The performance of cursive script recognition systems may be improved by applying higher level knowledge in the form of syntax or semantics. A fundamental part of such an approach is the creation of a lexical database containing the relevant information. However, to create a semantic lexicon by hand for a large vocabulary is a considerable task, which is a major reason why so many semantic theories fail to scale up from the small, artificial domains in which they were developed. An alternative approach is to use existing sources of semantic information, such as machine-readable dictionaries (which contain definitions and domain information) and text corpora (from which collocations and domain information may be derived). The development of techniques for acquiring semantic knowledge from such resources and applying it to large vocabulary cursive script recognition is described.< <ETX xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink">></ETX>