Causality, subjectivity and mental spaces: Insights from on-line discourse processing

1 Introduction

The notion of Mental Spaces (Fauconnier 1994; Sweetser and Fauconnier 1996) has shown to be a fruitful cognitive linguistic contribution to both linguistically and cognitively inspired studies of discourse representation. Mental space configurations are taken to be dynamic and to change from one moment in discourse to the next. During discourse interpretation, various linguistic expressions set up new spaces, reactivate existing spaces and add to already active spaces. These cognitive operations have been used in cognitive linguistic theory to provide explanations for the actual cognitive operations. For instance, the meaning and use of various types of discourse connectives has been explained in terms of mental spaces. Contrastive connectives have been analyzed as elements that block inferences (such as but) (Fauconnier 1994; Spooren 1989; Verhagen 2005), conditional constructions like if … then, can function as space-builders that typically establish new mental spaces (Dancygier and Sweetser 2005), and causal connectives can set up epistemic spaces which represent reasoning of a speaker or author (Sanders et al. 2009, 2012; Verhagen 2005). Here, we focus on causal connectives and how they drive cognitive operations. We set out to establish falsifiable claims about how a Mental Spaces account would be reflected in language processing. We investigate whether different cognitive operations that are proposed in Mental Spaces Theory can be observed in online language processing.

We use precise psycholinguistic experiments to investigate the role of causal connectives during on-line discourse processing. This is done against the background of an increasing number of studies showing the on-line function of connectives. Connectives are often regarded as processing instructions, providing information on the coherence relation that connects segments. Eye-tracking and reading time studies have shown how the integration of the segments is sped up in the presence of causal connectives (Cozijn et al. 2011; van Silfhout et al. 2014). Still, this general function of ‘integration’ does hardly do justice to the many systematic differences linguists have identified in the connective lexicons of various languages (Stukker and Sanders 2012; Li et al. 2013). In the case of causal connectives, there is a vast amount of theoretical and corpus work showing that connectives vary in the amount of subjectivity that is expressed. Eye-tracking studies indicate that language users use this information immediately when processing a causal relation (Canestrelli et al. 2013; Wei et al. 2019). Objective and subjective connectives provide different processing instructions, causing readers to slow down when encountering a subjective connective. In the present study we delve deeper in the exact processing instructions these connectives give and how these instructions might translate into different cognitive operations. In two eye-tracking studies, we compare processing times of Dutch forward causal relations marked with the subjective connective dus ‘so’ or the objective connective daardoor ‘as a result’. Below, we first discuss the differences between subjective and objective relations and connectives. Then, we relate these differences to Mental Spaces Theory to develop a dynamic account of discourse processing which is tested in two experimental studies.

2 Causality and subjectivity in discourse

According to the subjectivity account, causal relations differ in subjectivity (Pander Maat and Sanders 2001; Pit 2003; Sanders et al. 2009). Objective relations like the Consequence-Cause relation in (1) are observable in the real world. Subjective relations like the epistemic Claim-Argument relation in (2) are not: they are created by some conscious mind, the Subject of Consciousness (SoC) (Pander Maat and Sanders 2001). The SoC is responsible for making the connection between P and Q.^[1]

The SoC can be the speaker,^[2] but a speaker may also introduce a different SoC as the person responsible for the relation. The level of subjectivity entrenched in the relation is defined as the degree to which the ‘speaker’ is responsible for connecting the two propositions (‘speaker-involvement’). A sequence is called maximally objective when there is no SoC at all. These relations are observable in the real world. They are not created by some conscious mind (1).

A sequence is maximally subjective when the distance between the SoC and the speaker is nonexistent (speaker = SoC). The relation is not observable but is constructed inside the speaker’s mind (2).

The situation is different when the speaker introduces another person as the SoC for the relation (SoC ≠ speaker). In (3), it is John who is reasoning that the neighbors are not at home based on the fact that their car is not in the driveway. Since John is now the SoC of the relation – and not the speaker – the distance between speaker and SoC is larger and therefore the relation is viewed as less subjective than (2). The speaker is reporting the subjective causal relation that is created by the SoC John.

A second aspect of subjectivity concerns the explicit mentioning of the SoC, as in (3). In such cases, the SoC is ‘put on stage’ (Langacker, 1991) and the subjectivity of the relation is more transparent. In contrast, when the SoC is implicit – as in (2) –the interpreter has to infer the involvement of the SoC. Then, subjectivity is not visible: the SoC is left ‘off stage’.

