Kids must be able to identify what deserves attention. For example, they have to be able to focus on the teacher, not on the voices out in the hall. Kids have to be able to stay reasonably attentive over time. This could be for a three-minute presentation or for a minute lecture. Kids should be able to shift attention briefly when important new information is introduced.
For example, they should be able to focus on a brief announcement on the intercom. Then they should be able to turn their attention back to the teacher. This is where new information is first held. Working memory is an active and fairly instant process. It allows us to use new and learned information while we are in the middle of an activity.
Imagine a social studies class. The teacher is talking about great explorers. As students listen, their working memory does things with the information they hear so it can have meaning and relevance. This often involves putting pieces of information in some kind of order.
That could be chronological order. For example, kids might picture Columbus before Pizarro on a timeline. They might put what they know about Lewis and Clark after Columbus.
When kids have working memory issues, the information they put into long-term storage may be disorganized in their minds. And so it may not be very useful. Vision Research , 43 — LaRocque, J. Decoding attended information in short-term memory: An EEG study. Journal of Cognitive Neuroscience , 25 1 , — Lavie, N. Distracted and confused? Trends in Cognitive Sciences , 9, 75— Load theory of selective attention and cognitive control. Lawrence, B.
Interference with spatial working memory: An eye movement is more than a shift of attention. The effects of eye and limb movements on working memory. Memory , 9, — Lepsien, J. Modulation of working-memory maintenance by directed attention. Neuropsychologia , 49, — Lewis-Peacock, J.
Neural evidence for a distinction between short-term memory and the focus of attention. Journal of Cognitive Neuroscience , 24, 61— The removal of information from working memory.
Annals of the New York Academy of Science , , 33— Liefooghe, B. Working memory costs of task switching. Logie, R. Feature binding in visual short-term memory is unaffected by task-irrelevant changes of location, shape, and color. Luria, R. The contralateral delay activity as a neural measure of visual working memory.
Neuroscience and Biobehavioral Reviews , 62, — Ma, W. Changing concepts of working memory. Nature Neuroscience Reviews , 17, — Mall, J. Visual selective attention is equally functional for individuals with low and high working memory capacity: Evidence from accuracy and eye movements.
Mallett, R. Behavioral decoding of working memory items inside and outside the focus of attention. Annals of the New York Academy of Sciences , 1 , — Marshall, L. Obligatory encoding of task-irrelevant features depletes working memory resources. Journal of Vision , 13, 1— McElree, B. Accessing recent events. In: B. Ross Ed. San Diego: Academic Press. McVay, J. Conducting the train of thought: Working memory capacity, goal neglect, and mind wandering in an executive-control task.
Why does working memory capacity predict variation in reading comprehension? On the influence of mind wandering and executive attention. Meiran, N. Powerful instructions: Automaticity without practice. Current Directions in Psychological Science , 26, — Mendoza-Halliday, D.
Neuronal population coding of perceived and memorized visual features in the lateral prefrontal cortex. Mongillo, G. Synaptic theory of working memory. Science , , — Monsell, S. Task switching. Trends in Cognitive Sciences , 7, — Morey, C. Visual short-term memory always requires general attention. Navon, D. On the economy of the human-processing system. Psychological Review , 86, — Queuing or sharing? A critical evaluation of the single-bottleneck notion. Cognitive Psychology , 44, — Nieuwenstein, M.
Beyond a mask and against the bottleneck: Retroactive dual-task interference during working memory consolidation of a masked visual target. Oberauer, K. Removing irrelevant information from working memory: A cognitive aging study with the modified Sternberg task.
Access to information in working memory: Exploring the focus of attention. Selective attention to elements in working memory. Experimental Psychology , 50 4 , — Control of the contents of working memory — A comparison of two paradigms and two age groups. Design for a working memory. The focus of attention in working memory — from metaphors to mechanisms. Removal of irrelevant information from working memory: Sometimes fast, sometimes slow, and sometimes not at all.
Annals of the New York Academy of Science , , — The role of long-term memory in a test of visual working memory: Proactive facilitation but no proactive interference. Serial and parallel processes in working memory after practice. Dissociating retention and access in working memory: An age-comparative study of mental arithmetic. What limits working memory capacity? Psychological Bulletin , , — Attention to information in working memory.
