Tavs Bjerre (University of Southern Denmark, Kolding):
Locative Prepositions and Adverbs in Danish

Danish has a small group of words traditionally referred to as static-motoric or locative-directional adverbs, exemplified in (1):

(1) a. Børnene       legede  ude  i   haven.
       Children_the  played  out  in  garden_the

    b. Hunden   løb  hjem  i   haven.
       Dog_the  ran  home  in  garden_the.

    c. Børnene       gik     hjemad.
       Children_the  walked  homewards

In my talk I shall focus on the semantics of these items as well as of (normal) locative prepositions to account for the combinatoric possibilities they exhibit. The basic idea is that all locative prepositions denote a state temporally (and i some cases causally) related to a process or state denoted by the verb. Thus in (1a) the process and the state are simultaneous, while in (1b) the state follows and is caused by the process. To account for the difference between the two directional forms, I introduce a distinction between extensional and intensional states. (1b) describes a complex situation where the dog ends up at home, while in (1c) the children end up at home only in case nothing intervenes. Furthermore, I shall argue that `adverbial' express a relation between the location described by the PP and another, contextually dependent location.

Robert D. Borsley (University of Essex, Colchester):
Agreement and Missing NPs in Welsh

In HPSG, as in some other frameworks, it has sometimes been assumed that null subject sentences and other missing NP constructions involve phonologically empty NP's. However, recent work has rejected this approach. In papers like Bouma, Malouf and Sag (1997) it has been proposed that missing NP's of various kinds arise where some NP in the ARG-ST list of a word is of a type which has no counterpart in any valence list.

Welsh has missing NPs licensed by a number of agreement phenomena which appear to be problematic for this conception. In Welsh, finite verbs agree with a following pronominal subject as in (1).

(1) Gwelsan nhw.
    saw-3PL they
    'They saw.'

(2) arnyn  nhw
    on-3PL they
    'on them'

(3) Wnaeth  Emrys eu  gweld nhw.
    did-3SG Emrys 3PL see   they
    'Emrys saw them.'

(4) eu  tad    nhw
    3PL father they
    'their father'

(5) Dywedodd Emrys eu  bod nhw  yna.
    said-3SG Emrys 3PL be  they there
    'Emrys said they were there.'

In all these contexts there is no agreement with a non-pronominal NP

These phenomena are sufficiently similar to suggest that a unified account is desirable. HPSG seems quite promising here. Borsley (1989) argues on the basis of some of the data that post-verbal subjects and possessors should be analyzed as the realization not of the single member of the SUBJ list, but of an extra member of the COMPS list. Building on this idea, Pollard and Sag (1994, ch.9) argue that the full set of agreement phenomena should be analyzed as agreement with the initial member of a COMPS list. It is not difficult to formalize this idea.

All these phenomena license a missing NP. Thus, the pronouns can be omitted from all the preceding examples. Assuming the recent HPSG approach to missing NP's, this means that agreement must refer not to the COMPS list but to the ARG-ST list. However, agreement will sometimes be with the first member of an ARG-ST list and sometimes with the second member. It seems, then, that this approach precludes a unified account of agreement.

One solution to this problem would be to return to the assumption that missing NP's are phonologically null NP's. An alternative would be to assume that they reflect not a difference between ARG-ST lists and valence lists but a difference between the latter and constituent structure. It is not easy to chose between these two approaches.

References

Bouma, G., R. Malouf and I.A. Sag (1997), 'Satisfying constraints on extraction and adjunction', unpublished paper, Stanford University.

Pollard, C. and I.A. Sag (1994), `Head-driven Phrase Structure Grammar', Chicago: University of Chicago Press.

Berthold Crysmann (DFKI Saarbrücken):
Local Linear Licensing

Most approaches to word order in HPSG currently assume that linearisation operates across domains larger than immediate constituents. In the formulations of Reape (1994) and Kathol (1995), complex order domains are built up by means of an order preserving list union operation ("shuffle") which guarantees that ordering on some domain list must be preserved on any domain list higher up the tree. Although, under these approaches, a tree's ultimate yield (e.g. adjacency properties) can only be read off the top-most node, domain lists are nevertheless sorted on every intermediate node, introducing some amount of redundancy into the system.

On the basis of order-sensitive licensing relations in the grammar of European Portuguese (clitic placement and negative concord; (cf. Crysmann, to appear; Branco and Crysmann, to appear)), I will motivate the use of implicational constraints on linear precedence and domain membership. As these licensing relations are strictly clause-bound, cutting across the distinction between finite clauses (=saturated verbal signs) and infinitives (=almost saturated signs), I will suggest that the domain of such licensing is best defined in terms of linear structure. Furthermore, I will argue that application of linearisation constraints to all domain lists is not only unnecessary, but it also limits the expressive power of word order constraints: in particular, the statement of order-sensitive inside-out effects as constraints on linearisation can be made far more perspicuous, once this assumption is given up. Building on Kathol's variant of order domains, I propose that order constraints only apply to the highest domain list before an order domain is closed. In parallel to "shuffle" and "compaction", application of order constraints will be tied to the constituent structure, typically enforcing application of order constraints on the DOM list of the non-head daughter.

