In section 3 when authors describe model, they assumed that the i-th row of P follows Dirichlet distribution. Where is their reference? Why do they assume that?
Article A Bayesian hidden Markov model for imperfect debugging
I think if you read up on mixture models you might get some insight...but the dirichlet distribution is a multivariate beta distribution. the dirichlet distribution is a conjugate pair when you have a multinomial distribution.