<<   >>   TOC

5.1 A-Structure

Since movement-analyses of scrambling lead into dead ends, the next obvious place to look for a solution is the lexicon or, more specifically, argument structure (A-structure). In the discussion of example (14) under 4.1 I already mentioned the suspicion that overt markings, i.e. case morphology for German, have something to do with scrambling possibilities.

Neeleman (1994) proposes that scrambling structures are identifiable as directly projected q-grids. Similar to Grimshaw (1990), he suggests that, first of all, q-grids follow a thematic hierarchy (Jackendoff 1972), and that secondly, overt case allows to unequivocally link an argument to its underlying position in the q-grid, even though "little is known about the linking rules [of case and q-roles]" (Neeleman: 420). A verb like zeigen ('show') would thus predictably have the following A-structure:

(23)

            AGENT > EXPERIENCER > THEME

                                       |                   

                NOM           DAT                ACC

Dative linking is encoded in the lexicon, whereas nominative and accusative are structural cases, such that nominative is assigned to the most prominent argument and accusative to the lowest one. While this is certainly true for the majority of German verbs, this formulation is too strong. The data in (22) show a verb that in a specific idiomatic environment projects the EXPERIENCER in front of the AGENT and strongly resists any reordering within its domain. Whereas the reordering resistance may be subject to independent conditions (e.g. specificity), it seems likely that the A-structure encoded in the lexicon reflects an argument ordering which violates the thematic hierarchy. Haider (1993) furthermore points out that there seems to be whole class of verbs with deviating default order, as exemplified by aussetzen in (24).

(24)

            a.         wenn man         die Kinder        der Gefahr        aussetzt.

                        if        oneNOM the childrenACC    the dangerDAT   exposes

            b.         ?wenn man der Gefahr die Kinder aussetzt.

                                                            (examples taken from Haider 1993: 197)

While the thematic hierarchy remains intact here, accusative is linked to the EXPERIENCER, whereas the THEME is marked dative. Linking of case to thematic roles does thus not provide a possible solution without further assumptions.

Haider (1993) elaborates on a proposal by Bierwisch (1988) and argues that case is linked directly to an underlying argument structure, such that a verb like geben, 'give', would have a lexicon entry containing information similar to the following:

(25)

            lz           ly          lx              [x CAUSE [y HAVE z]]           semantic form

             |              |             |    

            f2             [+Dat]      d, f1                                                                                  syntactic form

The l-operators[1] are mapped into the hierarchical syntactic structure by l-conversion, i.e. the leftmost operator will yield the most embedded argument.[2] Associated with the l-operators is the syntactic form, which includes the q-grid (not further specified by Haider) and case information: [+Dat], as morphological case, is immediately expressable, whereas f1 and f2 represent structural cases. F1  is marked d, for „designated“, and will thus receive nominative case, whereas f2 will by default be marked accusative (cp. Haider 1993: 110 ff., Frey 1994: 38ff.).[3] A verb like aussetzen may thus have the following A-structure representation:

(26)

            lz           ly        lx     [x CAUSE [y BE-SUBJECT-TO z]]  semantic form

               |            |           |      

            [+Dat]     f2         d, f1                                                                         syntactic form

Since case is thus specified in the lexicon, Haider assumes that arguments are in principle not tied to specific positions within the VP, hence the 'open' VP structure shown in (6). All arguments are adjoined to VP at the same level and licensed through government by the verb (government defined as m-command: the verb thus governs everything dominated and adjoined to VP). Since arguments are structurally almost indistinguishable, positional information in the form of c-command relationships between arguments is only used for identification if case does not provide enough information.

If the syntactic form allows unequivocal identification of overt DPs with underlying argument positions in the semantic form (as is the case for 25 and 26), scrambling should thus in principle be well-formed. As Neeleman (1994) points out, this predicts that A-structures without sufficient distinctions in the syntactic form make scrambling at least very unlikely, which is born out by the facts . (27) shows a possible A-structure for lehren ('teach'):

(27)

            lz           ly          lx              [x CAUSE [y KNOW z]]         semantic form

               |            |             |    

              f3           f2           d, f1                                                                                  syntactic form

Since neither f2 nor f3 are marked for inherent case, they will both surface with accusative case. The syntactic form does not allow to draw correct inferences about the semantic form, unless via overt argument position. (28b) will thus yield a garden path effect, yielding the (strange) interpretation in (28c).

(28)

            a.         ..., daß der Lehrer        den Schüler      die Sprache         lehrt.

                             that the teacherNOM the studentACC    the languageACC  teaches.         

            b.         ?..., daß der Lehrer die Sprache den Schüler lehrt.

            c.         that the teacher teaches the student to the language.

                                                                        (cp. Neeleman 1994: 421)

The garden path effect in (28) is relatively easy to overcome, which makes one hesitate to mark (28b) as ungrammatical. But notice that (28) allows straightforward disambiguation through world knowledge, a strategy not available in (29):

(29)

            a.         ..., daß ich einen Schund einen Verbrecher heisse.

                             that I    a       punkACC a       criminalACC call.           

            b.         ..., daß ich einen Verbrecher einen Schund heisse.

            c.         ...that I call a punk a criminal.

            d.         ...that I call a criminal a punk.

It is very hard to get scrambling interpretations for either (29a) or (29b), i.e. (29a) will be interpreted as (29c) and (29b) as (29d).

The situation is further complicated by homophones in the German case system. There is, for example, no overt distinction between nominative and accusative for feminine nouns, such that the garden path interpretation for both (30a) and (30b) would be (30c) and (30d) respectively:

(30)

            a.         ..., daß die Verkäuferin            die Kundin                    angelächelt hat.

                             that the saleswomanACC/NOM the customerACC/NOM smiled-at    has.

            b.         ..., daß die Kundin die Verkäuferin angelächelt hat.

            c.         that the saleswoman smiled at the customer.

            d.         that the customer smiled at the saleswoman.

Here, it should be pointed out that the assumption of A-structures as presented by Haider (1993) does not in itself provide a mechanism for predicting the well-formedness or acceptability of scrambling structures. As I see it, VP-internal scrambling is always possible as far as licensing is concerned. However, a complex set of output conditions determines inhowfar a given structure is acceptable.

To pick up a point raised under 3 now: since scrambling is largely dependent on possibly idiosyncratic encodings in the lexicon and a number of additional more or less weak output conditions[4], variability in native speaker judgments is to be expected.

<<   >>   TOC 

© Philipp Strazny 1997


[1]               Nothing rests on the question whether Montague grammar/l-calculus is an appropriate format to capture natural language grammer. Here, it is a convenient notation to capture the notion that argument ordering is not necessarily a function of (propositional) logical form. l-conversion, as used here, simply provides a presyntactic mechanism assuring a specific predicate's default argument order.

[2]               This mechanism of mapping an argument list into syntax is in essence similar to the one proposed by Grimshaw (1990), cp. also Li (1990).

[3]               cp. Neeleman's (1994) treatment of case

[4]               One such condition would be the well-known 'definiteness effect': German strongly prefers definite objects to appear before an indefinite object. This is possibly linked with a 'specificity effect': similar to Turkish (Enç 1991), German seems to have a strong adjacency requirement for unspecific objects and the verb.