• Keine Ergebnisse gefunden

Relative Clause Extraposition as a Nonlocal Dependency

5.3 Relative Clause Extraposition

5.3.2 Relative Clause Extraposition as a Nonlocal Dependency

The most prevalent approach to relative clause extraposition is the analysis in terms of a nonlocal dependency developed by Keller (1994) and Müller (1999) for German, with an extension to English in Keller (1995). An adaptation to Dutch is given by Bouma (1996), and a variant for German is proposed by Penka (2000). Rather than summarizing every single analysis proposed and discussing the differences between them, I will present the basic concepts that all of these analyses treating extraposition as a nonlocal dependency have (more or less) in common.

It should be noted that the theories do not provide separate analyses for relative clause extraposition, but subsume both adjunct (including relative clauses and PP adjuncts) and complement extraposition under a general analysis of extraposition, implementing relevant differences where necessary (for example for the introduction of the nonlocal dependencies).

In the discussion to follow, I will only present the details relevant to relative clause extrapo-sition and ignore the fact that some (but not all) of the mechanisms introduced also apply to extraposition of the other types of adjuncts and complements.

Since extraposition is different from movement to the left (e.g., it is clause-bounded and may violate island constraints), a new nonlocal featureEXTRAanalogous toSLASH is intro-duced to establish the connection between the extraposed relative clause and its antecedent.25

24But see Yatabe (2001, 340–341), who proposes a variant of Kathol and Pollard’s linearization-based theory incorporating a theory of semantic composition that is claimed to provide an explanation of Guéron’s (1980) observation that extraposition may affect semantic scope relations.

25Gazdar (1981) develops a GPSG analysis for relative clause extraposition that uses the same SLASH

-CHAPTER 5. RELATIVE CLAUSES IN HPSG 116 The value ofEXTRAis a set of elements of type local.26 TheEXTRAset of a sign contains lo-cal information about all extraposed relative clauses whose antecedents are contained within the sign and which have not been bound off yet. The values are percolated in the same way as the other nonlocal values, i.e., by means of the Nonlocal Feature Principle (Pollard and Sag, 1994, 400), which must now comprise theEXTRAfeature as well (see Keller (1994, 21) for the new formulation of the principle).

The nonlocal dependency of an extraposed relative clause can be introduced into the

EXTRA set by means of a trace, a unary schema, or a lexical rule (Müller, 1999, 228–232, Keller, 1994, 22–31).27 The trace, which is equivalent to the trace assumed in Pollard and Sag (1994, 164) for movement to the left (see (188) above), is a phonologically empty ele-ment that specifies itsLOCAL value in itsEXTRA set (see Müller (1999, 229) for a lexical specification). However, the use of traces has been criticized in recent work in HPSG, and attempts are made to avoid invisible elements (see Pollard and Sag (1994, Ch. 9), Sag and Fodor (1994), Sag (1997), among others). For further problems concerning the use of traces to introduce nonlocal information in adjunct extraposition, see Keller (1994, 30–31).

Alternatively, one can use a unary schema that introduces the required information into the EXTRA set. Müller (1999, 229–230) provides an example of such a schema for com-plement extraposition. A second schema would be required for relative clause (adjunct) extraposition.

Finally, a lexical rule can be used that introduces the nonlocal dependency of an extra-posed relative clause into theNONLOCAL|INHERITED|EXTRA set of a noun. Such a rule is shown in (204) from Keller (1994, 23).28 The extraposed relative clause selects the modi-fied noun by identifying itsMOD|LOC value with theSYNSEM|LOC value of the noun. Note how the semantic contribution of the extraposed relative clause is incorporated into this lex-ical rule. TheCONTENT value of the output of the lexical rule is structure-shared with the

CONTENT value of the extraposed element (3). An (extraposed) relative clause forms its

CONTENT as described in Pollard and Sag (1994, Ch. 5): via its MOD value, it has access to the LOCAL|CONTENT value of the modified nominal; it identifies its INDEX value with that of the modified noun and unifies its set of restrictions with the set of restrictions of the modified nominal. Thus, the semantics of the output of the lexical rule corresponds to the semantics of the extraposed relative clause. This CONTENT value is then passed up to the mother in accordance with the Semantics Principle (Pollard and Sag, 1994, 56). Note that

mechanism as employed for movement to the left.

