Understanding Temporal Relations in Mandarin Chinese: An ERP Investigation

Temporal connectives play a crucial role in marking the sequence of events during language comprehension. Although existing studies have shown that sentence comprehension can be modulated by temporal connectives, they have mainly focused on languages with grammatical tense such as English. It thus remains unclear how temporal information is processed in tenseless languages. The present study used event-related potentials (ERPs) to examine how world knowledge is retrieved and integrated in sentences linked by zhiqian (before) and zhihou (after) in Mandarin Chinese (e.g., After/Before going to the countryside, Grandpa went to the city because the air there was fresh and pure). The critical words (e.g., fresh) were either congruent or incongruent with world knowledge. Relative to the after-congruent sentences, the after-incongruent sentences evoked a P600 on critical words and a negativity on sentence-final words, whereas relative to before-congruent sentences, before-incongruent sentences showed no significant difference on critical words but a sustained negativity on sentence-final words. Additionally, before-congruent sentences elicited a larger sustained positivity (P600) than after-congruent sentences. The results suggest that before is more difficult to process than after in Mandarin Chinese, supporting the iconicity account of temporal relations.


Introduction
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 Mandarin, 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 Comprehension of Before and After
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]. Table 1. Before and After in semantic accounts.

Veridicality
Iconicity Polarity p before q non-veridical iconic positive q after p veridical non-iconic negative Before q, p.
non-veridical non-iconic positive After p, q. veridical iconic negative q zhiqian p non-veridical non-iconic positive p zhihou q veridical iconic negative Note: Veridicality refers to the degree to which an experience, perception, or interpretation accurately represents reality.
(1a) Before the scientist submitted the article, the journal changed its policy. (1b) After the scientist submitted the article, the journal changed its policy.
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 sentenceinitial 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.
(2a) The journal changed its policy before the scientist submitted the article. (2b) The journal changed its policy after the scientist submitted the article. 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.

How World Knowledge Is Activated during Temporal Connectives Processing
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 negativegoing 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 Brain Sci. 2022, 12, 474 4 of 17 (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.
(3a) After the global economic crisis, securing a mortgage was harder. (3b) After the global economic crisis, securing a mortgage was easy. (3c) Before the global economic crisis, securing a mortgage was easy. (3d) Before the global economic crisis, securing a mortgage was harder.
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.

