(C) PLOS One
This story was originally published by PLOS One and is unaltered.
 . . . . . . . . . .



Negation mitigates rather than inverts the neural representations of adjectives [1]

['Arianna Zuanazzi', 'Department Of Psychology', 'New York University', 'New York', 'United States Of America', 'Pablo Ripollés', 'Music', 'Audio Research Lab', 'Marl', 'Center For Language']

Date: 2024-06 

Second, we explored the temporal dynamics of adjective representation as a function of negation (i.e., from the presentation of word 1 to the final interpretation; lines in Fig 2C ). While mouse trajectories of affirmative phrases branch towards either side of the scale and remain on that side until the final interpretation (lines in the left, gray, zoomed-in panel in Fig 2C ), trajectories of negated phrases first deviate towards the side of the adjective and then towards the side of the antonym, to reach the final interpretation (i.e., “not low” first towards “low” and then towards “high”; right, gray, zoomed-in panel in Fig 2C ; see S1 Fig for each adjective dimension separately). To characterize the degree of deviation towards each side of the scale, we performed regression analyses with antonyms as the predictor and mouse trajectories as the dependent variable (see Materials and methods ). The results confirm this observation, showing that (1) in affirmative phrases, betas are positive (i.e., mouse trajectories moving towards the adjective) starting at 300 ms from adjective onset (p < 0.001, green line in Fig 2D ); and that (2) in negated phrases, betas are positive between 450 and 580 ms from adjective onset (i.e., mouse trajectories moving towards the adjective, p = 0.04), and only become negative (i.e., mouse trajectories moving towards the antonym, p < 0.001) from 700 ms from adjective onset (red line in Fig 2D ). Note that beta values of negated phrases are smaller than that for affirmative phrases, again suggesting that negation does not invert the interpretation of the adjective to that of the antonym.

To quantify how the final interpretation of scalar adjectives changes as a function of negation, we first performed a 2 (antonym: low versus high) × 2 (negation: negated versus affirmative) repeated-measures ANOVA for participants’ ends of trajectories (filled circles in Fig 2B ), which reveal a significant main effect of antonyms (F(1,77) = 338.57, p < 0.001, η p 2 = 0.83), a significant main effect of negation (F(1,77) = 65.50, p < 0.001, η p 2 = 0.46), and a significant antonyms by negation interaction (F(1,77) = 1346.07, p < 0.001, η p 2 = 0.95). Post hoc tests show that the final interpretation of negated phrases is located at a more central portion on the semantic scale than that of affirmative phrases (affirmative low < negated high, and affirmative high > negated low, p holm < 0.001). Furthermore, the final interpretation of negated phrases is significantly more variable (measured as standard deviations) than that of affirmative phrases (F(1,77) = 78.14, p < 0.001, η p 2 = 0.50). Taken together, these results suggest that negation shifts the final interpretation of adjectives towards the antonyms, but never to a degree that overlaps with the interpretation of the affirmative antonym.

(A) Reaction times results for the online behavioral study (N = 78). Bars represent the participants’ mean ± SEM, and dots represent individual participants. Participants were faster for high adjectives (e.g., “good”) than for low adjectives (e.g., “bad”) and for affirmative phrases (e.g., “really really good”) than for negated phrases (e.g., “really not good”). The results support previous behavioral data showing that negation is associated with increased processing difficulty. ( B) Final interpretations (i.e., end of trajectories) of each phrase, represented by filled circles (purple = low, orange = high), averaged across adjective dimensions and participants, showing that negation never inverts the interpretation of adjectives to that of their antonyms. ( C) Mouse trajectories for low (purple) and high (orange) antonyms, for each modifier (shades of orange and purple) and for affirmative (left panel) and negated (right panel) phrases. Zoomed-in panels at the bottom demonstrate that mouse trajectories of affirmative phrases branch towards the adjective’s side of the scale and remain on that side until the final interpretation; in contrast, the trajectories of negated phrases first deviate towards the side of the adjective and subsequently towards the side of the antonym. This result is confirmed by linear models fitted to the data at each time point in D . ( D ) Beta values (average over 78 participants) over time, separately for affirmative and negated phrases. Thicker lines indicate significant time windows. ( C , D ) Black vertical dashed lines indicate the presentation onset of each word: modifier 1, modifier 2 and adjective; each line and shading represent participants’ mean ± SEM. ( A , B , D ) *** p < 0.001; * p < 0.05. Data are available on the Open Science Framework https://doi.org/10.17605/OSF.IO/5YS6B .

