Adverb position and verb raising

[Bouma G.]

The discussion on adverb position and verb raising has concentrated so
far on the [ V Adv NP] pattern. It is perhaps interesting to consider the
mirror image [NP Adv V] pattern as well. This order occurs in languages
like German or Dutch, and has been used as an argument for NP-movement
rather than V-movement.

In V-final languages like German and Dutch, you find the pattern NP Adv V in
subordinate clauses:

dat Kim haar moeder regelmatig bezoekt
that Kim her mother regularly  visits

In transformational frameworks (see Webelhuth's diss, various papers by de Hoop,
and others), this is generally accounted for by assuming a `scrambling'
operation, which moves a complement NP to a position left of the adverb
(where it is adjoined to some verbal projection).

Has anything like this been proposed for LFG?

The position of adjuncts poses interesting puzzles if you consider the interaction
with cross-serial word order in Dutch. As is well-known, Dutch perception
verbs can induce cross-serial word order:

dat Kim Anne een lied hoorde zingen
that Kim Anne a song heard sing

dat Kim Anne het scherm zag bestuderen
that Kim Anne the screen saw stare-at

Adjuncts preceding such a verb cluster may take wide or narrow scope w.r.t. the
upstrairs verb (see e.g. Haegeman and van Riemsdijk, LI 1986, for more discussion):

dat Kim Anne niet een lied hoorde zingen
that Kim Anne not a song heard sing
`that Kim didn't hear Anne sing a song'

dat Kim Anne aandachtig het scherm zag bestuderen
that Kim Anne intensely the screen saw stare-at
`that Kim saw Anne stare at the screen intensely'

Although scope of adjuncts can be hard to determine, the negation `niet' is
most naturally interpreted as taking scope over the matrix verb, whereas the
manner adverb  `aandachtig' is  most naturally interpreted as scoping over the
embedded verb.

In both cases, `scrambling' of the NP is possible:

dat Kim Anne het lied niet hoorde zingen
that Kim Anne the song not heard sing

dat Kim Anne het scherm aandachtig zag bestuderen

With negation, scrambling seems to require a definite NP.

I would be interested to know what the LFG account of such data would be.
(The classic analysis of Bresnan et al seems to be able to cover these cases,
but then Johnson's diss notes that this analysis overgenerates.  I seem to recall
there was more work on Dutch cross-serial dependencies by Kaplan and Zaenen using
functional uncertainty, but I don't think this adressed adjuncts.)

The transformational literature has to assume both (long-distance) NP-scrambling
and V-raising to account for these data.

In Bouma, Malouf, and Sag, NLLT 2001, data like these are taken as evidence that
adjuncts can be selected lexically, using a lexical `argument structure
extension' operation.  Consequently, adjuncts can be selected like complements.
The scope of an adjunct is determined by the lexical head that selects for it.
Cross-serial word order is accounted for by means of argument inheritance,
following an idea of Hinrichs and Nakazawa. I think the idea that argument
structure in the lexicon may not be fixed, but the consequence of interacting
constraints, is well-known in LFG. Argument inheritance seems to correspond to
something like predicate merger in LFG. It seems then, that a `lexicalist' account
of these data could also be formulated in LFG.

best,

Gosse Bouma.






--
Gosse Bouma, Alfa-informatica, RUG, Postbus 716, 9700 AS Groningen
gosse at let.rug.nl      tel. +31-50-3635937      fax  +31-50-3636855