The Present Study
In the current study, we aimed to investigate how world knowledge information was retrieved and integrated into clauses which were linked by temporal connectives zhiqian (before) and zhihou (after) in Mandarin Chinese.
Different from English in which temporal information (e.g., past, present, and future) can be expressed by inflectional morphology, temporal information in Chinese is expressed by lexical means such as time adverbials (e.g., yesterday), viewpoint aspect morphemes (e.g., le, zai), and discourse contexts (connectives, before, after). The lack of a tense marking system will influence the perception of temporal sequence in Chinese. In the framework proposed by Reichenbach [34], temporal information in language involves three time points: speech time (the time of the utterance), reference time (the time of perception by the listener), and event time (the time when the event occurs). While processing temporal information, it is necessary to grasp the relationship between speech time and reference time, which is linguistically encoded by tense. In a language which lacks tense marking such as Mandarin, where the relationship between speech time and reference time is usually ambiguous, it tends to be more difficult to comprehend reference time misalignment (RTM). Li et al. [35], for example, found that RTM between subordinate clauses and main clauses in English could be covertly detected by native English speakers even when they were not overtly noticed, as evidenced by an N400 effect, but even highly proficient Chinese-English bilinguals could not detect RTM, and showed no N400 effect.
Although zhiqian and zhihou share the property of their English counterparts before and after in expressing temporal information, they differ in a number of aspects, especially for zhiqian. Zhiqian in its current use still retains ancient characteristics, which is different from before in English. Zhi in ancient Chinese can be interpreted as the structural auxiliary de as in modern Chinese, meaning in front of when it is merged with qian. Zhiqian in its current use has maintained some of the ancient Chinese meaning, which is quite different from before in English. Zhiqian shares the properties of zhi in ancient usage and also retains the possessive property in the temporal proximity, and zhiqian provides the time reference to locate the target event in the recent past, which is close to the speech time [36]. Zhihou has developed maturely in modern Chinese, losing the ancient Chinese boundary proximity meaning of zhi to some degree, and zhihou is commonly regarded as a whole, and provides the time reference to locate the target event in the future, which may be far from the speech time.
As shown in Table 2, the experiment sentences have the structure "NP human goes to place A + before/after + going to place B, because + (the typical feature) + there + [critical word] + [sentence-final word]". As discussed in the previous section, [p zhiqian q] is noniconic because in the sentence (4c), the event 'going to the countryside' happens before the event 'going to the city', and thus the order of real event is contrary to the order of mention. Similarly, [p zhihou q] is iconic in the sentence (4a) because the order of mention is consistent with the real-world event sequence. In addition, le is generally viewed as a perfective marker [37]. Therefore, in the construction "p zhihou q" (4a/b), the event sequence should be (p, q), and p is veridical as le in the clause q denotes that the events (both going to countryside and going to the city) actually happened. q entails p, thus after-sentences are veridical. In contrast, in the construction "p zhiqian q" (4c/d), the event sequence should be (q, p), and q does not entail p (Table 1). Therefore, whether the before-sentences are veridical or not remains ambiguous. Although the locative pronoun nali (there) could be interpreted as referring to a closer or more distant place (Noun), the clauses would be acceptable if because licenses a reason for the movement (e.g., Grandpa liked to go to a place because the air there was pure and fresh). In other words, world knowledge (e.g., the city is usually a place where air pollution is serious, whereas the countryside is normally a place with fresh air.) helps to causally connect the main clause with the subordinate clause. Mandarin speakers have a preference to interpret the locative pronoun as referring to the closer noun in both after and before sentences. According to this preference, sentences 4a and 4c are acceptable because the referential interpretation in these two conditions (i.e., countryside) is consistent with common sense that the air in the countryside is fresher and purer. As for sentences 4b or 4d, however, due to the two places being swapped, the closer referential interpretation of the locative pronoun (i.e., city) is inconsistent with readers' world knowledge. Thus, the experiment had a 2 (connective type: before vs. after) × 2 (world knowledge congruency: congruent vs. incongruent) factorial design with both connective type and world knowledge information being manipulated.
The aim of the present study is to test the neural correlates of processing different temporal connectives and their potential influence on truth value judgments. Based on previous research [12][13][14], we had two lines of predictions. Firstly, if the iconicity account holds in processing Chinese temporal relations as in previous studies (e.g., [12]), a chronological temporal relation should be easier to process than a non-chronological relation. Therefore, pragmatic anomalies in zhihou (after) sentences should be quickly detected and initiate a repair process to re-interpret the incongruency, which will possibly be manifested by an N400 or N400/P600 effect. As for zhiqian (before) sentences, however, due to the additional process of reversing a non-chronological temporal sequence, the comprehension system may not be rapidly sensitive to the anomalies until a later processing stage in which more constraints are available. Under such circumstances, the pragmatic anomalies may induce reduced or even null N400/P600 effects (see [13]). However, according to the polaritybased account and the form-meaning mapping account [24][25][26], if after-sentences are more difficult to process than before-sentences, as observed in previous studies (e.g., [9,26]), then a reverse pattern of ERP modulations should be expected; that is, the above-mentioned ERP modulations should be obtained in before sentences instead of after sentences. In particular, given the important role of working memory in establishing a non-chronological temporal representation (Münte et al. [12]), we might find a significant correlation between the ERP modulations obtained at the connectives and readers' working memory score. Finally, in addition to the critical words, the ERP effects underlying processing different temporal connectives should also be manifested at the connectives [12,14].

Participants
Thirty-two undergraduate or postgraduate native speakers of Mandarin (13 male, mean age = 21.8, SD = 2.92, range 18-26) took part in the ERP experiment. Four participants were excluded from data analysis because of excessive EEG artifacts. All participants were right-handed and had normal or corrected-to-normal vision and none of them had a history of neurological or psychiatric impairments. Each participant gave informed consent before the formal experiment and was compensated for participation.

