Temporal Coherence

Abstract

This chapter consists of four main sections and ends with a summary. First, it provides a general account of coherence relations as cognitive entities, and focus on the status of temporal relations. Second, in order to support the proposal that temporal relations are cognitively motivated, it addresses the results of a series of online and offline experiments carried out in order to test the processing and conscious evaluation of temporal relations, as expressed by the French Passé Composé versus the Passé Simple, and by the temporal connectives ensuite and puis. Third, the notion of cognitive temporal coherence is discussed, by arguing that the cohesive ties investigated in this research are cognitively motivated categories, and by exploring a psycholinguistic account of coherence according to which mental representations of discourse segments are structured and coherent.

6.1 Coherence Relations

For Hobbs (1979, 69), coherence is the ‘mortar with which extended discourse is constructed’. In other words, a discourse has structure, and this structure is provided by various types of relations that bind contiguous segments of a text, transforming that text into a whole. For example, the segments in (565), given by Hobbs (1985, 1), illustrate this postulate:

(565)	(a) I would like now to consider the so-called “innateness hypothesis”
	(b) to identify some elements in it that are or should be controversial,
	(c) and to sketch some of the problems that arise as we try to resolve the controversy.
	(d) Then, we may try to see what can be said about the nature and the exercise of the linguistic competence that has been acquired, along with some related matters. (Chomsky, Reflections on Language, p. 13)

Between segments (a) and (d) there is a temporal relation, which is overly marked by then , linking the two topics the author wants to discuss. Clauses (b) and (c) elaborate on the first topic by breaking it into two subtopics. The additive relation between (b) and (c) is overtly marked by and. In discourses, numerous other types of relations can be identified, such as causal or adversative ones, as in (566) and (567) respectively.

(566)	John fell because the floor was slippery.
(567)	John fell but he did not hurt himself.

Many scholars have pointed out that such relations exist, and have attempted to classify them. For example, Halliday and Hasan (1976) speak about conjunctive relations and classify them into four main categories (additive, temporal, causal and adversative); Longacre (1983) speaks about combination of predicates or types of paragraphs, and distinguishes four categories (conjoining, temporal, implication and alternation). Hobbs (1979, 1985) speaks about coherence relations, and gives formal definitions for a set of relations rooted in the operations of a computational inferential system — that is, the procedures that apply to the represented data. For Hobbs, one of the crucial questions to be answered on coherence is why is discourse coherent in the first place (Hobbs 1985, 69). The answer he proposes resembles a pragmatic model of verbal communication, such as that proposed by Gricean and post-Gricean pragmatics (cf. Sects. 2.2 and 2.3). According to Hobbs, the function of coherence relations should be linked to the speaker’s goal of communicating his ideas using the imperfect (and underdetermined) medium of language to a hearer who undergoes the comprehension process under certain processing constraints. The speaker aims to have the hearer understand him — that is, to identify his informative intention by drawing the right inferences and arriving at his intended meaning of the utterance or series of utterances. Therefore, the speaker seeks to ease the hearer’s processing load by implicitly or explicitly structuring his message — or, as it might be formulated in relevance-theoretic terms, by making use of procedural expressions which specify paths to follow during the comprehension process. Likewise, the hearer will try to free himself of the load of underdetermined language, in order to construct the speaker’s intended meaning contextually.

For other scholars, what helps language users to process a discourse is to connect discourse segments, by inferring coherence relations on the basis of cognitive principles. This is the proposal made by Sanders et al. (1992, 1993), who did the grounding for the Cognitive approach Coherence Relations (CCR). For them, an accurate classification of discourse relations must be descriptively adequate (it must cover various types of naturally occurring data) and psychologically plausible (it must be based on cognitively plausible principles).

A series of papers by Sanders and colleagues (among many others, Sanders et al. 1992, 1993; Sanders 1997, 2005; Sanders and Noordman 2000) developed the CCR framework, in which psychological plausibility is a central aspect of coherence relations. For Sanders et al. (1992, 3), coherence applies to mental representations which hearers build when they hear, process and understand a discourse. In their view, coherence relations should be considered not as discursive entities but rather as cognitive entities, as they write:

A discourse structure approach is not necessarily restricted to descriptive analyses of discourse, because coherence relations should be considered as cognitive entities. Such a claim leads to the prediction that coherence relations and their linguistic marking affect the cognitive representation of a discourse (i.e., discourse understanding).

This prediction has been confirmed by numerous online experimental studies, from Haberlandt (1982) right up to more recent studies (among many others, Cozijn et al. 2011; Canestrelli et al. 2013; Mak and Sanders 2013; Van Silfhout et al. 2014, 2015; Zufferey 2014; Zufferey and Gygax 2016). For example, Sanders and Noordman (2000) show through experimental work that coherence relations and their linguistic marking (explicit vs. implicit) affect text processing of expository texts, which are considered more complex and less stereotypical than narrative texts. They tested the hypotheses that, because coherence relations play a crucial role in text understanding different relations (e.g. causal vs. additive vs. contrastive) result in different mental representations , and overt marking of relations influences processing in this kind of texts. They expected overt marking to facilitate the online construction of the mental representation, because the marker makes the coherence relations between text segments explicit. Once the representation has been built, overt markers are not expected to influence later access to the representation, as in recall tasks, for example. They found that different coherence relations are processed differently. In particular, the problem-solution relation structure, which is a causal relation, was processed faster and verified faster and more accurately than the list relation, which is an additive coherence relation. Additionally, they found that overly marked relations lead to faster processing of the text segment immediately following. These results indicate that discourse relations and their explicit marking affect the processing of, building of and access to mental representations of the content given of in discourse segments.

In the CCR framework, coherence relations have two characteristics. Firstly, they satisfy the relational criterion, according to which a coherence relation refers to the informational surplus which it adds to the interpretation of the discourse segments in isolation (Sanders et al. 1992, 5). Additionally, because coherence relations connect mental representations of discourse segments, the meaning of the segments must be compatible with the discourse relation (be it implicit or overtly marked using a compatible connective). Secondly, coherence relations are classified following a taxonomy consisting of four primitive fine-grained features: basic operation (causal or additive); source of coherence (semantic or pragmatic); order of segments (basic or nonbasic); and polarity (positive or negative). So, the procedure of defining a coherence relation consists of (i) identifying two discourse segments S₁ and S₂ in expressing two propositions P and Q, (ii) determining whether P and Q are related by a causal or an additive relation, (iii) identifying the source of coherence as semantic (the propositional content of the segments) or pragmatic (the illocutionary content of one or both segments), (iv) detecting whether the order of P and Q is basic (S₁ → S₂) or nonbasic (S₂ → S₁), and (v) determining whether the relation is positive, in the sense that P or Q follow the basic order, or negative, in the sense that ⌐P or ⌐Q follow the basic order. Using this taxonomy, Sanders et al. (1992, 11) identify twelve prototypical coherence relations, of which eight are causal, and four additive, as in Table 6.1 (cf. the original paper for examples and extensive discussions of these relations).

Table 6.1 Overview of the taxonomy and prototypical relations

For Sanders et al., during the language comprehension process, the hearer checks the primitives of this taxonomy, and does or does not infer a certain prototypical relation as a result. Moreover, when cognitive relations are marked linguistically, they are identified and processed faster (Haberlandt 1982). This taxonomy predicts which relations are more likely to remain implicit, such as cause-consequence, or to be overtly marked, such as concession: the former follow the basic order and whereas the latter follow the nonbasic order. This prediction is based on the causality-by-default hypothesis (Sanders 2005), according to which hearers by default expect two segments in a discourse to be causally related. This hypothesis complements the relational principle mentioned above, which states that the propositional content of the segments must be compatible with the inferred relation. As such, the basic categories postulated in the CCR framework correspond to highly expected discourse relations , and are more likely to be expressed implicitly than non-basic ones.

Within the same framework, Hoek and Zufferey (2015), argue that the rate of implicitation (that is, the optionality of overt marking) of discourse relations is governed by both the listener’s expectations about discourse (Segal et al. 1991; Murray 1995, 1997; Kaiser and Trueswell 2004; Rohde et al. 2006; Ferretti et al. 2009) and cognitive complexity. Segal et al. (1991) speak about the continuity hypothesis, which postulates that readers by default expect a discourse segment to be both causally and temporally continuous with the preceding segment. In the CCR framework, relations with a positive polarity are continuous, and those with a negative polarity are discontinuous. Hoek and Zufferey (2015), like Asr and Demberg (2012) before them, found that unexpected and discontinuous relations are overtly marked more often than expected and continuous relations. Taken together, these two hypotheses seem to describe the reality for language users accurately. However, they raise the questions of the cognitive status of temporal relations. In the CCR framework, temporality was not considered as a basic categorizing principle for two reasons. Their first reason is that temporal meaning is too dependent on the referential content of the segments, and temporality cannot be ignored by language users whereas causality can. Their second reason is that it is not a categorizing principle as productive as causality and additivity. Since Sanders and colleagues rejected temporality as a categorizing principle for discourse relations , they did not consider it as a fundamental cognitive principle. Nevertheless, they believe that temporal relations belong to the class of additive relations. Temporal relations can be distinguished from other additive relations based on “the referential meaning of individual segments” (1992, 28). Research in psychology, and more recent annotation and processing studies, provide evidence that temporal relations do play a role in constructing mental representations of situations, and should therefore be considered as cognitive relations (cf. discussion in Sect. 6.2).

The cognitive processes that people use to infer coherence relations, and thus to establish coherence at both discursive and cognitive levels, have also been studied from a psycholinguistic perspective. Three principal models try to explain how hearers build mental models during comprehension, on the basis of whether pragmatic inferences are drawn (i) after the utterance has been processed in its entirety (Garnham and Oakhill 1985, 1996; Garnham et al. 1996), (ii) during the utterance being processed, when cues are integrated as they become available (Kintsch and van Dijk 1978; van Dijk and Kintsch 1983; Kintsch 1995, 2005), or (iii) before the utterance has been processed in its entirety, making use of expectations about upcoming discourse (Kaiser and Trueswell 2004; Rohde et al. 2006; Ferretti et al. 2009; Rohde and Horton 2014).

The last two models both make proposals that have been validated experimentally. One the one hand, Kintsch and colleagues’ model of integrating cues (also called the mental or situation model ) states that speakers build simple and multithreaded mental representations of situations described in a discourse. A crucial property of these mental representations is that they are structured and coherent (Gernsbacher and Givón 195; Graesser et al. 1997). In this model, language is seen as encoding processing instructions on how to construct mental representations of the situations described (Zwaan and Radvansky 1998). (I will come back to this model in Sect. 6.4.2, in which I will speak about cognitive temporal coherence.) On the other hand, the defendants of the anticipatory model of language processing have shown that the establishment of coherence relations is sensitive to a variety of linguistic cues, including connectives (Knott and Dale 1994; Prasad et al. 2008; Asr and Demberg 2012; Koornneef and Sanders 2013; Mak and Sanders 2013), verb class and verb aspect (Koornneef and Van Berkum 2006; Kehler et al. 2008; Ferretti et al. 2009), coreference (Kehler and Rohde 2013), and the preceding coherence relation (Simner and Pickering 2005), among others.

To conclude, studies indicate that several factors should be considered when investigating coherence relations, such as the causality-by-default hypothesis, the continuity hypothesis, the hearers’ expectations during language comprehension and, more generally, the cognitive complexity of coherence relations.

6.2 The Cognitive Status of Temporal Relations

Temporal relations come in several subtypes. They can be classified as sequential chronological, where the order of eventualities in the discourse corresponds to their chronological order as in (568), (569), (570) and (571), sequential anti-chronological, where the order of eventualities does not correspond to their chronological order as in (572) and (573), and synchronous, where the two eventualities occur simultaneously as in (574).

(568)	Mary arrived home before her husband called her.
(569)	Mary arrived home, then her husband called her.
(570)	Mary arrived home. Ø Her husband called her.
(571)	Mary had arrived home when her husband called her.
(572)	Mary arrived home after her husband called her.
(573)	Mary arrived home. Ø Her husband had called her.
(574)	Mary was just entering the house when her husband called her.

