Oxford University Working Papers in Linguistics, Philology & Phonetics, Lecture notes of Philology

Download Oxford University Working Papers in Linguistics, Philology & Phonetics and more Lecture notes Philology in PDF only on Docsity!

Oxford University Working Papers in Linguistics,

Philology & Phonetics

Anna McNay (Editor)

Some Grammatical

Abbreviations

abl. ablative acc. accusative act. active dat. dative decl. declarative f. feminine foc. focus fut. future fut.pf. future perfect gen. genitive impf. imperfect impv. imperative ind. indicative infin. infinitive m. masculine mid. middle n. neuter nom. nominative pass. passive pf. perfect pl. plural plpf. pluperfect pple participle pres. present refl. reflexive sg. singular subj. subjunctive top. topic voc. vocative

Contents

Adjectives and Headedness Robert Truswell.............................. 1

The Internal Structure of the Korean DP: Evidence from prenominal and postnominal classifiers Inji Choi.................................. 19

Split Topicalisation – Motivating the Split Anna McNay................................ 39

Tracking the Progress of a Polarity Shift in Romanian Ioana Costache.............................. 59

Nonsubject Agreement and Discourse Roles Mary Dalrymple, Irina Nikolaeva..................... 71

*The Syntax and Semantics of Denominative - ye/o - Verbs in Ancient Greek Peter Barber................................ 93

The Modern Japanese Complementisers no and koto and their Old Japanese Precursors: A diachronic explanation for free variation Janick Wrona............................... 119

On the Nature of Subjective Modality Lars Ingemar Larm............................ 137

The Loss of Ergativity in Dari Modal Verbs Annahita Farudi, Maziar Doustdar Toosarvandani........... 149

wh -Question Formation in Nguni Joachim Sabel, Jochen Zeller....................... 161

Pronouns and procedural meaning: The relevance of spaghetti code and paranoid delusion David Cram, Paul Hedley......................... 179

2 Robert Truswell

semantics, but rather that there is a parallel syntactic selectional requirement for attributive adjectives to have a nominal sister. This section will also sketch one way of accommodating such a requirement within current syntactic theory. Finally, § 4 will show some further empirical advantages of this syntactic treatment.

2 Two Semantic Analyses of Attributive Adjectives

2.1 The Montagovian Analysis

Within the framework of Montague Grammar, the assumption that attributive adjectives are second order functions is absolutely natural. Distributionally, an Adj–N group is identical to a common noun in isolation. If the syntactic category of common nouns is CN, then, the natural analysis of attributive adjectives is as being CN/CN elements. On the assumption that common nouns are of type 〈e, t〉, then, attributive adjectives are of type 〈〈e, t〉, 〈e, t〉〉, their identity type reflecting their syntactic transparency. Indeed, modulo intensionality, this is exactly the proposal of Lewis 1972, Montague 1974 and Kamp 1975. Kamp defines several subclasses of attributive adjective on the basis of the relations that hold between [[Adj]], [[N]]and [Adj]. For example, an adjective F is intersective^2 if:

(1) ‘there is a property Q such that for each property P and each w ∈ W [the set of possible worlds], F(P)(w) = P(w) ∩ Q(w)’ (Kamp 1975:124);

and an adjective is affirmative if:

(2) ‘For each P and w, F(P)(w) ⊆ P(w)’ (Kamp 1975:125).

Crucially, though, these properties of classes of adjectives are irrelevant to their combinatorial properties: they are meaning postulates, or more specifically, state- ments about our knowledge of the relation between elements in the domain and the range of a given adjectival function.

2.2 On Semantics and All that Followed

The Montagovian treatment of attributive adjectives posits a uniform method of syntactic and semantic combination whereby adjectives are functions from proper- ty-denoting common nouns into property-denoting common nouns. The different entailment relations that different attributive adjectives give rise to must then be treated as meaning postulates, as in (1–2), concerning the relation between the domain and range of a given adjective.

(^2) I substitute the more current term intersective for Kamp’s predicative , which is used in a distinct sense below.

