... ofa distribution p2if for every event A, we have1λ≤p1(A)p2(A)≤ λ (13) For any feature function f(z) and any twosets of parameters θ2 and θ1 for G and for anymarginal q(x), if ... Linguistics, pages 150 2–1 511,Uppsala, Sweden, 11-16 July 2010. c 2010 Association for Computational LinguisticsViterbi Training for PCFGs:Hardness Results and Competitiveness of Uniform InitializationShay ... following Lemmas 1 and 2, is to state the decision problem for Viter-biTrain as “given G and x1, . . . , xn and α ≥ 0,is the optimized value of the objective functionL(θ, z) ≥ α?” and use α =...