References

Crysmann, B. (to appear), "Syntactic transparency of pronominal affixes," in R. Cann, C. Grover and P. Miller, eds., Grammatical Interfaces in Head-driven Phrase Structure Grammar, Studies in Constraint-based Lexicalism, CSLI Publications, Stanford.

Kathol, A. (1995), "Linearization-based German syntax," Ph.D. thesis, Ohio State University.

Reape, M. (1994), "Domain union and word order variation in German," in J. Nerbonne, K. Netter and C. Pollard, eds., German in Head-Driven Phrase Structure Grammar, no. 46 in Lecture Notes, pp. 151-197, CSLI Publications, Stanford University.

Kordula De Kuthy (Universität Tübingen):
Context effects and discontinuous NPs in German

  (1)   Über Syntax     hat   Sarah   sich  [ein  Buch]   ausgeliehen.
        about syntax    has   Sarah   self   a    book    borrowed
        `Sarah borrowed a book on syntax.'

  (2)   * Über  Syntax    hat  Sarah    [ein  Buch]   geklaut.
          on    syntax    has  Sarah     a    book    stolen
          `Sarah stole a book on syntax.'

  (3)   Gestern  wurde  in  der  Bibliothek  eine  Anzahl  von  Linguistikbüchern
        geklaut. Vor allem Semantikbücher verschwanden dabei.

        `Yesterday, a number of linguistics books were stolen from the 
         library. Mostly books on semantic disappeared.'

        a.  Über  Syntax     wurde    jedoch    [nur   ein   einziges   Buch]   geklaut.
            on    syntax     was      however    only  one   single     book    stolen
            `There was, however, only one book on syntax stolen.'

Danièle Godard (CNRS, Université Lille 3):
Complex predicates in Romance languages: a classification of their properties

A systematic comparative approach brings to light a new classification of the properties into three sets. The first one groups all the properties which show complex predicates to be associated with a monoclausality effect. The most famous is 'clitic climbing', which serves a the criterion for the phenomenon. The others are: the 'long' passives (middle-passive, and periphrastic), occurrence in bounded dependencies, and absence of sentential negation on the downstairs verb. Complex predicates do not behave homogeneously wrt these properties (except for negation), but each of them is associated with at least another one. In HPSG, monoclausality can be represented using argument composition (Hinrichs & Nakazawa).

This leaves open the question of the structure, which has to do with the second set of properties: the sequence of the predicate complement does not behave as a constituent wrt extraction, the occurrence (or non-occurrence) of a subject or an adverb between the two predicates, the possible (or impossible) wide scope of the head over the conjunction of the predicate complement and its subcategorized complements. After discussing the data, and stressing the importance of the last property which is generally overlooked, we propose two groups of predicates (heads of a complex predicate): (i) French, Italian and Portuguese predicates plus the Romanian modals, and (ii) Catalan, Spanish predicates and the tense and aspectual Romanian auxiliaries. The first are the head of a 'flat' structure while the others make up a verbal complex with the complement predicate.

In our view, the third set of properties does not tell us anything about either monoclausality or structure, but are idiosyncratic properties of predicates, although much has been made of them in the literature. Such are the possibility of lexical alternations (how many complement structures is the predicate associated with), the occurrence or non-occurrence of certain clitics on the downstairs verb (in the case of composition, see F causatives and R aux's), the possibility of a nul complement (see P), the requirement of adjacency (outside the verbal complex, see croire, etc. with a lot of variation), and the 'change' of auxiliary of restructuring verbs in Italian.

Stephan Kepser (Universität Tübingen):
Coalgebraic Modelling of Head-Driven Phrase Structure Grammar

Tibor Kiss (Ruhr-Universität Bochum):
A parametric binding theory

Two extensions of the basic theory will be discussed as well: First, anaphors seem to be generally prohibited in adjuncts in English, but can be found in German adjunct construction. In including anaphoric adjuncts into binding theory, we have to cope with the role of the adjuncts in an obliqueness hierarchy.

Second, we will investigate how the proposed Principle A relates to long-distance anaphora. We will propose that in general Principle A should not care for long-distance anaphora, because long-distance anaphora are generally morphologically distinct from 'true' anaphors. This rule, however, is not exceptionless.

Bob Levine (The Ohio State University & Universität Tübingen):
Weak UDCs and strong crossover: type covariation between token-distinct signs?

