An approuch to estimation of text recognition algorithm parameters is proposed, 
based upon the learning sample that consists of samples of text line images 
and corresponding letter sequences and doesn't need segmentations of these 
images into images of separate letters. Formulating task in such way - with partial 
information from teacher - allows to construct learning samples with less effort 
by reducing this construction to typing of the text that corresponds to learning sample. 
Along with independent use for text line recognition, an application of the algorithm as 
an auxiliary for building segmentations of images into images of letters is proposed.