... Rnhu×ddepends on theposition of the ithword in s(·), with respect to theposition posw of the word of interest, and with re-spect to the position posv of the verb of interest:Ci= C(i − posw, ... Analysis of our MLPWhile we gave an intuitive justification of the archi-tecture choices of our model in Section 4, we nowgive a systematic empirical study of those choices.First of all, providing ... partlybecause of the parse tree component, and partly be-cause of the use of Support Vector Machines (Boseret al., 1992), which have linear complexity in test-ing time with respect to the number of training...