In English, backward objective and subjective causality can be expressed by the same connective: because. Interestingly, several languages have causal connectives that are prototypically used to express either objective or subjective relations. Corpus studies have shown that in Dutch (Sanders and Spooren 2015; Verhagen 2005), French (Zufferey 2012), German (Stukker and Sanders 2012), and Mandarin Chinese (Li et al. 2013), language users have systematic preferences for one connective over another. These systematic preferences vary with the subjectivity of the relation. In Dutch, connective use ‘cuts up’ the domain of causal relations: Example (4) – the Dutch version of (1) – would typically be expressed by doordat ‘as a result of’ a connective which specializes in non-volitional objective relations. Example (5) – the Dutch version of (2) – on the other hand can only be expressed by want, a connective that prefers subjective relations.

A third common connective, omdat ‘because’, typically expresses the reason for an intentional action, as in (6).

Although there is no one-to-one relation between these relations and the connectives, these prototypical uses of the Dutch connectives have been shown to be robust (Degand and Pander Maat 2003; Pit 2003; Sanders and Spooren 2009, 2015; Sanders et al. 2009). A similar mechanism accounts for forward causals, in which the direction of the causality goes from cause to consequence (e.g., “The neighbours’ car is not in the driveway so they are not at home”) (see Table 1; Pander Maat and Sanders 2001; Stukker et al. 2009). In this paper, we investigate the cognitive representations of these systematic differences.

3 Mental representations of causal relations

According to the subjectivity account, examples (1), (2), and (3) are intrinsically different. These differences can be made apparent by visualizing the mental representations of each of these relations. Figures 1 and 2 are representations of objective and subjective causal relations; the figures are derived from the Basic Communicative Spaces Network (Sanders et al. 2009, 2012). Every causal relation originates from the ‘knowledge base’, which represents a language user’s “encyclopedic knowledge, pragmatic knowledge and human reasoning, as well as the lexicon of the language that is used to express the causal relations” (Sanders et al. 2009: 28).^[3] It licenses the utterance of a relation.

Figure 1:

Mental space configuration of an objective causal relation (e.g., The temperature rose, because the sun was shining).

Figure 2:

Mental space configuration of an implicit subjective causal relation (e.g., The neighbours were not at home, because the lights were off).

From the knowledge base, the relation is projected into the ‘linguistic base’, which represents the linguistic realization of the causal relation. Since objective relations are simply reported by the speaker, there is a direct connection from the knowledge base to the linguistic realization (the solid arrow in Figure 1). The speaker does not intervene, which is represented by the fact that the arrow does not pass through the SoC (the smiley face). The linguistic realization is presented in the linguistic base space and not in the SoC-space which holds the thoughts and beliefs of the SoC.^[4]

In subjective relations, the SoC is involved and in Figure 2 the SoC is the speaker. Rather than a direct representation of the relation between P and Q as present in the knowledge base, the linguistic realization is an interpretation of this relation (the lowercase letters). The relation passes through the SoC and comes out slightly altered (the dashed arrow). It must be interpreted as the belief of the SoC and therefore it must be placed inside the SoC-space. Since the SoC is not explicitly mentioned in this example, the SoC-space is represented with dashed lines, indicating that it must be inferred.

When the SoC is explicitly mentioned, no such inference is necessary. The subjectivity of the relation is already marked in such cases. SoC-phrases^[5] like John thought in (3), instruct the reader to set up an SoC-space. The influence of such a space builder on the mental space configuration of subjective relations is visualized in Figure 3. Due to the SoC-phrase, the involvement of the SoC is now apparent and does not have to be inferred: the smiley face and SoC-space are now represented by solid lines. In addition, believes that is capitalized to indicate that this is also linguistically realized. Although the realized relation is still a construal of the SoC (depicted by the dashed arrow), this is now apparent from the start.

Figure 3:

Mental space configuration of an explicit subjective causal relation (e.g., John thought the neighbours were not at home because the lights were off).

4 Constructing mental representations on-line

How are such representations constructed on-line? Traxler et al. (1997a, 1997b) found that subjective (in their terms: diagnostic) causal relations are processed slower than objective relations. To be precise: readers slow down at the point where an objective interpretation is no longer tenable. In case of the consequence-cause relation in (7) and the claim-argument relation in (8), readers slow down at stalled in (8) compared to (7). At this point, readers “make the inference that the assertion is about a possibility or a belief rather than about some straightforward state of the world” (Traxler et al. 1997b: 89). In other words: readers revise their mental representation from an objective to a subjective relation (Figure 1 → Figure 2). Readers must open a space for the SoC and assign the relation to this space. In (8), readers must do this without an explicit signal in the text.

