Estimating valence from the sound of a word: Computational, experimental, and cross-linguistic evidence