These examples illustrate three important aspects of temporal relations. The first is the fact that they can be expressed with or without a temporal connective , and can thus be explicit, as in (568), (569), (571) and (574), or implicit, as in (570) and (572). The second is that there is a lack of one-to-one correspondence between temporal connectives and temporal relations, as seen with the connective when, which can be used as a sequential anti-chronological relation, as in (571), or a synchronous relation, as in (574). The third is that verbal tenses play an important role in determining the temporal relation. For instance, the past perfect in (571) directs the interpretation towards a sequential relation, whereas the past continuous in (574) orients the interpretation towards a synchronous relation; both verbal tenses are compatible with the temporal connective when.

As I noted in Sect. 6.1, for Sanders et al. (1992), temporal relations say were not afforded the status of cognitive temporal relations, firstly because temporal relations do not seem to correspond to the relational principle, and secondly because temporality might not be a productive principle with respect to the taxonomy, like causality and additivity. However, Evers-Vermeul et al. (2017) argue that temporality, which is one of the prominent features determining coherence in a discourse, is a relational rather than segment-specific notion, and is cognitively plausible. In their study, they show that temporal relations meet the requirements of Sanders et al.’s (1992) taxonomy, and they adapt the annotation scheme used within this framework to account for temporal relations in addition. Evers-Vermeul et al. (2017, section 3) provide three types of evidence to determine whether temporal relations are distinct cognitive entities (compared to causal and additive relations): (i) evidence from the linguistic system; (ii) evidence from language acquisition; and (iii) evidence from language processing. Firstly, as I have shown in examples (568)-(574), there are linguistic markers that may be used to mark temporal relations overtly. As noted by Evers-Vermeul et al., citing previous work by Knott and Dale (1994) and by Stukker and Sanders (2012), linguistic markers are pointers to cognitive processes, such as the actual use of coherence relations to construct mental representations . Knott and Dale (1994) gather a corpus of around 200 relational cue phrases from academic articles and books. For example, relational cues to locate eventualities in time and to express temporal relations are given in Fig. 6.1.

Fig. 6.1

Temporal situation. (Knott and Dale 1994, 60)

As observed by Evers-Vermeul et al., this list of temporal relational cues exhibits two characteristics: the underspecification of these markers (that is, the connective is used to convey a relation that does not fully correspond to its encoded meaning); and their polysemy (that is, a connective can be used to express more than one coherence relation). For Evers-Vermeul et al., these characteristics do not provide evidence against the idea that temporal relations have specific linguistic markers, mainly because this is the case for markers generally considered contrastive, such as but, or additive markers, such as and. As the relevance-theoretic framework argues, underdetermination is a characteristic of language in general (cf. Sect. 2.3), and is not therefore a problem specific to temporal connectives .

Besides the linguistic markers identified by Knott and Dale, there are also the temporal categories of Tense , Aspect and Aktionsart. As I discussed in Sect. 1.1, these categories are generally considered as referential, and thus as playing a role in the location of eventualities in time. Several researchers (Reichenbach 1947; Kamp and Rohrer 1983; Partee 1973, 1984; Dowty 1982, 1986; Hinrichs 1986; Nerbonne 1986; Webber 1988; Moens and Steedman 1988) have sought to explain the temporal relations triggered by Tense by considering this category as anaphoric. The main idea is that the interpretation of temporal progression, corresponding to chronological sequential relations, is linked to the introduction of a new point of reference R, either by Tense (Reichenbach 1947; Kamp and Rohrer 1983; Nerbonne 1986) or by two of the four aspectual classes referred to as Aktionsart, namely accomplishments and achievements (Hinrichs 1986; Dowty 1986). Conversely, the interpretation of the temporal overlap of two eventualities, corresponding to synchronous temporal relations, is linked to the use of a verbal tense which does not introduce a new R, to verbs expressed by the imperfect Aspect or/and to the use of unbounded aspectual classes, i.e. activities and states.

Additionally, Hinrichs (1986, 63), following Partee (1973), points to the richness of the relational nature of the temporal localization of events using temporal expressions. He identifies seven possible cases: tense morpheme-tense morpheme, as in (575); tense morpheme-temporal adverbial, as in (576); temporal adverbial-temporal adverbial, as in (577); temporal adverbial-temporal connective , as in (578); tense morpheme-temporal connective , as in (579); temporal connective-temporal connective, as in (580); and temporal connective-temporal adverbial, as in (581). In each case, the target temporal expressions are written in italics.

(575)	He took off his clothes, went into the bathroom, took a shower and went to bed.
(576)	They wheeled me into the operating room and put me under sedation. Three hours later I woke up.
(577)	This week I toured London. On Thursday I saw the Tower.
(578)	Last Saturday when the State Fair started, all hotel in town were booked.
(579)	They ordered two Italian salads and a bottle of Frascati. When the waiter brought the wine, they noticed that they had forgotten their checkbook.
(580)	When all the cars poured out of the parking lot after the concert was over, a big traffic jam developed.
(581)	When Melissa left the party, Bill left 5 min later.

The first and fifth cases, illustrated in (575) and (579), have received the greatest amount of attention in linguistic and pragmatic studies (Kamp and Rohrer 1983; Grice 1989; Levinson 2000; Moeschler et al. 1998; 2000a, b; de Saussure 2003; Molendijk et al. 2004; Verkuyl et al. 2004; Borillo et al. 2004; Grisot and Moeschler 2014; Grisot 2015; Grisot and Blochowiak 2015, 2017, among others). To provide a comprehensive understanding of these rich interrelations among temporal relations, each of these cases should be extensively investigated. In Grisot and Blochowiak (2015, 2017) (discussed below in Sect. 6.3), we assessed the role of verbal tenses (the French Passé Composé and Passé Simple ) at the same time as the role of temporal connectives (ensuite ‘then’ and puis ‘then’) as instructions for processing temporal relations. Two cases of temporal ordering were examined: first, the case when temporal ordering is undetermined (Grisot and Blochowiak 2015); and second, the case of chronological order (Grisot and Blochowiak 2017). We found that, for both cases of temporal ordering, overtly marked temporal relations demanded processing time similar to that of implicit relations. In addition to this, we did not find significant difference regarding the roles played by the Passé Composé and Passé Simple. However, offline data from acceptability experiments indicated that participants preferred the implicit versions to the explicit ones. Additionally, we found that participants preferred the occurrence of the Passé Composé with ensuite to puis . So, based on these observations, I follow Evers-Vermeul et al. (2017) in their proposal that temporal relations have specific linguistic markers which point to their cognitive status.

The second type of evidence provided by Evers-Vermeul et al. (2017) to support the proposal that temporal relations are cognitive entities comes from language acquisition. As they note, language acquisition studies have shown that children acquire temporal connectives after additives, such as and, and before causals, such as because (Bloom et al. 1980 for English , Evers-Vermeul and Sanders 2009 for Dutch). Studies have also found a discrepancy between production and full comprehension of temporal connectives: children use temporal connectives before they are able to comprehend them fully (Bever 1970; Blything et al. 2015). Additionally, differences have been observed with respect to the various subtypes of temporal relations, following a clearly identifiable developmental path (Clark 1971). During the first stage of acquisition of temporal connectives (around the age of three), children are not able correctly to interpret the temporal order of events provided by before and after. Instead, they use the order-of-mention strategy, according to which the event mentioned first is interpreted as the event that took place first. Around the age of four, children are able correctly to interpret chronological sequential relations marked by before, and around the age of five they interpret correctly anti-chronological sequential relations marked by after. This evidence from acquisition indicates that temporal relations occupy an important place in children’s language acquisition processes, that they are directly linked to the order of the acquisition of temporal connectives , and that children’s comprehension of temporal relations and temporal connectives is facilitated by a chronological order of events.

The third type of evidence given by Evers-Vermeul et al. (2017) comes from data on adults’ processing. Since the eighties, studies in psychology have shown that temporal information is encoded in readers’ mental representations of eventualities (Mandler 1986; Zwaan 1996; Townsend 1983; Gennari 2004; van der Meer et al. 2002). Unlike children, adults are able to interpret temporal relations correctly, regardless of order of the segments. However, chronological order seems to facilitate processing, whereas anti-chronological order places more cognitive load on the brain, because it requires additional discourse-level computations (Münte et al. 1998; Ye et al. 2012; Politzer-Ahles et al. 2017). For example, it has been shown that sequential chronological temporal relations are remembered better than anti-chronological ones (Clark and Clark 1968; Townsend 1983; Baker 1978; Zwaan et al. 2001). As noted by Evers-Vermeul et al. (2017, 9) these studies show that “the temporal order of clauses affects how well the relation is encoded in the mental representation”.

Scholars have also observed that temporal relations are processed differently from causal ones. For example, Mandler (1986) found that a chronological order of eventualities facilitates the processing of temporal relations, whereas it does not play an important role in causal relations. For Mandler, this difference is due to the fact that readers have prior knowledge about the relation between a cause and an effect, whereas for temporally linked eventualities, readers have to determine the relation in context . In contrast, other researchers have found similarities between causal and temporal relations. For example, expectations that people have about relations holding between discourse segments when reading a text arise with respect to both causal and sequential temporal relations: Segal et al. (1991) and Murray (1997) argue that readers expect a sentence to be causally (canonical order) and temporally (sequential chronological) linked to its preceding context.

It might be possible that these expectations are particularly strong when comprehenders deal with narrative texts (stories or literary stories). Zwaan et al. (1995a, b) proposed an event indexing model to account for the reader’s construction of a multithreaded situation model while reading simple stories and literary short texts. According to this model, the reader of a narrative text expects a great degree of continuity with respect to five conceptual dimensions: the protagonist, temporality, causality, spatiality and intentionality (that is, the characters’ goals). This means that temporal discontinuity (that is, when the incoming event occurs much later in time, or earlier in time as in flashbacks) is less expected by the reader. There is experimental evidence that temporal discontinuities, such as anti-chronological sequential relations, impede comprehension and slow down reading times (Mandler 1986). One of the strategies writers use is overtly marking temporal discontinuities using temporal locating adverbials (such as back at the ranch, the previous summer or the next morning), temporal connectives , and the verbal categories Tense, Aspect and Aktionsart (Graesser et al. 1997).

6.3 Experimental Study on Processing Implicit and Explicit Sequential Relations

6.3.1 “Ensuite” and “Puis ” as Temporal Connectives

Defining temporal connectives has proven to be a rather difficult task in the literature; consequently, there is no agreement with respect to which linguistic markers should be included in this category. For example, Gosselin (2007) proposed that a temporal linguistic marker could be included in the category of temporal connectives when it conveys a specific temporal relation with the previous sentence(s). According to Gosselin, French markers such as et ‘and’, puis ‘then ’, alors ‘then, so’, ensuite ‘then’, après ‘after’, plus tard ‘later’, aussitôt ‘as soon as’ and dès cet instant ‘from this moment’ should be grouped under the label of temporal connectives. Nevertheless, as I will discuss below, this description is not generally accepted in classical grammars, such as Le Grand Robert de la langue française (Robert 2016) or Grevisse’s Le bon usage (Grevisse 2016), nor by a series of other studies.

In Robert (2016), the basic usage of ensuite is to express temporal succession, as in (582), where corresponds to English afterwards, and in (583), where corresponds to both then and afterwards. In addition to temporal succession, ensuite can express spatial succession, as in (584).

(582)	On appelle aîné le premier enfant, puîné celui qui naît ensuite. (Robert 2016)
	‘We call firstborn the child who is born first, younger sibling the one who is born afterwards.’
(583)	Paul s’est. rendu à Paris en décembre 1997. Ensuite, il y a habité pendant plus d’une année. (de Saussure 2011)
	‘Paul went to Paris in December 1997. Then he lived there for more than a year.’
(584)	La fanfare marchait en tête, ensuite venait le cortège. (Robert 2016)
	‘The brass band walked in front, then came the procession.’

Ensuite is described by Gosselin (2007) as a temporal connective, conveying a temporal relation of temporal succession, which excludes the relation of simultaneity (like puis and unlike et). More precisely, ensuite instructs the hearer to relate the final boundary of the first eventuality E₁ to the first boundary of the second eventuality E₂ by a relation of precedence but not immediate vicinity. In other words, there is a linguistically relevant interval between the end of E₁ and the beginning of E₂, as is the case for two other adverbs, après ‘afterwards’ and plus tard ‘later’. Gosselin points out that this is the fundamental difference between puis and ensuite, since puis instructs for a relation of the optional immediate precedence type (i.e. the first boundary of E₂ can coincide with the second boundary of E₁).

