Our conclusion would be incorrect if the combinatory power of GEN is sensitive to thesubcategorization requirements of morphemes, such that A can be conjoined with B only of Aselects B or vice versa (as proposed, e.g., by Stabler 1996, or as is true of HPSG). In such amodel, GEN would failto apply to certain inputs. There are no a priori reasons against this view,to the extent that these models of merging elements are motivated (rather than the more familiarones). Interpreting core selection requirements as principles of EVAL is, however, unconvincing,since they are never violated in any language.
Ineffability in Grammar19in the present and past tenses (26) 15 . This behavior is due to the fact that finite verbs must be placed into second position matrix clauses, and that verbal particles must be stranded in this process according to certain conditions involving stress. Stressed particles are stranded (compare dass er heute an-fängt 'that he today begins' with er fängt heute an, *er anfängt heute ), while unstressed ones are pied-piped (compare dass er heute ent-flieht 'that he today escapes' with er entflieht heute , * er flieht heute ent ). (26) a. dass wir das Stück ur-auf-führen that we the play ptc-ptc-lead 'that we premiere the play' b. *wir uraufführen das Stück c. *wir aufführen das Stück ur d. *wir führen das Stück urauf e. *wir urführen das Stück auf Although ur never actually appears in a stranded position, it can be argued that this particle would have to be stranded according to the general laws of German. Ur bears stress, to which strandability is linked. Furthermore, the placement of the infinitive morpheme zu in uraufzuführen and the placement of the ge- prefix of the participle in uraufgeführt implies that urauf involves strandable material. From these considerations, it follows that (26d) is the winner of the competition between (26b-e), but is blocked in a control component that requires that ur must be phonetically adjoined to a verbal category. There seems to be no alternative to this treatment, in particular because the rare examples of verbs that involve only ur ( urzeugen, urformen ) show exactly the same stranding difficulties. The principled case for the existence of lexical control effects in the syntax has thus been made. 16 While there are, thus, instances of lexical control effects that go beyond mere phonological consequences, it seems that not all lexicon-related instances of ineffability in syntax are control based. We show this for a particular aspect of lexical features here, and return to the general point in section 4. One of the properties that must be specified with particular lexical entries is non- structural exceptional Case. German transitive verbs assign accusative case to their objects, but some are constructed with the dative ( helfen 'help') or the genitive ( gedenken 'commemorate'). Likewise, gefallen 'please' requires a dative object. It is a commonly held view that this lexical specification leads to ineffability problems in a 15 For those speakers who allow do
- Fall '16