CoChat
Sign Up
PreprintCited by 33

The inhibition of automatic imitation: a meta-analysis and synthesis of fMRI studies

Darda, Kohinoor M., Ramsey, Richard·2018·bioRxiv
DOI 10.1101/334938

Preprint — published as 10.1016/j.neuroimage.2019.04.059

Open this paper in CoChat — free account

Abstract

Humans copy other people without their conscious awareness, a behaviour known as automatic imitation. Although automatic imitation forms a key part of daily social interactions, we do not copy other people indiscriminately. Instead, we control imitative tendencies by prioritising some actions and inhibiting others. To date, neuroimaging studies investigating the control of automatic imitation have produced inconsistent findings. Some studies suggest that imitation control relies on a domain-specific neural circuit related to social cognition (the theory-of-mind network). In contrast, other studies show engagement of a domain-general neural circuit that is engaged during a diverse range of cognitive control tasks (the multiple demand network). Given the inconsistency of prior findings, in the current paper we avoided problems associated with interpreting individual studies by performing a meta-analysis. To do so, we used a multi-level kernel density analysis to quantitatively identify consistent patterns of activation across functional magnetic resonance imaging studies investigating the control of imitation. Our results show clear and consistent evidence across studies that the control of automatic imitation is guided by brain regions in the multiple demand network including dorsolateral frontoparietal cortex. In contrast, there was only limited evidence that regions in the theory of mind network were engaged. Indeed, medial prefrontal cortex showed no consistent engagement and right temporoparietal junction engagement may reflect spatial rather than imitative control. As such, the current meta-analysis reinforces the role of domain-general control mechanisms and provides limited evidence in support of the role of domain-specific processes in regulating imitative tendencies. Consequently, neurocognitive models of imitation need updating to place more emphasis on domain-general control mechanisms, as well as to consider more complex organisational structures of control, which may involve contributions from multiple cognitive systems.

AI snapshot

Written from the paper’s full text.

TL;DR

This meta-analysis of fMRI studies found that inhibiting automatic imitation consistently engages domain-general multiple demand (MD) network regions, especially dorsolateral frontoparietal cortex, rather than the theory-of-mind (ToM) network. Medial prefrontal cortex showed no consistent engagement, and right temporoparietal junction engagement was limited and may reflect spatial rather than imitative control.

Key findings

  • Across 13 contrasts from 12 studies (300 participants, 142 foci), the general compatibility effect consistently activated right inferior parietal lobule, right supramarginal gyrus, right superior temporal gyrus, and right temporoparietal junction; right superior frontal gyrus and right middle frontal gyrus survived height and less stringent extent-based thresholding (p<.01), insula survived more stringent extent-based thresholding (p<.001), and right IFG survived less stringent extent-based thresholding (p<.01).
  • All regions passing height- or extent-based thresholding overlapped with the MD network; one general-compatibility cluster also overlapped with the rTPJ node of the ToM network, but there was no overlap with mPFC for any compatibility type.
  • For spatial compatibility (4 contrasts, 4 studies, 42 foci), left IPL and right SFG survived height-based thresholding, with no regions surviving cluster-based thresholding; for imitative compatibility (3 contrasts, 3 studies, 20 foci), left IPL survived height-based thresholding and right IPL survived height-based plus less stringent extent-based thresholding (p<.01).
  • Only 28.6% (4/14) of fMRI studies on imitation control to date showed any evidence for rTPJ involvement, and rTPJ overlapped with general and spatial compatibility maps but not with imitative compatibility maps.

Methods

  • The authors searched PubMed and Google Scholar for fMRI studies of imitation inhibition prior to January 2019, found 15 studies, excluded 3, and included 12 studies with 300 participants; exclusions were for non-fMRI methods, unavailable stereotactic coordinates, ROI-only analyses without whole-brain coordinates, or child/atypical populations without separate control coordinates.
  • They performed multi-level kernel density analysis (MKDA) in MatlabR2015b using the Wager toolbox, with a random-effects model; Talairach coordinates were converted to MNI, peaks were convolved with a 10 mm spherical kernel, density maps were weighted by square root of sample size and by fixed/random-effects adjustment (1 for random, .75 for fixed), and significance was assessed with Monte Carlo simulation (5000 iterations) using FWE-corrected height- and cluster-extent thresholds.
  • The main analysis examined the general compatibility effect, with exploratory separate MKDAs for spatial and imitative compatibility; resulting maps were overlaid with ToM and MD network masks.