26Müller (1999, 225n82) uses lists instead of sets as values of theEXTRAfeature. Bouma (1996) assumes that elements of theEXTRAset are of type synsem to account for the interaction of extraposition and fronting in Dutch; see Müller (1999, 237) for discussion.

27An alternative, which dispenses with traces and lexical rules, is proposed by Penka (2000) for extraposition in German. This constraint-based, head-driven analysis is based on and analogous to the traceless unified analysis of complement and adjunct extraction developed by Bouma et al. (2001). I will not provide the details here but refer the reader to the work cited.

28Keller (1995) provides a generalized version of this rule that also introduces PP adjuncts as nonlocal dependencies.

in order to allow for multiple extraposed relative clauses, the Relative Clause Extraposition Lexical Rule is a recursive lexical rule, which is rather unusual in standard HPSG.29

(204) Relative Clause Extraposition Lexical Rule (Keller, 1994, 23):30

"

SYNSEM

"

LOC 2|CAT|HEAD noun

NLOC|INHER|EXTRA 1

##

SYNSEM

LOC|CONT 3

NLOC|INHER|EXTRA 1

CAT

HEAD relativizerh

MOD|LOC 2

i

SUBCAT hi

CONT 3

The attachment of a relative clause in extraposed position is licensed by a new schema, which is similar to the head-filler schema and binds off the nonlocal dependency in the

EXTRA set. The schemata formulated in the different analyses mentioned above differ in some relevant details. Since most of the analyses are developed to capture generalizations of relative clause extraposition in German, which differ from the generalizations in English in relevant aspects, I will not provide the details here. For exemplification, I present the head-extra schema proposed by Keller (1995), whose analysis accounts for English as well as for German:31

(205) Head-Extra Schema (Keller, 1995, 4):

SYNSEM

"

LOC|PER extra

NLOC|INHER|EXTRA{}

#

DTRS

H-DTR|SYNSEM

LOC|PER right

NLOC

"

INHER|EXTRA loc(1)

TO-BIND|EXTRA loc(1)

#

EXTRA-DTRS 1

Keller introduces this new immediate dominance schema to build head-extra-structures, which bear the feature EXTRA-DTRS. The INHER|EXTRA value of the head daughter is

29It should be pointed out that Keller (1994) tries to subsume both complement and adjunct extraposition under a general analysis of extraposition. While the lexical rule in (204) can be easily amended to account for extraposition of PP adjuncts as well, as shown in Keller (1995), a separate lexical rule would be needed for complement extraposition. To avoid the postulation of several lexical rules, Keller (1994, 24–28) also proposes an alternative that makes it possible to introduce the nonlocal dependencies for both adjunct and complement extraposition in terms of a single lexical rule. This analysis presupposes that adjuncts appear inCOMPSlists, to which they are added by a lexical rule as proposed by Van Noord and Bouma (1994). Keller then formulates a lexical rule that removes an element from theCOMPSlist and introduces it into theEXTRAset. Cf. Bouma (1996) for a variant and discussion of this proposal.

30The lexical rule in (204) is taken directly from Keller (1994). In my own proposal, I use the symbol

“7→” between the input and the output description of a lexical rule. I use the symbol “⇒” for implicational constraints, following, for example, Ginzburg and Sag (2001).

31Recall that Keller’s analysis accounts for both adjunct extraposition, including PP adjuncts, and comple-ment extraposition. The head-extra schema applies to all of these cases.

