Inputs to parsers are generally strings of symbols. Traditionally, the strings are known as `words' in computer science. However, in an empirical computational linguistic model, the strings and the symbols can represent any kind of unit involving concatenation; for example:
Inputs from speech recognition devices may be complex word hypothesis lattices (word hypothesis graphs) in which words may be interlinked with many possible successors and can therefore represent many possible strings.
Parsing techniques have been thoroughy researched from the point of data structures and algorithms in the area of computer science known as `compiler construction', and a variety of motivational metaphors have been attached to more or less complex combinations of these techniques, such as some of those listed previously.