Limitations

  • Stated: the meta-analysis excluded Brass et al. (2009) because whole-brain data were unavailable; the MD network is larger than the ToM network, which may bias analyses toward MD findings; the spatial and imitative compatibility analyses were based on small subsets (N=4 and N=3 studies, respectively); and fMRI findings may diverge from neurostimulation and patient studies, whose evidence for rTPJ is limited to a few small-sample studies lacking formal power analyses and replications.
  • Evident from the text: many included studies had small samples (e.g., 10–25 participants in studies implicating mPFC/rTPJ; Darda et al. 2018 used N=28 and N=50), and the authors note that most previous fMRI studies were limited by low statistical power and few direct replications.
  • Evident from the text: the meta-analysis synthesizes univariate activation patterns and cannot resolve functional connectivity or more complex interactions between domain-general and domain-specific systems.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.1101/334938

Cites

112 of 112 references verified0 citation errors · 0 not checkable

Statistics

Nothing to recompute.

Claims the paper makes

15 of 40 cited sources checked · 11 supported · 0 flagged · accuracy not yet measured

  • Premisereasoning

    The tasks used in the studies included in the meta-analysis are conceptually, visually and cognitively similar to each other, with only minor differences across all studies.

    “However, the tasks are all conceptually, visually and cognitively similar to each other with only minor differences across all studies.”
    Quote in the paperSubstantial claim2 of 2 sources checked
    • cites Darda et al., (2018 source

      The Design and Procedure section (page unavailable) says the automatic imitation task in Experiment 2 was similar to that in Experiment 1, with two changes, including a different hand stimulus. That supports similarity between this paper’s two experiments, but it does not establish that tasks across all studies in the citing document’s meta-analysis were conceptually, visually and cognitively similar, with only minor differences. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
    • cites Wang et al. (2011) source

      Materials and Methods (page unavailable) says this study used a stimulus-response compatibility paradigm used in earlier studies and stimuli identical to those in Wang et al. (2011). That supports similarity with some prior work, but the paper does not compare the tasks in all studies included in the citing document’s meta-analysis or establish that their conceptual, visual, and cognitive differences were only minor. (LLM, conf=0.83)

      Experimental — accuracy not yet measured
  • Interpretationown data

    The absence of mPFC engagement for imitation control across studies is inconsistent with the hypothesis that a specific neural system related to social cognition is also engaged during the inhibition of automatic imitation.

    “An absence of mPFC engagement for imitation control across studies is thus inconsistent with the hypothesis that a specific neural system related to social cognition is also engaged during the inhibition of automatic imitation (Brass et al., 2009).”
    Quote in the paperSubstantial claim1 of 1 sources checked
    • cites Brass et al., 2009 source
  • Premisecitation

    Dampening activity in the rTPJ with repetitive transcranial magnetic stimulation interfered with imitative, but not spatial, responses.

    “Using repetitive transcranial magnetic stimulation (TMS), dampening of activity in the rTPJ interfered with imitative, but not spatial responses (Hogeveen et al., 2014;Sowden & Catmur, 2015)”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites Hogeveen et al., 2014 source

      The Discussion (page unavailable) reports that repetitive TMS to the TPJ impaired imitation inhibition but did not affect spatial compatibility, supporting the stated behavioral contrast. It does not say that the TMS dampened rTPJ activity. The Procedures (page unavailable) describe this paper’s own intervention as excitatory anodal tDCS, not TMS. (LLM, conf=0.86)

      Experimental — accuracy not yet measured
    • cites Sowden & Catmur, 2015

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Study design and conductcitation

    Each contrast in the MKDA is weighted by sample size and study quality (fixed or random effects model).

    “Further, each contrast is weighted by the sample size and study quality (i.e.whether the study used fixed or random effects model; Wager et al., 2007;Kober and Wager, 2010).”
    Quote in the paperNot a substantial claim2 of 2 sources checked
    • cites Wager et al., 2007 source
    • cites Kober and Wager, 2010 source
  • Premisecitation

    Neurostimulation studies have implicated the rTPJ in imitation control.

    “For instance, neurostimulation studies have implicated rTPJ in imitation control (Santiesteban et al., 2015;Bardi et al., 2017).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites Santiesteban et al., 2015 source
    • cites Bardi et al., 2017

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim. 8
  • May combine two claims. 24

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Study design and conductcitation

    The meta-analysis followed recent guidelines for meta-analysing neuroimaging studies.