De Saussure (2007, 2011) also argues in favour of a procedural account of ensuite. However, he does not follow Gosselin in his analysis that ensuite is a temporal connective, and instead argues that ensuite should be considered as a procedural serial connective rather than a temporal one. For him, the basic semantic meaning of ensuite, just like d’abord ‘firstly’ and enfin ‘finally’, is to order various types of elements. By way of pragmatic enrichment, the ordering of these elements can be specified to temporal order, as in (582)-(584), argumentative order, as in (585), and discursive order, as in (586).

(585)	Je ne sortirai pas. D’abord je suis fatigué, ensuite aller au restaurant est. la dernière chose qui me ferait plaisir. Enfin, il y a un match à la télé ce soir. (de Saussure 2007)
	‘I will not go out. First, I am tired, then going to a restaurant is the last thing that would make me pleasure. Finally, there is a game at the TV tonight.’
(586)	Il y a plein de cas où tu dois faire une sauvegarde supplémentaire. D’abord, si tu ouvres un fichier reçu par email. Ensuite, si tu dois transférer le fichier à un collègue qui utilise une autre plate-forme. Et puis surtout, chaque fois que tu fais une modification sur le fichier original. (de Saussure 2007)
	‘There are plenty of cases when you have to make an extra back-up. First, if you open a file that you received by email. Then, if you have to transfer the file to a colleague who uses a different platform. And then especially, every time when you make a modification to the original file.’

For de Saussure (2011), identifying a possible conceptual origin in diachrony, such as suite ‘followingness’ for ensuite, is not enough to justify the assumption that it encodes conceptual information, because all usages of ensuite can not be entirely predicted on the basis of this conceptual content. Ensuite has a temporal interpretation ‘only as a specification of a broader notion of “new element in a series”’ (2011, 69, 70), as in (587).

(587)	Marc a fait le repassage. Ensuite, il s’est. reposé sur le canapé.
	‘Marc ironed. Then, he had a rest on the couch.’

The linguistically relevant interval between the end of E₁ and the beginning of E₂, identified by Gosselin (2007) as distinguishing ensuite from puis , is described by Kozlowska (1996) and de Saussure (2011) as the non-adjacency interpretation imposed by ensuite. This is shown by the acceptability of ensuite in (589), and its unacceptability in (588).

(588)	Le vase est. tombé. *Ensuite il s’est. brisé.
	‘The vase fell.? Then it broke.’
(598)	La fenêtre s’est. ouverte. Ensuite, le courant d’air s’est. engouffré.
	‘The window opened. Then [afterwards but not immediately] the draught rushed in.’

In contrast to de Saussure’s treatment of ensuite, Kozlowska (1996) points out that this adverb is used to link bounded telic and atelic eventualities , thus excluding states. Following Dowty (1986) — who observed that bounded eventualities are usually interpreted sequentially where unbounded ones are usually interpreted to be temporally simultaneous — Kozlowska makes the hypothesis that ensuite is a formal means of overtly marking chronological sequential relations. She writes¹ (1996, 255):

Ensuite est directement lié à l’ordre temporel, i.e. à la progression temporelle en avant (E₁ se produit avant E₂). Par conséquent, ensuite est compatible avec les phrases traduisant de l’ordre temporel et il n’est pas compatible avec les phrases traduisant d’autres rapports temporel tel que : inversion causale, recouvrement, indétermination temporelle. Ainsi, ensuite doit être considéré comme un moyen formel de marquer l’ordre temporel.

So, as I have shown, scholars are split between accepting or rejecting ensuite as a member of the category of temporal connectives. A similar state of affairs is observed for puis . Puis is also described in classical grammars as indicating temporal succession (Grevisse 2016; Robert 2016; cf. Bras et al. 2001), as in (590), from Robert (2016). In this usage, puis corresponds to English then or afterwards. Succession can also be understood with respect to a spatio-temporal dimension, thus expressing it from the view of an observer (Robert 2016), as in (591).

(590)	Dieu nous prête un moment les prés et les fontaines […] Puis il nous les retire. Il souffle notre flamme. (V. Hugo, Les rayons et les Ombres)
	‘God lends us for a moment the meadows and the fountains […] Then he takes them back. He blows out our flame.’
(591)	En bas, des fleurs rouges, jaunes […] puis c’étaient les jasmins, les glycines. Puis voici une lande. La forêt… et puis un damier de plaines.
	‘Below, red, yellow flowers […] then there were the jasmines, the wisteria. Then here is a moor. The forest […] and then a tartan of fields.’

As noted by Bras et al. (2001), citing classical grammars, the notion of temporal succession can disappear, and be replaced by the meaning of logical succession. In this case, the meaning of puis corresponds to the English besides or moreover as in (592).

(592)	On trouvait à Yonville qu’il avait des manières comme il faut. Il écoutait raisonner les gens mûrs […] Puis il possédait des talents. (Flaubert, Madame Bovary)
	‘People from Yonville thought that he had manners as it should be. He used to listen to mature people reasoning […] Besides he was talented.’

Certain scholars, such as Hansen (1995) and Reyle (1998), have suggested that the meaning and the discursive function of puis have evolved from the basic temporal value to the enumerative and argumentative value, and that the temporal interpretation is only inferred by default in narrative contexts (cf. discussion in Bras et al. 2001). Bras et al. (2001) argue against this proposal, pointing to the fact that when puis links to past events expressed with puis, other temporal interpretations (such as simultaneity or temporal regression) are not possible. For them, puis is an adverbial marking temporal succession which acts, syntactically speaking, as a temporal connective. For others as well, such as Gosselin (2007) and de Saussure (2007), puis can be considered, semantically speaking, as a temporal connective that marks the temporal succession of an utterance with respect to the following utterance.

6.3.2 Hypotheses and Predictions

In Grisot and Blochowiak (2015, 2017), we investigated implicit and explicit underdetermined and chronological sequential relations holding between discourse segments in which the Passé Composé or Passé Simple was used. The explicit relations were overtly marked using ensuite and puis . This is illustrated in examples (593)-(594), which are the French translations of the English sentence given in (595), and in examples (596)-(599), which are the French translations of (600). The first series of past events is temporally undetermined (that is, they can either be interpreted sequentially or simultaneously), whereas the second series is interpreted sequentially.

(593)	Un homme entra dans le bar. Il vit son frère.
(594)	Un homme est entré dans le bar. Il a vu son frère.
(595)	A man entered the bar. He saw his brother.
(596)	Un homme entra dans le bar, il commanda une bière, il alla s’assoir au fond de la salle.
(597)	Un homme est entré dans le bar, il a commandé une bière, il est allé s’assoir au fond de la salle.
(598)	Un homme entra dans le bar, il commanda une bière, ensuite/puis il alla s’assoir au fond de la salle.
(599)	Un homme est entré dans le bar, il a commandé une bière, ensuite/puis il est allé s’assoir au fond de la salle.
(600)	A man entered the bar, he ordered a beer, (then) he went to sit in the back of the room.

As these examples show, a series of past events can be expressed using either the Passé Simple, as in (593), (596) and (598), or the Passé Composé , as in (594), (597) and (599). Additionally, the temporal relations can either remain implicit or be overtly marked using a temporal connective , as in (598) and (599).

As discussed in Sect. 1.1, these two verbal tenses are both described as perfective; however, the first one presents the eventuality from a past time reference point, whereas the second presents the eventuality from the present and expresses a resultative state that holds at the moment of speech. Semantic discourse theories such as DRT (cf. Sect. 2.1) have suggested that the Passé Simple and Passé Composé do not have the same role with respect to expressing temporal relations: the Passé Simple instructs the hearer to establish a sequential relation between two past events (time advances from the first event to the second), whereas the Passé Composé is undetermined (that is, it does not provide the comprehender with any information with respect to the sequential relations between past events).

The two temporal connectives encode similar but not identical procedural meanings (cf. Sect. 6.3.1). The procedural meaning of ensuite is to construct a precedence, but not a sequential relation of immediate vicinity (in other words, there is a gap between the final boundary of the first event and the initial boundary of the second event) (Gosselin 2007). In contrast, the procedural meaning of puis is to construct an optional sequential relation of immediate precedence (in other words, there might be a gap between the two events). Taking into account these semantic and pragmatic differences between the Passé Composé and Passé Simple on the one hand, and between ensuite and puis on the other hand, the first two research questions can be formulated: What is the role of the verbal tense in processing temporal relations? and Is there an interaction between verbal tenses and temporal connectives ?

Finally, the third research question refers to the implicit vs. explicit status of sequential temporal relations. Previous experimental studies on connectives as processing instructions for causal and contrastive relations (Haberlandt 1982; Britton and Gernsbacher 1994; Traxler et al. 1997; Sanders and Spooren 2009; Cozijn et al. 2011; Zufferey 2014; Cain and Nash 2011; Canestrelli 2013; Canestrelli et al. 2013) have found that connectives facilitate the processing of the immediately following region when their meaning correlates with the meaning of the discourse region processed. So, the third research question is: Do these two temporal connectives have the same impact on processing as causal and contrastive connectives?

In order to answer these three research questions, a series of scenarios can be formulated, with subsequent predictions regarding main and interaction effects between these independent variables. The notion of effect (main and interaction) refers to significantly shorter or longer reading times for the target segment in a given experimental condition (for online experiments), and to significantly lower or higher acceptability rates for experimental items in a given condition (for offline evaluation experiments). The scenarios and their subsequent conditions are summarized in Fig. 6.2.

Fig. 6.2

Possible scenarios for the independent variables playing a role in the expression of sequential temporal relations

The first set of hypotheses and predictions concerns the meanings of the verbal tenses tested, and their role in the expression of temporal relations. Firstly, the Passé Composé and Passé Simple encode different procedural content regarding sequential relations: the Passé Composé is undetermined with respect to sequential relations, whereas the Passé Simple instructs the hearer to establish a default sequential relation (Kamp and Rohrer 1983; de Saussure 2003). This scenario leads to the prediction that when comprehenders need to handle a series of past events that should be understood sequentially, we would expect a main effect for the verbal tense variable. Shorter reading times should be measured when expressed using the Passé Simple than when expressed using the Passé Composé. Secondly, the meanings of the Passé Composé and Passé Simple are contextually determined (Moeschler 2000a, b, 2002b; Moeschler et al. 2012; Grisot and Moeschler 2014). Consequently, when comprehenders need to handle a series of past events that should be understood sequentially, we would expect no main effect for the verbal tense variable. In other words, no difference in reading times is expected when past events are expressed with the Passé Composé or Passé Simple. This would equally be the case for undetermined series of past events, for which both sequential and simultaneous temporal relations are likewise possible.

The second set of hypotheses and predictions regards the roles of ensuite and puis as temporal connectives which can be used to mark sequential temporal relations overtly. Firstly, temporal connectives are processing instructions , as previously found for causal or contrastive connectives. In other words, processing a temporal relation is facilitated by the connective, unlike when the temporal relation is implicit. As such, we would expect a main effect of the explicit/implicit status of the temporal relation to take the form of short reading times and higher acceptability rates when the temporal relation is overtly marked, compared to when it is implicit. This would be the case both for undetermined and sequential temporal relations. The subsequent prediction is that we would expect a main effect of the connective, due to the fact that this difference in meaning is relevant for the cognitive processing of sequential temporal relations. Secondly, ensuite and puis are both sequential temporal connectives with similar meanings, which result in similar effects for the cognitive processing of sequential temporal relations. So, if this is the case, we would not expect a main effect of the connective. Thirdly, these two connectives have been described as having different meanings, at the level of a fine-grained semantic analysis: precedence but not immediate vicinity (i.e., there is a gap between the final boundary of the first event and the initial boundary of the second event) for ensuite; and optional immediate precedence (the final boundary of the first event might be the same as the initial boundary of the second event) for puis (Kozlowska 1998b; de Saussure 2003; cf. Sect. 6.3.1).