To evaluate the effect of antonyms and of negation on reaction times in behavioral Experiment 1, we performed a 2 (antonym: low versus high) × 2 (negation: negated versus affirmative) repeated-measures ANOVA. The results reveal a significant main effect of antonyms (F(1,77) = 60.83, p < 0.001, η p 2 = 0.44) and a significant main effect of negation (F(1,77) = 104.21, p < 0.001, η p 2 = 0.57, Fig 2A ). No significant crossover interaction between antonyms and negation was observed (p > 0.05). Participants were faster for high adjectives (e.g., “good”) than for low adjectives (e.g., “bad”) and for affirmative phrases (e.g., “really really good”) than for negated phrases (e.g., “really not good”). These results support previous behavioral data showing that negation is associated with increased processing difficulty [ 15 , 16 ]. A further analysis including the number of modifiers as factor (i.e., complexity) indicates that participants were faster for phrases with 2 modifiers, e.g., “not really,” than phrases with one modifier, e.g., “not ###” (F(1,77) = 16.02, p < 0.001, η p 2 = 0.17; see A Table in S3 Table for pairwise comparisons between each pair of modifiers), suggesting that the placeholder “###” may induce some processing slow-down. To confirm this hypothesis, further research should investigate the specific effect of placeholders (e.g., “###” or “xkq”) on word and phrase representation and semantic composition.

Experiment 1 (online behavioral experiment; N = 78) aimed to track changes in representation over time of scalar adjectives in affirmative and negated phrases. Participants read 2-to-3-word phrases comprising 1 or 2 modifiers (“not” and “really”) and a scalar adjective (e.g., “really really good,” “really not quiet,” “not ### fast”). The number and position of modifiers were manipulated to allow for a characterization of negation in simple and complex phrasal contexts, above and beyond single word processing. Adjectives were selected to represent opposite poles (i.e., antonyms) of the respective semantic scales: low pole of the scale (e.g., “bad,” “ugly,” “sad,” “cold,” “slow,” and “small”) and high pole of the scale (e.g., “good,” “beautiful,” “happy,” “hot,” “fast,” and “big”). A sequence of dashes was used to indicate the absence of a modifier. Fig 1A and S1 Table provide a comprehensive list of the linguistic stimuli. On every trial, participants rated the overall meaning of each phrase on a scale defined by each antonym pair ( Fig 1A ). Feedback was provided at the end of each trial (to which 1 and 0 were assigned to compute the average feedback score). We analyzed reaction times and continuous mouse trajectories, which consist of the positions of the participant’s mouse cursor while rating the phrase meaning. Continuous mouse trajectories offer the opportunity to measure the unfolding of word and phrase comprehension over time, thus providing time-resolved dynamic data that reflect changes in meaning representation [ 15 , 45 , 47 ].

The replication of Experiment 1 illustrates the robustness of the behavioral mouse tracking findings, even in the absence of feedback. Taken together, these results suggest that participants initially interpreted negated phrases as affirmative (e.g., “not good” interpreted along the “good” side of the scale) and later as a mitigated interpretation of the opposite meaning (e.g., the antonym “bad”).