If these cognitive operations are in fact responsible for the observed processing asymmetry, the extra processing time that is found should diminish when a subjective mental space configuration is already in place. If the SoC-space is already set up before encountering disambiguating propositional content like stalled, then the causal relation can immediately be interpreted within the SoC-space. Consequently, the processing time should diminish.

Traxler et al. (1997b) compared explicit subjective relations with implicit subjective and objective relations. They inserted SoC-phrases into the S₁ of the subjective relations (9). As a result, the subjectivity was already visible before the reader encountered the critical region stalled and no revision should be necessary (Figure 3). The insertion had the predicted effect: reading times for explicit subjective relations did not differ from objective relations.

A similar effect was found for Dutch by Canestrelli et al. (2013). They used Dutch translations of the materials of Traxler and colleagues but changed one critical aspect. Traxler and colleagues used the connective because in both subjective and objective relations (see [1] and [2]), but Canestrelli et al. (2013) used the prototypically objective connective omdat (10) and the prototypically subjective connective want (11). Unlike because, these connectives encode information on the subjectivity of the relation.

Canestrelli et al. (2013) showed that readers use all this information directly on-line. In line with Traxler et al. (1997a), they found a delay in reading times for subjective relations compared to objective relations. However, this delay transpired immediately after the connective. At this point, the readers did not have enough semantic information to determine the type of causal relation. The delay was, therefore, caused by the connective and not by the propositional content. The moment readers encountered the subjective connective want, they slowed down. They did not have to wait for disambiguating content to revise their mental representation. Thus, the processing asymmetry showed up earlier, directly following the connective.

To test whether their effect was based on the same principles as the effect found earlier, Canestrelli et al. also included explicit subjective relations in their experiment (12). For Dutch, the processing asymmetry also disappeared when an SoC-phrase was added to the beginning of the sequence. The SoC-phrase signals the presence of an SoC and that the upcoming information is part of the SoC’s beliefs. The subjective connective want now confirms the subjectivity of the information, rather than introducing the subjectivity of the relation (cf. [11]). At the point of the connective, the SoC-space is already open and the subjectivity of the relation is known.

In another experiment, Canestrelli showed that processing asymmetries can also be found in forward relations like (13) and (14) (Canestrelli 2013). Like in backward relations, readers slowed down after reading a subjective connective (dus).

Thus, when it comes to introducing an SoC, subjective connectives and SoC-phrases have the same function. Dutch subjective connectives signal the presence of an SoC and cue the reader to open an SoC-space, just like an SoC-phrase does. Finally, in an experiment employing the visual world paradigm, Wei et al. (2019) have recently shown how subjective connectives make people focus their attention on a responsible person who is the source of the information.

5 The cognitive processes involved in building the mental spaces representation

The results of the processing studies outlined above show that additional processing is required in the interpretation of subjective causal relations. When it becomes apparent that a relation is the belief of an SoC – rather than an observable truth – readers need to revise their mental representation. The presence of an SoC can be cued: before encountering the relation, by a subjective connective or by content. According to Mental Spaces Theory, readers should open an SoC-space representing the relation as part of the thoughts and beliefs of the SoC. However, it is still unclear which specific cognitive operations contribute to processing costs: is it just the opening of an SoC-space? If so, having an open SoC-space should eliminate all extra processing costs. Or does assigning a relation to an open space also contribute to the processing costs? If so, having an open SoC-space will eliminate some but not all extra processing costs.

In the backward causal relations used in previous processing studies, opening and assigning coincide in time. In (11), there is no signal to evoke an SoC-space before encountering the connective want and therefore no SoC-space is open at that point in time. When reading want, the connective signals the reader to open an SoC-space. In addition, it shows that the previous segment was in fact a belief (a claim). At that point the reader must open an SoC-space and revise the previous representation by assigning the S₁ to the just opened space. Thus, opening and assigning happen at the same moment in time. When we use an SoC-phrase as in (12), the SoC-phrase signals the reader to open an SoC-space and the S₁ is assigned to that space immediately. Opening the space and assigning the claim now happen during the processing of S₁, instead of during the processing of the connective. Now, when the reader encounters want, the SoC-space has already been opened and the claim is already assigned to it. Hence there is no extra processing cost.