    “In the current paper, we follow recent guidelines put forward for metaanalysing neuroimaging studies (Muller et al., 2018).”
    Quote in the paperNot a substantial claim1 of 1 sources checked
    • cites Muller et al., 2018 source

  • May combine two claims.

    Premisecitation

    rTPJ is involved in both social and non-social processes, making it a functionally heterogenous region.

    “Second, rTPJ is involved in both social and non-social processes, which makes it a functionally heterogenous region (Corbetta et al., 2008;Krall et al., 2015Krall et al., , 2016;;Lee & McCarthy, 2016;Schuwerk et al., 2017).”
    Quote in the paperSubstantial claim2 of 3 sources checked
    • cites Corbetta et al., 2008 source
    • cites Krall et al., 2015 source
    • cites Lee & McCarthy, 2016 source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The neural substrates for person and action perception span the fusiform gyrus, occipitotemporal cortex, and posterior superior temporal sulcus, as well as the mirror neuron system.

    “The neural substrates for person and action perception span the fusiform gyrus, occipitotemporal cortex, and posterior superior temporal sulcus, as well as the mirror neuron system (Kanwisher, 2010;Caspers, et al., 2010).”
    Quote in the paperNot a substantial claim1 of 2 sources checked
    • cites Kanwisher, 2010 source
    • cites Caspers, et al., 2010 source

  • May combine two claims.

    Premisecitation

    More recent behavioural evidence suggests that imitation may not vary as a function of autistic-like traits or empathy, questioning the reliance of imitation inhibition on a distinctly social mechanism.

    “However, more recent behavioural evidence suggests that imitation may not vary as a function of autistic-like traits or empathy, thus questioning the reliance of imitation inhibition on a distinctly social mechanism (Butler et al., 2015;Cracco et al., 2018;Genschow et al., 2017).”
    Quote in the paperSubstantial claim1 of 3 sources checked
    • cites Butler et al., 2015 source
    • cites Cracco et al., 2018 source
    • cites Genschow et al., 2017 source

      Queued — the background check reaches it, usually within a day.

  • May combine two claims.

    Findingown data

    For imitative compatibility, across 3 contrasts from 3 studies, consistent activation in left IPL survived height-based thresholding; activation in right IPL withstood height-based and less stringent extent-based thresholding (p<.01).

    “For imitative compatibility, across 3 contrasts from 3 studies, we found consistent activation in the left IPL that survived height-based (see Table 3; Figure 2C).Activation was also found in the right IPL, which withstood height-based as well as the less stringent extent-based thresholding (p<.01).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Brain regions in the MD network are also engaged for imitative (bilateral IPL) and spatial compatibility effects (left IPL, right SFG).

    “Brain regions in the MD network are also engaged for imitative (bilateral IPL) and spatial compatibility effects (left IPL, right SFG).”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    In the current meta-analysis, when unthresholded spatial and imitative compatibility effect maps were explored separately, there was partial overlap between the spatial compatibility effect and rTPJ, but no overlap between the imitative compatibility effect and rTPJ.

    “Similarly, in the current meta-analysis, when we explored the unthresholded spatial and imitative compatibility effect maps separately, there was partial overlap between the spatial compatibility effect and rTPJ, but no overlap between the imitative compatibility effect and rTPJ.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    Only 28.6% of fMRI studies on imitation control to date (4/14) show any evidence in support of a role of rTPJ in imitation control, and these studies do not dissociate between spatial and imitative effects.

    “Thus, we find that only 28.6% of fMRI studies on imitation control to date (4/14) show any evidence in support of a role of rTPJ in imitation control, and these studies do not dissociate between spatial and imitative effects.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Findingown data

    For the general compatibility effect, activation in left and right insula survived the more stringent extent-based threshold (p<.001) but not the height-based threshold.

    “Activation in the left and right insula survived the more stringent extent-based threshold (p<.001), but not the height-based threshold.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    For spatial compatibility, across 4 contrasts from 4 studies, consistent activation withstood height-based thresholding in left IPL and right SFG; no regions withstood cluster-based thresholding.

    “For spatial compatibility, across 4 contrasts from 4 studies, we found consistent activation that withstood the height-based thresholding in the left IPL and the right SFG (see Table 3; Figure 2B).No regions withstood cluster-based thresholding.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    For the general compatibility effect, activation in right IFG survived the less stringent extent-based threshold (p<.01) but not the height-based threshold.