The 2 (antonym: low versus high) × 2 (negation: negated versus affirmative) repeated-measures ANOVA for participants’ final interpretations reveal a significant main effect of antonyms (F(1,54) = 166.40, p < 0.001, η p 2 = 0.75), a significant main effect of negation (F(1,54) = 48.62, p < 0.001, η p 2 = 0.47), and a significant interaction between antonyms and negation (F(1,54) = 210.13, p < 0.001, η p 2 = 0.80). Post hoc tests show that the final interpretation of negated phrases was located at a more central portion of the semantic scale than that of affirmative phrases (affirmative low < negated high, and affirmative high > negated low, p holm < 0.001, Fig 3B ), indicating that negation never inverts the interpretation of adjectives to that of their antonyms. Results also show that the final interpretations of negated phrases was significantly more variable (measured as standard deviations) than that of affirmative phrases (F(1,54) = 15.43, p < 0.001, η p 2 = 0.22). These results again replicate Experiment 1. As for Experiment 1, we then performed regression analyses with antonyms as the predictor and mouse trajectories as the dependent variable. For this analysis, trials with “not not” were not included as, in this experiment, the trajectories pattern was different compared to the other conditions with negation ( Fig 3C ). The results of the regression analyses show that (1) in affirmative phrases, betas are positive (i.e., mouse trajectories moving towards the adjective) starting from 400 ms from the adjective onset (p < 0.001, green line in Fig 3D ); and that (2) in negated phrases, betas are positive (i.e., mouse trajectories moving towards the adjective) between 400 and 650 ms from the adjective onset (p = 0.02), and only became negative (i.e., mouse trajectories moving towards the antonym) from 910 ms from the adjective onset (p = 0.003, i.e., red line in Fig 3D ). This pattern replicates that of Experiment 1.

The 2 (antonym: low versus high) × 2 (negation: negated versus affirmative) repeated-measures ANOVA reveal a significant main effect of antonyms (F(1,54) = 36.90, p < 0.001, η p 2 = 0.40) and a significant main effect of negation (F(1,54) = 73.04, p < 0.001, η p 2 = 0.57). Moreover, a significant crossover interaction between antonyms and negation was found (F(1,54) = 16.40, p < 0.001, η p 2 = 0.23, Fig 3A ). These results replicate Experiment 1, showing that participants were faster for high adjectives (e.g., “good”) than for low adjectives (e.g., “bad”) and for affirmative phrases (e.g., “really really good”) than for negated phrases (e.g., “really not good”). Results on complexity reveal that participants were faster for phrases with 2 modifiers, e.g., “not really,” than phrases with 1 modifier, e.g., “not ###” (F(1,54) = 28.87, p < 0.001, η p 2 = 0.35, especially in affirmative phrases: complexity by negation interaction F(1,54) = 6.26, p = 0.015, η p 2 = 0.10), again replicating results of Experiment 1 (see Table B in S3 Table for pairwise comparisons between each pair of modifiers).

(A) Reaction times results for the online behavioral study (N = 55). Bars represent the participants’ mean ± SEM, and dots represent individual participants. Participants were faster for high adjectives (e.g., “good”) than for low adjectives (e.g., “bad”) and for affirmative phrases (e.g., “really really good”) than for negated phrases (e.g., “really not good”). These results replicate Experiment 1. ( B) Final interpretations (i.e., end of trajectories) of each phrase, represented by filled circles (purple = low, orange = high), averaged across adjective dimensions and participants, showing that negation never inverts the interpretation of adjectives to that of their antonyms. ( C) Mouse trajectories for low (purple) and high (orange) antonyms, for each modifier (shades of orange and purple) and for affirmative (left panel) and negated (right panel) phrases. Zoomed-in panels at the bottom demonstrate that mouse trajectories of affirmative phrases branch towards the adjective’s side of the scale and remain on that side until the final interpretation; in contrast, the trajectories of negated phrases first deviate towards the side of the adjective and subsequently towards the side of the antonym (except for “not not”). This result is confirmed by linear models fitted to the data at each time point in D . These results also replicate Experiment 1. ( D ) Beta values (average over 55 participants) over time, separately for affirmative and negated phrases. Thicker lines indicate significant time windows. Trials with “not not” were not included in this analysis as the trajectories pattern was different compared to the other conditions with negation. ( C , D ) Black vertical dashed lines indicate the presentation onset of each word: modifier 1, modifier 2 and adjective; each line and shading represent participants’ mean ± SEM. ( A , B , D ) *** p < 0.001; ** p < 0.01; * p < 0.05. Data are available on the Open Science Framework https://doi.org/10.17605/OSF.IO/5YS6B .

