<<   >>   TOC

8. The syntax of names

 

In the previous section I argued that names are underspecified for syntactically relevant features like definiteness and specificity. Underspecification allows for the fact that (a) proper names can occur in spite of a constraint which disallows phrases marked for a particular feature or that (b) they cannot occur where a particular feature is required. One nagging question immediately reappears: could it be that there is a categorial difference between common nouns and proper names after all, namely to be found in the fact that common nouns are specified for certain features, while names are not? Again, because of free conversion, such a categorial difference is highly unlikely. Moreover, common nouns can, of course, be used as specifics or nonspecifics, as definites or indefinites, which clearly suggests that the relevant features are not inherent in the lexical categories, but are rather supplied syntactically and therefore visible to syntactic rules.

        Within the Principles & Parameters framework one would, of course, hope to be able to find general principles that the language specific syntactic characteristics of proper names can be derived from. I doubt, however, that this is possible. In English, the distribution of proper names is intuitively straightforward: proper names are words that do not take a determiner in the singular. Nevertheless, as I showed in the context of example (9), determiners of all kinds can full well cooccur with proper names even in English, and in certain environments a determiner is even required (e.g. when the name is modified see below).

        In a nonrepresentative and eurocentric survey of languages that use overt determiners, one finds that determiner cooccurrences with proper names are suspiciously regular: in English, German, French, and Spanish, proper names stand by default without determiners. An indefinite determiner may be used to convert the proper name into a generic or an explicitly novel predicative expression, but - at least in the standard dialects - a definite article with a simple unmodified name is to be considered ungrammatical. However, there are particular syntactic environments which allow or even require the use of a definite article with proper names:

(22)       

English:

a.          I met the/* old John.     

b.         I met the/* John who I introduced to you yesterday.

German:

a.          Ich traf den/* alten Johannes.

b.         Ich traf den /* Johannes, den ich dir gestern vorgestellt habe.

French:

a.          Cest ainsi que Rolland pousa la/* belle Aude.

             it-is   so     that   Rolland  married the beautiful Aude.

b.          Le/* Pierre que jaime nest plus.

            The Pierre who I-love not-is any-longer.  

(Gary-Prieur 96)

Spanish:

a.          El/* buen Alberto se fue a la/* hermosa Maracaibo.

            The good Alberto REFL went to the beautiful M.

b.         El/* Alberto que estudia Filosofia es muy agradable.

           The Alberto who studies Philosophy  is very likable.

 

In all four languages, a definite article thus appears when the proper name is restrictively modified. With nonrestrictive modification proper names can still optionally occur without determiner, but since nonrestrictive modifiers are usually set apart by phonological means (e.g. a pause between the head noun and the nonrestrictive relative clause), one may argue that they are not really part of the syntax. Restrictive modifiers, on the other hand, are phonologically merged with the modified element. If the Chomskyan view is correct that sentences are construed derivationally and that there is a point Spell-Out at which the structure established so far is made available to the phonological component PF, one could claim that restrictive modifiers are base generated in their given position and are thus pre-Spell-Out components of the syntax. The phonological component will thus treat them as integral parts of the syntactic structure. Nonrestrictive modifiers, in contrast, are separated from the main structure by the phonological module, which suggests that they may actually be inserted at the level of PF. Since this possibility exists, I will for concentrate on restrictive modification here.[1]

Gary-Prieur (1994) explains the requirement for definite articles with restrictive modifiers on semantic grounds: when a proper name in Gary-Prieurs view a semantic category denoting an individual is restrictively modified, this triggers a type conversion process. While a name, by itself, would be assigned type <e>, the structure [ [ N ] modifier ] would be <<e, t>, t>. Being thus a set, this structure is treated like any other set expression by the syntax, i.e. an overt determiner is required to reestablish uniqueness.

