The Post-verbal Effect of Negators in Mongolian Contradictory Negations Provides Support for the Fusion Model

Xu, Qinghong; Zhang, Shujun; Li, Jie; Wu, Baizhou; Qiu, Helin; Qilimuge,

doi:10.3389/fpsyg.2021.603075

ORIGINAL RESEARCH article

Front. Psychol., 20 May 2021

Sec. Psychology of Language

Volume 12 - 2021 | https://doi.org/10.3389/fpsyg.2021.603075

The Post-verbal Effect of Negators in Mongolian Contradictory Negations Provides Support for the Fusion Model

$\nQinghong Xu,,&#x;$ Qinghong Xu^1,2,3^†

Shujun Zhang^2,4^†

Jie Li^1,2^*

Baizhou Wu^2,5

Helin Qiu²

Qilimuge²

¹The Key Laboratory of Psychology, Inner Mongolia Normal University, Hohhot, China
²School of Psychology, Inner Mongolia Normal University, Hohhot, China
³School of Foreign Languages, Yulin University, Yulin, China
⁴Baotou Municipal Party School, Baotou, China
⁵Institute of Psychology, Chinese Academy of Science, Beijing, China

There are two contending models regarding the processing of negation: the fusion model and the schema-plus-tag model. Most previous studies have centered on negation in languages such as English and Mandarin, where negators are positioned before predicates. Mongolian, quite uniquely, is a language whose negators are post-verbal, making them natural replicas of the schema-plus-tag model. The present study aims to investigate the representation process of Mongolian contradictory negative sentences to shed light on the debate between the models, meanwhile verifying the post-verbal effect of negators. A series of experiments using the sentence–picture verification paradigm supports the fusion model: (i) Mongolian contradictory negative sentences were processed by representing the actual conditions rather than the negated state of affairs at 250 ISI (interstimulus interval of 250 ms), and (ii) despite the fact that a post-verbal effect of negators was measured at 250 ISI when Mongolian and Mandarin negative sentences were compared, Mongolian–Mandarin bilinguals adopted the same representational strategy for contradictory negation in both languages.

Introduction

For syntactic processing, negative stand-alone sentences can be divided into two types: contradictory negative sentences from which syntactic features the actual conditions can be explicitly inferred and non-contradictory negative sentences from which syntactic features the conditions cannot be inferred. In other words, contradictory negative sentences have only one alternative state such as in The window is not open. Though no explicit information about the window's condition has been shown, the sentence provides enough information to generate the actual representation, that is The window is closed. In comparison, non-contradictory negative sentences have multiple alternative states, such as in The skirt is not red, where multiple representations such as The skirt is blue/brown/yellow, etc. are plausible.

Comprehension of negative sentence statements is cognitively more demanding than that of affirmative sentence statements, as latencies are longer and more error prone. At present, two models are discussed concerning the processing of negations: dual- and single-step processing models.

Both models differ concerning how the negation tag and the core supposition are processed. In the fusion model, they are spontaneously fused (e.g., Brewer and Lichtenstein, 1974; Horn, 1989; MacDonald and Just, 1989; Gannon and Ostrom, 1996; Lea and Mulligan, 2002; Mayo et al., 2004), and in the schema-plus-tag model, they are separately processed (e.g., Fiedler et al., 1996; Rubaltelli and Slovic, 2008). Put differently, the dual-step model assumes that a negation such as The window is not open is processed by the negation-incongruent schema and an open window by the core supposition. In contrast, a single-step model states that in the case of a closed window, a negation-congruent schema will be spontaneously activated. This distinction is important as both schemas are embedded in distinct cognitive networks. Both hypotheses have been supported by relevant studies, though we note a trend toward the dual-step model (Kaup and Zwaan, 2003; Giora et al., 2004; Hasson and Glucksberg, 2006; Kaup et al., 2006, 2007a,b; Li et al., 2007; Lüdtke et al., 2008).

