The integration of knowledge sources for word recognition

Cursive script recognition is problematic because of the great variability between writers and in the writing of a single individual as well as difficulties in segmenting characters. One way of depicting the reading process is to make a distinction between the letter level, the word level and the meta-word level, and to consider the different sources of information which can be utilized at each of these levels. This paper presents some results arising from work concerned with the application of information at the word level. A number of different sources of information, such as word frequency and word length, have been chosen as being used at the word level.<
>