Horizontal methods take a linguistic stratum and attempt to define it as comprehensively as possible. This method has been applied primarily to lexicogrammar, giving rise to re-usable surface realization components such as Nigel, MUMBLE, and SURGE for English, all mentioned above. All of these components are available for converting input specifications to English sentences.
Large generation lexicogrammars for languages other than English still lag behind those for English but are catching up fast. There are also some interesting differences in generators for differing languages that arise out of the different research contexts in the countries involved. Generators of French include more sentence generators (e.g., [Nogier: 1991]) that consider their input to be a conceptual graph in the sense of Sowa [Sowa: 1983]; generators of German are often based on structural models of lexicogrammar strongly influenced by formal natural language analysis work, although larger-scale work relativizes this somewhat. Larger generation lexicogrammars of German include: SUTRA [Busemann: 1988], SEMSYN [Rösner: 1988], POPEL [Reithinger: 1991], KOMET [Teich: 1992], DISCO [Uszkoreit, Backofen, Busemann, Diagne, Hinkelman, Kasper, Kiefer, Krieger, Netter, Neumann, Oepen and Spackman: 1994]. A number of generators provide good coverage of restricted areas of other languages--for example, temporal expressions in French [Gagnon and Lapalme: 1996]--rather than broad coverage of the respective language as a whole. There are also cross-overs between various approaches, and some of the work of the coming decade will undoubtedly involve a significant consolidation of results. The role of multilingual generation (i.e., generation systems that are capable of generating texts in more than one language) will also increase and push further the development of significant lexicogrammatical resources for diverse languages; an extensive overview of multilingual NLG is given in Bateman, Matthiessen and Zeng [Bateman, Matthiessen and Zeng: 1999].
More recently, there have been similar attempts to achieve broad coverage for NLG in the areas of ideational semantics [Bateman, Henschel and Rinaldi: 1995], and rhetorical relations (cf. [Maier and Hovy: 1993,Knott and Dale: 1996]).