Capturing motion verb generalizations in synchronous tree adjoining grammars

CAPTURING MOTION VERB GENERALIZATIONS IN SYNCHRONOUS TREE ADJOINING GRAMMARS Martha Palmer and Joseph Rosenzweig Department of Computer Science University of Pennsylvania Philadelphia PA 19104-6389 mpalmer/josephr@linc.cis.upenn.edu Abstract This paper describes the use of verb class memberships as a means of capturing gener- alizations about manner-of-motion verbs in Synchronous Tree Adjoining Grammars, STAGs, [20, 21, 22]. This approach allows STAGs, which are essentially transfer-based, to take advan- tage of the same types of generalizations which are generally thought of as wholly the domain of interlingua systems - without giving up any of the lexical specificity unique to transfer-based systems. In this way a machine translation system based on STAGs can respond with seamless flexibility to a wide spectrum of phenomena being presented for translation ranging from idioms and idiosyncratic lexical items to well-behaved verbs that follow lexical rules. 1 Introduction This paper describes the use of verb class memberships as a means of capturing generalizations about manner-of-motion verbs in Synchronous Tree Adjoining Grammars, STAGs, [20, 21, 22]. This approach allows STAGs, which are essentially transfer-based, to take advantage of the same types of generalizations which are generally thought of as wholly the domain of interlingua systems - without giving up any of the lexical specificity unique to transfer-based systems. In this way a machine translation system based on STAGs can respond with seamless flexibility to a wide spectrum of phenomena being presented for translation ranging from idioms and idiosyncratic lexical items to well-behaved verbs that follow lexical rules. This paper begins with a review of the role played by lexical semantics in natural language processing and machine translation. Lexical Conceptual Structure (LCS) as an independent level of lexical semantic representation are introduced, with reference to their implementation in text processing systems and as an interlingua. The use of linking rules as a means of communication between syntactic and semantic levels of representation is discussed and then Lexicalized Tree- Adjoining Grammars (LTAGs) are described. The inherent advantages of lexical trees for the representation of lexical semantic information, as compared with Lexical Conceptual Structures, are emphasized. An LTAG-based approach to machine translation (MT) is presented, and its handling of cross-linguistic lexical divergences is contrasted with an interlingual approach based on Lexical Conceptual Structures. A comparison is also made with the ACQUILEX MT framework, which uses typed feature structures and unification to represent a categorial grammar. 2 Lexical-semantic representation formalisms A primary task of lexical semantics is to find correct correspondences between the underlying semantic representation of a verb and its alternative syntactic realizations. Given any particular syntactic realization, the appropriate semantic representation is then available. A lexical seman- tic representation formalism should encode the constraints governing these syntax-semantics correspondences transparently and efficiently, enabling applications such as machine translation 76 systems to retrieve semantic information from syntactic structure as quickly and directly as possible. 2.1 Lexical Conceptual Structures Lexical Conceptual Structures (LCS), Jackendoff's seminal contribution to the representation of lexical semantic information, have been utilized in a number of natural language process- ing systems. One …

Content truncated to 500 words