In this paper we report the work done on the hypothesisation of word boundaries in Hindi speech. Firstly, the language related clues useful in word boundary hypothesisation are identified. These clues are then grouped into two classes (i) clues describing conditions to be satisfied at word boundaries, called lexical clues and (ii) clues that consist of frequently occurring patterns in text, called pattern clues. Two measures, frequency and correctness, are used to evaluate these clues. The results led to the organization of these clues into hypothesisation and verification clues. Experiments with the word boundary hypothesiser on a 300 sentence text yielded two results: (i) upto 50% of word boundaries in the text are hypothesised and (ii) considerable speed improvement in lexical search is achieved due to the hypothesisation of these boundaries.