*Robin-i was tough for us to persuade him-i to allow us to nominate __-i for a knighthood.

which were first noted in Cinque's 1990 monograph manifest a typical strong crossover effect, parallel to `It was Robin who-i we had a hard time persuading him-i to allow us to nominate __i for a knighthood'. But the Pollard-Sag account of `tough' predicts that no such effect should arise, since the content type of the gap, based on the lexical specifications for the SLASH value borne by the infinitival VP, is ppro, and personal pronouns never induce strong crossover in English. This problem can be overcome, as argued in the Levine-Hukari paper given at the 1999 Colloque de Syntax et Semantique at Paris 7, by (i) altering binding principle C slightly to exempt coindexation between the gap and an element outside the SLASH path, along lines first proposed in Chomsky's 1986 binding theory in Knowledge of Language, and (ii) redefining the content type of the SLASH value introduced in `tough' VP complements as npro, leading to the induction of SCO effects. So far, so good.

But this solution turns out to be insufficient. I will offer evidence at the workshop that whether or not SCO effects appear depends entirely on the content type of the constituent with whose index the index of the `tough' SLASH value is identified. SCO effects only appear when the coindexed NP in a missing object construction is itself an npro; when it is a ppro or an ana element, no SCO effects appear, and the examples are well-formed as long as the relevant binding principles are satisfied. This pattern holds not only for `tough', but for clefts as well. The Pollard-Sag story about clefts parallels that for missing object constructions: the facts appear to independently call for the same revision of principle C, but it turns out that in addition the type of the gap must covary with that of the constituents coindexed with the cleft SLASH specification.

We thus have a somewhat unusual situation: in both varieties of weak UDC, the SLASH value must, for familiar reason, be token-distinct from the LOC specification of the coindexed element, yet the *type* of the CONT value of this SLASH specification must be the same as that of the the element SLASH is coindexed with. This might be implemented by making the CONT specification, not merely the index, structure-shared between the weak UDC pseudo-filler and the SLASH specification, or by some other formal restriction directly expressing the covaration in type among the relevant objects. An empirically interesting fallout of these considerations is that the `pit-stop' reflexive cases discussed by Pollard and Sag cannot be accomodated by their solution (presented in section 6.5 of P&S94), since that solution predicts that data will be well-formed which instead reflect obvious SCO ill-formedness effects. Rather, in addition to the lexical item speakers of the pit-stop reflexive variant of English share with non-pit-stop English speakers, a separate lexical entry for `be' is required to license the pit-stop cases, with the focal element specified as [ana].

Paola Monachesi (Universiteit Utrecht):
Romanian auxiliary(-like) constructions

These constructions are characterized by the presence (or absence) of clitic climbing and by the possibility (or impossibility) for certain elements to intervene between the auxiliary and the nonfinite verb: a whole range of variation is attested. In particular, tense auxiliaries trigger clitic climbing, but nothing, except for monosyllabic intensifiers, can intervene between the auxiliary and the nonfinite verb. Similarly, the modal verb `a putea' triggers clitic climbing, but adverbs and pronominal subjects can separate the modal from the infinitival. As for the colloquial future, no elements can occur between the auxiliary and the embedded verb, as in the case of tense auxiliaries, but contrary to tense auxiliaries, clitic climbing is not allowed. Romanian raising verbs do not trigger clitic climbing either. However, they allow subjects to intervene between the relevant verbs, as in the case of the modal verb `a putea', while adverbs and quantifiers cannot occur in this position on a par with the tense auxiliaries.

My claim is that the interaction of the lexical mechanism of argument composition with the various modes in which elements can compose allow for a unitary analysis of these constructions.

Stefan Müller (DFKI Saarbrücken):
Depictive Predicates in German

It is known that depictive secondary predicates in German behave like adjuncts. I will examine the possible antecedents of depictive predicates and provide evidence that the subjects of depictive predicates may be coreferent with nominatives, accusatives, and datives although the last possability is frequently denied (Rothstein, 1985; Haider, 1985).

It is clear that the reference to datives is more marked than that to nominatives or accusatives. I explain this by the different accessability of the appropriate grammatical functions on the obliqueness hierarchy that is known since Keenan and Comrie (1977):

SUBJECT => DIRECT => INDIRECT => OBLIQUES => GENITIVES => OBJECTS OF 
           OBJECT    OBJECT                               COMPARISON 

Depictive predicates pose interesting challanges for the architecture of grammar since it is possible that they refer to elements that are neither expressed at the surface nor unrealized subjects as in control constructions.

(1) Jedes nackt geputzte Fenster muß  extra      bezahlt werden.
    every naked cleaned  window  must separately payed   get
    `Every window that has been cleaned naked has to be paid separately.'