CHAPTER 5. RELATIVE CLAUSES IN HPSG 118 bound off by identification with the local features of the extraposed relative clause(s) and by introducing it asTO-BIND|EXTRA of the head daughter.32 In accordance with the Nonlocal Feature Principle, this prevents theEXTRAfeature from continuing to be passed up the tree and hence being bound again higher in the tree. Note that the value of EXTRA-DTRS is a non-empty list of sign. Since the mother is specified as [INHER|EXTRA { }], all elements ofEXTRA have to be bound at the same level; hence, extraposed relative clauses originating from the same nominal are realized as syntactic sisters.

Instead of assuming a fixed landing site for extraposed constituents, Keller proposes that an extraposed relative clause is attached to a phrase that introduces material that intervenes between the extraposed element and its antecedent.33 This is to explain the interaction be-tween extraposition and fronting and to prevent spurious ambiguities. In order to account for this, Keller employs the notion of periphery, for which he introduces the featurePERIPHERY

(PER), located underLOCAL. A phrase with a non-emptyINHER|EXTRA set is marked [PER

left] if there is no material that could intervene between the extraposed element and its an-tecedent. Additionally, for English, all lexical entries are assumed to be marked [PER left].

Extraposed relative clauses can only be attached to phrases that are marked [PER right]. For that reason, the head daughter in the head-extra schema is specified as [PERright].34

To account for the upward-boundedness of relative clause extraposition, Keller (1995, 6) proposes that a sentence be specified as [INHER|EXTRA { }].

Analyzing relative clause extraposition as a nonlocal dependency is not without prob-lems. Although a relative clause is not actually moved to its extraposed position, it is related to its antecedent (or its canonical position) through structure sharing which, in effect, is a simulation of movement. For this reason, this analysis has the same difficulties as the right-ward movement theories formulated within Government and Binding Theory that I have reviewed in Chapter 4.1.

First of all, it cannot account for relative clauses with split antecedents. No matter whether the nonlocal dependency is introduced by a trace, a unary schema, or a lexical rule, it is always introduced by the modified noun (in the case of a trace, the trace must be combined with the modified noun) whose LOCAL value is identified by the element introduced into

32In (205), loc(x)is “a function which takes asxa list of sign and returns a set of loc containing theLOC

values of the elements ofx” (Keller, 1995, 4n8).

33This is reminiscent of the locality constraints formulated as generalizations by Akmajian (1975, 119), Asakawa (1979, 505), Baltin (2006, 241), and others, which state that an extraposed element is adjoined to the first maximal projection that dominates its antecedent. See the discussion of locality constraints on extrapo-sition in Section 2.2.1. Keller’s analysis was inspired by a similar locality constraint proposed by Wiltschko (1994).

34The mother of the head-extra schema is specified as [PERextra] in order to prevent adjuncts from being adjoined higher than extraposed relative clauses, as in (i):

(i) *An entirely new band rings today, [RC several of whom are members of the congregation] at Great Torrington.

Adjuncts are specified as [MOD|LOC|PER non-extra], so that they cannot be adjoined on top of a head-extra structure.

theEXTRA set (via the MOD feature). TheLOCAL value of this element is structure-shared with theLOCAL value of the extraposed relative clause. In the case of a relative clause with two (split) antecedents, it is not clear how the nonlocal dependency would be introduced, since neither of the two singular nominals can introduce a nonlocal dependency that can be structure-shared with a relative clause that requires a plural antecedent.

Furthermore, since the semantic contribution of the relative clause is incorporated into the lexical rule (or the unary schema or the trace) that introduces the nonlocal dependency, an extraposed relative clause receives the same “low” interpretation as its in-situ equivalent.

But we have seen evidence from binding and scope that extraposed relative clauses must be interpreted differently from in-situ relative clauses. These interpretive effects cannot be captured by this analysis.

Since the noun introduces the nonlocal dependency, it is not clear how to account for determiners with obligatory relative clauses.

Finally, Kiss (2005, 318) criticizes the theory for neglecting the fact that modifier extra-position is completely different from complement extraextra-position, for example with respect to island constraints. Since the same feature EXTRA is employed to account for both of these processes, it is unclear how to capture their different behavior.35

Kiss proposes an alternative theory of relative clause extraposition which does not rely on movement. I will present his theory in the following section.