My objection is essentially the same as the one formulated earlier: it is not clear how this type conversion is executed. The modifier would somehow subcategorize for an individual, but return a set. Since one and the same modifier can cooccur with proper names and common nouns, the semantic subcategorization frame would redundantly have to be specified for two different types as complements.

It is my task here to explain the requirement for the definite article without appealing to semantic type (since the noun-modifier construction does not differ in type from a simple noun), and a syntactic explanation seems to be the closest alternative. The suspicion that syntax plays a role is supported by the fact that it does not seem to be an issue whether the modifer precedes or follows the noun, as in English the old John vs. the John who... If semantics was solely governing these (surface) structures, as seems to be suggested by Steedman (1996), one may wonder why e.g. a following relative clause cannot modify a [DP D [NP N ] ] structure in which the determiner is optionally left empty.[2] This structure is ruled out by the subcategorization requirement of the matrix verb which expects a DP. A base generated structure like [[DP D [NP N ] ] who...] could be argued to be a CP, however. Kayne (1994), in fact, argues that the modified NP is generated in [Spec, CP] of the modifying relative clause, but he also crucially assumes that is wrapped in a DP.       

For bare proper names, I suggested the possibility of a D head, which is phonologically empty but semantically contentful insofar as its presence plays a role in the pragmatic intepretation. Many linguists are uncomfortable with the proliferation of empty heads in current syntactic theory, since the mere possibility of empty heads seems to make the theory far too powerful. I would thus suggest a constraint on the licensing of empty heads:  

(EHC)  Empty Head Constraint (preliminary version)

Empty functional heads are only licensed by adjoined lexical heads.[3]

It remains to be shown that unmodified proper names can undergo head-to-head movement to D, but that this kind of movement is ruled out as soon as a restrictive modifer is present. The first seems straightforward the head noun simply moves up to the next higher head position (see 23). This movement is motivated by the licensing requirement of the EHC. I do not want to speculate about Greed vs. Enlightened Greed, or Affect-alpha vs. Move-alpha, since the status of the relevant principles seems very unclear to me at this point.

(23)      

Since this movement is blocked in DPs with restrictive modifiers, I suggest that it is the relationship between the restrictive modifier and the modified noun which rules out such movements in more complex DPs. Since the modifier introduces a predication over the modified, I assume in the tradition of Generative Semantics that AP-N or N-relative constructions contain an underlying predicate phrase:

(24)      

a.          [VP pro  [V (is) [ AP ] ]   

b.         [CP pro  [C who [ ... ] ]   

Contrary to transformationalist views, I assume that these predicative phrases are base generated in an adjoined position to the modified NP (I do not have anything to say about the directionality of adjunction here). There, they are lexically governed by the modified noun, such that pro in Spec position of the adjunct is identified with the governing N:

(25)

 

If such an NP is selected by a D head, the D node needs to be interpretable at PF, i.e. it requires P-features. P-features can be made available in two ways: either D is supplied with independent phonological realization (by insertion of an overt determiner), or it triggers movement of the governed lexical head, i.e. the N. As already mentioned in footnote 5 above, this is very similar to the relationship between INFL and the verb. In English, INFL is not independently saturated, so the verb is moved up. In Bantu languages, on the other hand, INFL is morphogically realized, such that the verb may stay in situ (cf. Myers 1990). As the realization of proper names in languages with overt determiners show, head-to-head movement of N to D is always a syntactic possibility. I believe that even in the structures shown in (25), N-to-D movement is syntactically possible. It is ruled out on semantic grounds, however: if pro is not in the immediate domain of an overt N head, its referent may be freely picked from the domain of discourse, such that pro fails to reliably receive the intended interpretation. 

        If the above analysis is correct, it should be applicable to other cases of null determiners. Here, indefinite plurals have to be mentioned, since they can occur with null determiners in spite of restrictive modification in English and German:

(27)      

English:

a.          There were dogs of all ages, but I saw only old dogs.

b.         ..., but I saw only dogs who were quite old.                 

