• Keine Ergebnisse gefunden

From Subject-to-Object Raising to Small-Clause Theory

5. Challenging small-clause analyses of secondary-predicate constructions constructions

5.1 From Subject-to-Object Raising to Small-Clause Theory

Generative syntax by and large neglected the [NP1 V NP2 XP]-pattern up to the 1980s (Aarts 1992: 9). The focus of the (Extended) Standard Theory was rather on the question of how the infinitival complement in Mary believes John to be foolish is related to the finite variant in Mary believes that John is foolish (Postal 1974: 4). Picking up an analysis originally sug-gested by Rosenbaum (1967), Postal and his followers advanced the proposition that such finite and infinitival clauses are identical underlyingly, but that, in the non-finite pattern, the deep-structure subject has been raised to the matrix object position at surface structure, with a concomitant change of the verb form from finite to infinitival (56) (Perlmutter and Soames 1979: 79, 81-3; Postal 1974: 40-1).

(56) S

NP1 VP

V S

NP2 VP

Mary believes John is foolish → to be foolish

This 'Subject-to-Object Raising' analysis tries to capture the characteristic facts also noticed in descriptive grammars (cf. 4.2) that the NP2 seems to function as the semantic subject of the infinitive but has the grammatical qualities of a direct object. The derivation sketched in (56), which was assumed to be triggered by verbs such as believe and consider (Postal 1974: 297-8), was regularly contrasted with the so-called 'Equi-NP-Deletion'-analysis suggested for verbs such as persuade and urge, which were taken to bear both a syntactic and a logical relation to their objects. In a sentence such as Mary forced John to leave, a second occurrence of the NP

John was posited in the subject position of the infinitival clause, which was thought to have been deleted at surface structure because it is identical with the object of the main verb forced (Borkin 1984: 4).

The Raising-analysis of (56) was extended from infinitival clauses to [NP2 as XP ]-complements as well: [NP2 as XP] was argued to form a clause at deep structure but to have been broken up at surface structure, where NP2functions as the derived object of the matrix verb in the same way as in the infinitival pattern (Postal 1974: 240-1). Like some descriptive

grammarians (cf. 4.3), Postal analysed as in She regards him as a fool as governing a verbal –ing-complement (as being a fool) underlyingly, with the copula having been deleted in the derived structure (1974: 240; cf. also Borkin 1984: 5-6; Hantson 1989: 212-3). The sugges-tion that a clause underlies the NP2 XP-pattern as well (cf. Menzel 1973: 16-7; Stockwell, Schachter and Partee 1973: 578) did not figure prominently in the Raising debate, though.

It was not before the framework of GB-theory was established (Chomsky 1981) that the [NP1 V NP2 XP]-structure captured the attention of generative linguists (Aarts 1992: 12) within the new concept of 'Small-Clause theory'. This concept posed a challenge to the Raising analysis because it tried to accommodate the phenomenon in a different way. NP2(to be/ as)

XP-strings were now analysed as clauses at all levels of derivation, including surface structure.

The deconstruction of the traditional Raising analysis had already begun ten years earlier, however, when Chomsky claimed that finite and infinitival structures form clausal constitu-ents both underlyingly and in the derived structure, i.e. that NP2 in the infinitival variant has not been moved to the object position of the superordinate clause (1972: 6-7). Chomsky's original criticisms of Subject-to-Object Raising were not always well-grounded and are largely obsolete today, but the spirit of his suggestion has nevertheless caught on. Small-Clause-theoreticians have pressed the matter even further, arguing that not only finite that-clauses and infinitival that-clauses, but verbless NP2 XP-strings as well should be treated as clauses at all levels of derivation, and that the three complement structures differ only in their internal form (finite verb, non-finite verb, no verb). The verbless 'NP2 XP'-string, or 'small clause' (SC), has assumed a determining role in GB-theory because the subject/predicate relationship in-stantiated by the NP2 XP-string is thought to be a reflection of its clausal status even in the ab-sence of an ordinary verb phrase (Aarts 1992: 21; Basilico 2003: 1; Stowell 1995: 272).39 The square brackets around Mary intelligent in (57a) indicate that an SC is analysed as one clausal constituent occupying the direct-object position in the matrix sentence on a par with finite and non-finite clauses (57b, c).

