A recent development in computer science is the elaboration of the theory of typed feature structures (TFS) [Carpenter 1992]. A TFS system was implemented at the ATR project [Emele & Zajac 1989], and subsequently applied to multilingual generation in Project Polygloss at the University of Stuttgart [Emele et al. 1990]. [Carpenter 1992] describes how both SFG system networks and HPSG feature structures and signs can be represented and manipulated in the TFS formalism.
[Bateman & Momma 1992] describe an experiment in which a small fragment of Penman was implemented in a TFS system. Part of the Upper Model (see Section 2.2.3), part of the system network, including realization rules, and part of the chooser/inquiry interface (see Section 2.2.2) were all translated into the TFS representation. After defining this fragment of Penman in TFS form, it was possible to run it in both directions, for generation and analysis. The analysis started with HPSG parsing to produce a minimum set of grammatical features needed to determine a unique semantic analysis.
The only sentence attempted was ``Kim devours every cookie.'' Therefore this experiment was extremely inconclusive. However, further research may result in further progress in finding possible ways to combine some of the strong points of SFG with some of the strong points of HPSG by using the TFS formalism.
If such progress can be achieved, it could become possible to develop some kind of integrated HPSG/SFG parser capable of extracting a wider range of factors from source texts. Up to now, none of the experiments in parsing with SFG have had great success. However, if an HPSG/SFG parser could automatically extract a wider range of factors from source texts, and if further factors not present in source texts could be obtained by means of inquiries, as in SFG-based text generation, the requirements of [Tsujii 1986] for improving the quality of machine translation, discussed in the Introduction, would be much closer to being met.