To see which operations add to the processing cost of subjective relations, we must separate these operations in time. An SoC-space must be evoked well before encountering the claim. Canestrelli (2013) took the first step by comparing forward causal relations (see Section 4). The difference between forward and backward configurations is how the relation is mapped onto the S₁ and S₂ (Sanders et al. 2012): for a forward subjective relation, the linguistic realization is not “q because p” but “p so q”. As with backward relations, the reader must infer the presence of a speaker as the SoC and evoke an SoC-space when processing a subjective relation. However, since the presentation order is different, the time course for on-line processing is also different. In forward relations, the claim is presented in the second segment of the relation; it follows the connective. As such, it should be possible to evoke an SoC-space before encountering the claim. By doing so, opening an SoC-space and assigning the claim are separated in time.

The present study uses the different time course of forward causal relations to separate the operations of opening an SoC-space from assigning information to an SoC-space. In two eye-tracking experiments, we will test which operations are costly and which come ‘for free’.

6 Experiment 1: Processing subjectivity in forward causal relations

Based on previous research and theoretical considerations, we expect different processing times when forward causal relations are marked with either a subjective or objective connective. For the four situations outlined below, we predict different time courses.

• Situation A: It is raining. The streets are wet. {implicit relation}

• Situation B: It is raining. As a result, the streets are wet. {objective connective}

• Situation C: It is raining. So, the streets are wet. {subjective connective}

• Situation D: I believe it is raining. So, the streets are wet. {SoC-phrase + subjective connective}

In Situation A, the relation is implicit. No cues are given with regards to the coherence relation between the two segments. When a connective is present (Situation B, C and D), we expect the connective to facilitate integration of the upcoming segment. As a result, the words following the connective (i.e., the streets) are processed faster in Situation B, C and D compared to A. In Situation B, the objective connective signals an objective relation. No SoC-space needs to be opened and there is no claim to assign to an SoC-space. In Situation C, the subjective connective signals a subjective relation and that a claim is following. It signals the presence of an SoC, which means the reader must open an SoC-space and assign the claim to it. Relative to Situation B, this will cause a delay: the words following the connective will be processed slower in Situation C compared to Situation B. In Situation D, the SoC-phrase “I believe” signals the presence of an SoC and at this point, the reader must open an SoC-phrase. Therefore, when the subjective connective is encountered, the SoC-space is already opened. As a result, the processing asymmetry between subjective and objective relations will disappear, if having an open SoC-space is enough to eliminate all processing difficulty, or diminish, if assigning a claim to an SoC-space also adds to the processing difficulty.

We summarize these hypotheses as follows:

The hypotheses were tested in an eye-tracking experiment.^[6] Processing times of Dutch forward causal relations marked with subjective and objective connectives were compared with each other and with an implicit version of the relation that functioned as a baseline.

6.3 Discussion

The results of Experiment 1 support H₁ and H₂. The subjective connective dus led to longer reading times than the objective connective daardoor. In addition, both conditions facilitated processing of the information directly following the connective as compared to the implicit condition.

The results do not support H_3a or H_3b. We hypothesized that an SoC-phrase in the S₁ of a forward causal relation would facilitate processing of subjective relations with dus since the SoC-phrase opens an SoC-space to which the relation can be assigned. This prediction was not borne out. On the contrary, the reading times for this condition were longer than in the condition without the SoC-phrase. Furthermore, the SoC-phrases eliminated the facilitating effect of the connective dus compared to the implicit relation. These results suggest that the SoC-phrases impeded instead of facilitated processing.

One possible explanation of this result is that SoC-phrases tend to have narrow scope (only over S₁). This may have caused readers to interpret S₁ as a claim, rather than a descriptive argument/fact. If so, they would expect the S₂ to contain an argument in support of the claim. However, the forward connective dus signals that the S₁ was an argument and the S₂ will contain the claim. Hence, the claim in S₂ may have come as a surprise. This could explain why reading times in the condition with an SoC-phrase were actually longer than in the condition without an SoC-phrase.