Current Directions in Psychological Science , 21, — Evidence against decay in verbal working memory. Further evidence against decay in working memory. Journal of Memory and Language , 73, 15— Modeling working memory: An interference model of complex span. An interference model of visual working memory. Psychological Review , , 21— Analogous mechanisms of selection and updating in declarative and procedural working memory: Experiments and a computational model.
Cognitive Psychology , 66, — Olivers, C. Interactions between visual working memory and visual attention. Frontiers in Bioscience , 13, — Different states in visual working memory: When it guides attention and when it does not. Trends in Cognitive Sciences , 15, — Park, S. Concurrent working memory load can facilitate selective attention: Evidence for specialized load. Pashler, H. Dual-task interference in simple tasks: Data and theory.
Plate, T. Convolution-based memory models. In: L. Nadel Ed. London: Nature Publishing Group. Ralph, A. Immunity to proactive interference is not a property of the focus of attention in working memory. Randall, J.
Mind-wandering, cognition, and performance: A theory-driven meta-analysis of attention regulation. Rerko, L. Retro-cue benefits in working memory without sustained attention. Ricker, T. The nature of short-term consolidation in visual working memory. Journal of Experimental Psychology: General. Rose, N. Reactivation of latent working memories with transcranial magnetic stimulation. Sander, M. Binding and strategic selection in working memory: A lifespan dissociation.
Psychology and Aging , 26, — Saults, J. A central capacity limit to the simultaneous storage of visual and auditory arrays in working memory. Schneider, W. Controlled and automatic human information processing: I. Detection, search, and attention. Psychological Review , 84, 1— Shiffrin, R. Controlled and automatic human information processing: II. Perceptual learning, automatic attending, and a general theory. Psychological Review , 84, — Shipstead, Z. The mechanisms of working memory capacity: Primary memory, secondary memory, and attention control.
Journal of Memory and Language , 72, — Singh, K. Working memory capacity mediates the relationship between removal and fluid intelligence. Journal of Memory and Language , , 18— Soto, D. Automatic guidance of attention from working memory. Trends in Cognitive Sciences , 12, — Souza, A. In search of the focus of attention in working memory: 13 years of the retro-cue effect. The contributions of visual and central attention to visual working memory.
Refreshing memory traces: Thinking of an item improves retrieval from visual working memory. Annals of the New York Academy of Sciences , , 20— Where to attend next: Guiding refreshing of visual, spatial, and verbal representations in working memory. Annals of the New York Academy of Science. Sprague, T. Restoring latent visual working memory representations in human cortex.
Neuron , 91, — Szmalec, A. Response selection involves executive control: Evidence from the selective interference paradigm. Tehan, G. Thalmann, M. Revisiting the attentional demands of rehearsal in working-memory tasks. Journal of Memory and Language , , 1— Theeuwes, J. Visual selection: Usually fast and automatic; seldom slow and volitional.
Journal of Cognition , 1, 1— Todd, J. Capacity limit of visual short-term memory in human posterior parietal cortex. Nature , , — Posterior parietal cortex activity predicts individual differences in visual short-term memory capacity. Tombu, M. A central capacity sharing model of dual-task performance. Tremblay, S. Rehearsal in serial memory for visual-spatial information: Evidence from eye movements. Tsubomi, H. Neural limits to representing objects still within view.
Ueno, T. Disruption of visual feature binding in working meemory. Unsworth, N. Consistency of attentional control as an important cognitive trait: A latent variable analysis. Intelligence , 49, — Working memory and fluid intelligence: Capacity, attention control, and secondary memory retrieval.
Cognitive Psychology , 71, 1— Van der Stigchel, S. The effects of a task-irrelevant visual event on spatial working memory. Rapid influences of cued visual memories on attentional guidance. Annals of the New York Academy of Sciences. Spatially selective alpha oscillations reveal moment-by-moment trade-offs between working memory and attention. Journal of Cognitive Neuroscience , 30, — In competition for the attentional template: Can multiple items within visual working memory guide attention?
Vergauwe, E. Do mental processes share a domain-general resource? Psychological Science , 21, — The impact of storage on processing: How is information maintained in working memory?
Vogel, E. Neural activity predicts individual differences in visual working memory capacity. Neural measures reveal individual differences in controlling access to working memory. Nature , 24 , — Watanabe, K. Neural mechanisms of dual-task interference and cognitive capacity limitation in the prefrontal cortex. Nature Neuroscience , 17, — Wickens, C.