Design and Materials
One hundred and twenty sets of two-clause sentences were developed (see Table 2 for example stimuli). The main clause described an event in which someone moved to one place before or after moving to another. The subordinate clauses were kept the same across conditions, so the differences occurred only in the main clause where the type of connective and the mention order of the two places were manipulated. Since the locative pronoun is preferentially interpreted as referring to the closer referent, such co-referential relation would cause the critical word to be congruent (the air in the countryside was fresh) or incongruent (the air in the city was fresh) with one's world knowledge in after sentences (4a/b) or before sentences (4c/d).
Before the EEG recording, the selected set of sentences underwent four pretests: a cloze probability test, a sentence rating test and two forced-choice tests. One hundred fifty quadruplets of sentences were developed for the cloze probability test, and 140 quadruplets were left for the other three pretests after removing 10 quadruplets with lower cloze probability scores.
In the cloze probability test, 36 participants were randomly assigned to one of four counterbalanced lists with the last two words truncated (e.g., After going to the city, Grandpa went to the countryside because the air there was . . . ) and were asked to record the first word that came to their mind and to make the completed sentence as natural as possible. Cloze value was calculated as the percentage of the participants who completed the sentences with the intended critical words [38]. Mean cloze probability scores per condition are shown in Table 3. The results revealed a significant main effect of congruency (F(1, 149) = 358.5, p < 0.001) without any interaction or other main effect, demonstrating that the critical words were equally predictable in after and before sentences. In the sentence acceptability test, another group of 31 students were assigned to one of four counterbalanced lists in which they were asked to rate the overall acceptability of the sentences on a 7-point Likert Scale (1 = highly unacceptable, 7 = highly acceptable). Mean scores and standard deviations Brain Sci. 2022, 12, 474 7 of 17 are given in Table 3. Statistical analysis revealed a significant main effect of congruency (F(1140) = 1557.62, p < 0.001) and a connective type × congruency interaction (F(1140) = 5.51, p < 0.05). Follow-up tests showed that the after-congruent sentences were more acceptable than before-congruent sentences (t(140) = 3.05, p < 0.01). As nali is a syntactically unmarked demonstrative pronoun in Chinese, it can refer to either the closer (place B) or the more distant place (place A) in a sentence, depending on the contextual preference [31,39,40]. To this end, two forced choice tests were conducted. The first test presented the complete sentences and 30 participants were then asked to judge whether nali refers to place A, place B or a third place unmentioned in the context. In the second forced choice test, the sentences were truncated after the demonstrative pronoun nali (e.g., after going to the city, Grandpa went to the countryside because there . . . ); another group of 37 participants were asked to select the actual referent of nali among the three potential choices, as carried out in the first forced choice test. As shown in Table 4, the locative pronoun nali was preferentially interpreted as referring to the closer location (place B) throughout the sentences. In the before sentences, the closer referential preference was stronger in the congruent than in the incongruent condition (t(140) = 4.86, p < 0.001), whereas in the after sentences, the closer referential bias was equally strong irrespective of the sentence congruency (85.7% vs. 83.2%, p > 0.5). Interestingly, in the congruent conditions, the closer preference was stronger in the before sentences than in the after sentences (t(140) = 3.12, p < 0.01) when the truncated information was added. On the contrary, in the incongruent conditions, the distant referents became more preferential than the closer ones. In the ERP study, the 120 experimental items selected based on the pretests (60 for each temporal connective type) were distributed across 4 different testing lists according to a Latin square procedure, with each list containing 30 items per condition. In addition, 120 two-clause filler items were interspersed randomly among the 120 experimental items to obscure the purpose of the experiment from participants. The 120 filler items were always consistent with real world knowledge, including 40 two-clause sentences connected by before or after; 40 sentences in which the two clauses were connected by other connectives (e.g., although, therefore), and 40 sentences without connectives. The experimental items were pseudo-randomly mixed with fillers to limit the succession of no more than three sentences of identical type and no more than three sentences had comprehension questions with the identical intended answer (e.g., Does Grandpa like the fresh air? and the participant press the YES button). Participants were randomly assigned to read one of the testing lists.