Adjectives and Headedness 3

An alternative view was proposed by Higginbotham 1985, arguing for an ex- plicit representation of these relations in the syntactic and semantic derivation. Higginbotham assumes that an attributive adjective, as well as a noun and an Adj– N group, denotes a property. Since functor–argument relations between two 〈e, t〉 elements as they stand are impossible, as neither element is of the correct type to serve as argument to the other, the immediate question is one of how the adjective and noun are to combine semantically. Higginbotham’s solution is to shift the relationship between adjective and noun into θ-theory. He assumes that his proposal that adjectives and nouns denote first- order functions translates into such elements bearing θ-roles, which must be dis- charged. There are two mechanisms by which discharge of adjectival θ-roles is achieved, namely θ -identification and autonymous θ -marking. These correspond, roughly, to Kamp’s properties of affirmativity and non-intersectivity respectively, but here, they are structurally encoded. Diagrammatically, these operations are represented by Higginbotham as follows:

(3) a. θ-identification:

(N’,< 1 >)

qqqqqqq MM

MMM MM

(A,<1,2>) (N,< 1 >)

b. Autonymous θ-marking:

(N’,<1,2*>)

qqqqqqq MM

MMM MM

(A,< 1 >)

(N,< 1 >)

(Higginbotham 1985:564–567)

We can then define three classes of adjective, according to the thematic relations into which a given adjective enters:

(4) a. An adjective whose θ-role is discharged through θ-identification is affir- mative and intersective.

b. An adjective whose θ-role is discharged through autonymous θ-marking is non-affirmative and non-intersective.

c. An adjective with two θ-roles, one discharged through θ-identification and one through autonymous θ-marking, is affirmative and non-intersec- tive.

Adjectives and Headedness 5

Clearly, the red in these two examples is not identical, as a red face and a red double-decker bus are probably not the same colour. This means that even a prototypically intersective adjective, such as red , does not simply denote a property in (6), but is relativised to the noun that it modifies. In other words, even adjectives such as red are not absolutely intersective, in the sense of (1). This casts some doubt on the existence of (5a) as a separate class of adjectives. It may be argued, though, again following Higginbotham 1985, that a red face is still a face , and a red double-decker bus is still a double-decker bus. In the above examples, this is undoubtedly true. However, consider the following sentences, all of which I would claim are true:

(7) a. A corn marigold is not an ox-eye daisy. [They have separate Latin names, for example]

b. A corn marigold is (just) a yellow ox-eye daisy.^5

c. Therefore, a yellow ox-eye daisy is not (necessarily) an ox-eye daisy.

(8) a. A marrow is not a courgette. [They are sold as separate vegetables by greengrocers, for example]

b. A marrow is (just) a big courgette.

c. Therefore, a big courgette is not (necessarily) a courgette.

(9) a. A Tonka truck is not a truck. [This will become evident if you attempt to drive one on a public highway]

b. A Tonka truck is a toy truck.

c. Therefore, a toy truck is not (necessarily) a truck.

In each case, the pattern is the same. The (a) sentences assert that the two noun phrases denote distinct properties, while the (b) sentences assert that one particular property, denoted by the adjective, distinguishes the properties denoted by the two noun phrases^6. This allows us to deduce, as in the (c) sentences, that the inference from Adj–N to N is not automatic. However, only (9) involves an adjective which would standardly be considered as non-affirmative, where affirmative is defined as in (2). (7–8) involve canonical

(^5) I have no explanation for why it should be that just facilitates interpretation of the (b) examples. (^6) There is a sense in which the (a) and (b) sentences are true at different levels of granularity. It is only at a fairly superficial level that the equation “ox-eye daisy + yellow = corn marigold” could be taken to be true. That is beside the point, here, touching on questions of how an audience evaluates the truth of such utterances. The fact that they could all be taken to be true, under quite unremarkable circumstances, already has serious implications for the notion that attributive adjectives can denote conjoined properties.

6 Robert Truswell

intersective and subsective adjectives, respectively.^7 Under a theory of attributive adjectives such as Higginbotham’s, the failure of the inference that “an Adj N is an N” is unexpected in such cases, as such inferences are inseparably linked, through θ-theory, to the fundamental mode of structural combination of such adjectives with nouns. However, if the combination of adjective and noun is independent of such entailment relations, which are instead represented as meaning postulates, as in the Montagovian theory, then their defeasibility in contexts such as (7–9) has less drastic theoretical consequences. Such examples must be taken, then, to strongly favour the uniform analysis of attributive adjectives as second-order func- tions, and to provide evidence against θ-identification as the mode of combination of attributive adjectives.