The analysis is part of a book about predication: http://hpsg.fu-berlin.de/~stefan/Pub/complex.html

Sabine Reinhard (Universität Osnabrück):
An Architecture for Morphology

Ivan Sag (Stanford University):
Some Wh-Constructions: a default analysis

Manfred Sailer (Universität Tübingen):
Idiomatic expressions in non-constructional HPSG

(1)     a. kick the bucket ('die')

        b. trip the light fantastic ('dance [nimbly]')

        c. den Löffel abgeben ('die')
           the spoon  away.give

The first option has been proposed in Riehemann 97 within the framework of construction types developed in Sag 97. In the talk, we will explore the second option. We introduce an attribute COLL which is defined on the sort sign. We assume a trivial COLL value for regular signs and a non-trivial COLL value for irregular signs. We will show that the regularity/irregularity dichotomy cuts across all subsorts of signs (words and phrases).

We will, then, turn to expressions whose occurrence is restricted to a very limited number of contexts (hapax legomena/'cran'-elements). We show that there are words that have this property:

(2)     a. headway: make headway

        b. Kohldampf: Kohldampf     haben  ('be ravenously hungry')
                      cabbage.steam have

The combinations in (2) can be analyzed as regular, with the exception that they contain a cran-element. The occurrence restrictions of these cran-elements cannot be expressed in terms of selectional requirements. Thus, they pose an additional problem for the architecture of Pollard and Sag 94. We show that the non-trivial COLL value can be used to account for these distributional irregularities as well.

Finally, we will demonstrate that there are also phrases which show distributional irregularities. These phrase have a non-trivial COLL value, because of their internal idiomatic properties. Again, we can use this COLL value to capture their distribution.

(3)     [wo Barthel den Most holt]: 

        wissen/*glauben/ *annehmen, [wo    Barthel den Most  holt   ]
        know/   believe/  assume    [where Barthel the cider fetches]
        ('know every trick in the book')

As we can use a non-trivial COLL value to account for distributional irregularities of internally irregular signs (words or phrases), the introduction of a new attribute instead of new sorts seems to be justified.

References

Sag, Ivan A. (1997). English Relative Clause Constructions. In: Journal of Linguistics, Vol 33, pp. 431-483.

Kiril Simov (LML, Bulgarian Academy of Sciences, Sofia):
Grammar Learning from an HPSG Corpus

Jesse Tseng (Universität Tübingen):
Syntactic Selection of Prepositions

In most analyses, PFORM takes the values to, with, of, etc., with the assumption that prepositions with distinct lexical (i.e., phonological) forms should have distinct PFORM values. I will discuss cases, however, where distinct prepositions take non-distinct PFORM values, and where prepositions with the same lexical form nevertheless have distinct PFORM specifications. In particular, the subtypes of pform must reflect not only lexical form, but also case government and valence properties.

Nathan Vaillette (The Ohio State University & Universität Tübingen):
Resumptive Pronoun Constructions as Unbounded Dependencies

References

Doron, Edit (1982) "On the syntax and semantics of resumptive pronouns." Texas Linguistic Forum 19:1--48.

Frank van Eynde (Katholieke Universiteit Leuven):
Some more minor prepositions in Dutch

Exploiting that possibility, I have claimed in Minor prepositions in Dutch (2000) that there are two types of prepositions: the major ones which can head a branching PP, and the minor ones which cannot. Not surprisingly, most of the prepositions belong to the former type, but there are also some which ---at least in some of their uses--- behave as nonhead sisters of verbal or nominal projections. A relevant example is the Dutch infinitival te (to). Treating it as the head of a PP leads to various problems, and so does its treatment as an affix, a marker, a head of VP[inf] or a head of CP. Treating it as a minor preposition, by contrast, solves all of these problems, and makes a number of further predictions, which turn out to be correct, i.e. the impossibility to be stressed, conjoined or topicalised, the leftward tendency, and the lack of autonomous content.

The aim of this talk is to demonstrate that Dutch has at least three other constructions in which the preposition has minor status, i.e. the use of van (of) in phrases like Er staan van die dure glazen in de kast (lit. There are of those expensive glasses in the cupboard), the use of voor (for) in the combination Wat voor boeken leest hij tegenwoordig? (lit. What for books reads he nowadays?), and the complementizer use of om (for), as in Hij had beloofd om de bloemen water te geven (lit. He had promised for the flowers water to give). Extrapolating on the basis of these examples and further refining the criteria, I will identify some further plausible candidates for a minor preposition treatment, both in Dutch, and in some other languages.

PS: Although this talk builds on a previous one with almost the same title, the presentation will be self-contained.

References

Van Eynde, F. (2000), Minor Prepositions in Dutch. In: Proceedings of the Berkeley Formal Grammar Conference. Berkeley, July 2000.