We replicated Experiment 1 in a new group of online participants (N = 55; Fig 3 ). The experimental procedure was the same as that of Experiment 1, except that no feedback was provided to participants based on the final interpretation, but only if the cursor’s movement violated the warnings provided during the familiarization phase (e.g., “you crossed the vertical borders”; see Materials and methods ). We performed the same data analyses performed for Experiment 1.

Experiment 2: MEG shows that negation weakens the representation of adjectives and recruits response inhibition networks

In this study (MEG experiment, N = 26), participants read adjective phrases comprising 1 or 2 modifiers (“not” and “really”) and scalar adjectives across different dimensions (e.g., “really really good,” “really not quiet,” “not ### dark”). Adjectives were selected to represent opposite poles (i.e., the antonyms) of the respective semantic scales: low pole of the scale (e.g., “bad,” “cool,” “quiet,” “dark”) and high pole of the scale (e.g., “good,” “warm,” “loud,” “bright”). A sequence of dashes was used to indicate the absence of a modifier. Fig 1B and S2 Table provide the comprehensive list of the linguistic stimuli. Participants were asked to indicate whether a probe (e.g., 6) represented the meaning of the phrase on a scale from “really really low” (0) to “really really high” (8) (yes/no answer, Fig 1B). Feedback consisted of a green or red cross, to which 1 and 0 was assigned to compute the average feedback score. Behavioral data of Experiment 2 replicate that of Experiment 1: Negated phrases are processed slower and with lower feedback score than affirmative phrases (main effect of negation for RTs: F(1,25) = 26.44, p < 0.001, η p 2 = 0.51; main effect of negation for feedback score: F(1,25) = 8.03, p = 0.009, η p 2 = 0.24).

The MEG analyses, using largely temporal and spatial decoding approaches [48], comprise 4 incremental steps: (1) we first identify the temporal correlates of simple word representation (i.e., the words “really” and “not” in the modifier position, and each pair of scalar adjectives in the second word position, i.e., the head position; see S2 Table); (2) we test lexical-semantic representations of adjectives over time beyond the single word level, by entering low (“bad,” “cool,” “quiet,” and “dark”) and high (“good,” “warm,” “loud,” and “bright”) antonyms in the same model (adjectives in purple versus orange in S2 Table). We then test the representation of the negation operator over time (modifiers in green versus red in S2 Table); (3) we then ask how negation operates on the representation of adjectives, by teasing apart 4 possible mechanisms (i.e., No effect, Mitigation, Inversion, Change; adjectives in purple versus orange for modifiers in green and red separately in S2 Table); (4) we explore changes in beta power as a function of negation (motivated by the literature implicating beta-band neural activity in linguistic processing).

(1) Temporal decoding of single word processing

The butterfly (bottom) and topography plots (top) in Fig 4A illustrate the grand average of the event-related fields elicited by the presentation of all words, as well as the probe, regardless of condition. Results of decoding analyses performed on these preprocessed MEG data (after performing linear dimensionality reduction; see Materials and methods) show that the temporal decoding of “really” versus “not” is significant between 120 and 430 ms and between 520 and 740 ms from the onset of the first modifier (dark gray shading, p < 0.001 and p = 0.001) and between 90 and 640 ms from the onset of the second modifier (light gray shading, p < 0.001, Fig 4B). Pairs of antonyms from different scales (regardless of specific modifier) were similarly decodable between 90 and 410 ms from adjective onset (quality: 110 to 200 ms, p = 0.002 and 290 to 370 ms, p = 0.018; temperature: 140 to 280 ms, p < 0.001; loudness: 110 to 410 ms, p < 0.001; brightness: 90 to 350 ms, p < 0.001, Fig 4C), reflecting time windows during which the brain represents visual, lexical, and semantic information (e.g., [7,49]). These results further show that single words can be decoded with relatively high accuracy (approximately 70%).