The third set of hypotheses concerns the co-occurrence of verbal tenses and temporal connectives . Firstly, one can assume that the Passé Composé and Passé Simple are equally compatible with both ensuite and puis . In this case, we would expect no effect of interaction between the independent variables of verbal tense and connective. In other words, we do not expect to find either of the two verbal tenses behaving differently when combined with ensuite than when combined with puis. Secondly, we could expect that, due to fine-grained meaning distinctions, the Passé Composé and Passé Simple are not equally compatible with both ensuite and puis. Consequently, we expect an interaction effect between the independent variables of verbal tense and connective.

The hypotheses issued from these possible scenarios have been tested by Grisot and Blochowiak (2015, 2017) in two series of experiments on the role played by the connective ensuite, and one series of experiments on the role played by the connective puis. In the online self-paced reading experiments, reading times were measured on the segment immeditately following the connective, which was an entire sentence consisting of subject, verb and object. More recently, in Grisot and Blochowiak (to appear) similar experiments were carried out and reading times were measured on smaller target regions, that is on the subject-verb region and on the object region separately. This measurement resulted into more fine-grained results and analyses. The reader may refer to Grisot and Blochowiak (to appear) for a discussion of these hypotheses and prediction based on this second type of analysis.

6.3.3 “Ensuite”, the Passé Composé and Undetermined Temporal Relations: A Self-Paced Reading Experiment

Our aim in this experiment, from Grisot and Blochowiak (2015), was to test whether the connective ensuite is useful for disambiguating indeterminate temporal relations — that is, forcing the sequential interpretation of examples, as in:

(601)	[Les enfants ont décoré le sapin]P. [Maman a préparé des bons gâteaux.]Q
	The children decorate.PC the Christmas tree. Mum cook.PC delicious biscuits.
	‘The children decorated the Christmas tree. Mum cooked delicious biscuits.’
(602)	[Les enfants ont décoré le sapin]P, [ensuite] [maman a préparé des bons gâteaux.]Q
	The children decorate.PC the Christmas tree, then mum cook.PC delicious biscuits.
	‘The children decorated the Christmas tree, then mum cooked delicious biscuits.’

The participants were 48 undergraduate students from the Faculties of Humanities of the Universities of Geneva and Neuchâtel in Switzerland (42 females, mean age: 22.47, range: 18–32). All participants were native speakers of French, studying French language and literature, or language sciences. Their participation in the experiments was voluntary and they were not paid for their participation.

We used 10 experimental items per condition, similar to those from examples (601) and (602), and 10 fillers similar to (603). For every experiment item, the target segment was the Q segment. There were three experimental conditions: the implicit condition P.Q; the explicit condition PensuiteQ; and the control condition Q. The task was to judge if the situation described was plausible (like the experimental items in (601) and (602)) or implausible (like the filler in (603)).

(603)	[Paul est. parti pêcher au lac]P. [Les poissons ont bu. sa bouteille de vin.]Q
	‘Paul went fishing. The fish drank his bottle of wine.’

Each group saw only one condition. The experimental items and the filler were presented with E-prime, in a random order. Participants answered on the keyboard by pressing one key for plausible and another key for implausible. The experiment consisted of a training phase using 4 experimental items and 4 fillers, followed by the genuine experimental phase. Participants’ accuracy was assessed according to their responses with respect to the experimental items (i.e. correct) and the fillers (i.e. incorrect).

The mean reading times for the target segment Q in each of the three experimental conditions are reported in Table 6.2.

Table 6.2 Reading times for the target segment in each condition

A one-way ANOVA was performed in order to check for an effect of the condition (between-subjects factor) on the RT of the target segment. A statistically significant effect of the condition on RT was found for the control condition (F(2,44) = 3.255, p < .05). Both the mean RT for the implicit condition PQ and for the explicit condition PensuiteQ were longer than those for the control condition Q. The difference between conditions PQ and PensuiteQ was not statistically significant, as Fig. 6.3 shows.

Fig. 6.3

Mean RT for the target Q segment in the three conditions

The results indicate that the target segment Q without cotext (the control condition) is processed faster than in a cotext, given by the previous sentence P in the implicit condition and by the previous sentence P and the connective ensuite in the explicit condition. Additionally, the presence of the connective does not produce a facilitation effect by reducing the processing time of the target segment.

These results confirmed prediction 3, given in Fig. 4.4. Firstly, it was shown that ensuite does not produce a facilitation effect where the hearer would be prevented from having to choose between the sequential and simultaneous interpretations by instructing him to choose the former over the latter . In other words, this experiment did not provide evidence in favour of a procedural account of this connective, in the sense of encoding instructions that constrain the hearer in the comprehension process (cf. the theoretical discussion in Sect. 2.3.2). Secondly, similar reading times were measured for the implicit and explicit conditions. This could indicate that the verbal tense is a sufficient cue to determine the temporal relation between two events, because it encodes the instruction “relate E₁ to E₂”. This means that this relation does not necessarily have to be overtly marked. I suggest that, if this was not the case, the PQ condition would have been read slower than the PensuiteQ condition. Thirdly, the differences between the control condition Q and the PensuiteQ and PQ conditions were statistically significant. This result might indicate that the processing of a sentence is shorter out of cotext than in a cotext, be it PQ or PensuiteQ. This is mainly because there is no relation to calculate (instructed by the verbal tense or by the temporal connective ). So, the sentence is processed faster than in the other two conditions.

6.3.4 “Ensuite”, the Passé Composé , the Passé Simple and Sequential Temporal Relations

In Grisot and Blochowiak (2017), we presented two experiments carried out on chronological sequential relations holding between segments in which the Passé Simple or Passé Composé is used. We tested the online processing and offline acceptability of implicit and explicit sequential relations.

Participants in these two experiments were 41 s- and third-year students from the University of Neuchâtel (35 females, mean age: 22.53, range 19–31). All participants were native speakers of French, studying language sciences or speech therapy. Their participation in the experiments was part of their activity for one class in linguistics, and they were not paid for their participation. These participants took part in this experiment (online), and in the offline evaluation experiment immediately afterwards.

For the online experiment, participants were divided into two groups. One group saw the explicit condition, in which the temporal relation was made explicit by the connective ensuite, as in (604). The second group saw the implicit condition, in which the temporal relation was implicit, as in (605). This distribution gave rise to an inter-subject analysis.

(604)	⦋Agnes a joué du piano⦌1, ⦋ensuite⦌2 ⦋elle a rangé sa bibliothèque⦌3, ⦋et après⦌4 elle est. allée promener son chien.⦌5
	‘Agnes played the piano, then she tidied up her bookcase, and then she walked her dog.’
(605)	⦋Agnes a joué du piano.⦌1 ⦋Elle a rangé sa bibliothèque.⦌2 ⦋Elle est allée promener son chien.⦌3
	‘Agnes played the piano. She tidied up her bookcase. She walked her dog.’

All the sentences were created with two different versions of the initial sentence: in one, the verbal tense used was the Passé Composé , and in the other the Passé Simple. The sentences were distributed into two lists containing one version of the initial sentence, in either the Passé Simple or Passé Composé. Each list contained a total of 16 items and 14 fillers having the same structure as the items. Each participant saw only one list. This distribution gave rise to an intra-subject analysis.

In the explicit condition, the critical segment was segment 3 – the segment immediately following the target connective (ensuite). Segments 4 and 5 were wrap up segments, intended to avoid a critical reading time measure at the end of an item, which cretes an effect due to the end of the task. In the implicit condition, the critical segment was segment 2, which was identical to the critical segment in the explicit condition. The reading times were measured for the critical segments and were compared between the two conditions. Additionally, the critical segments in the Passé Simple and Passé Composé were compared using an intra-subject analysis.

Experiments were designed with the E-prime software. The different segments appeared on the screen one after another upon pressing the space bar, each disappearing from the screen as the readers went on to the next. This design allowed the participants to read each segment individually, and stopped them pressing the space bar in order to see all the segments before starting to read. Having read the series of 5 segments (in the explicit condition) or the series of 3 segments (in the implicit condition), participants had to answer the likelihood question by pressing one key for likely and another key for unlikely to record their answers.

The mean reading times for the target region, in each of the four experimental conditions, are reported in Table 6.3.

Table 6.3 Reading times for the target segment in each condition- ensuite

A mixed ANOVA performed on Log10 mean values showed no main effect of the within-group variable Verbal Tense (VT) (F(1,39) = 2.641, p > .05, η² = .063), nor an interaction effect of Verbal Tense*Implicitness (F(1,39) = .076, p > .05, η² = .002). In other words, Log10 mean reading times for the Passé Composé in the explicit condition (M = 3.19, SD = .098) are comparable with those in the implicit condition (M = 3.19, SD = .105). Similarly, the Log10 mean reading times of the Passé Simple in the explicit condition (M = 3.21, SD = .086) are comparable with those in the implicit condition (M = 3.21, SD = .113). This can be seen in Fig. 6.4.

Fig. 6.4

The Passé Composé and the Passé Simple in the explicit (using ensuite) and implicit conditions

For the offline experiment, sentences from the online experiment were presented in four variants, corresponding to the four experimental conditions from the previous experiment: Passé Composé implicit; Passé Simple implicit; Passé Composé explicit; and Passé Simple explicit. These conditions were created by manipulating two within-group factors, Verbal Tense and Implicitness. To create the explicit version of the items, the connective ensuite was used to mark the sequential temporal relation overtly. There were no fillers used in this experiment. Participants were told that they would have to participate in a second experiment, in which they would see the same sentences as the previous experiment, and that each sentence would occur in four variants. The 16 groups of four variants (corresponding to the 16 items from the online experiment) were distributed into two lists. Each list contained 8 groups of sentences, and each participant saw one of the two lists. Participants were asked to rate the acceptability of each variant on a 4 point Likert scale. They were allowed to use each of 4 values of the Likert scale only once within each group of sentences.

In order to analyze the data, we calculated the median value for each variant across all the participants who saw that variant. Values were organized according to condition, resulting in 16 observations per condition. The results are shown in the bar chart in Fig. 6.5.

Fig. 6.5

The mean median values in the four conditions – ensuite

A repeated measures ANOVA was performed, to test whether the differences between several mean median values of the dependent variable (the participants’ acceptability judgments) depended on the independent variable tested: the status of the temporal relation and the verbal tense. The various groups of mean values of the dependent variable were formed by the within-group factors Verbal Tense and Implicitness.

The results of the ANOVA showed a significant main effect of the Implicitness factor (F(1, 15), 30.533, p < .05, η² = .671), according to which the variants where the temporal relation was implicit were scored higher (M = 3.078, SE = .101) than those where it was overtly marked using ensuite (M = 2.047, SE = .103). The factor Verbal Tense was not statistically significant (F(1, 15), 2.047, p > .05, η² = .120), and nor was the interaction Verbal Tense*Implicitness (F(1,15), 1.337, p > .05, η² = .082). These results indicate that participants prefer the sequential temporal relation to remain implicit, but do not show a preference when the Passé Composé or the Passé Simple is used.

The results of these two experiments on ensuite, the Passé Composé, the Passé Simple and their role in the expression of sequential temporal relations from Sect. 6.3.3 provided evidence for predictions 1 and 4, given in section Fig. 6.2. The results of the online experiment showed no main effect of the explicit/implicit status of the temporal relation. In other words, ensuite does not produce a facilitation effect with respect to marking a sequential temporal relation overtly. This result is in line with the findings of the previous experiment, in which undetermined relations were tested, and confirms prediction 1. One of the scenarios was that this temporal connective encodes procedural content which constrains the hearer in the comprehension process, by signalling the speaker’s intended interpretation with respect to the type of temporal relation. Evidently, this seems not to be the case when it comes to the online processing of sentences. In contrast, in the offline experiment, a main effect of the Implicitness factor was found, according to which participants preferred the sequential temporal relation to remain implicit over its overt marking by ensuite. According to these results, prediction 4 and its subsequent scenario need to be adjusted. In particular, a main effect of the implicit/explicit status of the temporal relation was only found in the offline acceptability tasks , in which participants did not prefer the explicit versions of the experimental items tested. This means that the presence of ensuite is consciously perceived as less acceptable, but this does not seem to play a role in terms of processing discourse segments.