The structure of attentional ressources. In: R. Nickerson Ed. Hillsdale, N. Wilhelm, O. What is working memory capacity, and how can we measure it?
Williams, M. Visual-spatial attention aids the maintenance of object representations in visual working memory. Wolff, M. Dynamic hidden states underlying working-memory-guided behavior. Nature Neuroscience , 20, — Woodman, G. Where do we store the memory representations that guide attention?
The role of working memory and long-term memory in visual search. Visual Cognition , 14, — Xu, Y. Dissociable neural mechanisms supporting visual short-term memory for objects.
Nature , , 91— Yeung, N. The neural basis of error detection: Conflict monitoring and the error-related negativity. Yu, Q. Occipital, parietal, and frontal cortices selectively maintain task-relevant features of multi-feature objects in visual working memory. NeuroImage , , 97— Journal of Cognition , 2 1 , p. Oberauer K. Journal of Cognition.
Journal of Cognition , 2 1 , Oberauer, Klaus. Journal of Cognition 2 1 : Journal of Cognition 2, no. Journal of Cognition , vol. Start Submission Become a Reviewer. Special Collection: Theoretical review with commentaries: attention and working memory. Abstract There is broad agreement that working memory is closely related to attention. Keywords: Working memory, Attention, Cognitive Control.
How to Cite: Oberauer, K. This was observed in single blue and in dual red task conditions. The single-task vs.
Panel d The coefficient of variations CV did not significantly differ between durations irrespective of single-task and dual-task conditions.
However, the CV significantly differed across single-task and dual-task conditions, with the CV in the dual-task being larger than in the single-task. These results, combined with results in panel b , suggest that the WM task may have an impact on the likelihood estimates of duration.
A main effect of Task was found Fig. Table 4 , Main effect of Task. Although we did not find a main effect of Duration Supp. Data are reported in Supp. Figure 1a and statistical contrasts in Supp. In other words, paying attention to duration lengthened subjective duration, but this lengthening did not scale with target duration.
In fact, this effect either increased or decreased with longer durations according to the experimental groups Supp. Given that the single-task condition was identical in both groups in this condition, there was no feedback provided to the No Feedback or to the Feedback groups , the variance observed here may be accounted for by random inter-individual variability. The pattern of results thus suggests that the internal representation of duration estimation would be somewhere between the blue and the grey data, with attention effectively acting as an up and down switch of this internal estimation; interestingly, the effect of paying attention to time did not robustly increase with the length of duration.
To directly test this, we computed the coefficient of variations CVs separately for each duration in the single-task and in the 0-back dual-task condition Fig. No significant differences were found as a function of Duration Supp. In the dual-task conditions including all n-back conditions , duration was overall significantly underestimated Fig. Table 5 , Main effect of Duration. Consistent with the above observations, significant differences between the single- and the dual-task conditions were observed Supp.
As could be expected, a significant interaction between Task and Duration was observed so that differences between single- and dual-task conditions increased with longer durations Supp. This effect was driven by the scaling effect observed in the dual-task condition as a function of duration. The full data for the regression model are reported in Supp. The scalar property of time estimation is the empirical observation that the variance of subjective durations scale with duration. In Fig. To test whether the scaling induced by WM affected the scalar property of timing during dual-task, we fitted an lme model using the variance computed from all subjective duration estimates as dependent variable.
As predicted, this analysis revealed a main effect of Duration Supp. We then computed the coefficient of variations CVs separately for each duration, in the single- and dual-task conditions Fig. To test whether the CVs changed as a function of single- or dual-task condition and as a function of duration, we performed an lme model including a by-subject random slope for the effect of Task Supp. We found that the CVs in the single-task condition were significantly smaller than in the dual-task conditions across all durations Supp.
This pattern suggests that WM may interfere with duration estimation in a manner consistent with the hypothesis that the precision of estimated duration held in WM memory may be modulated To test this hypothesis, we thus turned to the effect of WM load on prospective duration estimation.
For this, we fitted an lme model using the above-mentioned factors as fixed effects. The underestimation of duration was found to increase with higher WM loads Supp. Table 8 , Main effect of WM load. A main effect of duration was found in both the Feedback and in the No Feedback groups Supp. Table 8 , Main effect of Duration.