(57) a. John considers [SC Mary intelligent].

b. John believes [that Mary is intelligent].

c. John believes [Mary to be intelligent].

What appears to be a completely unprecedented and unorthodox proposal had one notable predecessor in the first half of the 20th century: the Dane Otto Jespersen. Jespersen draws a

39 The term 'SC' is also frequently applied to structures where XP is an uninflected verb such as a bare infinitive, an –ing-form or a past participle (e.g. Mary made John leave the house; The waitress saw him drinking his coffee; Mary got Fred invited to her party) (e.g. Haegeman and Guéron 1999: 111-2). Since these verbal SCs differ crucially from non-verbal SCs both syntactically and semantically (Aarts 1992: 189; Basilico 2003: 2-3), they do not fall within the purview of this dissertation.

distinction between what he calls a 'junction' and a 'nexus'; the former structure expresses a single idea through the combination of a core semantic element and an adjunct enriching its description (e.g. a beautiful rose), while the latter construction contains two ideas that stand in a subject/predicate relationship with each other (e.g. the rose is beautiful): "A junction is like a picture, a nexus like a process or a drama" (Jespersen 1924: 116). Jespersen then goes on to distinguish between nexus structures which occur independently — essentially simple sentences — and those which cannot occur on their own (1940: 6). Such 'dependent nexus structures' include finite subordinate clauses (e.g. I think that the rose is beautiful), non-finite structures (e.g. I like roses to be beautiful), and, Jespersen claims, predicative NP2 XP-strings (e.g. John considers the rose beautiful) (1924: 117-23, 1940: 7-8). Jespersen provides the following abstract functional representations of nexus-objects, where 'S2' stands for the second subject in the sentence and 'P' for the predicate of the dependent nexus structure. Like modern descriptive grammars, Jespersen differentiates between current objects (58a) and resultant objects, which he represents as 'Or' (58b) (1924: 122, 1940: 8). (58c) and (58d) are the de-scriptions of the functionally equivalent infinitival and finite clauses, respectively, and (58e) the corresponding structure with the 'particle' (p) as before the secondary predicate (1932:

375, 1940: 12). The four structures (S2 P), (S2 I P), (S2 V2 P) and (S2 p P) are consequently treated by Jespersen as functionally equivalent expressions that only differ in the formal reali-sation of the dependent subject/predicate relationship (1924: 117-23).

(58) a. They considered Tom a happy man: S V O(S2 P)

b. They elected Tom President: S V Or(S2 P)

c. I considered Tom to be a happy man: S V O(S2 I P) d. I believed that Tom was a happy man: S V O(S2V2P) e. He regarded Tom as a happy man: S V O(S2 p P)

The idea that the direct objects in (58a, b) and (58e) are realised by subordinate verbless clauses, which establish a predicative link of their own, was and still is anything but conven-tional. Yet Jespersen went one step further still in bucking traditional analyses: he argues that if in a sentence such as They made him President the postverbal elements form a nexus-object, then the corresponding passive sentence He was made President must contain a dis-continuous nexus-subject, namely "He...President" (1924: 123, 1933: 312).40 SC-theory has adopted Jespersen's verbless dependent nexus structures in essence if not detail. While the

40 Not everyone of Jespersen's followers was prepared to go all the way with him. The German linguist Ham-merich, who on the whole espoused Jespersen's account of dependent nexus structures, was a bit more cautious about the analysis of passive sentences. While he admits that in a sentence like "der Sohn wurde Johannes ge-tauft" the logical subject should be the nexus "der Sohn Johannes", he feels urged to concede: "Aber das Sprachbewusstsein fasst den Zusammenhang sicher anders auf, nämlich als der Sohn (S) wurde Johannes getauft (P)" (1930: 307).

treatment of active sentences is almost identical, the modern distinction between deep and surface structure made it possible to replace Jespersen's clumsy notion of discontinuous nexus subjects in passive sentences with the more elegant analysis of (59). The NP He has been raised from the subject position in the SC to be the subject of the main clause, leaving behind a trace t. The moved constituent He and the trace are co-indexed (cf. Aarts 1992: 21).