2.4 Relating Attributive and Predicative Uses of Adjectives

A consequence of Higginbotham’s claim that adjectives canonically denote properties is that we might then expect a primary function of theirs to be forming propositions by taking type e arguments. Indeed, the evidence from copular and predicative constructions suggests that we should preserve this intuition, as Partee 1987 demonstrates that adjectives in such constructions alternate with indefinite DPs, and can be co-ordinated with them:

(10) Mary considers John competent in semantics and an authority on unicorns. (Partee 1987:119)

On the other hand, the evidence reviewed above from attributive uses suggests that adjectives can be second-order functions, of type 〈〈e, t〉, 〈e, t〉〉. If this is the case, we may assume that an operator is available of type 〈〈e, t〉, 〈〈e, t〉, 〈e, t〉〉〉, that is, an operator which takes a first-order property as its argument and outputs a second-order function from properties to properties, shifting from the predicative to attributive use of an adjective. Postulation of such an operator permits explanation of several facts concern- ing the distribution of adjectives. For example, it allows us to suggest a semantic reformulation of the distinction proposed in Bernstein 1993 and Alexiadou 2001 between a class of adjectives that are X^0 s and a class that are XPs: in Bernstein’s analysis, the A^0 s are distinguished by never occurring in predicative constructions and by not taking modifiers such as very , for example:

(11) a. ‘That car is big’

b. ‘A very big car’

c. * ‘That car is former’ (^7) A subsective adjective is here defined as a non-intersective, affirmative adjective, that is, the class (5c).

8 Robert Truswell

This representation captures several essential facts about attributive adjectives: the modification is potentially recursive because the type of the noun is unchanged by merging an adjective; the modification is optional because what is usually es- sential for a DP is that it provides an argument to the clause which contains it, and this is neither helped nor hindered by merging an adjective because there is nothing in the nominal semantics which requires the merging of an adjective (the require- ment comes instead from JOIN, which requires two 〈e, t〉 elements as arguments); and scope is represented because the most recently merged adjective directly mod- ifies the group consisting of the noun and any adjectives merged earlier, and this group is basically semantically opaque (i.e. an Adj–N group behaves exactly like a bare noun as far as any further computation is concerned). It is important to note that postulation of JOIN means that, even in cases such as English where the phonetic forms of adjectives used predicatively and attribut- ively are identical, they make distinct contributions to the semantic representation. In particular, a predicative adjective is first-order and an attributive adjective is second-order. Note also that JOIN must be formulated in such a way as to take the adjective as its first argument, in order to capture the data in (11). A possible alternative formulation of JOIN which took the adjectival and nominal arguments in the opposite order would necessarily assume a uniform type for adjectives, rul- ing out the possibility that certain adjectives could be lexically specified as type 〈〈e, t〉, 〈e, t〉〉. These considerations will have repercussions when we turn our at- tention to syntax in the following section.

3 English Attributive Adjectives are Heads

We have seen that attributive adjectives denote functions from properties to properties, and that they are derived from first-order properties by means of the two-place relation JOIN. However, not every property constitutes an acceptable input to such a relation. For instance, we also saw in §2.4 above that there is reason to believe that predicative adjectives denote properties, yet an attributive adjective cannot modify a predicative adjective:

(14) a. * The car was big red.

b. (JOIN(big′))(red′) = λy.(bigAttr.(red′))(y)

Furthermore, JOIN must only be able to take an adjectival property as its first argument, otherwise the following should be deriveable, by JOIN taking two nom- inal arguments:

(15) a. * The car vehicle raced down the street.

b. (JOIN(car′))(vehicle′) = λy.(carAttr.(vehicle′))(y)

Adjectives and Headedness 9

As the only elements modified by attributive adjectives are nominals (that is, a noun with any complements, possibly modified by other, lower, adjuncts), and as an attributive adjective with no nominal sister is marked, if not ungrammatical (compare null nominal constructions such as (16)), it appears that an attributive adjective selects an N projection as its sister.^9

(16) # The red is good.

Now, under a standard conception of attributive adjectives as adjuncts, the nom- inal projects, by definition. For instance, Chomsky 2000 writes that:

‘Adjunction has an inherent asymmetry: X is adjoined to Y. Exploit- ing that property, let us take the distinction between substitution and adjunction to be the (minimal) distinction between the set {α, β} and the ordered pair 〈α, β〉, α adjoined to β... [In the latter case,] Given the asymmetry, it is natural to conclude that the adjoined element α leaves the category type unchanged: the target β projects.’ (Chomsky 2000:133)

Leaving aside the question of whether JOIN has a syntactic reality, or instead perhaps describes a lexical redundancy rule, this would give a tree for an Adj–N constituent as follows:

(17) (^) car

qqqqqqq MM

MMM MM

redAttr. car

With regard to projection in cases of substitution, Chomsky writes:

‘Set-Merge typically has an inherent asymmetry. When α, β merge, it is to satisfy (selectional) requirements of one (the selector ) but not both. Fairly generally, furthermore, the selector is uniquely deter- mined for a pair (α, β)... In this case too, then... the label of the se- lector projects.’ (Chomsky 2000:133–4)

These two statements, however, should give rise to doubts concerning the va- lidity of the substitution–adjunction distinction, formulated in this way. It is the in- herent asymmetry in adjunction which leads Chomsky to propose that adjunction is a pair-forming operation. However, there is an inherent asymmetry in substitution too. It is not clear why these two asymmetries should be represented in different ways (ordering of elements and projection in the case of adjunction, as opposed

(^9) Note that such a suggestion is only plausible if attributive and predicative uses of adjectives are syntactically distinct. Predicative adjectives do not subcategorise for a sister at all, instead standardly occurring as the sister of a predicative head.