We also found a significant interaction between WM load and Duration Supp. This interaction suggested that the differences in the underestimation of duration across target durations increased with higher WM load. Figure 3a provides a synthetic view of the main effects of attention and WM load combined in both experimental groups. Figure 3 provides the same data sorted as a function of the Feedback and the No Feedback group.
WM load incrementally affects duration estimation. All data illustrated here combined the Feedback and No Feedback groups who showed comparable effects. Their separate analysis is provided in Supp. Figure 1. Hues are durations with lighter darker hues marking shorter longer durations.
The dashed line represents the ideal observer so that positive values indicate a subjective overestimation of duration and negative values indicate a subjective underestimation of duration. Irrespective of feedback cf. Figure 1 , durations were overestimated in single-task and underestimated in dual-tasks.
Remarkably, the underestimation of duration in dual-task condition systematically increased with WM load. In both single and dual-tasks, the response variance significantly increased with duration in agreement with the scalar property.
Interestingly the peak and width of the distributions varied as a function of WM load. The distribution observed in the single-task blue are also provided for comparison. The shift in the peak distribution during the dual-task can readily be seen to vary with the n-back task across all three durations. We thus asked whether the CVs were affected by WM load. Figure 3c reports the CVs as a function of WM load. The CVs were fitted with an lme model using WM load as fixed effect.
A main effect of WM load was found Supp. This showed that WM-load affected the CVs irrespective of the duration. Our behavioral study explored the effect of attention and working memory load on prospective duration estimation. Our main findings are that, i paying attention to the estimation of duration lengthens subjective duration; ii splitting attention to a concurrent WM task shortens perceived duration; iii the magnitude of attentional over-estimation was comparable across durations in single-task; iv attention did not affect CVs; v performing a concurrent WM task shortened perceived duration proportionally to the WM load; vi the effect of WM scaled with duration so that shorter durations were less affected by WM than longer durations; vii WM load affected timing precision equally across the three target durations so that an increase in WM load also increased the coefficient of variations CVs.
We discuss the implications of these findings in the context of a Bayesian framework for time estimation with the main effects synthesized and compiled in Fig. Summary of attention and WM load interferences from the perspective of Bayesian time estimation. First, attention blue may systematically shift the likelihood distribution of the original target duration prior grey : paying attention to time may shift the distribution towards larger estimates resulting in an overestimation of the duration upper blue, posterior whereas diverting attention away from time may shift the distribution towards shorter estimates bottom blue, posterior.
Importantly, our results suggests that the precision of under- and over-estimations of duration due to attention is comparable across the full duration range. To the contrary, WM may skew the likelihood distribution as a function of duration so that WM load would affect both the mean and the width of likelihood distributions in addition to scaling with duration.
This would result in increasingly wider posterior distributions shifted towards shorter durations Fig. Three major features of time perception are the central tendency, the range effect, and the scalar variability The central tendency effect is now considered a signature of Bayesian computations in the estimation of magnitudes such as duration 35 , 37 , 38 , 39 , 40 , Second , given a range of durations, the central tendency is more pronounced for longer than for shorter durations 37 , suggesting that the estimation of duration is shaped by context.
For instance, recent findings have demonstrated the existence of carry-over effects in duration estimation, which were accounted for by the combination of perceptual and decisional biases induced by the preceding context 42 , Altogether, regression to the mean and range effects have been suggested to emerge from the need to minimize errors in a noisy decision process governed by Bayesian computations 37 , 38 , 39 , 40 , 43 , The estimation of duration or interval timing in this context is realized by taking into account the likelihood estimates of the elapsed time i.
In our study, manipulating attention to time appeared to have an effect comparable to error minimization due to context: overall, shorter durations tended to be overestimated and longer durations tended to be underestimated whether paying full attention to time in the single-task or being in a dual-task condition Fig.
Additionally, paying attention to time affected duration in a comparable manner across target durations with no change in precision as assessed by CVs.
One possible explanation, consistent with a general role of attention as gain modulation 45 , would be that the likelihood estimates for a given duration may be shifted towards longer durations when paying attention to time, but towards shorter durations when attention is diverted away from it Fig. This pattern held, irrespective of target durations: in the single-task condition, paying attention to time contributed to an overestimation of duration with a comparable precision as paying attention away from time, which yielded underestimation.
