Latest paper from me & @DianaTamir, now in press at Cortex: "People represent mental states in terms of rationality, social impact, and valence: Validating the 3d Mind Model" sciencedirect.com/science/articl… (preprint available @PsyArXiv: psyarxiv.com/akhpq/)
In a 2016 paper @PNASNews, we found initial evidence 3 dimensions, which we called rationality (vs. emotionality), social impact (states likely or not to impact others), and valence (+/-), describe how the brain represents other people's mental states: pnas.org/content/113/1/…
Here we expanded on those findings by aiming to answer 3 questions about this '3d Mind Model':
1) Is it robust?
2) Is it complete?
3) Is it generalizable?
1) Is it robust?
2) Is it complete?
3) Is it generalizable?
To answer the 1st question, we conducted a mega-analysis of 113 participants in 4 separate fMRI studies. We found that all three dimensions were significant predictors of neural pattern similarity in this much larger and more heterogeneous sample: 
In addition to testing this effect across the social brain network as a whole (see above) we also used a @neurosynth-based parcellation to assess the spatial distribution of these effects across the whole brain:
To answer the 2nd question, "Is the 3d Mind Model complete?" we collected ratings of mental state terms across a set of 58 dimensions. We then reduced the dimensionality of these ratings to 11 components using a bi-cross-validated PCA:
We used a cross-validated model selection procedure to determine which subset of these PCs achieved optimal out-of-sample prediction of the neural pattern similarity between mental states. We found that PCs 1-3, which map closely onto the 3d Mind Model, achieved this optimum.
As part of this model selection procedure, we also computed a noise ceiling: an estimate of the best performance which could be expected, given the reliability of the data. Using the noise ceiling we estimated that the 3d Mind Model achieves 80% of the best possible performance.
These findings suggest that the 3d Mind Model explains much of how people think about others mental states. However, there are important caveats on this: the model does not explain idiosyncrasies of mental state representation specific to individual people, contexts, or content.
Finally, we aimed to answer the 3rd question "Does the 3d Mind Model generalize?" We considered generalization in three senses: 1) to new measures, beyond neural pattern similarity, 2) to new participants, drawn from a different population, and 3) to new mental states.
To this end we tested the ability of the 3d Mind Model to predict two new measures: behavioral judgements of which of two mental states is more like a third, and semantic similarity between mental state words, estimated using computational text analysis.
The former data came from volunteer on MySocialBrain.org, my online research platform (@OurSocialBrain), participating in this experiment: mysocialbrain.org/msdim_info.html
The latter data came from the English-language fastText word embedding of the @CommonCrawl: fasttext.cc/docs/en/englis…. The populations underlying both these measures were quite different from the college student and MTurk samples we'd used previously.
Additionally, in both cases we examined a larger set of 166 mental state terms, rather than the smaller sets of 60 or less used in the fMRI studies.
Results indicated that the 3d Mind Model did successfully generalize, with rationality, social impact, and valence each significantly predicting both similarity judgements and text semantics.
Together, the results suggested in this paper suggest that the 3d Mind Model is a robust, comprehensive, and generalizable account of our shared concepts of mental states.
In papers we are currently writing, we examine whether the 3d Mind Model generalizes to other geographically, linguistically, and historically distinct cultures, and examine the computational origins and functions of rationality, social impact, and valence.
Data and code supporting the present publication are freely available online @OSFramework: osf.io/6zmnc/. Thanks to @pkanske & Ryan Murray for organizing the special issue on "Understanding Others" at Cortex, of which is article is a part.