Procedures
Participants were seated comfortably in a chair in a dimly lit and sound-attenuated room where stimuli were presented on a computer screen. Each trial started with a fixation cross ("+") at the center of the screen for 500 ms. After a 500 ms blank screen, each sentence was presented word-by-word with a single word duration of 400 ms followed by a 400 ms blank screen. Participants were then required to answer a yes/no comprehension question by pressing one of two keys as quickly and accurately as possible. The questions were designed to probe participants' understanding of the sentences. About 60% of the trials were followed by a comprehension question. For the remaining trials, the word "blank" was presented following the trials and participants had to press the space bar to continue. The assignment of left/right hand to yes/no response was counterbalanced across participants. Each participant performed 20 practice trials at the beginning of the experiment, which were similar in structure to the test stimuli. The stimuli were divided into four blocks, and the participant had a break of about 5 min between each block. The whole experiment lasted about 2 h per participant, including electrode preparation, instructions, and practice. At the end of the ERP experiment, the participants were asked to complete a working memory test (reading span). In the working memory test, participants pressed a key to judge whether the sentence presented on the screen was true or not based on their common sense and, at the same time, memorized the last word of each sentence. After a series of sentences (the number of sentences gradually increasing), participants needed to type in the memorized words in order. Reading span capacity scores were calculated by summing the number of words recalled in correct sequence with correct judgment of each sentence [41,42].

EEG Recording and Analysis
The EEG response was recorded from 31 electrodes held in place by an elastic cap (Electro-cap International). Additional electrodes were placed above the right eye and at the outer canthus of the left eye to monitor the vertical and horizontal electrooculogram. The EEG was referenced online to the tip of the nose and re-referenced offline to the algebraic average activity measured in the left and right mastoids (TP9 and TP10). Electrode impedance was kept below 5 kΩ. EEG signals were filtered using a band-pass of 0.016-70 Hz and digitized at a sampling rate of 500 Hz to exclude slow drifts and then was segmented into epochs from −200 to 900 ms around the critical words (e.g., fresh) or the sentence-final words (e.g., pure), with a prestimulus baseline of 200 ms. The signal was averaged for each condition and each participant before grand averages were computed across all participants. An independent component analysis (ICA) was applied to correct ocular artifacts. After correction, remaining artifacts (EOG, movement, technical) were marked manually and dropped from data analysis.
Based on the relevant literature in sentence/discourse processing [39,43,44], statistical analyses of the ERPs were carried out in the time window of 300-800 ms. ANOVAs were conducted on mean ERP amplitudes in this time window with temporal connectives (before vs. after), congruency (congruent vs. incongruent) and topographical factors as within-participant variables. For the midline analysis, the topographic factor was electrode  [45]). Mean amplitudes over electrodes in each region of interest were entered into ANOVAs. The Greenhouse-Geisser correction was applied when appropriate [46]. In the correlation analysis on connectives, we chose the 3000-4500 ms time window, which, according to a previous relevant study [12], showed the most pronounced effect for temporal connectives.

Behavioral Results
The mean accuracy in the comprehension task was 93.9%, suggesting that the participants could understand the sentences well. The average accuracy was 95.4% for after-congruent sentences and 92.2% for after-incongruent sentences, whereas the accuracy was 93.8% for before-congruent sentences and 94.0% for before-incongruent sentences. A 2 × 2 repeated measures ANOVA showed a marginally significant main effect of congruency (p = 0.054), indicating that incongruent sentences were more difficult to comprehend than congruent sentences.

ERP Results
The grand-average ERPs, time-locked to the critical words and the sentence-final words, are shown in Figures 1 and 2, respectively. As can be seen in Figure 1, afterincongruent sentences elicited a sustained larger positivity (P600) relative to the aftercongruent sentences, whereas there was no difference between the incongruent and congruent conditions in before sentences. Moreover, before-congruent sentences elicited a larger positivity than after-congruent sentences. At the sentence-final position, incongruent sentences elicited greater negativities than congruent sentences irrespective of the connective types, as demonstrated in Figure 2.
x FOR PEER REVIEW 10 of 18 In ERP data analysis, we carried out two omnibus repeated-measures ANOVAs in which the mean amplitude (ROI-based) from the lateral electrodes and the midline elec-

ERP Responses to the Critical Words (300-800 ms Time Window/P600)
In ERP data analysis, we carried out two omnibus repeated-measures ANOVAs in which the mean amplitude (ROI-based) from the lateral electrodes and the midline electrodes were analyzed separately. This approach has been adopted by a large number of ERP studies including those focusing on concessive relations [31,47]. Below, we mainly report the statistically significant results. For the full ANOVA results, please refer to the Appendix A (Table A1 presents the results for the critical words and Table A2 the results for the sentence-final words).

