Some statistical-estimation methods for stochastic finite-state transducers

Formal translations constitute a suitable framework for dealing with many p roblems in pattern recognition and computational linguistics. The applicati on of formal transducers to these areas requires a stochastic extension for dealing with noisy, distorted patterns with high variability. In this pape r, some estimation criteria are proposed and developed for the parameter es timation of regular syntax-directed translation schemata. These criteria ar e: maximum likelihood estimation, minimum conditional entropy estimation an d conditional maximum likelihood estimation. The last two criteria were pro posed in order to deal with situations when training data is sparse. These criteria take into account the possibility of ambiguity in the translations : i.e., there can be different output strings for a single input string. In this case, the final goal of the stochastic framework is to find the highe st probability translation of a given input string. These criteria were tes ted on a translation task which has a high degree of ambiguity.