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 4. Evoked activity and temporal decoding of modifiers and adjectives as letter strings. (A) The butterfly (bottom) and topo plots (top) illustrate the event-related fields elicited by the presentation of each word as well as the probe, with a primarily visual distribution of neural activity right after visual onset (i.e., letter string processing). We performed multivariate decoding analyses on these preprocessed MEG data, after performing linear dimensionality reduction (see Materials and methods). Detector distribution of MEG system in inset box. fT: femtoTesla magnetic field strength. (B) We estimated the ability of the decoder to discriminate “really” vs. “not” separately in the first and second modifier’s position, from all MEG sensors. We contrasted phrases with modifiers “really ###” and “not ###,” and phrases with modifiers “### not” and “### really.” (C) We evaluated whether the brain encodes representational differences between each pair of antonyms (e.g., “bad” vs. “good”), in each of the 4 dimensions (quality, temperature, loudness, and brightness). The mean across adjective pairs is represented as a solid black line; significant windows are indicated by horizontal solid lines below. (B and C) AUC = area under the receiver operating characteristic curve, chance = 0.5 (black horizontal dashed line); for all panels: black vertical dashed lines indicate the presentation onset of each word: modifier 1, modifier 2, and adjective; each line and shading represent participants’ mean ± SEM. Data are available on the Open Science Framework https://doi.org/10.17605/OSF.IO/5YS6B. https://doi.org/10.1371/journal.pbio.3002622.g004

(2) Temporal and spatial decoding of adjectives and negation

After establishing that single words’ features can be successfully decoded in sensible time windows (see Fig 4), we moved beyond single word representation and clarified the temporal patterns of adjective and negation representation independently from their interaction and identified temporal windows where to expect changes in adjective representation as a function of negation. First, we selectively evaluated lexical-semantic differences between low (“bad,” “cool,” “quiet,” and “dark”) and high (“good,” “warm,” “loud,” and “bright”) adjectives, regardless of the specific scale (i.e., pooling over quality, temperature, loudness, and brightness) and by pooling over all modifiers. Temporal decoding analyses (see Materials and methods) reveal significant decodability of low versus high antonyms in 3 time windows between 140 and 560 ms from adjective onset (140 to 280 ms, p < 0.001; 370 to 460 ms: p = 0.009; 500 to 560 ms: p = 0.044, purple shading in Fig 5A). No significant differences in lexical-semantic representation between low and high antonyms were observed in later time windows (i.e., after 560 ms from adjective onset). The spatial decoding analysis illustrated in Fig 5B (limited to 50 to 650 ms from adjective onset; see Materials and methods) show that decoding accuracy for low versus high antonyms is significantly above chance in a widespread left-lateralized brain network, encompassing the anterior portion of the superior temporal lobe, the middle, and the inferior temporal lobe (purple shading in Fig 5B, significant clusters are indicated by a black contour: left temporal lobe cluster, p = 0.002). A significant cluster was also found in the right temporal pole, into the insula (p = 0.007). Moreover, we found significant clusters in the bilateral cingulate gyri (posterior and isthmus) and precunei (left precuneus/cingulate cluster, p = 0.009; right precuneus/cingulate cluster, p = 0.037). Overall, these regions are part of the (predominantly left-lateralized) frontotemporal brain network that underpins lexical-semantic representation and composition [7,8,46,49–55].

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 5. Temporal and spatial decoding of antonyms across all scales and temporal decoding of negation. (A) Decoding accuracy (purple line) of lexical-semantic differences between antonyms across all scales (i.e., pooling over “bad,” “cool,” “quiet,” and “dark”; and “good,” “warm,” “loud,” and “bright” before fitting the estimators) over time, regardless of modifier; significant time windows are indicated by purple shading. (B) Decoding accuracy (shades of purple) for antonyms across all scales over brain sources (after pooling over the 4 dimensions), between 50 and 650 ms from adjective onset. Significant spatial clusters are indicated by a black contour. (C) Decoding accuracy of negation over time, as a function of the number of modifiers (1 modifier: dark red line and shading; 2 modifiers: light red line and shading). 1 modifier: “really ###,” “### really,” “not ###,” “### not”; 2 modifiers: “really really,” “really not,” “not really,” “not not.” Significant time windows are indicated by dark red (1 modifier) and light red (2 modifiers) shading. For all panels: AUC: area under the receiver operating characteristic curve, chance = 0.5 (black horizontal dashed line); black vertical dashed lines indicate the presentation onset of each word: modifier1, modifier2, and adjective; each line and shading represent participants’ mean ± SEM; aff = affirmative, neg = negated; LH = left hemisphere; RH = right hemisphere. Data are available on the Open Science Framework https://doi.org/10.17605/OSF.IO/5YS6B. https://doi.org/10.1371/journal.pbio.3002622.g005

