• Keine Ergebnisse gefunden

Linear precedence rules

Im Dokument NP-Arguments in NPs (Seite 83-86)

Phrase Structure Grammar

2.5 Rules, principles, constraints

2.5.4 Linear precedence rules

The ID-schemata presented in the previous section determines only the hierar-chical relations between objects. In order to constrain the possible positions of constituents, HPSG (as well as GPSG Gazdar et al. cf. 1985: 44ff) makes use of LP-rules. The example (41) shows the 6 possible configurations for the specifier der, the head-noun Gewinn, and its complement phrase der WM, of which only (41a) is grammatical.

(41) a. [der]spec the

Gewinn win

[der WM]comp

of the World Championship b. * [der WM]comp

of the World Championship

Gewinn win

[der]spec the c. * Gewinn

win

[der]spec the

[der WM]comp

of the World Championship d. * Gewinn

win

[der WM]comp

of the World Championship

[der]spec the e. * [der]spec

the

[der WM]comp

of the World Championship

Gewinn win f. * [der WM]comp

of the World Championship

[der]spec the

Gewinn win

Due to the binarity constraint, stated in the ID-schemata, and the Head-Specifier Schema, which states that a specifier can only adjoin to a head with an empty comps list, the structures in (41c) and (41f) are ruled out, but (41b), (41d), and (41e) are still not being ruled out by the grammatical system, despite their ungramaticality. LP-rules constraining the order of constituents inside phrases help to get the right results in these cases. With LP-rules (cf. Pollard and Sag, 1987: 169ff), it is possible to constrain the order of:

• elements which contain a certain feature-configuration, or

• elements with a certain syntactic function, e.g. specifier or argument, (cf. (43)), or

• elements with a certain feature-configuration fulfilling a certain syntactic function (cf. (42)).

ID/LP grammars have an advantage towards a strong universal interpretation of the X-schema (cf. Kayne, 1994: 33ff). It is more natural to posit universal rules to

constrain hierarchical relations between words and phrases (our ID-schemata), but if these hierarchical constraints have to adjust constituent order as well, they actu-ally loose their universal potential, since – typologicactu-ally seen – linear order is much more variable than hierarchical relations. Therefore, a universal X-schema, which reflects also matters of constituent order, must assume a movement mechanism to achieve the distinct linearity patterns existent in the world languages, making the grammatical system of some languages unnecessarily complex.72 Whereas, our ID/LP grammar can postulate universals with respect to structural hierarchy, and let linear order be generalised by means of LP-rules avoiding a mechanism such as movement. Furthermore, universals on constituent order can be as well postulated without being forced – but with the possibility – to make claims about hierarchical structure (cf. Gazdar et al., 1985: 49f), so for example for the so-called “Greenberg universals” (cf. Greenberg, 1963: 58ff).

Going back to our phrases in (41), inserting a new attribute initial(ini), whose values are +or −,73 as a head attribute the position of arguments with respect to its head can be modelled either with the LP-rule in (42a) if the head precedes the argument, or with (42b) if the arguments precede the head (cf. Müller 1999: 163ff and Müller 2013a: 132f). This new attribute is needed, since German NPs are head-initial, but its VPs are head-final. Hence, one cannot make a generalisation with only one rule for all heads in German, but generalisations for heads with the value ini+or ini−, respectively.74

(42) LP-rule 1: Head – Argument a. head[ini+] <argument b. argument <head[ini−]

In the case of (41), the rule in (42a) would apply, ruling out all NPs in which the complement phrase precedes the head noun, that is (41b), (41e), and (41f).

72See for instance the derivational complexity for some possible constituent orders within NPs in Cinque (2005) and Georgi and Müller (2010). This complexity is due to the assumption of a rigid universal pattern “Specifier-Head-Complement” à la Kayne (1994). It is controversial, whether the derivational complexity is reflected in some way in cognitive complexity – and until now, there is no psycholinguistic evidence for that – or whether it should only be regarded as an artefact created by the axioms of the theory.

73To be more precise, the value ofiniis of typeboolean(bool), which has two subtypes+and−.

74In contrast to pure head-initial languages like English, or pure head-final languages like Japanese (cf. Pollard and Sag, 1987: 172).

Similarly, in order to rule out phrases in which the head precedes the specifier (cf. (41b), (41c), and (41d)), an LP-rule that refers to this head-specifier relation, such as (43), is needed.

(43) LP-rule 2: Head – Specifier specifier <head

As a result, the LP-rules in (42) and (43), along with the ID-schemata Head-Complement Schema given in (37) and Head-Specifier Schema given in (38) would yield grammatical structures licensed by constraints such as (44), (45), and (46).

(44) Head < Argument

Since in (44) the value of the ini attribute of thehd-dtr is +, the phon value of the whole structure is computed by the concatenation of thehd-dtr’sphon value first, and the nh-dtr’sphonvalue next. This is the case of our only grammatical example (41a) with respect to the head-complement combination.

(45) Argument <Head

⎡⎢

In (45), the phon value of the whole structure is the other way around: the nh-dtr’s phon value, followed by the hd-dtr’s phon value. This is needed in

German for the combination of verbs with their arguments. Being German a verb-final language, nouns and verbs must encode differentinivalues, +for the former, and −for the latter.

(46) Specifier < Head

In (46), theinivalues are not relevant. The LP-rule constraining the linearity of the head and its specifier only refers to the syntactic categories, and not to internal attribute-value pairs of them. The result is the concatenation of the specifier, i.e. of the nh-dtr’s phon value, with the head, i.e. the hd-dtr’s phon value.

Furthermore, the ID-schema in (38) specifies that the hd-dtr’s comps list must be empty in order for the two elements to be concatenated.

Finally, the interplay of ID-schemata and LP-rules give as the only possible grammatical structure the one in (41a), as it was intended to be.

Im Dokument NP-Arguments in NPs (Seite 83-86)