    “Activation in the right IFG survived the less stringent extent-based threshold (p<.01) but not the height-based threshold.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Findingown data

    In exploratory analysis, there was overlap between contrast indicator maps for general compatibility and spatial compatibility with the right TPJ node of the ToM network mask, but no overlap between imitative compatibility and the same right TPJ mask.

    “Figure 4 shows overlap between contrast indicator maps for general compatibility and spatial compatibility with the right TPJ node of the ToM network mask.By contrast, there is no overlap between contrast indicator maps for imitative compatibility and the same right TPJ mask.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The results provide unambiguous support for engagement of a domain-general neural network during control of imitation, and only limited evidence for engagement of a domain-specific neural network tied to social cognition.

    “Thus, our results provide unambiguous support for the engagement of a domaingeneral neural network during the control of imitation, and only limited evidence for the engagement of a domain-specific neural network that is tied to social cognition.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The meta-analysis found less evidence for a specialist view of imitation control relying on the ToM network; there was no engagement of mPFC across studies and no clear evidence regarding rTPJ, with only suggestive evidence that rTPJ may reflect spatial rather than social control.

    “We found less evidence for a”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductcitation

    Both the MD and ToM networks were defined accurately based on prior work that used large samples of participants.

    “Second, both networks were defined accurately based on prior work, which used large samples of participants.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The multiple demand (MD) network is engaged across a diversity of mental operations.

    “This network is also called the multiple demand (MD) network as it is engaged across a diversity of mental operations (Duncan et al., 2010).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Duncan et al., 2010

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Classified as probably not a substantial claim.

    Study design and conductown data

    The meta-analysis used two thresholding techniques (height-based and cluster-size-based) to maximise sensitivity in testing hypotheses.

    “In order to maximise sensitivity in testing our hypotheses, we report results using two thresholding techniques.One thresholding technique is based on height and the other is based on cluster size.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductnone given

    The regions of interest in the ToM network were not particularly small: the mPFC regions included several portions of the dorsal, middle, and ventral mPFC, and the rTPJ covered a relatively large area of cortex.

    “First, regions of interest in the ToM network were not particularly small areas.The mPFC regions included several portions of the dorsal, middle, and ventral mPFC, and the rTPJ covered a relatively large area of cortex.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning

    MKDA uses a random effects model to assess convergence across studies, so results are not biased by a small number of individual studies.

    “MKDA is an analysis technique that uses a random effects model to assess convergence across studies.This allows for assessing convergence across studies as opposed to between individual foci (as implemented in classical meta-analysis techniques that use fixed effects analyses).Thus, results are not biased by a small number of individual studies.”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationown data

    The meta-analysis results supported a generalist view of imitation control: clear engagement of dorsolateral frontoparietal cortices when observing an action that conflicted with a current motor intention, and these regions overlapped with the MD network.

    “Our results supported a”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Study design and conductreasoning

    Because the MD network spans a much larger area than the ToM network, the meta-analysis may be biased toward finding results in the MD network over the ToM network.

    “Given that the MD network spans a much larger area than the ToM network, our analysis may be biased toward finding results in the MD network over the ToM network.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The current meta-analysis indicates that the selection mechanism when inhibiting automatic imitative tendencies is guided by the regions of the domain-general multiple demand network, which is also engaged during the inhibition of other non-social external influences, rather than by a domain-specific system related to social cognition.

    “The current meta-analysis suggests that the selection mechanism in imitation inhibition is guided by a domain-general multiple demand system, which is also engaged during the inhibition of other non-social external influences.”
    • Also stated in: § Theoretical implications · hedged
      “In conclusion, the current meta-analysis provides evidence that the selection mechanism when inhibiting automatic imitative tendencies is guided by the regions of the domain-general multiple demand network rather than a domain-specific system related to social cognition.”
    Quote in the paperSubstantial claim

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based). May combine two claims.

    Premisereasoning

    The input in the imitation inhibition task can be identified as emanating from a social entity, a human hand, so the observed input is social in that the observed agent offers opportunity for social interaction.

    “Of course, the input in the imitation inhibition task can be readily identified as emanating from a social entity i.e. a human hand.Thus, the observed input is clearly social in the sense that the observed agent offers opportunity for social interaction.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Premisecitation

    The computational model of Cooper et al. (2012) shows that spatial and imitative compatibility effects depend on similar cognitive processes, with any behavioural differences accounted for by different sets of input nodes for spatial and imitative effects in a general dual-route framework.

