Richard Hudson (r.hudson@linguistics.ucl.ac.uk)
Tue, 25 Jul 1995 10:49:30 -0600 (CST)
Many thanks for the helpful reactions to my query about German partial VP fronting from dependency grammarian. I'm following the ongoing debate with interest. Meanwhile, HPSG experts may be interested in a view from one version of DG - Word Grammar. a. DG and HPSG Some important similarities - stress on headedness, on lexical valency (subcategorisation) and on structural sharing. Differences relevant to this debate: 1. No constituent structure (except for coordination), so no VP. Therefore: no structural asymmetry between subject and object, so no structural reason why subjects should be treated differently from objects. 2. All grammatical functions are distinguished by labelling; this is equivalent to a version of HPSG in which `object', `xcomp' etc are labelled slots alongside `subject', replacing the subcat list (whose insights need to be captured in some other way). 3. If an element (word) is shared by two larger structures, it has two dependencies, only one of which is `surface'. It is the surface dependency that is relevant to word order (and that must avoid tangling with other dependencies). E.g. (using arrows to show dependencies): <-subj- -xcomp-> -obj-> -comp-> SURFACE DEPENDENCIES John was eating an apple. <-----subj------ DEEP DEPENDENCY b. German Unlike English, German has `universal raising' for non-finite verbs; i.e. every dependent of a relevant non-finite verb is raised, i.e. also fills a slot in the structure of the higher clause. (`Relevant' verbs are presumably just those that are xcomps, e.g. those which depend on an auxiliary verb.) E.g. "Er hat einen Apfel gegessen", `He has eaten an apple'. ----xcomp--> <-s-- <-o-- Er hat Aepfel gegessen. -?--> <------------subj----------- In this example, the dependency labelled `?' is the `raised' dependency between "Aepfel", the object of "gegessen" and "hat". It is semantically empty (just like the raised subject dependency), but in this example it's also irrelevant to word order. When a word has two heads, either of them can be taken as the `surface' one, but in this case it makes no difference: so the object link could equally well have been shown as the `deep' relation, with the `?' link as the surface one. Now, turning to VP fronting: <-o- <-x-- -s-> Aepfel gegessen hat er. <-------?----- ----s---> This is what happens if the object dependency is surface. And this is what you get if the `?' dependency is surface: --?---> <-x- -s-> Gegessen hat er Aepfel. ------o---> ----s--> Likewise for all the other examples discussed, in which the lower verb's subject is preposed: -adjunct-> <----s----- <-x-- -?-> -c-> Ein Fehler unterlaufen ist ihr nie. <--------s--------------- -io-----> This doesn't explain the restrictions on the preposed subject (e.g. for definiteness), but at least it offers a structural analysis. Any comments welcome! ============================================================================ Prof Richard Hudson Tel: +44 171 387 7050 ext 3152 E-mail: r.hudson@ling.ucl.ac.uk Dept. of Phonetics and Linguistics Tel: +44 171 380 7172 Fax: +44 171 383 4108 UCL Gower Street London WC1E 6BT UK
This archive was generated by hypermail 2.0b3 on Fri Dec 18 1998 - 20:33:03 PST