ERP Responses to the Sentence-Final Word (300-800 ms time window/negativity)
At lateral sites, ANOVA revealed a significant main effect of congruency (F (1, 27) = 6.01, p = 0.021, η 2 = 0.182) and an interaction between congruency and hemisphere (F (1, 27) = 5.00, p = 0.034, η 2 = 0.156). No other interesting main effects or interactions were observed. The follow-up analysis to resolve the interaction showed that incongruent sentences elicited larger negativities than congruent sentences in the right hemisphere (t (1, The follow-up analysis to resolve the interaction showed that incongruent sentences elicited larger negativities than congruent sentences in the right hemisphere (t (1, 27) = 3.24, p < 0.01) but not in the left. No significant effects were found in the midline analyses.

Correlation Analyses
Following the previous study [12], we calculated the ERP difference effects at the connectives and carried out correlation analysis between the ERP difference effects and the individuals' working memory span (see Figure 4). The difference effect at the connectives was the amplitude difference between before sentences (averaged over all before sentences) and after sentences (averaged over all after sentences). In the time window of 3000-4500 ms, the ERP difference amplitude (before vs. after) in the left posterior area (P3, P7, O1) was positively correlated with individuals' working memory span (r(28) = 0.441, p = 0.019).
tives was the amplitude difference between before sentences (averaged tences) and after sentences (averaged over all after sentences). In the tim 4500 ms, the ERP difference amplitude (before vs. after) in the left pos O1) was positively correlated with individuals' working memory spa 0.019).

Discussion
This study investigated how world knowledge is modulated by nectives before and after in Mandarin Chinese. Behavioral results showe sion accuracy for incongruent sentences was lower than those for the c regardless of the connective type. As for the ERP findings, critical w gruent sentences evoked a larger positivity than critical words in a tences, whereas the critical words in before-incongruent and before-cong vealed no significant difference. For congruent sentences, critical word ent sentences elicited larger positivity than critical words in after-congr Figure 1). Moreover, at the connectives, the ERP difference between b after sentences was positively correlated with individuals' working m 4). As for sentence-final words, incongruent sentences elicited larger ne gruent sentences irrespective of temporal relations (see Figure 2).

Effect of Temporal Connectives on Pragmatic Processing at the Critical W
Pragmatic incongruency has different ERP modulations on before After sentences with pragmatic anomalies evoked a P600 effect, but b not. In addition, we also observed a significant positive correlation be ference at the connectives (before minus after) and individuals' work consistent with Münte et al. [12] and Xiang et al. [33], suggesting tha higher working memory capacity are better at dealing with the reverse This seems to support the iconicity account, namely that comprehend clauses is more difficult than understanding after temporal clauses.

Discussion
This study investigated how world knowledge is modulated by the temporal connectives before and after in Mandarin Chinese. Behavioral results showed that comprehension accuracy for incongruent sentences was lower than those for the congruent sentences regardless of the connective type. As for the ERP findings, critical words in after-incongruent sentences evoked a larger positivity than critical words in after-congruent sentences, whereas the critical words in before-incongruent and before-congruent sentences revealed no significant difference. For congruent sentences, critical words in before-congruent sentences elicited larger positivity than critical words in after-congruent sentences (see Figure 1). Moreover, at the connectives, the ERP difference between before sentences and after sentences was positively correlated with individuals' working memory (see Figure 4). As for sentence-final words, incongruent sentences elicited larger negativities than congruent sentences irrespective of temporal relations (see Figure 2).

