“…Substringto-substring edit operations -or equivalently, (monotone) many-to-many alignments -have appeared in the NLP context, e.g., in (Deligne et al, 1995), (Brill and Moore, 2000), (Jiampojamarn et al, 2007), (Bisani and Ney, 2008), (Jiampojamarn et al, 2010), or, significantly earlier, in (Ukkonen, 1985, (Véronis, 1988). Learning edit distance/monotone alignments in an unsupervised manner has been the topic of, e.g., (Ristad and Yianilos, 1998), (Cotterell et al, 2014), besides the works already mentioned. All of these approaches are special cases of our unigram model outlined in Section 2 -i.e., they consider particular S (most prominently, S = {(1, 0), (0, 1), (1, 1)}) and/or restrict attention to only N = 2 strings.…”