German (parallel constructions)

c.          Da waren Hunde jeden Alters, aber ich say nur alte Hunde.

d.         ..., aber ich sah nur Hunde, die schon ziemlich alt waren.

Both English and German have a lexical gap in the indefinite plural of determiners. However, the lexical gap can obviously not serve as an explanation, but rather needs to be explained itself. The Spanish unos perros or the French des chiens do not seem to contain more information than English dogs or German Hunde. Thus, one could argue that unos or des do not convey any information that is not already contained in the plural nouns.

        Nouns denote sets. A singular marked noun denotes an individual from a given set, i.e. in my view: a subset with cardinality one. For the listener, definiteness vs. indefiniteness is relevant information, because it certainly makes a difference whether a particular individual is meant (definiteness entails specificity) or just any individual from the given set. In the singular, all four languages thus mark for definiteness and indefiniteness (notice that proper names default to a definite interpretation, see below).

        If a noun is marked for plural, it denotes a subset of unspecified cardinality. Similar to definiteness and indefiniteness (cf. Heim 1991), singular and plural form a scalar implicature: if a noun is marked for singular, it donotes a subset with cardinality of exactly one; if a noun is not marked for singular, it does not denote a subset with a cardinality of exactly one. Thus, plural means a cardinality higher than one. This scalar implicature is a pragmatic expectation. Therefore, the continuation of the scenario in (27) given in (28) may be somewhat puzzling to a listener, but the second sentence does not falsify the first:

(28)       ..., but I saw only old dogs. To be honest, I saw only one dog and that was old.          

        In any case, a noun marked for plural carries the implication that the denoted set contains more than one element. Which elements exactly are to be included is not relevant. For all purposes, a plural noun is thus indefinite, unless specifically marked otherwise. Spanish unos or French des are thus redundancies that simply support the indefinite interpretation of the plural noun. Note that redundancies in overt expression are not on par with redundancies in e.g. subcategorization frames. The first is a computational service to the listener, whereas the second would introduce unnecessary computational difficulties.

        The fact that English and German do not require overt determiners with restrictively modified indefinite plurals poses a problem for the EHC as formulated so far: even though the indefinite article would be redundant, one would expect it to be required, since N-to-D movement is not an option given the structural assumptions made above. Thus, either my structural assumptions are wrong, or the EHC is too strong. I opt for the latter and propose a weakened version:

(EHC2) Empty Head Constraint (final version)

Empty functional heads can only add semantic information iff they are licensed by an adjoined lexical head.

This new version of the EHC allows for indefinite determiners to be phonologically null if the modified noun is already indefinite, e.g. pluralized. This weaker EHC has the welcome effect for syntacticians to allow for empty heads with a purely syntactic function, but it still constrains the theory so far that random introduction of semantic information via empty categories is ruled out.

<<   >>   TOC

Philipp Strazny 1998


[1]           Nonrestrictive adjectival constructions often do not exhibit a phonological separation, but rather the opposite, i.e. an complete phonological merger. Consider German Klein-Erna (little Erna) or English Little John. Both in English and in German, these adjective-noun constructions can occur without determiner, i.e. they have the distribution of a regular proper name. There is an intuitive sense that adjective and noun form a tight unit that is different from typical nominals with adjectival adjuncts. I believe that this intuition stems from the fact that these words are, in fact, not nominal phrases, but rather compounds formed in the lexicon. Thus, such names are not counterexamples to my view that only restrictive modification is a pre-Spell-Out syntactic process.

[2]           I believe that nonrestrictive modifiers can actually yield structures like the ones ruled out for restrictive modifiers: however, the relevant Merge occurs at PF such that the subcategorization frame of the verb is not affected.

[3]           Notice that I hedged by claiming this licensing requirement only for functional heads. Thus, it will apply to e.g. INFL, C and D, but not to lexical heads such as PRO, pro, NP-traces, or light verbs.