Time is an unbounded dimension. To explore time, people usually rely on a reference point or an anchor for location such as space, albeit abstractly [1,2,3]. Languages often employ two devices to locate situations in time: the grammatical device tense and the lexical device temporal connectives [4]. Temporal connectives play a pivotal role in marking temporal location, particularly in tenseless languages which lack overt inflectional marking of tense. Mandarin Chinese, for instance, is a tenseless language which does not mark past, present, or future with dedicated morphemes, yet Mandarin speakers successfully comprehend temporal information, largely depending on temporal adverbials (e.g., yesterday, last week), viewpoint aspect morphemes (e.g., ‘le’ termination or completion, ‘zài’ ongoing state), and temporal connectives (e.g., ‘zhiqian’ before, ‘zhihou’ after [5,6,7]). In M^{[1][2][3][4][5][6][7][8][9][10][11][12][13][14][15][16][17][18][19][20][21][22][23][24][25][26][27][28][29][30][31][32][33]}andarin, the temporal connectives zhiqian (before) and zhihou (after) involve the selection of an event over a period of time, which provides important cues to set a time frame [8]. These temporal connectives have been demonstrated to be more difficult to understand than other connectives (e.g., causal connective because) [9], yet it remains unclear whether these temporal connectives are processed differently. Although a number of studies have investigated how sentence comprehension is modulated by temporal connectives such as before and after [10,11,12,13,14,15,16], they have mainly focused on tense languages such as English. It thus remains unclear how temporal connectives are processed in tenseless languages such as Mandarin Chinese.

The temporal connectives before and after are different in expressing iconicity. Linguistically, iconicity is the conceived similarity between the linguistic form and the meaning in the real world [17]. In other words, if the sequencing of clauses described in a sentence is consistent with the sequencing of events occurring in the real world, it is viewed as iconic; otherwise, it is viewed as non-iconic. Suppose p occurs before q: this temporal sequence can be described by an after construction [After p, q] in which the temporal order of the event is consistent with the chronological order, and therefore is regarded as iconic. The same event sequence, however, can also be described by a before construction [Before q, p], in which the sequence of the described clauses is inconsistent with the sequence of occurrence in the real world. It is therefore considered to be non-iconic. Similarly, in a construction in which the temporal connective occurs in the sentence-middle position, [p before q] is iconic whereas [q after p] is non-iconic (see Table 1). Previous studies adopting various approaches have shown that a non-iconic sequence is more difficult to understand than an iconic sequence. For example, Mandler [11] found that participants spent more time reading sentences when the order of occurrence mismatched the order in which they were presented. Münte et al. [12], using the event-related potentials (ERPs) technique, observed a larger sustained negative ERP modulation for non-chronological before-sentences (1a) than chronological after-sentences (1b) over the left frontal scalp. Interestingly, the ERP difference was significantly correlated with a working memory score. Individuals with higher working memory span showed a more pronounced negative difference between before and after sentences, suggesting that computing a non-iconic temporal relation places high demands on the working memory system. Ye et al. [16] found that the caudate nucleus and middle frontal gyrus were more strongly activated for processing before-clauses than after-clauses. Moreover, another line of studies has revealed that sentence-initial sentences headed by before are more difficult to understand than those headed by after for young children [18] as well as Parkinson patients [19,20].

The comprehension of before and after could also be influenced by their position in a sentence. Politzer-Ahles et al. [14] compared the neural correlates of processing before and after in both sentence-initial (e.g., 1a/b) and sentence-middle positions (e.g., 2a/b). They found that before elicited more negative ERP responses than after in the sentence-initial position but more positive ones in the sentence-middle position. The result in the sentence-initial context is consistent with previous studies (e.g., [12]), but the sentence-middle result shows the opposite, indicating that ERP modulations elicited by temporal connectives are related to the order in which events are presented.

The studies addressed so far seem to support the idea that a non-iconic temporal relation is semantically/pragmatically more costly to understand than iconic temporal relation (regardless of connective type). However, despite the difference in expression iconicity, temporal connectives also differ in frequency and polarity, and the explanation based on frequency and polarity will have a different prediction for processing temporal connectives such as before and after.

