id	author	title	date	pages	extension	mime	words	sentences	flesch	summary	cache	txt
work_zecvwmdvwvacdcb7oko24oq6fy	Ryan Cotterell	Modeling Word Forms Using Latent Underlying Morphs and Phonology	2015	15	.pdf	application/pdf	10202	1044	65	Modeling Word Forms Using Latent Underlying Morphs and Phonology involves loopy belief propagation in a natural directed graphical model whose variables are unknown strings and whose conditional distributions are encoded as finitestate machines with trainable weights. In fact, generative linguists traditionally posit that each morpheme of a language has a single representation Figure 1: Our model as a Bayesian network, in which surface forms arise from applying phonology to a concatenation of SR at layer 3 is generated using the phonology model Sθ (a probabilistic finite-state transducer). morph M(a) ∈M as an IID sample from a probability distribution Mφ(m).3 This model describes Figure 2: Illustration of a contextual edit process as it pronounces the English word wetter by transducing the underlying /wEt#@r/ (after erasing #) to the surface [wER@r]. We are given a training set of surface word forms word types of a language, we sample a training set	./cache/work_zecvwmdvwvacdcb7oko24oq6fy.pdf	./txt/work_zecvwmdvwvacdcb7oko24oq6fy.txt