Next, we turn to representations of negation over time. We performed a temporal decoding analysis for phrases containing “not” versus phrases not containing “not,” separately for phrases with 1 and 2 modifiers (to account for phrase complexity; see S2 Table for a list of all trials). For phrases with 1 modifier, the decoding of negation is significantly higher than chance throughout word 1 (−580 to −500 ms from adjective onset, p = 0.005), then again throughout word 2 (−470 to 0 ms from adjective onset, p < 0.001). After the presentation of the adjective, negation decodability is again significantly above chance between 0 and 40 ms (p = 0.034) and between 230 and 290 ms from adjective onset (p = 0.018; dark red line and shading in Fig 5C). Similarly, for phrases with 2 modifiers, the decoding of negation is significantly higher than chance throughout word 1 (−580 to −410 ms from adjective onset, p = 0.002), throughout word 2 (−400 to 0 ms from adjective onset, p < 0.001), and for a longer time window from adjective onset compared to phrases with one modifier, i.e., between 0 and 720 ms (0 to 430 ms, p < 0.001; 440 to 500 ms, p = 0.030; 500 to 610 ms, p < 0.001; 620 to 720 ms, p < 0.001; light red line and shading in Fig 5C). The same analysis time-locked to the onset of the probe shows that negation is once again significantly decodable between 230 and 930 ms after the probe, likely being reinstated when participants perform the task (S2 Fig).

Cumulatively, these results suggest that the brain encodes negation every time a “not” is presented and maintains this information up to 720 ms after adjective onset. Further, they show that the duration of negation maintenance is amplified by the presence of a second modifier, highlighting combinatoric effects [2,6,56].

(3) Effect of negation on lexical-semantic representations of antonyms over time

The temporal decoding analyses performed separately for adjectives and for negation demonstrate that the brain maintains the representation of the modifiers available throughout the presentation of the adjective. Here we ask how negation operates on the representation of the antonyms at the neural level, leveraging theoretical accounts of negation [11,12,42–44], behavioral results of Experiment 1, and 2 complementary decoding approaches. We test 4 hypotheses (see Predictions in Fig 6A): (1) No effect of negation: Negation does not change the representation of adjectives (i.e., “not low” = “low”). We included this hypothesis based on the 2-step theory of negation, wherein the initial representation of negated adjectives would not be affected by negation [27]. (2) Mitigation: Negation weakens the representation of adjectives (i.e., “not low” < “low”). (3) Inversion: Negation inverts the representation of adjectives (i.e., “not low” = “high”). Hypotheses (2) and (3) are derived from previous linguistics and psycholinguistics accounts on comprehension of negated adjectives [42–44]. Finally, (4) Change: We evaluated the possibility that negation might change the representation of adjectives to another representation outside the semantic scale defined by the 2 antonyms (e.g., “not low” = e.g., “fair”). Importantly, these predictions focus on how negation affects representations rather than on when. Thus, a combination of mechanisms may be observed over time (e.g., first no effect and then inversion).

