Using Learned Conditional Distributions as Edit Distance
In order to achieve pattern recognition tasks, we aim at learning an unbiased stochastic edit distance, in the form of a finite-state transducer, from a corpus of (input,output) pairs of strings. Contrary to the state of the art methods, we learn a transducer independently on the marginal probability distribution of the input strings. Such an unbiased way to proceed requires to optimize the parameters of a conditional transducer instead of a joint one. This transducer can be very useful in pattern recognition particularly in the presence of noisy data. Two types of experiments are carried out in this article. The first one aims at showing that our algorithm is able to correctly assess simulated theoretical target distributions. The second one shows its practical interest in a handwritten character recognition task, in comparison with a standard edit distance using a priori fixed edit costs.
KeywordsEdit Distance Input String Edit Operation Handwritten Digit Input Alphabet
Unable to display preview. Download preview PDF.
- 2.Ristad, E.S., Yianilos, P.N.: Finite growth models. Technical Report CS-TR-533-96, Princeton University Computer Science Department (1996)Google Scholar
- 3.Casacuberta, F.: Probabilistic estimation of stochastic regular syntax-directed translation schemes. In: Proceedings of VIth Spanish Symposium on Pattern Recognition and Image Analysis, pp. 201–207 (1995)Google Scholar
- 4.Clark, A.: Memory-based learning of morphology with stochastic transducers. In: Proceedings of the Annual meeting of the association for computational linguistic (2002)Google Scholar
- 5.Eisner, J.: Parameter estimation for probabilistic finite-state transducers. In: Proceedings of the Annual meeting of the association for computational linguistic, pp. 1–8 (2002)Google Scholar
- 6.Gómez, E., Micó, L., Oncina, J.: Testing the linear approximating eliminating search algorithm in handwritten character recognition tasks. In: VI Symposium Nacional de reconocimiento de Formas y Análisis de Imágenes, pp. 212–217 (1995)Google Scholar