Adjectives and Headedness 11

This is surprising, given the evidence that the adjective is the head of an Adj–N constituent. This phenomenon, whereby syntactic characteristics are inherited from the non-head, is described extensively in Williams 1994, where the non-head is called a relativised head. Williams’ proposal is as follows:

‘Suppose first that there is an “absolute” head. Then, “head with re- spect to F” is defined as follows:

(45) X is the head with respect to F of Y if X is marked for a value of F, and either X is the absolute head of Y, or the absolute head of Y is not marked for F.

This provides a sort of unification that always succeeds, for in cases of conflict the head wins.’ (Williams 1994:46)

Assume that the criterion of identity is encoded by some feature. As this is essentially a variant of the long tradition of describing lexical categories in terms of categorial features, let us call the feature [N]. Then nouns have this feature, and adjectives don’t, as a direct consequence of Baker’s theory. Furthermore, the [N] feature on a nominal complement can only be inherited by a head if that head is not verbal, as no category can be both nominal and verbal on Baker’s theory. In that case, in the construction [AP A [NP N]], A can inherit N’s [N] feature, as A is defined as a lexical category lacking the defining characteristics of both nouns and verbs. This contrasts with cases where, for example, a verb has a nominal complement. As [V] and [N] features are incompatible, there is a conflict, and the absolute head alone determines the category of [VP V DP]. We must note, however, that it is possible to subcategorise for an adjectival projection, as in the case of too , for example, which only occurs modifying ad- jectival and adverbial projections. This means that an adjective cannot simply be analysed as a lexical category without [N] and [V] features, as, in that case, there will be nothing for too to subcategorise for. However, there is evidence that it is only possible to subcategorise for predicative adjectives in this way: the sequence D– too –A–N is at least marginal in English, and adjectives argued above to be lex- ically specified as type 〈〈e, t〉, 〈e, t〉〉 are unable to be modified by too :

(20) a. ?? A too happy man

b. * Too former / too solar

This constitutes further evidence that the syntactic nature of predicative and attributive adjectives is radically different: to the best of my knowledge, there is no clear evidence that anything subcategorises for a specifically attributive adjective in English. The categorial status of an attributive adjective with respect to the noun that it modifies is reminiscent, then, of the status of functional heads: although

12 Robert Truswell

they head their own projections, these still form part of the extended projection (in the sense of Grimshaw 2003) of the lexical item most locally c-commanded by the functional head. This paper claims that attributive adjectives are heads in the nominal extended projection, even if predicative adjectives are associated with some distinct categorial feature, which permits elements such as too and very to subcategorise for them. The theory of projection and category membership sketched here amounts to saying that, no matter which daughter projects, if the mother can function as a criterion of identity, then it will have the distribution of a noun phrase. And only nominal extended projections (by definition) and projections of attributive adjec- tives (where the ability to function as an expression of sortality is inherited from the nominal complement, there being no conflict between an attributive adjective and this nominal characteristic) can function in this way. At least in English, one welcome immediate consequence of such an analysis is that the general Adj–N order then coincides with the general head–complement order. However, DP-internal adjectives are not always pre-nominal in English: no- tably, adjectives with PP complements occur post-nominally, and the same applies to occurrences within the noun phrase of adjectives modified by too :

(21) a. A proud man

b. * A man proud

c. * A proud of his children man

d. A man proud of his children

(22) a. * A too proud man

b. A man too proud

Interestingly, Williams himself assumes that certain adjectives, notably alleged , stand in a head–complement relation to the nouns they modify. One reason for sin- gling out alleged for such an analysis is that it has distinct semantic characteristics when in post-nominal position. While an alleged murderer is not necessarily a murderer, an instance of murderer modified by post-nominal alleged necessarily does denote a murderer:

(23) a. * The murderer alleged since yesterday...

b. The murderer alleged to have stolen the car... (Williams 1994:92)

However, given the discussion in §2.3, we may now expect to find that such patterns hold more generally. And it seems that this is true: the following ex- amples must denote ox-eye daisies and courgettes, respectively, in contrast to the pre-nominal cases: