... nontrivial taskthat involves nonliteral uses of such words.Moreover, we show that visual and textual in- formation are tapping on different aspects ofmeaning, and indeed combining them in mul-timodal ... that vi-sual and textual information are tapping on differentaspects of meaning, such that they are complemen-tary sources of information, and indeed combiningthem in multimodal models often ... until now in the lit-erature, we are missing important semantic informa-tion, such as the color information that we can minewith LAB.Again we find that hybrid models do very well, in fact in this...