In the so-called "coherent construction", certain matrix verbs and the verb of their embedded clause appear to behave as a single verb, licensing a range of phenomena such as long scrambling, the "third construction" (transposition of matrix and embedded verbs), matrix interpretation for negation, and so on. The following example illustrates both long scrambling and the third construction:
...weil es keiner den Kindern versucht hat zu gebenFollowing Evers's transformation analysis (1975), the coherent construction has been virtually universally interpreted as being derived by special mechanisms, namely verb complex formation (forming a single lexical entity from the two words with a combined argument list) and/or clause union (dissolution of the clause boundary). Thus, this analysis uses derivational machinery which is markedly different from that used in the analysis of the corresponding construction in other languages such as English.
...because it no-one the children tried has to give
...because no-one tried to give it to the children
I start out by arguing that the coherent construction provides a challenge to computational approaches relying on lexicalized or lexicalist linguistic theories, including lexicalized mathematical frameworks such as Tree Adjoining Grammar (TAG). I present a formalism related to TAG which is based on the notion of tree descriptions, "D-Tree Substitution Grammar" (DSG). In DSG, domination in trees can be underspecified, and exploiting underspecification we can account for the data in a natural manner. The DSG analysis makes correct empirical predictions which the clause-union analysis fails to capture.
In addition, DSG allows us to model lexical derivations in the same formalism as syntactic derivations. In DSG, the derivations for the corresponding constructions in English and German are essentially similar. There is no special lexical mechanism required for German, and the different syntactic behavior is an effect of independently required syntactic restrictions on word order variation in English. The effects of verb complex formation and clause union can be seen as epiphenomenal in a DSG derivation.
Not needing special mechanism for German has important formal and computational ramifications. I briefly discuss the formal properties of DSG, and show that it is a computationally "benign" formalism which has limited generative capacity and admits polynomial parsing algorithms. Furthermore, the parallelism between lexical and syntactic derivation means that DSG can be the basis for lexically-oriented parsers. Thus, DSG is interesting both as a metalanguage for formal linguistics and a framework for applications.
I finish by discussing the relation between DSG and several other linguistic frameworks, namely dependency grammar, HPSG, and LFG.