(59) Heiwas made [SCtipresident].

Jespersen's suggestion to analyse NP2 XP-strings as clauses both semantically and syntactically had lain dormant for almost 50 years, when SC-theory arrived at similar conclusions within a completely different syntactic framework. Stowell (1981, 1983) was the first to revisit the possibility that NP2 XP-strings constitute verbless clauses at all syntactic levels, and has thereby touched off one of the liveliest discussions in generative grammar to date. The GB-framework offered a fresh conceptual basis for this proposal because many of its principles barred the descriptive complex-transitive analysis or a hybrid analysis such as the Raising proposal, where a clausal constituent is assumed to be present at D-structure and two separate constituents are postulated for S-structure. To keep our discussion within bounds, I will out-line the basic ideas behind these GB-principles only.

The lexical entry of a verb is generally thought to spell out the number and nature of the arguments selected by that verb (Grewendorf 2002: 18-9; Haegeman and Guéron 1999: 25).

According to the Theta-Criterion, one of the central principles of GB-theory, each thematic or theta role (θ-role) associated with a verb must be realised by one and only one argument in the syntax, and each argument may be assigned one and only one θ-role (Aarts 1992: 19;

Haegeman and Guéron 1999: 28). This "biunique relation between arguments and θ-roles"

(Haegeman and Guéron 1999: 30) prohibits the complex-transitive analysis, in which NP2 is both the direct object of the main verb and the subject of the predicative complement because it would be assigned two θ-roles. The Raising analysis would not violate this principle be-cause NP2 is thought by Postal and his followers to be assigned a thematic role only by the subordinate predicate (XP) and not to bear a thematic relation with the matrix verb.

As a result of the Theta-Criterion, it must be argued by proponents of GB-theory that a matrix verb such as consider does not select a direct-object NP, but assigns a propositional θ-role to the whole string [John foolish] (Chomsky 1986a: 90-1; Kaplan 1988: 78). This seman-tic claim follows mainly from the uniqueness requirement of the Theta-Criterion, whereas the following independent semantic arguments given in the literature are rather feeble:

(i) Rafel surmises that the postverbal string must form a semantic clause because the sub-ject/predicate relationship between NP2 and XP "can be independently established within a

copular sentence" (2002: 475). As has already been noted in 4.1.3, the copula sentence John is a fool is not necessarily entailed by Mary considers John a fool, so there is no sound semantic basis for arguing that it must be assumed to underlie the latter structure.

(ii) Proponents of SC-theory frequently mention the semantic affinity between the NP2 XP-string and infinitival or finite subordinate clauses and claim that these are alternative for-mal realisations of the same semantic unit (e.g. Aarts 1992: 70; Rothstein 1995: 32; Staud-inger 1996: 28). Since consider selects a subordinate clause in Mary considers that John is a fool and in Mary considers John to be a fool, the "truth-conditionally equivalent" (Rothstein 1995: 32) or "interpretive similar[.]" (Haegeman and Guéron 1999: 110) structure Mary con-siders John a fool is argued to contain a (verbless) subordinate clause as well (Chomsky 1981: 106; Haegeman and Guéron 1999: 109; Rothstein 1995: 29). Hoekstra goes on to claim that "[t]he minimal hypothesis would be that the same thematic roles are assigned ..., the sin-gle difference being that the internal role is now borne by a small clause instead of a full clause" in the NP2 XP-pattern (1992: 147; similarly, Staudinger 1996: 123).41 Two objections must be raised against this 'minimal hypothesis', though. First of all, the three structures claimed to be paraphrases of one another are not synonymous, and one cannot felicitously be exchanged for the other in most cases. If I consider John a fool and I find John a fool are sim-ply paraphrased by I consider that John is a fool and I find that John is a fool, important meaning distinctions have to a considerable degree fallen by the wayside (see chapter 11).