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 6. Predictions, decoding approaches, and results of the effect of negation on the representation of adjectives. (A) We tested 4 possible effects of negation on the representation of adjectives: (1) No effect; (2) Mitigation; (3) Inversion; (4) Change (left column). Note that we depicted predictions of (3) Inversion on the extremes of the scale, but a combination of inversion and mitigation would have the same expected outcomes. We performed 2 sets of decoding analyses (right column): (i) We trained estimators on low (purple) vs. high (orange) antonyms in affirmative phrases and predicted model accuracy and probability estimates of low vs. high antonyms in negated phrases (light and dark red bars). (ii) We trained estimators on low vs. high antonyms in affirmative and negated phrases together and predicted model accuracy and probability estimates in affirmative (light and dark green bars) and negated phrases (light and dark red bars) separately. (B) Decoding accuracy (red line) over time of antonyms for negated phrases, as a result of decoding approach (i). Significant time windows are indicated by red shading and horizontal solid lines. (C) Decoding accuracy of antonyms over time for affirmative (green line) and negated (red line) phrases, as a result of decoding approach (ii). Significant time windows for affirmative and negated phrases are indicated by green and red shading and horizontal solid lines. The significant time window of the difference between affirmative and negated phrases is indicated by a black horizontal solid line. (D) Probability estimates for low (light red) and high (dark red) negated antonyms averaged across the significant time windows depicted in B. Bars represent the participants’ mean ± SEM and dots represent individual participants. (E) Probability estimates for low (light green) and high (dark green) affirmative adjectives and for low (light red) and high (dark red) negated adjectives, averaged across the significant time window depicted as a black horizontal line in C. Chance level of probability estimates was computed by averaging probability estimates of the respective baseline (note that the baseline differs from 0.5 due to the different number of trials for each class in the training set of decoding approach (i)). Bars represent the participants’ mean ± SEM and dots represent individual participants. (B and C) AUC: area under the receiver operating characteristic curve, chance = 0.5 (black horizontal dashed line); each line and shading represent participants’ mean ± SEM. (B–E) The black vertical dashed line indicates the presentation onset of the adjective; green = affirmative phrases, red = negated phrases. Data are available on the Open Science Framework https://doi.org/10.17605/OSF.IO/5YS6B. https://doi.org/10.1371/journal.pbio.3002622.g006

To adjudicate between these 4 hypotheses, we performed 2 complementary sets of decoding analyses. Decoding approach (i): we computed the accuracy with which estimators trained on low versus high antonyms in affirmative phrases (e.g., “really really bad” versus “really really good”) generalize to the representation of low versus high antonyms in negated phrases (e.g., “really not bad” versus “really not good”) at each time sample time-locked to adjective onset (see Materials and methods); decoding approach (ii): we trained estimators on low versus high antonyms in affirmative and negated phrases together (in 90% of the trials) and computed the accuracy of the model in predicting the representation of low versus high antonyms in affirmative and negated phrases separately (in the remaining 10% of the trials; see Materials and methods). Decoding approach (ii) allows for a direct comparison between AUC and probability estimates in affirmative and negated phrases and to disentangle predictions (1) No effect from (2) Mitigation. Expected probability estimates (i.e., the averaged class probabilities for low and high classes) as a result of decoding approach (i) and (ii) are depicted as light and dark, green and red bars under Decoding approach in Fig 6A.

Temporal decoding approach (i) reveals that the estimators trained on the representation of low versus high antonyms in affirmative phrases significantly generalize to the representation of low versus high antonyms in negated phrases, in 4 time windows between 130 and 550 ms from adjective onset (130 to 190 ms, p = 0.039; 200 to 270 ms: p = 0.003; 380 to 500 ms: p < 0.001; 500 to 550 ms: p = 0.008; red shading in Fig 6B). Fig 6D depicts the probability estimates averaged over the significant time windows for low and high antonyms in negated phrases. These results only support predictions (1) No effect and (2) Mitigation, thus invalidating predictions (3) Inversion and (4) Change. S3 Fig illustrates a different approach that similarly leads to the exclusion of prediction (3) Inversion.

Temporal decoding approach (ii) shows significant above chance decoding accuracy for affirmative phrases between 130 and 280 ms (p < 0.001) and between 370 and 420 ms (p = 0.035) from adjective onset. Conversely, decoding accuracy for negated phrases is significantly above chance only between 380 and 450 ms after the onset of the adjective (p = 0.004). Strikingly, negated phrases are associated with significantly lower decoding accuracy than affirmative phrases in the time window between 130 and 190 ms from adjective onset (p = 0.040; black horizontal line in Fig 6C). Fig 6E represents the probability estimates averaged over this 130 to 190 ms significant time window for low and high antonyms, separately in affirmative and negated phrases, illustrating reduced probability estimates for negated compared to affirmative phrases. No significant difference between decoding accuracy of affirmative and negative phrases was found for later time windows (500 to 1,000 ms from adjective onset, p > 0.05). A follow-up analysis where we trained and tested on low versus high antonyms in affirmative and negated phrases separately shows similar results (S4A Fig). Furthermore, the analysis including all trials, regardless of feedback score, also shows similar results (S4B Fig).

