectives (group 1 and 2), while the two other groups were given the stories without connectives (group 3 and 4). Recall questions were manipulated between subjects. The first and third group were asked recall questions about the content relations, the second and fourth group about the epistemic relations. Both content and epistemic relations were present in all texts.
Four recall questions were asked per story, so participants had to answer 16 questions in total. The analysis was run on the number of correct answers participants gave out of the possible 16.

Results
An analysis of variance was conducted with age, type of relation and presence of connective as between subjects variables. There was a main effect of connective; relations with connective (mean = 12.93) were easier to recall than relations without connective (mean = 11.57) (F(1, 527) = 62.6; p < .0001).
As expected, there was a main effect of age with the youngest children (mean = 10.20) doing worse than the older children (mean = 12.34), who did in turn worse than the adults (mean = 14.21) (F(2, 527) = 176.17; p < .0001).
There was an interaction effect between age and coherence relation (F(2, 527) = 10.98; p < .0001), with the interesting finding that the difficulty associated with epistemic relations (compared to content relations) decreases as participants get older (see Figure 1).
Finally, there was an interaction effect between connective and coherence relation (F(1, 527) = 7.35; p < .01) to the extent that connectives play a larger role in epistemic relations compared to content relations (see Figure 2).

Figure 1
Interaction between type of coherence relation and age

Type of coherence relation
The experiment revealed that content relations are better recalled than epistemic relations for the younger age groups. Content relations are easier to recall because there is a one-to-one correspondence with reality. Epistemic relations are harder because they require extra steps. They are no description of reality but present a form of thinking about reality. The experiment has shown that content relations are easier to recall than epistemic relations for children. It is the first time that this has been unequivocally shown. Sanders and Noordman (2000) found a difference between the recall of problem-solution coherence relations and list coherence relations. They found this effect in a free recall task. It seems not so surprising that relations that are joined by a causal relation are better recalled than simply lists of things. However, our study has for the first time shown that two groups of coherence relations can be successfully distinguished as regards recall. Moreover, this had been done by a task consisting of precise questions concerning the coherence relations. This means that the possible interfering factor of participants only recalling what they thought relevant (which is often the case in a free recall task) was ruled out in our task.

Age effect
Children encounter more difficulties with epistemic relations than adults. Interestingly, the difficulty associated with epistemic relations (compared to content relations) decreases as participants get older. This can be interpreted as in line with the study of Martins et al. (2006) who showed that experts performed better than novices because they possess more general knowledge about events etc. Needless to say, the older children become, the more they are aware of the general workings of the world and the more they are used to inferring information from texts (see also Denhière & Baudet, 1992). On the basis of this recall study we have been able to show that content relations appear to be easier to understand and recall for younger children. As children grow older, there is a gradual increase in the recall of epistemic relations. Apparently, it takes some time before children have fully acquired the epistemic relations. These data from language acquisition converge with data from language processing and diachronic linguistic research. The fact that the content meaning seems to be the one that is acquired first, is in line with results from other sources: content readings are more syntactically entrenched (Dancygier, 1998;Haegeman, 1991;Haegeman, 2003;Verbrugge & Smessaert, 2011), more easily processed (Noordman & de Blijzer, 2000) and also diachronically the central meaning from which the other classes are meta-phorical extensions (cf evidence such as Hopper & Traugott, 1993/2003Traugott, 1989).

Presence or absence of connective: implicit versus explicit
Relations with connective were easier to recall than relations without connective. The results obtained in the recall task run counter to the existing (cf Martins et al., 2006;Millis et al., 1993;Sanders & Noordman, 2000) literature that states that connectives only influence the reading process itself, but not the recall of the relations afterwards. Interestingly, connectives were shown to play a larger role for participants to be able to recall epistemic compared to content relations. Precisely because these epistemic relations are more complex, connectives carry a greater beneficial effect in the recall of these relations. While literature was in doubt about recall, we have been able to show that connectives play a vital role for better remembering coherence relations.
The claim that the positive effects of connectives are only transient, only playing a role during the reading of the texts, does not hold.

General discussion
In this study, we have focused on content and epistemic coherence relations involving the connective 'so' (dus) and on the developmental aspect of the acquisition of content and epistemic relations. We have studied the interaction with connectives, something which has not been done in the past. This led to surprising results. While coherence of a text is a crucial factor for a proper understanding, previous studies were not able to draw a conclusive picture of the (un)beneficial aspect of connectives for the recall of information in texts.
By means of the recall task we have been able to show that connectors play a larger role for the recall of epistemic relations compared to content relations and that younger participants encounter more difficulties with epistemic relations than adults. We did not make use of free recall, because in that case there is no control whether the participant could not recall a particular part of a sentence or simply did not think it relevant. By asking specific questions we were able to compare all conditions.
The results of the recall study are in line with previous research involving reading times (Noordman & de Blijzer, 2000;Traxler, Bybee, & Pickering, 1997). Epistemic relations require longer reading times than content relations, showing that they need more time in order to be properly processed. This complexity of epistemic relations was found in the recall study in that they linger worse in memory than content relations do. A novel finding was that connectives play a more important role for the recall of epistemic than content relations. Apparently, adding connectives to epistemic relations makes them easier to understand and recall. There was also an effect for content relations, but smaller, showing that content relations are often understandable enough without further explicitation or elucidation.
In conclusion, we can say that this study has shown that the precise type of relation that is expressed is an important factor for its ease of recall. Moreover, this variable was shown to interact with the age of the participants and the presence or absence of a connective. Specific small additions in language can have a huge effect for memory. This way, this study fits into the tradition in reasoning that has developed over the last thirty years showing that content and coherence effects play a huge role in reasoning with and understanding of sentences and texts.