The current approach is defined as an on-line algorithm in the sense, that it does not initialize the chart for each individual element of the input before these elements are then combined, but rather creates new item sets on demand.
This is especially useful in the case of generation, because in general the generator cannot know in advanced which lexical elements belong to which part of the semantic expression. Of course, our algorithm could easily be adapted to such kind of initialization step, if a process is available, that can assign to each part of the semantic expression a corresponding lexical entry. However, at least in the case of generation, we prefer an on-line approach, because only then syntactic information can be taken appropriately into consideration to restrict lexical access during early stages of processing.
Furthermore, using an on-line approach together with top-down prediction is necessary to satisfy the valid prefix property as discussed for example in [Schabes1990] and it is also an important basis for incremental processing which we will consider in more detail in the final chapter of this thesis.