Another result of this experiment is that no main effect of the Verbal Tense variable was found: the observed mean differences in reading times of the target segment between the Passé Composé and Passé Simple were not statistically significant. As with the online data, the Verbal Tense factor was not statistically significant in the offline acceptability experiment. These findings validate prediction 1, according to which the meanings of the Passé Composé and Passé Simple are contextually determined, and invalidate scenario 1, according to which the different meanings of these two verbal tenses play significant roles in terms of processing and conscious acceptability tasks. These outcomes are at odds with the assumptions made by previous theoretical studies of French verbal tenses, according to which significant meaning differences exist between the Passé Composé and Passé Simple (cf. the discussions in Sects. 1.1.1, 1.1.3, 2.1 and 2.3.3). That is to say, the Passé Simple is a perfective verbal tense whose meaning can be described using the configuration of Reichenbachian temporal coordinates E = R < S, and which instructs the hearer to establish a sequential temporal relation. The Passé Composé , on the other hand, is a perfect verbal tense, whose meaning can be described using the configuration of Reichenbachian temporal coordinates E < R = S, and which guides the hearer towards the identification of a resultative state relevant at S.

In contrast, Moeschler et al. (2012) and Grisot (2015) advance the hypothesis that the meanings of the Passé Composé and Passé Simple are contextually determined at two levels. The first level is conceptual — that is, building an ad hoc concept regarding the localization of an event with respect to S (cf. the discussion in Sect. 5.2.2). This was confirmed in the annotation experiment described in Sect. 4.2.2, in which participants had to identify the localization of de-temporalized events expressed with verbs in their infinitive form in the past or non-past (present or future). The second level is procedural — that is, using R to locate events with respect to one another (cf. the discussion in Sect. 5.2.3). This property of the Passé Composé and Passé Simple, termed narrativity , was tested in the annotation experiment discussed in Sect. 4.2.2, in which it was shown that they both allow sequential and simultaneous temporal relations (corresponding to their narrative and non-narrative usages respectively).

6.3.5 “Puis ”, the Passé Composé , the Passé Simple and Sequential Temporal Relations

In Grisot and Blochowiak (2017), we also presented two experiments in which the connective ensuite was replaced with puis. The experimental design, the procedure and the experimental items were practically the same, with the exception of the target connective. For both series of experiments, the verbal tense was a within-subjects variable, and implicitness a between-subjects variable. The participants were not the same, but the two groups had similar mean ages and educational backgrounds.

For the puis experiment, participants were 43 s- and third-year students from the University of Neuchâtel (38 females, mean age: 20.93, range 19–25). All participants were native speakers of French, studying language sciences or speech therapy. As before, their participation in the experiments was part of their activity for one class in linguistics, and they were not paid for their participation. These participants took part in the online experiment, and, immediately afterwards, in the offline evaluation experiment.

The mean reading times for the target region in each experimental condition are reported in Table 6.4.

Table 6.4 Reading times for the target segment in each condition- puis

A mixed ANOVA performed on Log10 mean values showed no main effect of the within-group variable Verbal Tense (F(1,41) = 3.305, p > .05, η² = .075), nor an interaction effect of Verbal Tense*Implicitness (F(1,41) = .691, p > .05, η² = .017). In other words, Log10 mean reading times for the Passé Composé in the explicit condition (M = 3.22, SD = .112) are comparable with those in the implicit condition (M = 3.25, SD = .111). Similarly, the Log10 mean reading times of the Passé Simple in the explicit condition (M = 3.21, SD = .119) are comparable with those in the implicit condition (M = 3.21, SD = .132). This can be seen in Fig. 6.6.

Fig. 6.6

The Passé Composé and the Passé Simple in the explicit (using puis ) and implicit conditions

For the offline experiment, sentences from the online experiment were presented in four variants, corresponding to the four experimental conditions from the previous experiment: Passé Composé implicit; Passé Simple implicit; Passé Composé explicit; and Passé Simple explicit. These conditions were created by manipulating two within-group factors, Verbal Tense and Implicitness. To create the explicit version of the items, the connective puis was used to mark the sequential temporal relation overtly.

The median value for each variant was calculated across all participants who saw that variant. Values were organized according to condition, resulting in 16 observations per condition.

Fig. 6.7

The mean median values for the four conditions – puis

A repeated measures ANOVA was performed, to test whether the differences between several mean median values of the dependent variable (the participants’ acceptability judgments) depended on the independent variable tested: the status of the temporal relation and the verbal tense. The various groups of mean values of the dependent variable were formed by the within-group factors Verbal Tense and Implicitness.

The results of the ANOVA did not show a main effect of the Implicitness factor (F(1, 15), .522, p > .05, η² = .034), nor of the factor Verbal Tense (F(1, 15), 1.588, p > .05, η² = .096). In other words, the variants in which the temporal relation was implicit received scores similar to the variants in which it was overtly marked using puis (M = 2.594, SE = .112 and M = 2.438, SE = .075 respectively). Similarly, the variants in which the Passé Composé was used received scores similar to the variants in which the Passé Simple was used (M = 2.422, SE = .075 and M = 2.609, SE = .085 respectively). Furthermore, the interaction Verbal Tense*Implicitness was not statistically significant (F(1,15), .769, p > .05, η² = .049). These results indicate that participants did not show a preference, be it for the explicit vs. implicit cases or for the Passé Composé vs. the Passé Simple.

The results of the two experiments on puis , the Passé Composé, the Passé Simple and their role in the expression of sequential temporal relations from Sect. 6.3.5 are in harmony with the outcomes of the experiments on ensuite and its role in the online processing of sequential relations, and differ with respect to offline acceptability evaluation. As before, we did not find a main effect of the Verbal Tense variable in this experiment. As such, the results of the previous experiment were replicated with respect to the roles of these two verbal tenses in the expression of sequential temporal relations; this confirms scenario 2, which posits that the meaning of verbal tense is contextually determined. As with the experiments on ensuite, in processing data the Implicitness factor was not statistically significant, therefore invalidating scenario 4 (Fig. 6.7). As will be indicated below, this conclusion is confirmed by the results of the mixed ANOVA performed on processing data.

Differences between the two connectives emerged in offline acceptability data. Where ensuite indicated a main effect of Implicitness, it was not true for puis . This means that participants preferred the implicit versions of the experimental items when the explicit versions used ensuite, but did not show this preference when puis was used. This outcome seems to support scenario 6, which theorizes that these two connectives have different meanings. This assumption is confirmed by the results of the mixed ANOVA performed on evaluation data, as I will show below.

6.3.6 “Ensuite” and “Puis ”-Mixed Statistical Analysis

In order to investigate the predictions formulated in Sect. 6.3.1 with respect to the role of both ensuite and puis, as well as their occurrence with the Passé Composé and Passé Simple, two mixed ANOVA analyses were performed. The first concerned the processing data from the experiments carried out on ensuite and puis , and the second concerned the acceptability rating data from experiments carried out on ensuite and puis. These analyses are possible because of two common characteristics. Firstly, exactly the same items were used in these two groups of experiments (with the exception of the connective). Secondly, the two groups of participants in these experiments were part of a larger group, whose members had comparable ages and educational backgrounds (native speakers of French, who were second- and third-year students studying language sciences or speech therapy at the University of Neuchâtel).

The mixed ANOVA performed on the processing data aimed to test whether the differences between several mean values of the dependent variable (the reading times of the target segment) depended on the independent variables tested. The various groups of mean values of the dependent variable were formed by the between-group factor Implicitness with two levels (explicit and implicit), the between-group factor Connective with two levels (ensuite and puis ), and the within-group factor Verbal tense with two levels (Passé Composé and Passé Simple).

The results of the mixed ANOVA did not show a significant effect of the within-group factor Verbal tense (F(1, 80), .350, p > .05, η² = .004), nor of the interaction Implicitness*Verbal Tense (F(1,80), .279, p > .05, η² = .003), There is a significant effect of the interaction Connective*Verbal Tense (F (1,80), 5.688, p < .05, η² = .066), according to which the Passé Composé is easier to process when it occurs with ensuite (M = 3.19, SD = .098) than when it occurs with puis (M = 3.22, SD = .112). This effect does not apply to the Passé Simple when it occurs with ensuite (M = 3.21, SD = .086) compared to when it occurs with puis (M = 3.21, SD = .119), as can be seen in Fig. 6.8.

Fig. 6.8

The effect of Connective*Verbal Tense in processing data

This mixed ANOVA analysis confirmed the previous conclusions, according to which the Passé Composé and Passé Simple do not trigger significant differences in mean reading times for the target segment. Additionally, no significant interaction effect was found between Implicitness and Verbal Tense. This suggests that the mean reading times for both the Passé Composé and Passé Simple do not significantly differ between cases where the temporal relation is implicit and cases where it is explicitly marked using a connective. Furthermore, the interaction effect Verbal Tense*Connective was significant, pointing to different behaviour of the verbal tense depending on the connective with which it occurs. In particular, the Passé Composé is easier to process when it occurs with ensuite than when it occurs with puis . As for the Passé Simple, its combination with ensuite or puis results in similar processing costs. This finding provides evidence for fine-grained semantic/pragmatic differences between ensuite and puis. The gap between the final boundary of event₁ and the initial boundary of event₂ seems to play a role in terms of the processing of events when they are expressed using the Passé Composé . This outcome supports scenario 7, and validates its subsequent prediction. However, this difference in meaning doesn’t seem to be generalizable, since we did not find a main effect of the variable Connective. As such, scenario 5 is partially validated, and needs to be adjusted to take into consideration the interaction effect Verbal Tense*Connective.

The mixed ANOVA performed on the acceptability data also aimed to test whether the differences between several mean median values of the dependent variable (the participants’ acceptability judgments) depended on the independent variable tested. The various groups of mean values of the dependent variable were formed by the within-group factors Verbal tense and Implicitness, and the between-groups factor Connective with two levels (ensuite and puis). The results of the mixed ANOVA showed a significant main effect of the Implicitness factor (F(1, 30), 17.273, p < .05, η² = .365), according to which the variants where the temporal relation was implicit were scored higher (M = 2.836, SE = .056) than the variants where it was overtly marked (M = 2.242, SE = .076). In contrast, the factor Verbal Tense was not statistically significant (F(1, 30), .021, p > .05, η² = .001), suggesting that the variants in which the Passé Composé was used received scores similar to the variants in which the Passé Simple was used (M = 2.547, SE = .058 and M = 2.531, SE = .061 respectively).

A significant interaction effect between Connective*Implicitness was found (F(1, 30), 9.378, p < .05, η² = .238), according to which the cases in which ensuite was used to mark the temporal relation overtly were rated lower (M = 2.047, SE = .108) than the cases in which puis was used (M = 2.438, SE = .108). Additionally, the cases in which the implicit variants were compared to variants overtly marked using ensuite were rated higher (M = 3.078, SE = .107) than the cases in which they were compared to variants overtly marked using puis, as shown in Fig. 6.9. In other words, if participants have to mark a sequential relation overtly, they prefer to do it with puis. Also, they rate implicit relations higher when they are in opposition to overtly marked relations using ensuite than when they are in opposition to overtly marked relations using puis.

Fig. 6.9

The interaction effect Connective*Implicitness in acceptability data

Furthermore, a tendency towards significance was found for the interaction Connective*Verbal tense (F(1, 30), 3.627, p = .06, η² = .108), according to which participants rated the Passé Composé occurring with ensuite (M = 2.672, SE = .081) higher than with puis (M = 2.422, SE = .081), and the Passé Simple occurring with ensuite (M = 2.453, SE = .086) lower than with puis (M = 2.609, SE = .086). This can be seen in Fig. 6.10.

Fig. 6.10

The interaction effect Connective*Verbal tense in acceptability data

This mixed ANOVA analysis reveals discrepancies between processing and conscious evaluation. Where the lack of main effect of Verbal Tense found in the mixed analysis on processing data discussed above is confirmed, a main effect of Implicitness was found in the offline data, according to which participants prefer implicit sequential temporal relations. Additionally, the offline data confirm the preference for the Passé Composé to occur with ensuite over puis , as measured in terms of shorter mean reading times. So, the shorter mean reading times are translated into the offline tasks as higher acceptability rates. These findings support scenario 8 and its subsequent prediction of an interaction effect between the verbal tense and the connective. Moreover, this mixed ANOVA revealed a significant interaction effect between Connective*Implicitness, according to which participants prefer to use puis to mark sequential relations overtly. This unpredicted outcome might indicate that it is more accurate to consider puis as a temporal sequential connective than ensuite.

Generally, the results of the processing data indicate that the reading times of the target segment were not influenced, either by the verbal tense or by the connective. Also, segments with the Passé Composé were easier to process when it occurred with ensuite than when it occurred with puis . In contrast, the Passé Simple did not show this kind of preference. As for the offline evaluation with acceptability task, participants preferred sentences in which the temporal relation is implicit. This applies to both the Passé Composé and Passé Simple. However, if the temporal relation is explicit, they prefer when it is overtly marked using puis. Finally, participants showed a tendency to prefer the Passé Composé occurring with ensuite and the Passé Simple with puis.

These results can be interpreted with respect to two issues. The first is that these results confirm the psycholinguistic findings that comprehenders have expectations about relations holding between segments when reading a text, which bias their inferential decisions during comprehension (Segal et al. 1991; Murray 1997; Sanders 2005; Asr and Demberg 2012). The second is that certain contextual cues raise comprehenders’ expectations of the upcoming discourse relation , and therefore, the preference for implicitness. For example, Rohde and Horton (2010) have shown in an eye tracking experiment that a sentence in which a consequence is expressed by an implicit causality verb (such as adore, inspire, humiliate) raises the expectation of a causal explanation in the following segment. For Levy and Jaeger (2007), this extra cue for a causal relation increases the possibility of omitting the causal connective and having an implicit causal relation. They proposed the Uniform Information Density hypothesis, according to which humans tend to spread information evenly across an utterance or series of utterances, thus reducing or omitting optional material when the information is highly predictable (Levy and Jaeger 2007; Frank and Jaeger 2008; Jaeger 2010). Similarly, Kehler et al. (2008) have shown that the grammatical aspect of the verb plays a role in determining the expected temporal relation: perfective bias for sequential temporal relations, and progressive for synchronous relations. In the experiments from this chapter, the presence of supplementary cues raises expectations regarding discourse relations , and therefore the preference for implicitness — that is, for content to be inferred instead of explicitly expressed by the speaker. For temporal relations, the relevant linguistic cues are verbal tenses and grammatical aspect (that is, the perfective/imperfective aspects).

These expectations affect sequential chronological temporal relations and canonical causal relations (cause – consequence). Where hearers expect segments to be presented chronologically, two immediate consequences for the cognitive processing of these temporal relations can be identified. Firstly, expected relations are claimed to be easier to process than unexpected relations, such as reverse causal and sequentially non-chronological temporal relations. Secondly, highly expected relations may be left implicit, in that comprehenders prefer passages with the connective omitted rather than included. As pointed out by Asr and Demberg (2012), at the level of discourse relations , this would mean that unexpected relations are more frequently expressed explicitly than expected ones. Rendering an expected relation explicit would be redundant, and therefore costlier for processing. The results of our experimental work on sequential chronological relations confirm Asr and Demberg’s prediction.

The second regards the relevance-theoretic notion of procedural meaning , described in the literature as having the role of constraining the inferential phase of the interpretative process by signalling the most relevant interpretative path (that is, accessing the appropriate contextual assumptions to obtain the interpretation intended by the speaker). Accordingly, connectives such as then , or French ensuite and puis , instruct the hearer to order mental representations of eventualities chronologically, whereas connectives such as before/avant que instruct for an anti-chronological temporal order between two eventualities. Based on these descriptions, we would expect connectives to guide the hearer in every case where the connective is used. However, this is not the case. Connectives facilitate processing only when their instruction is compatible with the content communicated in the discourse; when this is not the case, the presence of the connective increases the cognitive effort (Canestrelli et al. 2013; Zufferey et al. 2015). The results of the experiments described in this book indicate that the facilitating effect caused by the procedural content of discourse connectives may be cancelled when comprehenders have higher expectations which bias their inferential process. Consequently, comprehenders prefer highly expected relations to be expressed implicitly rather than marked overtly. However, as we will discuss below, this seems to depend on the temporal connective used.

Thirdly, in terms of psycholinguistic models of comprehension, the pragmatic approach to temporal relations advanced in this book corresponds to a model in which the interpretation process is incremental (e.g. Gibbs 2002; Koornneef and van Berkum 2006), allowing for the integration of cues as they become available. In relevance theoretic terms, the comprehension process consists of several subtasks that take place in parallel. More precisely, the logical form encoded by an utterance containing incomplete conceptual representations is dealt with by the inferential process in three ways: constructing the explicit content via decoding, disambiguation, reference resolution and other pragmatic enrichment processes (e.g. narrowing and loosening); constructing an appropriate hypothesis about the intended contextual assumptions; and constructing an appropriate hypothesis about the intended contextual implicatures (Wilson and Sperber 2004, 615). Wilson and Sperber point out that there is no sequential order in which these subtasks of the comprehension process take place, as comprehension is an online process. As they take place in parallel, the resulting hypotheses are, if needed, revised or elaborated as the utterance unfolds. According to this framework, linguistic expressions encoding procedural information, such as connectives and verbal tenses among others, play a crucial role because they guide the comprehender in this process by directing him towards the speaker’s intended meaning.

6.4 What Is “Cognitive Temporal Coherence”?

Based on previous proposals that coherence relations are cognitive entities and relate mental representations of discourse segments, I propose the notion of cognitive temporal coherence. I suggest that hearers integrate temporal information provided not only by various temporal cohesion cues, such as Tense, Aspect, Aktionsart, temporal connectives , but also by temporal adverbials, such as yesterday, last/next year, in 2010 etc., into the mental representations that they build for discourse segments. So, in order to describe the notion of cognitive temporal coherence, I will give two types of argument. Firstly, the categories of Tense, Aktionsart and Aspect are cognitively relevant, whereas the generic notion of verbal tense (as used in linguistic and pragmatic studies, cf. Sect. 1.1 and Chap. 2) is not (Grisot, submitted). Secondly, drawing on the psycholinguistic account of coherence, according to which mental representations of discourse segments are structured and coherent (Givón 1995; Graesser et al. 1997), I propose that the cohesion ties investigated in this book are pointers to this temporal mental coherence which speakers establish during the comprehension process of utterances. I will develop these two types of arguments below.

6.4.1 Temporal Cohesion Ties Are Cognitively Motivated

In Grisot (submitted), I argue that the generic notion of verbal tense is not cognitively motivated, mainly because it is a generic notion used to refer to single underlying temporal and aspectual categories. A cognitively motivated linguistic category is a category that plays a role in language processing, and in the construction and storage of mental representations. More specifically, the experimental manipulation of a cognitively motivated category produces an effect — that is, a change in the participant’s behaviour — which is observable and measureable (Rossi 1997). Observable measures such as reaction times, reading times, answers to questionnaires, pragmatic or grammatical judgments, the choice of an image, eye movements, etc., often indicate the cognitive processes at work as utterances are dealt with.

In Chap. 4, I discussed a series of offline experiments in which native speakers were asked to judge the meaning of these categories. Cross-linguistically, the results have shown that participants were able to evaluate and judge their meanings consciously, but the rate of inter-annotator agreement varied as a function of the type of encoded meaning (that is, procedural or conceptual). The experiments from Sect. 4.2 indicated that Reichenbachian coordinates accurately describe the meaning of Tense and its functions at the conceptual and procedural levels. In particular, the results of these experiments clearly showed that two systematic patterns arise when participants are asked consciously to evaluate the contribution of verbal tenses to the interpretative process. The first is the ease of the task and the high rate of inter-annotator agreement when dealing with the past/non-past distinction. The second is the greater difficulty of the task and the lower rates when dealing with the temporal ordering of eventualities. As noted in Sect. 4.1, based on Wilson and Sperber’s (1993, 2012) cognitive foundations of the conceptual/procedural distinction, these two patterns are explained in terms of Tense encoding both conceptual information (the past/non-past distinction via the E/S configuration) and procedural information (the localization of eventualities with respect to one another, making use of the R coordinate, and corresponding to the temporal relations holding between eventualities) (cf. detailed discussion in Sect. 5.1). Similar patterns were found when participants dealt with aspectual information. The experiments from Sect. 4.3 revealed that aspectual information related to the actual realization of Aktionsart — that is, boundedness — is easily accessible to consciousness, and results in high levels of inter-annotator agreement . In contrast, consciously identifying grammatical perfective or imperfective viewpoint is a more difficult task, and results in lower levels of inter-annotator agreement. As with Tense, based on Wilson and Sperber’s (1993, 2012) cognitive foundations of the conceptual/procedural distinction, I have argued in Sect. 5.1 that the categories of Aktionsart and Aspect are encoded at the conceptual and procedural levels of language meaning respectively. During the comprehension process, Aspect imposes constraints on Aktionsart: the conceptual representations of eventualities are viewed from the speaker’s point of view as completed or in progress.

Furthermore, a considerable number of studies in psychology, psycholinguistics and neurolinguistics have shown that Tense, Aspect and Aktionsart have an impact at the cognitive level. Research has shown that these categories are processed online, that they determine the construction of the ongoing and subsequent mental representations , that they influence perception and the memory of events, that they bias the interpretation of a series of events, and that they become dysfunctional in cases of brain damage (for example, Radvansky et al. 1998; Todorova et al. 2000; de Vega et al. 2004; Therriault and Raney 2007 and Dery and Koenig 2015 for Aktionsart; Carreiras et al. 1997; Magliano and Schleich 2000; Stavrakaki and Kouvava 2003; Rohde et al. 2006; Pickering et al. 2006; Ferretti et al. 2009; Madden and Ferretti 2009 and Mozuraitis et al. 2013 for Aspect; Mandler 1986; Segal et al. 1991; Murray 1997; Gibbs and Moise 1997; Radvansky et al. 1998; Bastiaanse 2008 and Bastiaanse et al. 2011 for Tense, regarding the localization of eventualities with respect to S and to one another).

These studies argue that, during comprehension, hearers build mental models of situations (Johnson-Laird 1983; cf. the discussion in Radvansky et al. 1998; Zwaan and Radvansky 1998) exploiting linguistic, pragmatic and general world knowledge (Glenberg et al. 1987). These mental models of situations are simple and multithreaded mental representations of situations described in a discourse. Mental models have a series of properties. First, they are multidimensional, in that they are temporal, spatial and referential. Second, they are coherent: each mental representation is integrated with the previous one, allowing the hearer to draw temporal and causal inferences, among others (Givón 1995; Graesser et al. 1997). Third, mental models are dynamic: they are updated and adjusted when necessary, depending on the new information processed during the hearing of auditory stimuli or the reading of written stimuli. Fourth, mental representations are stored in the memory, and accessed at a later point when they are needed. In this model, language is seen as encoding processing instructions on how to construct mental representations of the situations described (Zwaan and Radvansky 1998).

Several researchers have shown that Aspect constrains the construction of mental representations of situations in several ways. Firstly, Magliano and Schleich (2000) tested the influence of grammatical aspect on the interpretation of a series of situations: English native speakers read stories in which the target eventuality was expressed with the progressive, such as was changing a tire or, with the perfective, changed a tire. This target eventuality was followed by three other eventualities, which could be understood as taking place either during or after the target situation. The results indicated that eventualities expressed by the imperfective aspect are understood as ongoing at the moment of speech, whereas eventualities expressed by the perfective aspect are understood as completed. Magliano and Schleich have also found that general world knowledge about situations, such as their duration, interacts with the information from Aspect, and in particular with the imperfective aspect. Their experiments revealed that situations with a long duration, such as writing a novel, are more frequently understood as ongoing at the moment of speech than short situations, such as write a letter. This effect is observed later in the story, and thus not immediately after the target situation. In other words, the effect of a situation’s duration on the interpretation of a series of situations is visible later during processing. In addition, the influence of the imperfective aspect on the duration of situations persists longer in memory than that of the perfective aspect. These observations indicate that comprehenders take into account the various types of information they receive, and build a coherent multithreaded structure of mental representations of situations.

Secondly, it has been shown that the imperfective aspect influences the activation of information stored in the working memory: situations expressed using the imperfective are more active and more accessible than those expressed using the perfective (Magliano and Schleich 2000). This also applies to the accessibility of people or characters in a story, entities, instruments, locations or various characteristics of situations (Carreiras et al. 1997; Madden and Zwaan 2003; Ferretti et al. 2007). Thirdly, Aspect also represents a linguistic cue regarding the expectations that comprehenders have for the continuation of a story. This was found to be true in relation to several phenomena, such as reference and coreference resolution (Rohde et al. 2006; Ferretti et al. 2009), and relative clause processing (Rohde et al. 2011). For example, participants in Rohde et al.’s study (2006) read sentences that included verbs of transfer presented in either their perfective or imperfective form, followed by ambiguous pronouns that could refer either to the Source or the Goal referent, as in example (606). The results demonstrated that participants proposed a significantly higher number of Goal continuations after a sentence including the perfective aspect than after one with the imperfective. This effect was confirmed by Kehler et al. (2008), who even found it with sentences that did not provide the ambiguous pronoun, as shown in (607).

(606)	John SOURCE handed/was handing a book to Bob GOAL. He ……….
(607)	John SOURCE handed/was handing a book to Bob GOAL.

Other studies have explored the role of aspectual classes in constructing mental representations of situations, as well as the interaction between Aktionsart and Aspect taking the form of aspectual coercion. For example, Piñango et al. (1999, 2006) and Todorova et al. (2000) demonstrated that coercion is cognitively costlier than the construction of a mental representation of a situation whose inherent temporal information is compatible with the constraints imposed by Aspect. For example, the verb hop in (608), which is ontologically an achievement, is not directly compatible with an adverb like until. The iterative interpretation is built through coercion. In contrast, no coercion is required for the activity expressed by the verb glide in (609) to occur with the adverb until.

(608)	The insect hopped effortlessly until it reached the far end of the garden that was hidden in the shade.
(609)	The insect glided effortlessly until it reached the far end of the garden that was hidden in the shade.

Piñango et al. (1999, 2006) had participants listen to sentences which either needed or did not need coercion, as in (608) and (609) respectively. As participants heard these sentences, they were also required to make a lexical decision about a word presented on the screen. This second task was supposed to compete with the primary comprehension task. They found significantly longer lexical decision times when the concomitant sentence required aspectual coercion than when it did not. Todorova et al. (2000) also found that it is aspectual coercion which is cognitively costly, and not the iterative interpretation of achievements such as hopped.

Other studies have found that comprehenders are particularly attentive to fine-grained ontological properties of eventualities, such as duration, telicity and boundedness . For example, Therriault and Raney (2007) examined how duration-related inconsistencies influenced processing time and processing strategies set up by comprehenders when building multithreaded mental representations of the situations described in a narrative text. They found that readers encode the durations of events online, and regularly monitor them. They are able to detect temporal inconsistencies between the expected and given durations, such as Sally brushed her teeth for 3 min compared to Sally brushed her teeth for 30 min. Processing these types of temporal inconsistencies is cognitively costlier than processing situations consistent with the expected duration. In another study, Yap et al. (2009) demonstrated that compatibility between telic eventualities expressed using the perfective aspect, and between atelic eventualities expressed using the imperfective aspect, facilitates processing in terms of reading times and accuracy of answers. More recently, Dery and Koenig (2015) explored the roles of boundedness and event complexity in determining the temporal relations holding between eventualities, corresponding to what they call the temporal update of mental representations of situations. They found that bounded eventualities more frequently trigger sequential temporal relations than unbounded ones (Magliano and Schleich 2000; Madden and Zwaan 2003). Additionally, they tested Dowty’s (1986) hypothesis, according to which a series of two states will be interpreted simultaneously rather than sequentially (cf. Sect. 2.1). They demonstrated that Dowty’s hypothesis is too coarse-grained to be accurate, and propose that finer-grained distinctions, such as temporary vs. permanent states, are necessary in order to investigate the role of Aktionsart in the expression of temporal relations. Indeed, they found that temporary states are much more likely to trigger sequential temporal relations than permanent states do. For them, this is due to the fact that temporary states are more easily represented as bounded than permanent states are.

The cognitive foundation of the category Tense has also received the attention of scholars, who have mainly focused on two lines of research. The first is the localization of eventualities with respect to the moment of speech, and the second is the localization of eventualities with respect to one another. Sections 5.2.3 and 6.2 were dedicated to the latter, and showed that this information plays a role in language processing, and in the construction and storage of the mental representations of situations. When comprehenders of various languages are asked to consciously evaluate this information, they systematically have difficulty carrying out this task. Nevertheless, they succeed, with inter-annotator agreement rates above the level of chance. As for processing, it has been shown that fine-grained distinctions between chronological and anti-chronological sequential relations are cognitively relevant. Specifically, chronological relations are cognitively less costly than anti-chronological ones (Mandler 1986; Segal et al. 1991; Murray 1997; Asr and Demberg 2012). Furthermore, it was found that comprehenders prefer overtly marked chronological sequential relations in offline evaluation tasks, but no difference was found between implicit and explicit relations in terms of processing (Grisot and Blochowiak 2015, 2017).

As for the localization of eventualities with respect to S (that is, the past/non-past distinction), several studies have shown that reference to past time using grammatical morphology is severely impaired in agrammatic aphasia,² whereas reference to present and future are spared by comparison. Several explanations have been suggested for this phenomenon, such as the Tree-Pruning Hypothesis (Friedmann and Grodzinsky 1997; Friedmann 2008), which makes use of a syntactic hierarchy of inflection nodes, the Impaired Interpretable Features model (Nanousi et al. 2006; Varlokosta et al. 2006), which is based on Chomsky’s (1995) distinction between the interpretable and uninterpretable features of functional categories, the Tense Underspecification Hypothesis (Wenzlaff and Clahsen 2004, 2005), and the PAst DIscourse LIinking Hypothesis (PADILIH) (Bastiaanse 2008; Bastiaanse et al. 2011). The neurolinguistic PADILIH model builds on previous analyses of Tense as an anaphoric device (Kamp 1979; Hinrichs 1986; Kamp and Rohrer 1983, Partee 1973, 1984; Nerbonne 1986; Webber 1988; and also, from a syntactic approach, Avrutin 2000, 2006; Zagona 2003, 2013). Zagona (2003) suggested that reference to present time should be considered as a kind of ‘binding relation’, based on the fact that temporal coordinates S, R and E are simultaneous. In the case of reference to past time, on the other hand, temporal coordinates do not coincide. Zagona argues in favour of a discourse linking relation between S, R and E regarding reference to past time. As far as reference to the future is concerned, Zagona (2013) argued that it is a subclass of the present, and therefore not discourse linked. Based on a series of experiments, Bastiaanse (2008) and Bastiaanse et al. (2011) observed that reference to the past is discourse linked not only when expressed by Tense but by periphrastic verb forms (‘has walked’) as well. Their suggestion is that reference to past time by verb inflection generally requires discourse linking, and is expected to be impaired in agrammatic speakers, due to the more complex forms. It has been shown (Faroqi-Shah and Dickey 2009) that reference to the past by verb inflection produces longer reaction times than verb forms referring to the present.

Further evidence for the discourse linking nature of past reference comes from event-related brain potential (ERP) and behavioural (reaction time and acceptability rating) data, from Dragoy et al.’s study (2012). Their study was designed to focus on the processing of time reference violations in which verbal tenses do not match a time frame previously set by the adverbial: a past time adverbial followed by a present time verbal tense, as in (610), and a present time adverbial followed by a past time verbal tense, as in (611).

(610)	De kelner	die zonet	de peper	malt	krijgt geen fooi.
	The waiter who just before the pepper grind.PRES gets				no	tip.
	*‘The waiter who is just before grinding the pepper doesn’t get a tip.’
(611)	De kelner	die nu	de peper	malde	krijgt geen fooi.
	The waiter who now the peper grind.PAST gets				no	tip.
	*‘The waiter who now ground the pepper doesn’t get a tip’.

Dragoy et al.’s research aimed to develop Baggio’s (2008) findings on the link between temporal and pronominal reference. Baggio’s study proved that processing present time reference marked on the verb in a past time reference context is accompanied by the same ERP effects as processing locally bound pronouns. Consequently, Dragoy and colleagues designed a study of the processing of past and present tense in incongruous contexts, hypothesizing that they rely on different neural processes. They investigated three types of measures: evoked brain responses (ERP); reaction times; and acceptability judgments. Brain responses evoked by time reference violations were explored according to several measures:

• P600 wave produced by the brain when it detects a morphosyntactic locally bound anomaly (usually 600 ms after the target word onset).

• N400 wave produced by the brain when it detects a lexical, semantic or conceptual anomaly (usually 400 ms after the target word onset).

• Left Anterior Negativity (LAN) wave produced by the brain when it processes a rule-governed compositional parsing of complex forms across linguistic domains, including both morphology and syntax (usually occurring during 300–500 ms after the target word onset).

• Numerous negative waves (other than N400) produced by the brain when it has difficulty finding a discourse-linked referent (for expressions such as ambiguous words, pronouns).

In terms of the results of Dragoy et al.’s study, the main findings can be summarized as follows. The analysis of the ERP data supports the idea that distinct neural areas process references to past and to present time, as signalled by different brain reaction patterns. The processing of a past time context disrupted by a present tense verb produced a P600 response triggered by the targeted verb.³ In contrast, the processing of present time context disrupted by a past time verb did not produce an immediate brain response. However, both past and present time reference produced sentence final negativity, which is a typical response to referential violations in general. Moreover, this ERP data is linked to behavioural data. Investigation of reaction times shows that present time reference violations by past tense verbs were detected later than past time references violated by present tense verbs, which produced an immediate P600 response. Furthermore, the acceptability rating showed that relative clauses with an adverb referring to the present and a verb referring to the past are considered less unacceptable than sentences with a past time context disrupted by a present tense verb. When a continuation of the relative clause is provided, participants find it easier to coerce the present time adverbial/past tense verb combination into a meaningful sentence than the past time adverbial/present tense verb combination. Dragoy and colleagues interpret the participants’ willingness to wait for further contextual information before judging the relative clause as unacceptable to be an indication of the discourse-linking view of past tense processing. They point out that participants notice the violation of the present/past context with a past/present tense verb, but respond to it in a qualitatively different manner. This response is showed by the negativity wave in ERP elicited by the end of the sentences.

Dragoy et al.’s study provided new evidence for the theoretical suggestion that time reference expressed by verbal inflections involves processing similar to pronominal reference (Partee 1973; Webber 1988), and that past time and present time reference involve different neural processes, a dissociation observed in both healthy and aphasic participants (Bastiaanse 2008; Bastiaanse et al. 2011; Faroqi-Shah and Thompson 2007). Moreover, this study supports Zagona’s (2003) suggestion that present tense processing requires the establishment of bound co-reference with the speech time (local binding, i.e. the present tense is deictic), while past tense processing requires the establishment of co-reference with another event time (discourse-linking, i.e. the past tense is anaphoric).

An important question that arises at this point of the discussion is whether these patterns of processing past and present time reference are directly linked to the processing of Tense, or are independent, and can therefore be observed in tenseless languages. Qiu and Zhou (2012) and Bastiaanse et al. (2011) investigated this question. Qiu and Zhou (2012) designed a study with features like those of Dragoy et al. (2012); they investigated brain responses to disagreements between a temporal context set by a temporal adverbial⁴ (jiangyao ‘to be going to’ for future time reference, and cengjing ‘in the past’ for past) or by the aspectual particle guo and temporal noun phrases, as in examples (613), (615) and (617).

(612)	Next month the United Nations V + jiangyao/ will dispatch a special investigation team.
(613)	*Last month the United Nations V + jiangyao/ will dispatch a special investigation team.
(614)	Last month the United Nations V + ceinging/dispatched a special investigation team.
(615)	*Next month the United Nations V + ceinging/dispatched a special investigation team.
(616)	Last month the United Nations V-guo/dispatched a special investigation team.
(617)	*Next month the United Nations V-guo/dispatched a special investigation team.

Temporal marking in Chinese has to rely on either lexical semantics and discourse principles (in the case of temporal adverbials) or morphosyntactic processing (e.g. suffixation of verbs by the aspectual particle –guo). The authors found similar patterns for time reference disruptions in Chinese as those found for tensed European languages. Disagreements between noun phrases and temporal adverbials or the aspectual particle produced a P600 wave signalling the morphosyntactic violation, and an additional N400 wave only for the temporal adverbials, due to their lexical nature. Moreover, a sustained negativity effect was found after the targeted words and the final words for all types of temporal markers, interpreted as the brain’s attempt to correct errors and create a coherent representation of the sentence.

Bastiaanse et al. (2011) also argued that impairments regarding reference to past time occur not only in the Tense morphology of tensed languages but also in tenseless languages, such as Chinese . Bastiaanse and colleagues designed a study where three typologically different languages were compared (Chinese, which expresses time reference by aspectual information; Turkish, which has very complex verb inflection paradigms; and English, which has a combination of free and bound morphemes), testing reference to past, present and future time. They used sentence production tasks and comprehension assessments, and tested healthy and Broca’s aphasia patients. The healthy speakers from the control group all attained the maximum possible score (i.e. normal scores). Their findings regarding the production and comprehension of aphasic patients can be summarized as follows:

Firstly, in all languages, the agrammatic speakers were impaired when producing the grammatical forms for reference to the past. English and Turkish speakers performed significantly worse for past than for present and future reference. Chinese speakers performed well for sentences which did not require a specific time reference, but poorly for past, present and future reference. The authors assume that this is due to the fact that aspectual adverbs are not obligatory, unlike English and Turkish verb inflection. A qualitative analysis of Chinese production shows that the aspectual adverb was most often omitted (the sentence remaining grammatical when a lexical adverb expressing the time frame also occurs); when included, the past (le) and future (yao) adverbs were substituted by the present marker zai.

Secondly, as far as production is concerned, there was no significant difference between the two tensed languages: speakers performed similarly, regardless of the complex verb inflection paradigm in Turkish and the use of periphrastic forms in English. Finally, in all languages, the agrammatic patients were impaired in terms of comprehension of sentences containing reference to the past. This was significantly worse than comprehension of sentences containing present time reference. The comprehension of future morphology is significantly worse than present time reference, but better than past time reference, for all patients (though for the Turkish patients, the difference between past and future was quite close to the significance threshold).

As such, Bastiaanse et al. (2011) showed that agrammatic speakers exhibit performance patterns that are the same for all three languages: past time reference is more impaired than present reference; past reference is as impaired as — or more impaired than —future reference; and future reference is more impaired than present reference. They suggest that these data should be interpreted at the morpho-semantic interface: temporal information about the event moment relative to the moment of speech must be encoded (in production) and decoded (in comprehension) grammatically. In other words, temporal localization by the relation of E to S is encoded information both in tensed and tenseless languages. Other studies using aphasiological data indicate that not only is past time reference worse than present time reference, but perfective aspect is also more impaired than imperfective aspect in agrammatic aphasia (Nanousi et al. 2006; Stavrakaki and Kouvava 2003). These studies suggest that the aphasia-related impairment of grammatical expressions for past and/or event completeness occurs regardless of the category conveying the temporal localization of eventualities (Tense and/or Aspect), and of the type of linguistic expressions (inflexions, auxiliaries, free or bounded morphemes), as Dragoy and Bastiaanse point out (2013, 114).

Additionally, studies in neurolinguistics have also pointed to similar deficits linked to Aspect and its interaction with Tense. Dragoy and Bastiaanse’s (2013) study of Russian aphasic patients investigated the hypothesis suggested in Bastiaanse et al. (2011) that verb forms expressing reference to past time or conveying perfective semantics are more impaired than verb forms expressing reference to the non-past or conveying imperfective semantics, for both production and comprehension. Dragoy and Bastiaanse (2013) point out that Tense, Aspect and Aktionsart deeply intertwined in Russian in Russian. They note that Russian children strongly prefer to use perfectives to refer to past time, and imperfectives to refer to the present, as suggested by Gagarina (2004).

Dragoy and Bastiaanse tested the following hypotheses: (i) past forms are more impaired than present forms (according to PADILIH); (ii) the production of perfective verbs is more impaired than that of imperfective verbs; and (iii) due to the interaction between time reference and aspect in Russian, non-past time reference is advantageous only for imperfective verbs, and past-time reference for perfective ones. They used sentence completion tasks and tested aphasic patients (both fluent/Wernicke’s aphasia and non-fluent/Broca’s aphasia). Their results showed a significant main effect of temporal reference (as predicted by PADILIH). There was no significant effect for Aspect alone (invalidating the second hypothesis), but there was a significant interaction of Tense and Aspect. In particular, reference to the non-past is better preserved than reference to the past, but only for imperfective verbs; in contrast, for perfective verbs, reference to the past is better preserved than reference to the non-past.

6.4.2 Coherent Mental Representations

As I will show below, Kintsch and colleagues’ psycholinguistic model of discourse comprehension (Kintsch and van Dijk 1978; van Dijk and Kintsch 1983; Kintsch 1995, 2005) fits perfectly with the relevance-theoretic cognitive pragmatics approach to language comprehension (Sperber and Wilson 1986, 1998; Wilson and Sperber 1993, 2004; Blakemore 1987, 2002; Escandell-Vidal et al. 2011; Wilson 2011). According to both of these approaches, the comprehender’s task can be considered as constructing a mental representation of the information provided by the speaker, using oral or written types of verbal communication. This mental representation is integrated with the comprehender’s existing knowledge, beliefs and intentions. Kintsch and colleagues make the distinction between the mental representation of the text itself (named a textbase) and the representation of the situation described by the text and integrated into the comprehender’s previous knowledge (named a situation model ). In this research, I am particularly interested in the latter, and I will argue that temporality is one of the dimensions of the situation model that the hearer monitors and shapes in a coherent manner during comprehension, making use of various linguistic and world knowledge cues.

For both the psycholinguistic and relevance-theoretic approaches, the mental representation built consists of a series of propositions consisting of conceptual representations (Sperber and Wilson 1998) forming an interrelated network,⁵ and which are manipulated in accordance to the procedural cues provided by linguistic expressions, such as pragmatic and logical connectives, the grammatical categories of tense, aspect and mood, modality, evidentiality and referential expressions, among others (Givón 1989, 1995; Blakemore 1987, 2002; cf. Escandell-Vidal et al. 2011). These processing instructions are useful for the construction of a coherent mental representation of a text. For example: referential expressions and pronouns serve to construct a referentially coherent representation; causal connectives such as because, thus and so, causal conceptual rules such as push-fall, and causal reasoning⁶ serve to construct a causally coherent representation; and the categories of Tense, Aspect and Aktionsart, as well as temporal connectives such as then , before and when, serve to construct a temporally coherent mental representation of a text.

According to the HD model of temporal reference , described in Sect. 5.1, the category of Tense both contributes to and constrains the construction of mental representations. On the one hand, by way of its conceptual content (that is, the localization of eventualities in the past vs. non-past) it contributes to constructing the conceptual mental representations; on the other hand, by way of its procedural content (that is, the localization of eventualities with respect to one another), it constrains the manipulation of conceptual mental representation by instructing to the comprehender to determine the exact relation (chronological sequential, anti-chronological sequential, simultaneous or indeterminate). Furthermore, aspectual information from Aktionsart and Aspect also contributes to and constrains the construction of mental representations respectively: Aktionsart provides the type of eventuality to be included in the conceptual mental representation (state, activity, achievement or accomplishment), whereas Aspect constrains this process by instructing the comprehender to represent the eventuality as completed or in progress. Finally, the speaker may choose to use temporal connectives to mark overtly the temporal relation he intends to establish between the mental representations of eventualities; he may also, however, choose to communicate the temporal relations implicitly. In the latter case, the hearer will have to infer the relation according to other linguistic cues and world knowledge.

Phenomena like aspectual coercion and certain usages of verbal tenses, such as the futural Passé Composé, the historical present or the narrative Imparfait (cf. Sects. 1.1 and 2.3), clearly indicate that comprehenders deal with apparent linguistic inconsistencies in a coherent manner, by deriving less frequent but completely plausible interpretations. Comprehenders make the effort to resolve apparent inconsistencies because of the presumption of relevance of utterances. As demonstrated by Yap et al. (2009), comprehenders prefer (in terms of processing effort) compatible co-occurrences (for example, between the temporal adverbial and the verbal tense, or between the telicity status of a situation and the grammatical aspect) . However, they are able to interpret incompatible co-occurrences, albeit at a higher cognitive cost, which might be explained by their need to establish temporal coherence between the pieces of information provided by linguistic cues.

So, my proposal is that conceptual and procedural types of information are key notions for successful language comprehension, which requires the recognition of the speaker’s intended meaning and its coherent representation at the cognitive level. As Hobbs (1979) and Sperber and Wilson (1986); Wilson and Sperber (2004) argue, the speaker aims to have the hearer understand him, and the hearer aims to identify the speaker’s intended meaning. Coherence ties that may encode conceptual or procedural information — or both of them — are cues which will help the speaker and the hearer in their tasks. Of course, the cues which will direct the hearer towards the intended meaning are selected by their contextual relevance (as it is understood in the relevance-theoretic framework; cf. Sect. 2.3.1). In order to acquire coherence at the mental level, these linguistic cues must be cognitively motivated. As I argued in Sects. 6.2 and 6.4.1, temporal categories and temporal relations have cognitive foundations, and are therefore pointers towards cognitive temporal coherence.

6.5 Summary

In this chapter, I have tackled the notion of temporal coherence, and more specifically of cognitive temporal coherence. To do so, I have given arguments in favour of the cognitive status of coherence relations in general, and of temporal relations in particular. Building on Hobb’s (1979, 1985), Sanders et al.’s (1992, 1993) and Evers-Vermeul et al.’s (2017) cognitive approach to discourse relations , I have argued that temporal relations are cognitively motivated for two reasons. The first is because they affect processing and language acquisition. The second is because the linguistic categories triggering them (Tense, Aspect and Aktionsart), along with temporal connectives and temporal adverbials, are themselves cognitively motivated, as has been found by numerous experimental studies carried out in psychology and neurolinguistics .

I discussed three online and two offline experiments, assessing the role of ensuite in the expression of sequential and indeterminate temporal relations holding between events expressed with the Passé Composé and Passé Simple, the role of puis in the expression of sequential relations, and the occurrence of these connectives with these verbal tenses (Grisot and Blochowiak 2015, 2017). Differences and similarities were found between online processing and offline acceptability judgments. Firstly, the participants’ preference for implicit relations in the offline task does not seem to translate into a facilitation of the processing of sentences which are not linked with a temporal connective . Secondly, the lack of main effect of the verbal tense seems to be common to both processing and offline acceptability judgments. As such, there is evidence for the cognitive status of temporal relations, and for the fine-grained distinction between chronological, anti-chronological, synchronous and undetermined types of configuration. Nevertheless, further research is required in order to explore the rich interrelations between the various temporal expressions, as well as to determine the exact role for processing language of temporality as a cognitive principle, the causality-by-default hypothesis, the continuity hypothesis, and comprehenders’ expectations during text comprehension.

The notion of cognitive temporal coherence was principally linked to the coherence established in the multithreaded mental representations that we build during language comprehension (Gernsbacher and Givón 195; Graesser et al. 1997). In this model, language is seen as encoding processing instructions on how to construct mental representations of the situations described (Zwaan and Radvansky 1998). According to the HD model of temporal reference , described in Sect. 5.1, the category of Tense both contributes to and constrains the construction of mental representations. Furthermore, aspectual information from Aktionsart and Aspect also contributes to and constrains the construction of mental representations respectively: Aktionsart provides the type of eventuality to be included in the conceptual mental representation (state, activity, achievement or accomplishment), whereas Aspect constrains this process by instructing the comprehender to represent the eventuality as completed or in progress. In this research, the role of the [±narrativity] and the [±boundedness] features in processing and translating temporal information was also tested, using automatic tools and machine translation systems. I will discuss this application in the next chapter.