28.10.2014 Views

Empirical Issues in Syntax and Semantics 9 (EISS 9 ... - CSSP - CNRS

Empirical Issues in Syntax and Semantics 9 (EISS 9 ... - CSSP - CNRS

Empirical Issues in Syntax and Semantics 9 (EISS 9 ... - CSSP - CNRS

SHOW MORE
SHOW LESS

You also want an ePaper? Increase the reach of your titles

YUMPU automatically turns print PDFs into web optimized ePapers that Google loves.

Class VSp<strong>in</strong>e<br />

syntactic dimension<br />

VP [AGR= 1]<br />

V⋄ [AGR= 1]<br />

Class n0V<br />

export: e<br />

use classes V 1 =VSp<strong>in</strong>e, N 1 =Subj<br />

identities: V 1 .V = N 1 .V, e=N 1 .e<br />

Class n0Vn1<br />

export: e<br />

use classes V 1 =n0V<br />

N 2 =DirObj<br />

identities: V 1 .V = N 2 .V, e=N 2 .e<br />

Class DOConstr<br />

use classes V 1 =n0Vn1, N 3 =IndirOjb<br />

identities: V 1 .V = N 3 .V<br />

semantic dimension<br />

⎡<br />

⎤<br />

causation<br />

EFFECTOR 1<br />

THEME 2<br />

GOAL 3<br />

[ ]<br />

V 1 .e<br />

activity<br />

CAUSE<br />

EFFECTOR 1<br />

⎡ ⎤<br />

change-of-poss<br />

⎢ ⎢ ⎥<br />

⎣EFFECT<br />

⎣THEME 2 ⎥<br />

⎦⎦<br />

RECIPIENT 3<br />

Figure 11: MG classes for transitive verbs <strong>and</strong> the DO construction<br />

actually mean conjunction <strong>and</strong> feature value equation. So far, our impression is that we need<br />

only a simple feature logic without quantification or negation.<br />

Now we comb<strong>in</strong>e our small tree fragments <strong>in</strong>to larger ones, thereby def<strong>in</strong><strong>in</strong>g further MG<br />

classes. We add a class VSp<strong>in</strong>e that takes care of the percolation of features (for <strong>in</strong>stance AGR)<br />

along the verbal sp<strong>in</strong>e. This class comb<strong>in</strong>es with the subject class <strong>in</strong>to the n0V class that <strong>in</strong><br />

turn comb<strong>in</strong>es with classes for further arguments. The def<strong>in</strong>ition of the class for active transitive<br />

verbs is shown <strong>in</strong> Fig. 11. 10 What is still miss<strong>in</strong>g here is a fully elaborate l<strong>in</strong>k<strong>in</strong>g theory that<br />

determ<strong>in</strong>es the possible comb<strong>in</strong>ations of semantic roles for a given unanchored syntactic tree,<br />

for example, along the l<strong>in</strong>es of Van Val<strong>in</strong> (2005). We leave this issue for future research.<br />

The further comb<strong>in</strong>ation with the class for the <strong>in</strong>direct object is shown <strong>in</strong> Fig. 11. The<br />

m<strong>in</strong>imal model of DOConstr is the unanchored tree from Fig. 9a. In addition to the frame<br />

shown <strong>in</strong> Fig. 9a, we <strong>in</strong>clude a specification of the thematic roles on the top level of the frame<br />

that serves to obta<strong>in</strong> the correct identifications of participants when unify<strong>in</strong>g with the frame of<br />

the lexical anchor. We will come back to this when treat<strong>in</strong>g lexical anchor<strong>in</strong>g <strong>in</strong> §5.<br />

Now let us consider the PO construction. Here, the n0Vn1 class is used aga<strong>in</strong>. For the third<br />

argument, we use the class DirPrepObj for a directional PP-argument. The PP contributes the<br />

goal of some change of location. The higher class POConstr arises from a comb<strong>in</strong>ation of the<br />

n0Vn1 class <strong>and</strong> the class for the directional PP (Fig. 12). The change of location frame contributed<br />

by the PP is embedded under the EFFECT attribute of the frame of the verb <strong>and</strong> it is<br />

enriched with a role THEME that is the event participant contributed by the direct object. F<strong>in</strong>ally,<br />

we can def<strong>in</strong>e a class DAltConstr as the disjunction of DOConstr <strong>and</strong> POConstr. This way, we<br />

obta<strong>in</strong> a s<strong>in</strong>gle tree family conta<strong>in</strong><strong>in</strong>g trees for both constructions. Depend<strong>in</strong>g on whether we<br />

have a PP or a direct object, only the correspond<strong>in</strong>g part of the family can be selected. The<br />

m<strong>in</strong>imal referent of the class DAltConstr conta<strong>in</strong>s the two trees from Fig. 9. 11<br />

10 Note that we assume that whenever we use a class, its meta-variables ( 0, 1 , etc.) get <strong>in</strong>stantiated with fresh<br />

values. This avoids un<strong>in</strong>tended unifications.<br />

11 As mentioned above, the classes correspond<strong>in</strong>g to elementary tree families usually have more than one m<strong>in</strong>imal<br />

referent s<strong>in</strong>ce all possible realizations of an argument (topicalization, extraposition, relativization, etc.) have to be<br />

taken <strong>in</strong>to account.<br />

180

Hooray! Your file is uploaded and ready to be published.

Saved successfully!

Ooh no, something went wrong!