    “In line with this, the computational model put forth by Cooper et al. (2012) further substantiates this notion by demonstrating that spatial and imitative compatibility effects depend on similar cognitive processes, and any behavioural differences are accounted for by different sets of input nodes for spatial and imitative effects in a general dual-route framework”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Cooper et al. (2012)

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    The dual-route model proposes that, like other stimulus-response compatibility tasks, imitation control is mediated by long-term stimulus-response associations that are a product of learning.

    “The model suggests that like other stimulusresponse compatibility tasks, imitation control is mediated by long-term stimulusresponse associations which are a product of learning.”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Heyes (2011) source

  • Probably not a substantial claim: may not pass S3 (load-bearing).

    Premisecitation

    Excitatory stimulation of the rTPJ by anodal transcranial direct current stimulation caused increased performance on the imitation task.

    “whereas excitatory stimulation of the rTPJ by anodal transcranial direct current stimulation (tDCS) caused increased performance on the imitation task (Santiesteban et al., 2012).”
    Quote in the paperNot a substantial claim0 of 1 sources checked
    • cites Santiesteban et al., 2012

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • Probably not a substantial claim: may not pass S3 (load-bearing), S4 (based).

    Interpretationnone given

    Other socio-perceptual circuits that extend beyond the MD network may also be involved when inhibiting automatic imitative tendencies.

    “In a similar manner, other socio-perceptual circuits, which extend beyond the MD network, may also be involved when inhibiting automatic imitative tendencies.”
    Quote in the paperNot a substantial claim

  • May combine two claims.

    Interpretationown data

    The lack of consistent activation in mPFC in the current meta-analysis and the difficulty in interpreting the role of the rTPJ have implications for specialist theories of imitation.

    “The lack of consistent activation in mPFC in the current meta-analysis and a difficulty in interpreting the role of rTPJ have implications for”
    Quote in the paperSubstantial claim

  • May combine two claims.

    Interpretationcitation

    Instead of a distinctly social mechanism, imitation control may involve domain-general cognitive control mechanisms, which are also engaged during the control of other non-social pre-potent response tendencies.

    “Instead of a distinctly social mechanism, imitation control may involve domain-general cognitive control mechanisms, which are also engaged during the control of other non-social pre-potent response tendencies (Heyes, 2011;Cooper et al., 2012).”
    Quote in the paperSubstantial claim0 of 2 sources checked
    • cites Heyes, 2011 source
    • cites Cooper et al., 2012

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    mPFC and rTPJ may play a regulatory role in imitation control and may be functionally connected to other networks without being directly engaged.

    “These findings suggest that mPFC and rTPJ may play a regulatory role in imitation control and may be functionally connected to other networks without being directly engaged (Burnett & Blakemore, 2009).”
    Quote in the paperSubstantial claim0 of 1 sources checked
    • cites Burnett & Blakemore, 2009

      Couldn’t be matched to a paper in Crossref, so it can’t be checked.

  • May combine two claims.

    Interpretationreasoning

    The sum total of evidence from neurostimulation and patient studies for rTPJ engagement in imitation control can only be judged suggestive at present, because it is based on a few studies with small sample sizes that lack formal power analyses and replications.

    “Under any yardstick, therefore, the sum total of evidence from neurostimulation and patient studies can only be judged to be suggestive at present.It is based on a few studies with small sample sizes that lack formal power analyses and replications.”
    Quote in the paperSubstantial claim
31 of 31

  • The authors state this as an assumption.

    On incompatible trials, expectations are violated.

    “Assuming that on incompatible trials expectations are violated, rTPJ may play a more generalised role of context updating in imitation and spatial control.”
1 of 1

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

  • Height-based threshold is described as both FDR- and FWE-corrected

    Minor

    The figure description labels the height-based threshold as FDR-corrected, whereas the methods describe it as FWE-corrected. These are different multiple-comparison corrections.

1 of 1

Cites

Flagged first, then by citations

Research works printed but not linked: 1

  1. 1987, Advances in psychology, 45, 235

Cited by

Sorted by citations

Related

Authors
Darda, K. M., & Ramsey, R.
Venue
bioRxiv
Publisher
openRxiv
Published
2018-05-30
DOI
10.1101/334938
Semantic Scholar
fc4cca7d783787c988bada0808caaad946084fe7
References
95
Type
Preprint

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

Claims read by deepseek-v4.1-flash under definition version 3 (extract-v3) on 2026-10-08.

Classified by jev-1.13-20260917 (gold-guideline-v0.2) on 2026-10-08.

Cited sources are checked by gpt-6-sol.

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review