Lexicon-directed Segmentation and Tagging of Sanskrit

We propose a methodology for Sanskrit processing by computer. The first layer of this software, which analyses the linear structure of a Sanskrit sentence as a set of possible interpretations under sandhi analysis, is operational. Each interpretation proposes a segmentation of the sentence as a list of tagged segments. The method, which is lexicon directed, is complete if the given (stem forms) lexicon is complete for the target corpus. It uses an original design for a finite-state transducers toolkit, based on functional programming principles. Further layers of this computational linguistics architecture are discussed. 1 COMPUTATIONAL LINGUISTICS AND SANSKRIT Descriptive linguistics is the study of natural language phenomena. Theoretical linguistics strives to provide formal models of linguistic activity. Noam Chomsky initiated modern theoretical linguistics with successive theories (context-free and transformational grammars, government and binding, minimalism). These theories usually provide a generative grammar paradigm, by which means any valid sentence in the language may be generated. With the advent of computers a still more formal approach is attempted, where executable programs process a digital representation of natural language into information structures, which provide some degree of understanding of the given text or discourse. In the pioneer days of the fifties, automated translation systems were thus attempted, although difficulties were largely underestimated, leading to doubts about the feasability of natural language understanding by computers. 50 years later, however, the situation