Effect of Temporal Connectives on Pragmatic Processing at the Critical Words
Pragmatic incongruency has different ERP modulations on before vs. after sentences. After sentences with pragmatic anomalies evoked a P600 effect, but before sentences did not. In addition, we also observed a significant positive correlation between the ERP difference at the connectives (before minus after) and individuals' working memory span, consistent with Münte et al. [12] and Xiang et al. [33], suggesting that participants with higher working memory capacity are better at dealing with the reverse temporal relations. This seems to support the iconicity account, namely that comprehending before temporal clauses is more difficult than understanding after temporal clauses.
While comprehending temporal relations, people tend to map the linear order of the temporal clause to the chronological order of the events in the real world. Since the event described in after sentences is consistent with such a preference, readers have little difficulty processing this temporal relation. They, therefore, can successfully detect the pragmatic anomalies and initiate a reinterpretation process to rationalize the inappropriate input, as reflected by a P600 effect [48,49]. As for before sentences, however, readers need to make more effort to build a new mental representation because the before construction not only reverses the linear order but also requires readers to integrate the context with world knowledge to verify its validity [40,50]. Since this process can be cognitively demanding, fewer cognitive resources are available for processing pragmatic anomalies, resulting in two indistinguishable positivities. Similar "insensitivity to semantic/pragmatic anomalies" have been repeatedly observed in studies in which complex sentence/discourse structures with ambiguous representations were adopted (e.g., reversed temporal relation, Nieuwland 2015; concessive relation [31,40]; counterfactual relation, [51,52]). Of particular relevance, Nieuwland [13] found that while pragmatic anomalous words elicited a greater N400 effect in after sentences, this difference effect was significantly attenuated in before sentences. Nonetheless, the absence of a P600 effect on the critical words in the current study (in the before construction) does not mean that the pragmatic incongruence exerts no influence on subsequent sentence processing. Indeed, it modulates the late stage of integrating sentence constituents into the discourse representation, as manifested by an increased negativity for the incongruent relative to the congruent sentences at the wrap-up position. We will return to this issue in the last section.
The different cognitive processes underpinning after and before sentences can also be demonstrated by directly comparing the two consistent conditions. The observation of a larger sustained positivity (P600) for before-congruent sentences than for after-congruent sentences at critical words indicates that truth value judgments under a reversed temporal relation are more difficult than a sequential temporal relation. However, although previous research on sentence processing has recorded a close relation between P600 effect and sentence-level prediction (e.g., the predictability of an upcoming word) and sentence acceptability (e.g., [53]), the larger P600 modulation for before sentences cannot simply be attributed to such effects because critical words are equally predictable and the sentences are equally acceptable between the two constructions (cloze probability: 48.7% vs. 49.7%; acceptability: 5.35 vs. 5.13). Instead, the P600 modulation is more likely related to the complex semantic and pragmatic properties denoted by the reverse temporal expression before compared to after. This is also consistent with Nieuwland [13], who observed a negative ERP effect (N400) in response to pragmatic inconsistencies in after sentences but not before sentences. However, one difference between the present study and Nieuwland's [13] study is that we found a P600 but they reported an N400. This could be related to the linguistic differences of expressing temporal relations between Chinese and English. As a tenseless language, time arguments in Chinese are regularly absent. Chinese speakers depend on a number of approaches, such as deictic patterns (e.g., next week) and context cues, to express temporality [6]. As a consequence, the truth value judgement in a temporal structure in Chinese is not simply based on the lexical semantics of a sentence, but rather depends on the explicit temporal devices, the contextual cues, and even readers' personal social cognition capacities [6,[54][55][56]. However, different from the present study, Nieuwland also reported a late positivity effect at the critical words in before sentences, which was interpreted as reflecting the involvement of monitoring processes [13]. These differences may be related to the recruitment of different computing processes. Firstly, the sentence position of the connectives was different between the studies, which can influence the ease of retrieving and keeping information in working memory, and may consequently influence the process of building a discourse representation. When a connective appears at the beginning of a sentence (e.g., [13]), it can provide immediate information about the temporal relationship of the statement at an early stage; thus, encountering pragmatic anomalies can immediately evoke an N400 in after sentences or a late P600 in before sentences.
When a connective appears in the sentence-middle position, however, due to its closer distance to subsequent critical words, repairing and integrating the incongruency into discourse could increase processing load and, therefore, prolong processing time. A similar prolonged reinterpretation process, as reflected by a sustained negativity in post-critical words, has been reported for other complex constructions containing reverse coherence connectives (e.g., even so construction in English, Xiang and Kuperberg [47]; lian . . . dou (even) construction in Chinese, Jiang et al. [29]). Secondly, different from Nieuwland [13] and Politzer-Ahles et al.'s [14] studies, the sentences in the current study are more complex. A co-referential relation must be established between the two clauses, as there is a locative pronoun (nali) in the subordinate clause which must be resolved towards one of the locations in the preceding clause. The co-referential relation is more difficult to establish in the reversed temporal relation than in the canonical temporal relation. This could also contribute to the delayed negative-going effect.