Considering that attention did not scale with estimated duration in the single-task, and did not affect the CVs in the single-task or in the 0-back dual-task, why the decision criterion would not scale with duration remain puzzling. Altogether, neither the central tendency nor the range effect seemed to be differentially impacted by attention in this task. Follow up studies would thus be very helpful in determining the conditions under which the effect of attention could scale with the range of durations being used and whether attention can be conceived as a bias or gain function of duration likelihood estimations.
The third property of magnitude estimation, and interval timing in particular, is the scalar variability consisting in the observation that variance scales with duration, i. Said differently, scalar variability is the property of noisy representations onto which the brain computes In the scalar timing theory 12 , scalar noise characterizes time representations, and more generally magnitude representations Consistent with this general property of timing, we report scalar variability in both single- and dual-tasks when using a numerical estimation of duration in the supra-second range.
Additionally, the WM load parametrically shortened perceived durations so that the higher the n-back, the more underestimated subjective durations were. This effect scaled with durations so that the WM load affected shorter durations less than it affected longer durations, suggesting that the WM load may not have only skewed the likelihood estimates of elapsed time but also affected their variance. In this study, this effect is quite distinct from the attentional effect. In a Bayesian framework, this would amount to the likelihood of the duration estimates scaling with memory load Fig.
According to timing models, and within a Bayesian framework, the multiplicative factor accounting for scalar variability could intervene at the likelihood estimation stage, or at the transferred posterior, which would be mechanistically equivalent to memory scaling. Previous modeling approaches of time estimation have, by default, assigned scalar variability at the stage of the likelihood estimation 37 , 38 , 39 , 40 , 43 , 44 although, as previously discussed, the Bayesian framework makes no assumption regarding the scalar property of timing In other words, the origin of scalar variability in Bayesian models of time estimation does not constitute a functionally relevant variable in time computation and rather been assigned as an ad-hoc property of interval timing.
Scalar variability has generally been considered as the noise of remembered representations which either originates from memory itself, or from its read out, but not from the measurement of the magnitude per se In scalar timing theory 48 , the scalar property would originate from a scaling factor which is multiplied to the experienced time interval and whose origin is also tied to memory Our results thus support the notion that manipulating WM load while timing scales with the estimation of duration.
Specifically, the effect of WM on duration estimation quantified here appears consistent with the notion of precision in the representation of duration: for instance, scalar property may emerge from the iterative assignment of the scalar factor in the course of the experienced duration, due to the maintenance of duration estimates in WM.
This could also occur at a later decisional stage through direct comparison between the stored duration and the WM output. The observed scaling of duration estimation with WM load is in line with the notion that the representation of duration scales with noise, but also with alternative interpretations suggesting that the representation of duration in memory may be subject to deterioration in precision with increased WM load In a recent study, the source of scalar variability was proposed to put a limit on the precision with which quantities may be represented in the brain 49 , and this proposal appears consistent with the observation that increased information held in WM would interfere with the precision of quantity representation.
The combination of computational approaches and neuroimaging would help make the case on the distinct effects of attention and WM load on time estimation.
Although many brain regions engaged during timing overlap with attention and memory networks 50 , 51 , recent fMRI evidence has also shown some selective engagement of cortical regions during temporal estimation tasks [e. Functionally, recent hypotheses have also emerged suggesting that oscillatory multiplexing may contribute to the precision of maintained duration estimation The prediction that behavior alone could not disentangle the possible physiological implementations of the scalar property was previously raised with the hypothesis that the scalar property may result from a computational sampling procedure between memorized event timing as opposed to duration retrieval Whether the attentional gain regulation, and the time-information scaling trade-off reported here implicate neural oscillations thus remain to be tested.
All were right-handed with corrected-to-normal vision, no history of psychological disorders and all were naive as to the purpose of the study. All participants were compensated for their participation. In the single-task condition, participants solely performed a duration estimation favoring the full deployment of attention towards timing. In the dual-task condition, participants concurrently performed a duration estimation and an n-back working memory WM task.
This experimental manipulation invited participants to split attention between timing and WM. In the dual-task, a typical trial consisted of a visual n-back WM task during the entire length of the duration trial. In the n-back WM task, visual stimuli were white point Arial font capital letters centered on a grey background.
