A Parser for LTAG and Frame Semantics

A Parser for LTAG and Frame Semantics David Arps, Simon Petitjean Heinrich-Heine-Universit¨at D¨usseldorf david.arps@uni-duesseldorf.de, petitjean@phil.uni-duesseldorf.de Abstract Since the idea of combining Lexicalized Tree Adjoining Grammars (LTAG) and frame semantics was proposed (Kallmeyer and Osswald, 2013), a set of resources of this type has been created. These grammars are composed of pairs of elementary trees and frames, where syntactic and semantic arguments are linked using unification variables. This allows to build semantic representations when parsing, by composing the frames according to the combination of the elementary trees. However, the lack of a parser using such grammars makes it complicated to check whether these resources are correct implementations of the theory or not. The development of larger resources, that is to say large-coverage grammars, is also conditioned by the existence of such a parser. In this paper, we present our solution to this problem, namely an extension of the TuLiPA parser with frame semantics. We also present the frameworks used to build the resources used by the parser: the theoretical framework, composed of LTAG and frame semantics, and the software framework, XMG2. Keywords: Parsing, Grammar, Syntax, Semantics, Tools, Systems, Applications. 1. Introduction The development of linguistic resources such as precision grammars and lexicons is a complex and time consuming task. Even though the challenges offered by these tasks often come from the size of the resources to develop, the difficulty can be increased by the lack of tools processing this type of data and allowing to test it. This is the case of grammars using both the formalism of Lexicalized Tree Adjoining Grammars and frames as semantic representa- tions. For example, the description of the syntax-semantics interface in (Zinova, 2017) uses a workaround (simulating the lexical insertion while creating the grammar, and not during parsing), as no parser for such a type of resource was available at the time. This paper introduces a parser which uses grammars of this type, based on an existing parser for LTAG: TuLiPA. In Section 2., we provide a short explanation of the frame- work and show how it interacts with semantic parsing. In Section 3., we summarize the compact description of our resources in a metagrammar. In Section 4., we summarize the architecture of the TuLiPA parser that we are using. In Section 5., we introduce our new extension of TuLiPA to handle frame semantics, and give an example of use of this parser in Section 6.. 2. Semantic Parsing with LTAG 2.1. Tree Adjoining Grammars A Tree Adjoining Grammar consists of a set of elementary trees, from which larger trees are built using substitution and adjunction. Nodes in elementary trees have terminal or non-terminal labels. In a lexicalized TAG, every elementary tree has at least one terminal node. In a fully derived tree, all leaf nodes have terminal labels and all internal nodes have non-terminal labels. Elementary trees are either ini- tial trees or auxiliary trees. The latter have one leaf node marked as a foot node by an asterisk. The foot node and the …

Content truncated to 500 words