Pragmatic Incongruence Effect in Sentence-Final Position
The processing cost of violating a sentence's truth value may not fully manifest on the critical word, but instead be delayed to a later stage at which multiple constraints are integrated into the discourse representation. Indeed, pragmatic incongruencies evoked larger negativities at the sentence-final position in both after and before constructions, even though only the after construction evoked a P600 effect at the critical word. However, although the pragmatic incongruence evoked a late negativity in both constructions, the underlying cognitive processes as well as the functional interpretation should not be the same. For after sentences, while the P600 reflects a conflict between language input and knowledge stored in long-term memory and initial attempts to repair such incongruency, the late negativity at the sentence-final position may reflect a prolonged second-pass process to reinterpret the incongruency before integrating it into discourse. A similar pattern has been reported by Xiang and Kuperberg [47] who found that pragmatic inconsistencies evoked a large, late negativity at the sentence-final position in addition to early modulations at the target, irrespective of whether the sentence was connected by a complex concessive connective (i.e., even so) or not. As for before sentences, pragmatic anomalies only elicited a late negativity at the sentence-final position. Nevertheless, different from the negativity in after sentences, the negativity in before sentences lasted longer and was distributed more extensively over the scalp, which is comparable to the sustained negativities obtained for complex event structures where a new event representation has to be updated (e.g., [29,31,[57][58][59]. Instead of simply initiating a repair process when encountering a non-iconic temporal relation, the comprehension system must first reorder a reverse temporal sequence and then reinterpret the pragmatic incongruency. Such a process could be cognitively demanding for the system and therefore delayed to the sentence wrap-up position.

Conclusions
The present ERP study examined how world knowledge is influenced by the temporal connectives before and after in Mandarin Chinese, a language without tense. It was found that at the critical word, after-incongruent sentences evoked a sustained late positivity relative to the after-congruent sentences, whereas before-incongruent sentences and beforecongruent sentences evoked indistinguishable ERP responses. An increased sustained negativity was elicited by sentence-final incongruent words compared to congruent ones irrespective of temporal relations. These results suggest that before is more difficult to comprehend than after in Mandarin, supporting an explanation of processing temporal sequence in terms of iconicity. In the future, it would be interesting to compare how temporal relations are processed in Chinese and other related languages (e.g., Korean, Japanese).
Author Contributions: Conceptualization, X.X. and L.C.; methodology, L.C., Y.L. and X.X.; formal analysis, Y.L., L.C. and X.X.; writing-original draft preparation, L.C., Y.L. and X.X.; supervision, X.X. and L.C. All authors have read and agreed to the published version of the manuscript. Institutional Review Board Statement: This study was conducted according to the guidelines of the Declaration of Helsinki, and approved by the Ethics Committee of Nanjing Normal University (protocol code 202007004 of 4 July 2020).

Informed Consent Statement:
Written informed consent has been obtained from the participants to publish this paper.

Data Availability Statement:
The datasets presented in this article will be provided on request.
Requests to access the datasets should be directed to the corresponding author.

Acknowledgments:
We thank Max Wolpert for his suggestions and comments concerning earlier versions of the manuscript.

Conflicts of Interest:
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Appendix A Table A1. The results of ANOVAs on mean amplitude at the critical words in the 300-800 ms time window.