A stream of letters was built by connecting several random-generated-letter chunks of 10 stimuli each and a target letter was placed in a pseudo-random position on each of these chunks to ensure a uniform deployment of attention on the n-back task.
In the 1, 2 and 3-back conditions, participants were asked to press the space-bar when the current stimulus and the stimulus in the n th position before it were identical the higher the n th position, the higher the memory task demands. After a first training block on the n-back task, participants were asked to estimate the length of time during which they were engaged in the task, and which was bounded by two red dots at the beginning and at the end of each duration trial.
No feedback was provided regarding subjective time estimates in the Feedback or in the No Feedback group. All possible combinations of n-back WM block and duration were tested four times per participants for a total of 48 trials per experimental condition Fig. Trials were presented in a pseudo-random order using a Latin square design. On each of these trials, only a fixation cross was displayed over a grey background.
All statistical analyses were carried out in the R programming language R Core Team, and RStudio environment RStudioTeam , using the lme4 54 , betareg 55 and lsmeans 56 software packages.
The logic of the statistical analyses reported in this study is described below for the different quantifications. For duration estimates and for the reaction times RTs in the WM task, we used linear mixed effect lme models, which can be thought of as a generalization of linear regression models. In lme models, data are not aggregated so that statistics are made on all empirical observations. Additionally, and unlike repeated measures ANOVAs in which comparisons are made between averaged data single-trial observations being lost , each observation was here taken into account and the inter-individual variability was considered as a random effect.
This approach increases statistical power without over-fitting the data. Separate regression models were fitted to the entire data set i. For all dependent variables duration estimates in the duration task, and reaction times in WM , the initial lme model started with the mean component, the random effect and the dependent variable; we then incrementally added the predictor variables e.
The AIC is a measure that optimizes model fit by taking into account the amount of explained variance as well as the degrees of freedom. This procedure ensures that the model achieves the best fit to the data with the minimum number of predictor variables. When two models are compared, the AIC provides information about whether the predictors added in the second model account for a significant amount of variance in the dependent variable. The best model corresponds to the minimal AIC.
For instance, in the reported tables e. Table 1 , the list of models is provided along with their respective AIC. The model that best fit the data is the one with the minimal AIC.
Consistent with this, the best models can also be found using Chi square. Chisq comparing one model in the list to the next e. The last comparison providing a significant effect points to the best model. For the hit and false alarm rates in the WM task, we used a beta regression model in which the analysis of dependent variables was expressed as a ratio assuming values in a standard unit interval 0, 1 57 , We used beta regression models because they can easily accommodate the asymmetry of heteroskedastic data such as hit and false alarm rates acquired here in the WM task , whose variability increases around the mean but decreases towards the lower and upper limits of the standard unit 57 , Each model was built including two sub-models: a regression model for the mean, and a regression model for the variability.
The later allowed information from the predictors to better estimate the non-normal distribution common in proportional data.
The procedure for the beta regression models was similar to the one used in the lme so that after the initial model using the mean component, we added the precision component and applied the standard step-wise procedure. After completion of this procedure, we tested whether the inclusion of the precision component was justified by comparing the AIC of the initial model with the more complex model including both the mean and the precision components.
This was tested with the likelihood ratio test. Statistical significance between regression coefficients in the lme and beta regression models were directly drawn from the selected final model, and tested with t-tests and Wald tests yielding Z, respectively. To avoid biases due to unbalanced data, tests of significance were made on the population marginal means 59 estimated from linear models using the lsmeans R package The mean values and standard errors SEs reported in the Results section correspond to those extracted from the linear models.
The values shown in the figures represent the arithmetic mean and standard errors SEs calculated from the empirical data, unless otherwise specified. For clarity, significance levels are sparingly used in Figures to highlight the main effects, but the full statistical effects are provided in the Results section and in Supplementary Tables. The specifics for the assessment of performance in the n-back WM and the duration estimation tasks are provided below.
The HR was the proportion of target letters participants accurately detected, and the FA was the proportion of non-target letters participants incorrectly responded to. RTs were computed from the onset of the displayed letter to the button press. For each trial, the average RT was computed only for the accurately detected n-back targets. To address the effect of feedback and WM load on the n-back task performance, we fitted one beta regression model for HR separately from another one for FA; both HR and FA were dependent variables.
0コメント