The sentence–picture verification paradigm became accepted as a common method for negation-processing studies along with manipulation of interstimulus intervals (ISIs). It was found that transient, image-based, and highly view-specific representations mediate only identical view matches at short ISIs, and abstract, durable representations mediate all other view matches at long ISIs (Lawson and Humphreys, 1996). Response times (RT) across studies for view-invariant matches fell in the range of 300–700 ms (e.g., Shepard and Metzler, 1971; Klatzky and Stoy, 1974; Jolicoeur, 1985; Ellis and Allport, 1986; Ellis et al., 1989). Therefore, Stanfield and Zwaan (2001) used 250 ISI to separate the span a subject needed for accessing view-specific or more view-invariant object representations. They examined the hypothesis derived from perceptual symbol theories that people mentally represent the orientation of an object implied by a verbal description. Results indicated that at 250 ISI, when a picture matched the object orientation described by a sentence, the latencies were significantly faster compared to a non-matching picture. These results confirm that 250 ISI measures the initial stage of negation processing during which the first cognitive unit will be formed.

In later negation-processing studies (e.g., Kaup et al., 2006, 2007a; Gao et al., 2011, 2017; Chen et al., 2014; Cui et al., 2016), 250 ISI was used as the time a subject needed to form a representation of a sentence's core supposition (akin to view-specific representation in the picture-matching paradigm). Measuring whether the representation facilitates the judgment on negation-congruent or negation-incongruent picture stimuli that follow after 250 ISI would elucidate whether the representation fused the negation operator with the core or not at this stage, as advocated respectively, by the single- or dual-step model.

By that same logic, Kaup et al. (2006) used 750 ISI and found a congruency effect with the actual conditions for affirmations but not for negations. At 1,500 ISI, Yan and Bai (2000) found that most subjects can complete negation processing at around 1,000 ms in the sentence–picture verification paradigm, so 1,500 ISI was taken to be the final stage in negation processing. A congruency effect with the actual conditions was found for negations but not for affirmations. The results were interpreted in favor of the dual-step model. However, Gao et al. (2011) who applied all three ISI conditions to Mandarin negatives found diverging evidence that supported the fusion model.

There are further disparities in results of negation-processing studies. For example, Kaup et al. (2006, 2007a), using the sentence–picture verification paradigm to investigate negation processing, indicated that for both contradictory and non-contradictory negative sentences, negation-incongruent schemas were represented at 250 ISI and negation-congruent schemas at longer ISIs, which supports the dual-step hypothesis. However, Kaup et al. used different experimental stimuli when investigating representations at different ISIs. For instance, to investigate negation representation at 250 ISI, non-contradictory negations, such as There was no eagle in the sky and There was no eagle in the nest, were used. Such negations render corresponding negation-congruent schema inexplicit (Kaup et al., 2007b). In contrast, the experimental materials adopted to investigate the negation representation at 750 and 1,500 ISI were contradictory negations in German such as Die/Der/Das A war nicht X/Y (meaning: The A was/was not X/Y; where X/Y stood for the contradictory states of affairs of the subject A), and here, the negation-congruent schema was explicit (Kaup et al., 2006). Therefore, the study results were inconclusive (Gao et al., 2011).

Mayo et al. (2004) used both contradictory and non-contradictory negative sentences as experimental materials. Their results indicated that processing of non-contradictory negative sentences (e.g., The skirt is not red) converges with the dual-step model. The core supposition (The skirt is red) is represented at 250 ISI. Then, it is followed by the negation-congruent representation formed by attaching a negation tag to the core supposition. In contrast, the processing of contradictory negations (such as The window is not open) supports the fusion model, where a negation-congruent schema and a closed window are spontaneously activated at 250 ISI. Similarly, Ping et al. (2014) found that for Mandarin contradictory negations, the actual conditions were available at 250 ISI, while the negated state was represented for non-contradictory negations.

