Extraction

The discussion in this chapter covers constructions that are analyzed as having wh-movement in GB, in particular, wh-questions and topicalization. Relative clauses, which could also be considered extractions, are discussed in Chapter 14. Extraction involves a constituent appearing in a linear position to the left of the clause with which it is interpreted. One clause argument position is empty. For example, the position filled by frisbee in the declarative in sentence ((131)) is empty in sentence ((132)). The wh-item what in sentence ((132)) is of the same syntactic category as frisbee in sentence ((131)) and fills the same role with respect to the subcategorization.

(130)0(130

(131)

Clove caught a frisbee.  (131)0(131

(132)

What_i did Clove catch $\epsilon_{i}$? 

The English XTAG grammar represents the connection between the extracted element and the empty position with co-indexing (as does GB). The <trace> feature is used to implement the co-indexing. In extraction trees in XTAG, the `empty' position is filled with an $\epsilon$. The extracted item always appears in these trees as a sister to the S_r tree, with both dominated by a S_q root node. The S_r subtrees in extraction trees have the same structure as the declarative tree in the same tree family. The additional structure in extraction trees of the S_q and NP nodes roughly corresponds to the CP and Spec of CP positions in GB. All sentential trees with extracted components (this does not include relative clause trees) are marked <extracted>=+ at the top S node, while sentential trees with no extracted components are marked <extracted>=-. Items that take embedded sentences, such as nouns, verbs and some prepositions can place restrictions on whether the embedded sentence is allowed to be extracted or not. For instance, sentential subjects and sentential complements of nouns and prepositions are not allowed to be extracted, while certain verbs may allow extracted sentential complements and others may not (e.g. sentences ((133))-((136))).

(132)0(132

(133)

The jury wondered [who killed Nicole].  (133)0(133

(134)

The jury wondered [who Simpson killed].  (134)0(134

(135)

The jury thought [Simpson killed Nicole].  (135)0(135

(136)

$\ast$The jury thought [who did Simpson kill]? 

The <extracted> feature is also used to block embedded topicalization in infinitival complement clauses as exemplified in ((137)).

(136)0(136

(137)

* John wants [ Bill_i [PRO to see t_i]] 

Verbs such as want that take non-wh infinitival complements specify that the <extracted> feature of their complement clause (i.e. of the foot S node) is -. Clauses that involve topicalization have + as the value of their <extracted> feature (i.e. of the root S node). Sentences like ((137)) are thus ruled out.

   

Figure: Transitive tree with object extraction: $\alpha$W1nx0Vnx1

The tree that is used to derive the embedded sentence in ((135)) in the English XTAG grammar is shown in Figure 13.1^13.1. The important features of extracted trees are:

• The subtree that has S_r as its root is identical to the declarative tree or a non-extracted passive tree, except for having one NP position in the VP filled by $\epsilon$.
• The root S node is S_q, which dominates NP and S_r.
• The <trace> feature of the $\epsilon$ filled NP is co-indexed with the <trace> feature of the NP daughter of S_q.
• The <case> and <agr> features are passed from the empty NP to the extracted NP. This is particularly important for extractions from subject NP's, since <case> can continue to be assigned from the verb to the subject NP position, and from there be passed to the extracted NP.
• The <inv> feature of S_r is co-indexed to the <wh> feature of NP through the use of the <invlink> feature in order to force subject-auxiliary inversion where needed (see section 13.1 for more discussion of the <inv>/<wh> co-indexing and the use of these trees for topicalization).