Fully incremental text processing means that the input to an NLS is given in piecemeal fashion and that arbitrary changes to the input are handled, e.g., deletion or modification of the input. For example, in [Wirén and Rönnquist1993] a method for fully incremental parsing is proposed that is based on the Earley algorithm and the modification for its use in an incremental environment reported in [Wirén1992]. Clearly, if such modifications to the text are allowed, the overall process is non-monotonic.
In [Neumann and Finkler1990, Reithinger1991, Harbusch et al.
1991] incremental
generation systems are described that allow changes of the generated
string, e.g., if an adjective modifier is to be inserted into an
already produced nominal phrase this can be handled by means of some sort of
repairing.
The basic motivation behind fully incremental text processing is that it can be used to support highly interactive text-processing tasks, e.g., checking of grammar, spelling and style immediately and in real-time within a text-editing environment.
Most important from our uniform perspective is that [Wirén and Rönnquist1993] motivate a combination of incremental parsing and incremental generation based on a reversible grammar in order to explore highly interactive text-processing facilities, like structure-editing operations, propagation of grammatical changes, or on-line translation, in which the target-language text is generated in parallel with the source-language. However, they only mention this kind of integrated incremental parsing and generation approach as a further important research direction.
The incremental monitoring mechanism developed in this thesis already performs a kind of incremental parse-and-generate approach using a reversible grammar. By means of the item sharing approach we have demonstrated how such an interleaving of parsing and generation can be done efficiently. However, the incrementality is restricted since it does not allow for input given in a piecemeal fashion, and changes to the input are forbidden, too. The first restriction is not that problematic because our algorithm processes a given input on-line, i.e. it does not require an initialization of the chart. For the second case, it might be possible to adapt the update-mechanisms described in [Wirén1992] and [Wirén and Rönnquist1993] for the case of generation as well. Thus, the integrated approach to parsing and generation developed in this thesis seems also to be an important contribution for the investigation of fully incremental text processing.