Another possibility is that the SoC-phrase did not evoke the right mental space configuration. While the configuration of backward subjective relations marked with an SoC-phrase resembles Figure 3, their forward counterparts may have evoked a configuration like Figure 4. The involvement of the speaker is partially visible, but not completely: for the realization of ‘P’, the involvement is visible. The SoC is responsible for the observation of ‘P’, which is a representation of a cause observable in the real world. In contrast, the involvement of the SoC in the construction of the causal relation is still concealed. As a result, the SoC-space still has to be inferred (as in implicit subjective relations). Moreover, Figure 4 is also different from the representation of an implicit subjective relation (see Figure 2), which may indicate why the dus-relations with SoC-phrases were processed even slower than implicit relations. The way subjectivity was expressed, may in fact have complicated the process.

Figure 4:

Mental space configuration of a subjective forward causal relation marked by an SoC-phrase (e.g., I believe that the lights are off, so the neighbours are not at home).

Hence, the SoC-phrases used in Experiment 1 may have been unsuccessful in evoking a subjective mental space configuration necessary to interpret subjective relations. As a result, the readers may not have been prepared for a subjective relation, which would explain why the expected difference between the dus- and So C - phrase  +  dus-condition was not found. Therefore, the results of Experiment 1 are inconclusive.

The SoC-phrases may have evoked a configuration in which the SoC was presented as the observer responsible for the argument and not as a reasoning entity responsible for the causal relation. To solve this problem, we must make sure that the entire causal relation is interpreted within the scope of, i.e., ‘the belief’ of, an SoC. It must be clear to readers that the SoC’s thoughts are represented here, and not his observations. One convincing and very natural way to achieve this, is to use free indirect speech (FIS) as in (15) (e.g., Bal 1990; Banfield 1982; Chafe 1994; Fludernik 1993; van Krieken et al. 2016; 2017).

(15)

Gosh, he was tired.

Like in indirect speech – which was used in Experiment 1 – this style evokes a configuration that includes a space for an SoC embedded in the space of the narrator (Sanders and Redeker 1996). Crucially, however, these representations differ with regard to the allocation of responsibility to the narrator and characters in the discourse.

The subject is responsible for the content, but when it comes to the wording, responsibility is less clear. Rather than a complete deictic shift – to subject or narrator – the deictic center in FIS seems to blend between the two of them (‘dual voice theory’; Banfield 1982; Sanders 2010; van Krieken et al. 2017). The result is that the innermost workings of a subject’s mind can be portrayed. In longer text fragments, this causes the reader to experience a stream of consciousness which represents the subject’s thought flow (16). Therefore, FIS establishes an open SoC-space to which each new proposition can be assigned automatically.

(16)

No. No. No. This could not be happening. Not now, not ever! He had worked too damn hard to lose it all now.

7 Experiment 2: Processing subjective causality in free indirect style

The Free Indirect Style evokes a configuration in which an SoC is already opened before encountering a subjective connective. Using eye movement registration, processing times of Dutch forward causal relations are captured. Processing times of sequences marked with subjective and objective connectives are compared with each other in a FIS-context and a neutral context. The following hypotheses are tested:

As in Experiment 1, we expect that subjective connectives will cause a processing delay compared to objective connectives. We hypothesize that a FIS-context will facilitate processing of a subjective forward causal relation: an SoC-space is kept open, so that the content of S₂ can easily be assigned to it. Like in Experiment 1, we do not know whether opening a space is enough to alleviate all processing difficulty (H_5a). If the processing asymmetry does not disappear but only diminishes, this indicates that only opening a space is not enough and that assigning the utterance to this space also adds to the processing costs (H_5b).

8 General discussion and conclusion

Our goal was to gain more insight in subjective mental spaces structures and the construction of causality. Focusing on the processing differences of subjective versus objective causal relations, we hypothesized which processes underlie the construction of the corresponding mental representations and how different cognitive operations that are proposed in MST are reflected in language processing. We started from the finding that backward subjective causal relations, such as (17), are processed slower than objective causal relations. The second segment makes clear that the first segment does not describe an objective fact, but that it is a speaker’s claim for which the second segment is the argument. We hypothesized that this processing effect might be an indication of setting up a Mental Space for the speaker. This analysis is in line with experimental results showing that the processing delay for subjective relations disappears when it is clear from the beginning that they should be interpreted as subjective claims or conclusions (18) because they are part of the SoC-Mental Space: According to John … or In my opinion … (Canestrelli et al. 2013; Traxler et al. 1997a, 1997b).

