It has been theorized that positive intergroup relations can reduce prejudice and facilitate peace. However, supporting empirical evidence is weak, particularly in the context of real-world conflict. Mousa randomized Christian Iraqi refugees to soccer teams that were composed of either all Christian players or a mixture of Christian and Muslim players (see the Perspective by Paluck and Clark). Playing on the same team as Muslims had positive effects on Christian players' attitudes and behaviors toward Muslims within the context of soccer, but these effects did not generalize to non-soccer contexts. These findings have implications for the potential benefits and limits of positive intergroup contact for achieving peace between groups. Science , this issue p. ; see also p.  Can intergroup contact build social cohesion after war? I randomly assigned Iraqi Christians displaced by the Islamic State of Iraq and Syria (ISIS) to an all-Christian soccer team or to a team mixed with Muslims. The intervention improved behaviors toward Muslim peers: Christians with Muslim teammates were more likely to vote for a Muslim (not on their team) to receive a sportsmanship award, register for a mixed team next season, and train with Muslims 6 months after the intervention. The intervention did not substantially affect behaviors in other social contexts, such as patronizing a restaurant in Muslim-dominated Mosul or attending a mixed social event, nor did it yield consistent effects on intergroup attitudes. Although contact can build tolerant behaviors toward peers within an intervention, building broader social cohesion outside of it is more challenging. : /lookup/doi/10.1126/science.abb3153 : /lookup/doi/10.1126/science.abb9990
We propose a novel neural topic model in the Wasserstein autoencoders (W AE) framework. Unlike existing variational autoencoder based models, we directly enforce Dirichlet prior on the latent document-topic vectors. We exploit the structure of the latent space and apply a suitable kernel in minimizing the Maximum Mean Discrepancy (MMD) to perform distribution matching. We discover that MMD performs much better than the Generative Adversarial Network (GAN) in matching high dimensional Dirichlet distribution. We further discover that incorporating randomness in the encoder output during training leads to significantly more coherent topics. To measure the diversity of the produced topics, we propose a simple topic uniqueness metric. Together with the widely used coherence measure NPMI, we offer a more wholistic evaluation of topic quality. Experiments on several real datasets show that our model produces significantly better topics than existing topic models.