According to the polarity-based explanation, before shares a similar meaning with in front of in the pair in front of–behind, in which in front of is positive as it is used to describe the visible area of perception, whereas its counterpart behind is negative as it describes the invisible area [21,22]. Linguists therefore distinguish them by labeling before as positive and after as negative. Indeed, psycholinguistic evidence has demonstrated that people tend to acquire positive words such as before more easily than negative words such as after [23]. In an early behavioral study, Clark [10] found that children had greater difficulty comprehending after-clauses than before-clauses.

With regard to frequency of use as temporal connectives, after is used less often than before in both adults and children’s language usage [24,25,26]. Besides functioning as a temporal connective, after is more frequently used as a preposition (like behind), and therefore has a less consistent mapping between form and meaning as compared to before in terms of temporal connective. This variability may lead to greater uncertainty in interpreting after sentences compared to before sentences, meaning that understanding after sentences could be cognitively more demanding.

Notably, there are cross-linguistic differences in expressing iconicity with temporal connectives. For example, the relationship between temporal connectives and iconicity in Mandarin Chinese is much clearer than in English, because in Mandarin, temporal connectives zhiqian (before) and zhihou (after) only occur at the non-initial position of a sentence. In this case, zhiqian generally expresses a non-chronological temporal relation whereas zhihou expresses a chronological temporal relation. This unambiguity provides a better opportunity to test the iconicity hypothesis.

In summary, regarding the different properties of before and after and especially their discrepancies between English and Chinese, it remains uncertain which of the two temporal connectives is cognitively more difficult to process. If iconicity is the most relevant factor for processing, then before should be more cognitively demanding to process than after. If, however, the value of polarity and form–meaning mapping (frequency) is more important, then after should be more cognitively demanding than before.

During language comprehension, readers not only rely on semantic and syntactic knowledge, but also depend on world knowledge to establish a coherent sentence/discourse model [27,28,29,30,31,32]. If the linguistic input is inconsistent with world knowledge, a negative-going brain response, appearing around 300–500 ms (i.e., N400), will be regularly elicited [27,28,29,30,31]. However, whether a semantically reasonable sentence is pragmatically valid could be constrained by the coherent markers, namely connectives. A number of recent studies have explored this issue. Using event-related potentials (ERPs), Xu et al. [31] investigated how world knowledge incongruency was modulated by causal and concessive connectives (‘yinwei’ because vs. ‘jinguan’ although). They found that causal and concessive connectives had different influences on pragmatic processing at different stages of sentence processing. During early retrieval processing, world knowledge anomalies (e.g., Harbin is warm in winter) evoked indistinguishable N400 effects in both causal and concessive structures; at the integration stage, however, they showed dissociable ERP effects: while violation of world knowledge elicited a P600 effect in causal structures, a late negativity (N600) effect was obtained in concessive structures.

Only a few studies have explored how different temporal information influences world knowledge processing. Rinck et al. [15] presented seven-sentence passages to their participants. In each passage, the temporal information conveyed at target sentence (e.g., Claudia was already waiting for him when he got off the train with his huge bag) was either consistent (e.g., Markus’s train arrived at Dresden Central Station 20 min after Claudia’s train.) or inconsistent (e.g., Markus’s train arrived at Dresden Central Station 20 min before Claudia’s train.) with the contextual sentences. The results showed that inconsistent sentences were processed longer than consistent ones. In an ERP study, Xiang et al. [33] presented not only ad hoc stimuli with ambiguous veridicality (similar to 2a/b in Münte et al. [12], but also real-world stimuli (e.g., Before/After Tiger Woods won the Masters, Jake started playing golf) which are veridical based on world knowledge. The result showed that before-sentences evoked a larger sustained negativity than after-sentences in ad hoc stimuli, but no effect was observed in real-event stimuli, suggesting that the difference underlying processing of these two types of temporal relations could be related to the uncertainty of the events. Nieuwland [13], using ERP techniques, also investigated how world knowledge was processed in before vs. after sentences.

The results demonstrated that while the world knowledge-inconsistent words (3b) elicited a larger N400 than the world knowledge-consistent words (3a) in after sentences, the N400 difference effect was largely attenuated in before sentences (3c/d). However, at the late time window, the inconsistent words (3d) elicited a larger P600 than the consistent words (3c) in before sentences. These results indicate that integrating mismatched pragmatic information into discourse could be delayed in a non-chronological temporal sequence in comparison to a chronological sequence.