What these studies did not reveal is whether it is only the operation of opening an SoC-space which requires extra processing time or whether other operations, like assigning a claim to such a space, are also costly operations. In order to answer this question, the operation of opening an SoC-space and that of assigning a proposition to it must be separated in time. While in backward relations these two necessarily coincide, in forward relations they do not. In (19), the claim-argument relation of (17) is reversed, to form an argument-claim relation. When the first segment of that relation is marked as subjective (20), it does not automatically imply that the second segment is also subjective. However, since the mental space has been opened, and (in Dutch) the connective dus ‘so’ at the start of the second segment marks it as subjective, processing the second segment as subjective may not require extra processing time. Thus, if only opening an SoC-space adds to the processing costs, the processing asymmetry should disappear when an SoC is introduced beforehand. If the asymmetry does not completely disappear, other processes must be at play.

Experiment 1 explored the possibility that having opened the mental space for the speaker means that new information can be assigned to this space without any extra processing cost. The results showed that having an open SoC-space to which the claim in the second segment can be assigned did not alleviate the processing difficulty of the subjective relation. The relations marked with the subjective connective dus led to longer reading times compared to the relations marked with the objective connective daardoor, irrespective of the presence of a phrase marking the first segment as a claim. Readers slowed down immediately after the connective. Thus, simply having an already opened SoC-space does not imply that subjective information will be ‘automatically’ assigned to that space. Contrary to expectation, having an SoC-space open did not diminish the processing asymmetry. We hypothesized that the factual nature of the argument to which the SoC-phrase was added might have interfered with the scope of the SoC-phrase. As a result, the SoC-phrase did not evoke a supportive mental space configuration and integration was not facilitated. In fact, it probably complicated matters by including factual observation and adding a perspective to the information.

For this reason, we conducted Experiment 2 in which we studied the processing of subjective relations in free indirect speech (FIS), which provides a natural opportunity to investigate our main questions, but has never been used to study these and related issues in a processing context – we consider this an innovative aspect of the current study. FIS represents the thoughts and experiences of a character as a continuous stream-of-consciousness and, once this stream is established, the narrative will be naturally interpreted as taking place within the mental space of this character. For that reason, this ‘subjective discourse’ should have an SoC-space open for interpretation, so that the entire causal relation can be seamlessly interpreted within the SoC-space. As such, it should diminish the processing delay found for subjective relations. The results of Experiment 2 confirmed this hypothesis. In a neutral context, subjective relations (again) took longer to process than objective relations. Crucially, this effect was diminished in subjective contexts created by free indirect speech. In FIS, it appeared easier for readers to interpret the causal relation within the mental space of the character. This finding is an important addition to earlier results; it implies that readers are indeed sensitive to this more subtle signal of the presence of a thinking subject, and that this contributes to the processing ease of inherently complex relations. In addition, the results show that complexity in language processing is less determined by number and connections between different spaces, but by having or not having a supportive configuration in place at the right time. Even seemingly complex mental space configurations, like FIS, can facilitate processing of language.

In contrast to earlier results (Canestrelli et al. 2013; Traxler et al. 1997b), the processing asymmetry did not completely disappear. This difference can be explained by the separation of opening an SoC-space from other operations, like assigning a proposition to that space. While these processes coincided in the earlier studies, they did not in our experiment. The SoC-space was opened and established well before the claim was encountered. It seems therefore that while opening an SoC-space is a costly procedure, assigning a proposition to an open space does not come for free and is associated with its own costs. In addition, reactivating an open SoC-space may also come at a cost.

Our results add to the findings of Wei et al. (2019), who showed that subjective connectives immediately activate the awareness of a responsible thinking person. The pattern that emerges is one in which readers constantly keep track of the Subject of Consciousness, the ‘mind’ that is responsible for the content of the text, updating their mental model according to the (sometimes subtle) signals in the text as to where the responsibility for the content lies. The current results are in line with the idea that readers keep track of a network of mental spaces that is constantly being updated as the discourse unfolds (Fauconnier 1994; Sanders et al. 2009, 2012; Sweetser and Fauconnier 1996; van Krieken et al. 2016). The results also illustrate that various cues influence this network and that some of them may be better space builders than others. Further research should reveal how mental space configurations are affected by various types of space builders, from local cues (such as causal connectives) to more global cues (such as the perspective of the text as a whole).

In general, these results provide further insights in the dynamics of discourse processing, as well as in the cognitive representation of authors, speakers and characters as providing mental spaces. The methodology of combining cognitive theory with precise processing experiments has shown to be fruitful, in that respect. In our view, it is worthwhile to use these and similar methods to make Cognitive Linguistic theories like MST more precise and put them to the test.