Moreover, infinitival or finite variants are not available for resultative and depictive (and even some qualifying) structures at all (cf. She shot John dead; She served the coffee hot; They re-gard him as a genius). My second objection is related to the first one and concerns the argu-ment that NP2 and XP must form a clause because there is a clause in the similar finite and in-finitival structures as well. This sort of reasoning is dubious and has already been criticised by Marty as "Fehler der Substitution eines verwandten Satzes an Stelle der unbefangenen Inter-pretation des gegebenen" (in Eisenmeier, Kastil and Kraus 1918: 290). From a Construction-grammar perspective, syntactically distinct constructions cannot simply be claimed to be

41 Chomsky (1981: 38) and other generative grammarians argue that the material contained in lexical entries can be restricted to semantic information, so that the categorial realisation of an argument ideally follows from its semantics alone. Verbs that take a propositional argument will have this argument realised as a clause, i.e. as a structurally defined subject/predicate relationship (Rothstein 1992: 119-21). When different formal realisations of an argument are possible, as is the case with the finite, infinitival and small-clause variants of the proposi-tional argument, only the so-called 'Canonical Structural Realisation' (CSR) automatically follows from the meaning of the argument, while alternative syntactic realisations must be specified in the lexical entry of a verb as marked options (Chomsky 1986a: 87-90; Kaplan 1988: 86; Levin and Rappaport-Hovav 1992: 249-50). Since the CSR of the 'propositional argument' is usually taken to be a full clause, infinitival clauses and particularly SCs must be explicitly stated in the lexical entries of the respective verbs as marked encodings of the proposi-tional θ-role (Aarts 1992: 70; Kaplan 1988: 79, 85-6).

mantically identical; to speak with Marty: "Derselbe Tatbestand kann Anlaß zu einer Mehr-heit inhaltlich verwandter Urteile ... geben, und auch wenn mit dem Gegebensein des einen ein anderes ... nahe liegt, so ist damit gar nicht gesagt, daß sie identisch seien" (in Eisenmeier, Kastil and Kraus 1918: 292).

(iii) If consider selects a propositional argument, it cannot assign a θ-role to the NP2, so this noun phrase must be shown to be not a semantic argument of the main verb. To prove this contention, it is claimed that in a sentence such as Mary considers John a fool, Mary does not consider 'John' but the proposition 'that John is a fool' (cf. Aarts 1992: 22; Chomsky 1986a:

91). Likewise, since a sentence such as I consider the problem difficult does not entail I con-sider the problem, the noun phrase the problem is thought to be an argument of the main verb in the latter, but not the former structure (Rafel 2001: 476; Rothstein 1995: 31, 2000: 243).

Yet the entailment relation between the two sentences with consider is only permissible if we suppose that we are dealing with the same meaning of the verb. However, it seems more rea-sonable to argue that consider in He's considering the problem is a dynamic verb denoting a thinking process, while in He considers the problem difficult it has a static meaning, express-ing a subjectively stored opinion (Schneider 1988b: 59-70, 346-50). As a corollary of this, it can be argued that the dynamic sememe is bivalent, whereas the static one is trivalent, so that an entailment relation cannot be felicitously construed. In a Construction-grammar frame-work, consider would be argued to fill the roles in two semantically different syntactic con-structions, and it is not legitimate to explain its behaviour in one construction with that in the other.

Although the claim that verbs such as consider select a propositional argument is no-where meticulously argued for and is not based on reliable semantic tests, the bivalent nature of such verbs is generally thought to be sufficiently established by the provisions of the Theta-Criterion.

Once the lexical requirements of a verb are determined, it must be examined how they are projected into syntax. The D-structure format of representation constitutes an interface between the lexicon and syntax because the syntactic configurations at that level directly mir-ror the thematic relations between a verb and its arguments (Grewendorf 2002: 15). The D-structure representation would parse our example sentence as [Mary] considers [John a fool].

