Text retrieval from early printed books

We describe a text indexing and retrieval technique that does not rely on word segmentation and is tolerant to errors in character segmentation. The method is designed to process early printed documents and we evaluate it on the well known Latin Gutenberg Bible. The approach relies on two main components. First, character objects (in most cases corresponding to individual characters) are extracted from the document and clustered together, so as to assign a symbolic class to each indexed object. Second, a query word is compared against the indexed character objects with a Dynamic Time Warping (DTW) based approach. The peculiarity of the matching technique described in this paper is the incorporation of sub-symbolic information in the string matching process. In particular, we take into account the estimated widths of potential subwords that are computed by accumulating lengths of partial matches in the DTW array.