UDRT-based semantics construction for LTAG – and what it tells us about the role of adjunction in LTAG

In this paper we will have a closer look at the role that adjunction plays for semantics composition in LTAG and examine whether the semantics construction approach based on derivation trees can be rescued. The contributions of the paper are manifold. First, we show that the semantics construction approach for LDGs presented by Cimiano and Reyle can in principle be applied to semantics construction for LTAG.We borrow from Cimiano and Reyle the idea of talking explicitly about labels and scope, which will turn out as a crucial component for principle-based semantics construction on the basis of the derivation tree. Second, we show that an adequate account of quantifier scope within LTAG is possible without necessarily leading to an extension of LTAG as proposed by Kallmeyer, who introduces multicomponent elementary trees and allows for restricted multiple adjoining in the case of quantiers. Third, we show that it is possible to define semantics from derivation trees in a principled way, thus overcoming the problems arising from the non-isomorphism of adjunction and syntactic dependencies discussed in. We show that by respecting the depth of embedding of constituents as well as dening semantic composition as an update function iteratively combining the semantic representations of a mother node and each child, the above non-isomorphism can be solved in a principled way. Further, we show that by resorting to an underspecified semantic representation language - UDRT in our case - we can handle handle scope phenomena without any complication of LTAG as well as provide a semantic counterpart for the (syntactic) operation of adjunction. Finally, from a more general perspective, our work helps to clarify the role of adjunction in LTAG. The paper is structured as follows: in Section 2 we briefly introduce the syntactic operations of substitution and adjunction, and the standard approach to semantics construction in LTAG.We then review an example discussed by Frank and van Genabith, in which the derivation tree does not contain certain dependencies that are necessary to define semantics compositionally, using standard methods. In Section 3 we present our UDRT-based approach to semantics construction based on the derivation tree. In Section 4 we show how the approach elegantly accounts for two problems which have been observed for semantics construction based on the derivation tree. In Section 5 we discuss related work. Section 6 concludes.