These findings point to the possibility that the negated state may not typically be represented when processing negations and that type of negation (contradictory or non-contradictory) is an important factor. Gao et al. (2011) focused on Mandarin contradictory negations with a sentence–picture verification paradigm. Their results indicated a congruency effect with the actual conditions was found at 250 ISI, which again supports the fusion model.

Studies have mostly focused on negations with S–V–O (subject–verb–object) or S–L–P (subject–link verb–predicate) sentences in languages such as Mandarin or English whose negators appear before verbs. Mongolian, in contrast, is one of the Altaic Mongol languages spoken in Mongolia, west of central Asia, north Siberia, provinces of northern China, and Inner Mongolia in particular, and Mongolian negations are formed with negators appearing at the end of S–O–V or S–L–P sentences as additional components (Yu, 2010).

Research on negations has placed limited focus on languages with post-verbal negators, with one exception being the study of Foroni and Semin (2013) on how comprehension of action negation in Dutch elicits inhibitory motor activity. Previous work focused on the representation of actions in language, and evidence supports that language stimuli referring to actions activate motor processes (e.g., Pulvermüller, 2004, 2005; Buccino et al., 2005; Pulvermüller et al., 2005a,b; Zwaan and Taylor, 2006; Filimon et al., 2007; Fischer and Zwaan, 2008; Hauk et al., 2008; Winkielman et al., 2008; Foroni and Semin, 2009, 2011). Do negations activate the motor system? Subjects were exposed to sentences in the affirmative and negative forms (e.g., I am smiling and I am smiling not) while the zygomatic muscle activity on the left side of the face was measured (electromyography technique: EMG). Change in activity compared to baseline was averaged over intervals of 200 ms, giving rise to five periods of 200 ms each during the time interval considered. An important extension of this work is how the comprehension of a negated action was represented. Results indicated that comprehension of negation entails a fast inhibition of the relevant muscle, effects of which started within the very first 200 ms. In other words, the negation-congruent representation had been formed by integrating the post-verbal negation operator with the core supposition within the first 200 ms. This seems to support the fusion model except that the manipulations may not be comparable with the sentence–picture verification paradigm.

Just like Dutch, the post-verbal feature of Mongolian negators makes Mongolian negations structurally comparable to the mental representation proposed by the dual-step model. Whether this feature of negators in Mongolian makes the processing of contradictory negations in Mongolian conform with the dual-step processing model—thus overturning the findings of Mayo et al. (2004) and Ping et al. (2014) that the processing of contradictory negations fits the fusion model—remains to be verified. More importantly, if post-verbal negators are integrated spontaneously with predicates to form negation-congruent schemas that represent the actual conditions at 250 ISI [despite Mongolian negations' structural equivalence to the dual-step model as suggested by the study of Foroni and Semin (2013) on Dutch], it would be a strong evidence in support of the fusion model.

The sentence–picture verification paradigm has been commonly used for the examination of processes involved in negations. To conform with previous studies' manipulations other than the post-verbal effects of negators, the sentence–picture verification paradigm was used in the present study to investigate the processing of Mongolian contradictory negations at different ISIs to investigate the post-verbal effect of negators on contradictory negation processing. In line with existing studies (Kaup et al., 2006; Gao et al., 2011; Chen et al., 2014; Cui et al., 2016), the present study defines 250 ISI after sentence reading as the initial stage of negation processing, 750 ISI as the middle stage, and 1,500 ISI as the late stage.

Experiment 1 investigates the mental representation of negations in Mandarin and Mongolian at 250 ISI. As Mongolian negations feature post-verbal negation markers, the logic of the experiment is as follows: its processing may follow a dual-step simulation of negation processing and not be completed within the initial stage as is the case with Mandarin negation (Gao et al., 2011). Thus, the predicted result of the experiment is that response time under the Mandarin sentence–picture matching condition will be significantly shorter than under the mismatching condition. RT under the Mongolian condition will be the opposite to Mandarin, which means the processing of Mongolian negations first simulates the negated state instead of the actual state, thus facilitating RT under the mismatching condition. Experiments 2 and 3 examine the mental representation of Mandarin and Mongolian negations, respectively, at 750 and 1,500 ISI. If the processing of Mongolian negations in the experiment conforms with the dual-step model assumptions, then the logic of experiment 2 and experiment 3 is as follows: if Mongolian negation processing is completed at 750 ISI, the RT under matching conditions will be significantly faster for both language conditions, and differences between the two language conditions would be insignificant, thus supporting the dual-step model for Mongolian negation and verifying the post-verbal effect of the negation marker. Experiment 3 would be a confirmation that the processing of negations is fully completed in prior stages, in that differences between the two language conditions would disappear at this stage. If experiment 2 does not show such a predicted effect, the purpose of experiment 3 is to examine further whether the processing of Mongolian negation is not completed until the late stage.

Experiment 1: Mental Representation at the Initial Stage of Negation Processing of Mongolian–Mandarin Bilinguals

Method

Participants

Participants across the three experiments in the present study are Mongolian–Mandarin bilingual college students from a university in Inner Mongolia Autonomous Region, China. Participants were brought up in Mongolian households and received Mongolian–Mandarin bilingual school education for at least 12 years starting from primary school to graduation from high school (as mandated by government regulations). Therefore, they were all fully competent in reading, writing, and comprehending both languages. The ideal choice for the present study would have been Mongolian and Mandarin monolingual participants as there could be potential influences from participants' proficiency difference between their native language and second language. However, decades of racial integration in Mainland China have dramatically reduced the number of Mongolian monolinguals. This research included therefore bilingual participants who grew up in Mongolian households and were not exposed to all-Mandarin surroundings until school age.

A total of 49 Mongolian–Mandarin bilingual college students (11 males and 38 females) aged between 18 and 24 years, from a university in Inner Mongolia Autonomous Region, China, participated in experiment 1. All subjects had normal (or corrected to normal) vision and no hearing or reading impairments. All volunteered to participate in the experiment and were paid the equivalent of $2.00 after the experiment. None of the participants in the formal experiment took part in the stimulus assessment.

Material Assessment

Eighteen proficient Mongolian and Mandarin bilinguals evaluated the contradictory states of multiple objects and selected 49 objects with explicit contradictory states. The rationality and familiarity of Mongolian and Mandarin affirmative/negative sentences involving the 49 objects were rated in a single rating on a seven-point scale [1 indicated completely unreasonable and unfamiliar, 7 indicated completely reasonable and familiar, 2–6 indicated consecutive increases in ratio and familiarity as per Gao et al. (2011)] by 22 proficient Mongolian–Mandarin bilinguals. A total of 44 groups of sentences with an average score above 5.0 were selected, from which eight groups were randomly chosen and transformed into 8 × 4 questions (e.g., Is the window closed?/Is the window open?/Is the window not closed?/Is the window not open?) in both languages to serve as filler trials.

Corresponding pictures depicting opposite states of each object (e.g., a closed window/an open window) were black and white with 314 pixels in length and width. Another group of 28 Mongolian college students were selected to evaluate the degree of congruency between sentences and pictures of the remaining 36 groups of sentences on a seven-point scale, and 32 sets of sentences with an average score above 5.0 were selected as formal experimental material. The 32 groups of sentences and their corresponding pictures comprised eight versions [2 (sentence type: affirmation/negation) × 2 (actual state of affair: e.g., open/closed window) × 2 (congruency condition: congruent/incongruent); see Table 1] of the experimental material. Results of repeated-measures analysis of variance (ANOVA) showed that the main effect of sentence type, congruent/incongruent conditions, and the interaction between them was not significant (p > 0.05), indicating consistent and homogenous stimulus material. Participants who took part in the stimulus control assessment were excluded from participating in the experimental trials.

TABLE 1

Table 1. An example of one object's eight versions in experimental material.

Stimuli

A total of 32 objects that have explicit contradictory states in Mandarin and Mongolian were used in the experiments. As for the negative sentences, the negator “not” appeared in the middle of the subject–predicate complement structure in Mandarin negations, that is, subject (S) + not + predicate (X/Y) (a linking verb is not needed here in Mandarin). For example, yes (the window is not open in Mandarin), while the negator “not” appeared at the end in Mongolian negations, that is, subject (S) + predicate (X/Y) + not (linking verb not needed here in Mongolian), for example, “ yes ” (the window is open not in Mongolian). The corresponding affirmation was subject (S) + predicate (X/Y) for both Mandarin and Mongolian, e.g., yes and “ yes .” X/Y stood for the contradictory states of the subject. There were 32 pairs of pictures in total, each pair displaying contradictory states of the same situation, such as an open window and a closed window. All images were black and white with length and width of 314 pixels.

There were 32 objects and eight versions of each, i.e., 2 (sentence type: affirmative and negative) × ss2 (sentence implied state: state X and state Y) × 2 (picture described state: state X and state Y). These were arranged by a balanced Latin square order and divided into eight sets, so that each set of experimental materials contained any one of the eight possible versions as well as 32 Mandarin sentences and 32 Mongolian sentences with the same meaning but different syntactic structures (affirmative or negative). Each participant was randomly assigned to one set.

We included fillers that had different syntactic structures to the experimental items to avoid that participants would develop strategies for parsing affirmative and negative sentences. Twenty of the 32 filler stimuli (transformed from eight randomly selected sets of sentences) with consistency ratings between sentence and pictures above 5.0 on the seven-point scale, followed by Yes or No choices, were added to the experiment. The filler stimuli were the same across the eight sets of materials and were presented in random order together with experimental stimuli.

It should be pointed out that Mandarin sentence materials were horizontally presented and Mongolian vertically, as is conventional in each language. Although the difference in printing format may cause unwanted influences, we believe such influences were minor as eye-tracking studies (e.g., Yan and Bai, 2000) have found that native Mandarin participants were able to read vertically printed sentence materials as quickly as horizontally printed texts.

Procedure

The experiment adopted a 2 (language: Mongolian/Mandarin) × 2 (sentence type: affirmative/negative) × 2 (sentence–picture congruency condition: congruent/incongruent) within-subjects design. The materials were presented with the sentence–picture verification paradigm. Participants would first see a sentence at the center of the screen. After reading a sentence, they were asked to press the space bar as soon as possible so that a “+” sign appeared at the central fixation point during a 250-ms ISI before a picture was presented. Participants were instructed to press the “J” key for sentence–picture congruent associations and the “F” key for incongruent associations. They were asked to do so as accurately and quickly as possible. After the response, the next set of materials would be automatically and randomly presented. The filler items included questions that required an affirmative or negative response by pressing the “J” or “F” key. Each experiment consisted of two blocks and was conducted in a balanced Latin square order, with the first half of all tasks in the order of Mandarin sentences first and then Mongolian, and the other half in a reversed order.

Results and Discussion

The finding of Kaup et al. (2007a) was that participants responded to the questions regarding affirmative sentences with a mean accuracy of 77% and those regarding negative sentences with a mean accuracy of 80%. Thus, in the present experiment, we excluded four participants whose accuracy rates were lower than 75% (accounting for 8.2% of all data), and where RTs were longer than 3,000 ms or shorter than 300 ms (1.32%) or beyond 2 standard deviations from a participant's mean (1.76%), these were also omitted from all analysis (as per Gao et al., 2011; Cui et al., 2016). Yan and Bai (2000) noted that the average fixation duration on a Mandarin reading material was ~0.3 s for both horizontal and vertical formats, and the number of words an average participant can see at one gaze is about two (normal Mandarin printing format is horizontal, but it was found that participants were able to read vertically printed material just as quickly). In contrast, they also pointed out that the reading speed of Mandarin characters ranged from 2.9 to 11.0/s. Sentence length in the current study was mostly four characters, with slight variation in length for corresponding Mongolian sentences. Thus, 300 ms was too short for participants to respond, while 3,000 ms was too long as participants were able to process a complete sentence on average in 1.5 s (Bai et al., 2011). A statistical analysis was conducted on the valid data of 45 participants. Mean RTs and standard deviations are shown in Table 2.

TABLE 2

Table 2. Means and standard deviations of response time (ms) and accuracy rate (%) at 250 ISI.

RT data were submitted to a mixed-effect model analysis, and SPSS 25.0 software was used for statistical analysis. Repeated-measures data of language type (Mongolian/Mandarin), sentence type (affirmation/negation), congruency condition (congruent/incongruent), and two-factor (language type × sentence type, language type × congruency condition, and sentence type × congruency condition) and three-factor interactions were entered into the model as fixed effects. A random subject-specific effect of treatment was then added and retained in the model. An assumption of equal correlation of residuals was plausible (within-subject design), so the compound symmetry structure was adopted for covariance structure.

Results revealed that the main effect of language type was not significant [F_{(1, 308)} = 3.727, p = 0.54]. The sentence type effect was significant, indicating that processing of negations took longer than affirmations [F_{(1, 308)} = 77.854, p < 0.001]. The main effect of congruency condition was significant, namely, the RT for pictures that had incongruent associations with sentences took significantly longer than congruent ones [F_{(1, 308)} = 8.872, p < 0.01]. Three-factor interaction among language type, sentence type, and sentence–picture congruency condition was not significant [F_{(1, 308)} = 2.166, p = 0.142]. The interaction of language type and sentence type was significant [F_{(1, 308)} = 3.892, p = 0.049], and a further analysis of their interaction showed that the processing of negative sentences in Mandarin was significantly faster than in Mongolian (see Figure 1). Moreover, the interaction between sentence type and congruency condition was also significant [F_{(1, 308)} = 5.993, p = 0.015]. A further analysis indicated that when processing affirmations, there was no significant difference between congruent and incongruent conditions, but RT in the congruent condition was significantly shorter than that in the incongruent condition when negations were processed (see Figure 2). The interaction between language type and congruency condition was not significant [F_{(1, 308)} = 0.267, p = 0.606].

FIGURE 1

Figure 1. The interaction between language type and sentence type at 250 ISI. Bars represent standard error.

FIGURE 2

Figure 2. The interaction between congruency condition and sentence type at 250 ISI. Bars represent standard error.

Accuracy data submitted to logistic regression showed that the Cox–Snell R² and Nagelkerke R² are 0.045 and 0.086, respectively. The significance of the Hosmer–Lemeshow test is 0.683, while the effect of language type, sentence type, congruency condition, and their interactions are all insignificant. These findings indicate that the variables and their interactions exerted no significant influence on accuracy and that the possibility of a speed–accuracy trade-off can be ruled out.

Unexpectedly, the results of experiment 1 indicate that when processing contradictory negative sentences, the actual state (rather than the negated state) was represented at 250 ISI in both Mandarin and Mongolian conditions. As expected, the processing of Mongolian negations was significantly slower than that of Mandarin negations at this stage, verifying the existence of the post-verbal effect of negators in Mongolian at the initial stage of negation processing. However, such an effect did not change the fact that Mongolian negation processing does not conform with the dual-step model. To confirm that the processing of Mongolian negations is fully completed during the first stage, we carried out experiment 2 to test if the differences between both language conditions would disappear at 750 ISI.

Experiment 2: Middle Stage Mental Representation of Negation Processing of Mongolian–Mandarin Bilinguals