Overall, the generalization of representation from affirmative to negated phrases and the higher decoding accuracy (and probability estimates) for affirmative than negated phrases within the first 500 ms from adjective onset (i.e., within the time window of lexical-semantic processing shown in Fig 5A) provide direct evidence in support of prediction (2) Mitigation, wherein negation weakens the representation of adjectives. The alternative hypotheses did not survive the different decoding approaches.

(4) Changes in beta power as a function of negation

We distinguished among 4 possible mechanisms of how negation could operate on the representation of adjectives and demonstrated that negation does not invert or change the representation of adjectives but rather weakens the decodability of low versus high antonyms within the first approximately 300 ms from adjective onset (Fig 6C; with AUC for affirmative and negated adjectives being significantly different for about 60 ms within this time window). The availability of negation upon the processing of the adjective (Fig 5A and 5C) and the reduced decoding accuracy for antonyms in negated phrases (Fig 6C) raise the question of whether negation operates through inhibitory mechanisms, as suggested by previous research employing action-related verbal material [35–37]. We therefore performed time-frequency analyses, focusing on beta power (including low-beta: 12 to 20 Hz, and high-beta: 20 to 30 Hz [57]; see Materials and methods), which has been previously associated with inhibitory control [58] (see S5 Fig for comprehensive time-frequency results). We reasoned that, if negation operates through general-purpose inhibitory systems, we should observe higher beta power for negated than affirmative phrases in sensorimotor brain regions.

Our results are consistent with this hypothesis, showing significantly higher low-beta power (from 229 to 350 ms from the onset of modifier1: p = 0.036; from 326 to 690 ms from adjective onset: p = 0.012; red line in Fig 7A) and high-beta power (from 98 to 271 ms from adjective onset: p = 0.044; yellow line in Fig 7A) for negated than affirmative phrases. S6 Fig further shows low- and high-beta power separately for negated and affirmative phrases, compared to phrases with no modifier (i.e., with “### ###”).

PPT PowerPoint slide

PNG larger image

TIFF original image Download: Fig 7. Differences in beta power over time between negated and affirmative phrases. (A) Differences in low (12–20 Hz, red) and high (21–30 Hz, yellow) beta power over time between negated (i.e., “### not,” “not ###,” “really not,” “not really,” “not not”) and affirmative phrases (i.e., “### really,” “really ###,” “really really”). Negated phrases show higher beta power compared to affirmative phrases throughout the presentation of the modifiers and for a sustained time window from adjective onset up to approximately 700 ms; significant time windows are indicated by red (low-beta) and yellow (high-beta) shading; black vertical dashed lines indicate the presentation onset of each word: modifier1, modifier2, and adjective; each line and shading represent participants’ mean ± SEM. (B) Differences (however not reaching statistical significance, α = 0.05) in high-beta power between negated and affirmative phrases (restricted between 97 and 271 ms from adjective onset, yellow cluster). (C) Significant differences in low-beta power between negated and affirmative phrases (restricted between 326 and 690 ms from adjective onset) in the left precentral, postcentral, and paracentral gyrus (red cluster). Note that no significant spatial clusters were found in the right hemisphere. Data are available on the Open Science Framework https://doi.org/10.17605/OSF.IO/5YS6B. https://doi.org/10.1371/journal.pbio.3002622.g007

Our whole-brain source localization analysis shows significantly higher low-beta power for negated than affirmative phrases in the left precentral, postcentral, and paracentral gyri (p = 0.012; between 326 and 690 ms from adjective onset, red cluster in Fig 7C). For high-beta power, similar (albeit not significant) sensorimotor spatial patterns emerge (yellow cluster in Fig 7B).

[END]
---
[1] Url: https://journals.plos.org/plosbiology/article?id=10.1371/journal.pbio.3002622

Published and (C) by PLOS One
Content appears here under this condition or license: Creative Commons - Attribution BY 4.0.

via Magical.Fish Gopher News Feeds:
gopher://magical.fish/1/feeds/news/plosone/