Interest in emotion recognition in conversations (ERC) has been increasing in various fields, because it can be used to analyze user behaviors and detect fake news. Many recent ERC methods use graph-based neural networks to take the relationships between the utterances of the speakers into account. In particular, the state-of-the-art method considers self-and inter-speaker dependencies in conversations by using relational graph attention networks (RGAT). However, graph-based neural networks do not take sequential information into account. In this paper, we propose relational position encodings that provide RGAT with sequential information reflecting the relational graph structure. Accordingly, our RGAT model can capture both the speaker dependency and the sequential information. Experiments on four ERC datasets show that our model is beneficial to recognizing emotions expressed in conversations. In addition, our approach empirically outperforms the state-ofthe-art on all of the benchmark datasets.

By using the renormalization group (RG) method, the interaction between balanced flows and Doppler-shifted inertia–gravity waves (GWs) is formulated for the hydrostatic Boussinesq equations on the f plane. The derived time-evolution equations [RG equations (RGEs)] describe the spontaneous GW radiation from the components slaved to the vortical flow through the quasi resonance, together with the GW radiation reaction on the large-scale flow. The quasi resonance occurs when the space–time scales of GWs are partially comparable to those of slaved components. This theory treats a coexistence system with slow time scales composed of GWs significantly Doppler-shifted by the vortical flow and the balanced flow that interact with each other. The theory includes five dependent variables having slow time scales: one slow variable (linear potential vorticity), two Doppler-shifted fast ones (GW components), and two diagnostic fast ones. Each fast component consists of horizontal divergence and ageostrophic vorticity. The spontaneously radiated GWs are regarded as superpositions of the GW components obtained as low-frequency eigenmodes of the fast variables in a given vortical flow. Slowly varying nonlinear terms of the fast variables are included as the diagnostic components, which are the sum of the slaved components and the GW radiation reactions. A comparison of the balanced adjustment equation (BAE) by Plougonven and Zhang with the linearized RGE shows that the RGE is formally reduced to the BAE by ignoring the GW radiation reaction, although the interpretation on the GW radiation mechanism is significantly different; GWs are radiated through the quasi resonance with a balanced flow because of the time-scale matching.

No abstract

Influences of time-dependent precipitation on water mass transformation and heat budgets in an idealized marginal sea are examined using theoretical and numerical models. The equations proposed by Spall in 2012 are extended to cases with time-dependent precipitation whose form is either a step function or a sinusoidal function. The theory predicts the differences in temperature and salinity between the convective water and the boundary current as well as the magnitudes of heat fluxes into the marginal sea and across the sea surface. Moreover, the theory reveals that there are three inherent time scales: relaxation time scales for temperature and salinity and a precipitation time scale. The relaxation time scales are determined by a steady solution of the theoretical model with steady precipitation. The relaxation time scale for temperature is always smaller than that for salinity as a result of not only the difference in the form of fluxes at the surface but also the variation in the eddy transport from the boundary current. These three time scales and the precipitation amplitude determine the strength of the ocean response to changes in precipitation and the phase relation between precipitation, changes in salinity and temperature, and changes in heat fluxes. It is demonstrated that the theoretical predictions agree qualitatively well with results from the eddy-resolving numerical model. This demonstrates the fundamental role of mesoscale eddies in the ocean response to time-dependent forcing and provides a framework with which to assess the extent to which observed variability in marginal sea convection and water mass transformation are consistent with an external forcing by variations in precipitation.

The renormalization group equations (RGEs) describing spontaneous inertia-gravity wave (GW) radiation from part of a balanced flow through a quasi resonance that were derived in a companion paper by Yasuda et al. are validated through numerical simulations of the vortex dipole using the Japan Meteorological Agency nonhydrostatic model (JMA-NHM). The RGEs are integrated for two vortical flow fields: the first is the initial condition that does not contain GWs used for the JMA-NHM simulations, and the second is the simulated thirtieth-day field by the JMA-NHM. The theoretically obtained GW distributions in both RGE integrations are consistent with the numerical simulations using the JMA-NHM. This result supports the validity of the RGE theory. GW radiation in the dipole is physically interpreted either as the mountain-wave-like mechanism proposed by McIntyre or as the velocity-variation mechanism proposed by Viúdez. The shear of the large-scale flow likely determines which mechanism is dominant. In addition, the distribution of GW momentum fluxes is examined based on the JMA-NHM simulation data. The GWs propagating upward from the jet have negative momentum fluxes, while those propagating downward have positive ones. The magnitude of momentum fluxes is approximately proportional to the sixth power of the Rossby number between 0.15 and 0.4.