In earlier versions of generative grammar, the derivational component could raise the NP John to the direct-object position of the main verb, so that consider would have two postverbal complements at surface structure: [Mary] considers [Johni] [ti a fool]. Yet the introduction of the Projection Principle in Chomsky 1981 (32-3) required a tight correspondence between the

argument structure as determined by the Theta-Criterion and syntactic representation: infor-mation given in the argument structure of a verb must be represented categorically at each syntactic level, i.e. not only at D-structure, but at S-structure (and LF) as well (Grewendorf 2002: 20; Kaplan 1988: 78). This means that if a verb selects a propositional argument, the verb must have a propositional argument at all levels of grammar; similarly, if the verb does not subcategorise for a direct-object argument, it cannot have a direct-object argument at any level, contra the Raising account (Aarts 1992: 23; Cook and Newson 1996: 166; Hoekstra 1988: 106). The SC-analysis is in keeping with the strict requirements of the Projection Prin-ciple because a verb such as consider assigns a θ-role to a proposition and this proposition remains intact as a syntactic constituent at S-structure. NP2 is not moved to a direct-object po-sition, which is not subcategorised for by the verb and cannot be created at S-structure only.

[Mary] considers [John a fool] is thus both the D-structure and the S-structure representation of the sentence (Chomsky 1981: 32-3, 1986a: 90-1).

One final reason for favouring SC-analyses in GB-theory lies in the requirements of the categorial component of grammar (for more details see 5.3). Both the complex-transitive analysis and the Raising analysis (at surface structure) assume that a verb such as consider has two phrases as sisters:

(60) V'

V NP XP/VP

consider John a fool/ to be a fool

Such ternary branching is prohibited by more recent developments in the theory of the cate-gorial component, however. The new postulate that branching must be binary is, like the Theta-Criterion, based on theoretical rather than empirical arguments and is explained slightly differently from author to author. Kayne, an early proponent of binary branching, argues that there is an 'unambiguous-path' requirement on government, which is violated by the configu-ration in (60). NP and XP/VP must both be governed by the verb in this structure; yet, Kayne claims, there is no unambiguous path from either NP or XP/VP to the governor V because if one traced the path from, say, NP, one would have to make a choice between the branch leading to

V and the branch leading to XP/VP. Since government is subject to the unambiguous-path re-quirement, the branch leading to XP/VP would be in violation of this injunction (Kayne 1984:

130-2). Kayne consequently argues for the alternative configuration (61), where V unambigu-ously governs the subordinate clause and XP/VP unambiguously governs its subject NP (1984:

133-5).

(61) V'

V S

NP XP

consider John a fool/ to be a fool

Whatever the particular theoretical reasons offered for the binary-branching requirement in GB-theory or the MP, it has become the consensus opinion in generative syntax (e.g. Wilder 1994: 220) and has conspired with the Theta-Criterion and the Projection Principle to disal-low the Raising analysis, or, for that matter, the complex-transitive complementation analysis, and to favour the SC-analysis.42

The treatment of NP2XP-strings as SCs has become the standard view in generative text-books43 (e.g. Haegeman and Guéron 1999: 109), and some syntacticians have taken up the notion so enthusiastically that they apply SC-analyses "to almost every other construction in one form or another" (Stowell 1995: 271; cf. also Moro 1995: 109). Yet despite the volumi-nous literature on SCs published in the last two decades, most of the details of the SC-analysis still command no consensus (Aarts 1992: 21; Basilico 2003: 1; Ogawa 1994: 447; Suzuki 1991: 27), and the Raising analysis surfaces in the literature from time to time despite its hav-ing behav-ing banished from mainstream generative grammar as a "heresy" (Koster 1984: 445) — a fact that is frequently lamented by Raising proponents (cf. Hantson 1989: 207; Koster 1984:

The treatment of NP2XP-strings as SCs has become the standard view in generative text-books43 (e.g. Haegeman and Guéron 1999: 109), and some syntacticians have taken up the notion so enthusiastically that they apply SC-analyses "to almost every other construction in one form or another" (Stowell 1995: 271; cf. also Moro 1995: 109). Yet despite the volumi-nous literature on SCs published in the last two decades, most of the details of the SC-analysis still command no consensus (Aarts 1992: 21; Basilico 2003: 1; Ogawa 1994: 447; Suzuki 1991: 27), and the Raising analysis surfaces in the literature from time to time despite its hav-ing behav-ing banished from mainstream generative grammar as a "heresy" (Koster 1984: 445) — a fact that is frequently lamented by Raising proponents (cf. Hantson 1989: 207; Koster 1984: