Argument Realization

Argument Realization presents a thorough survey of current theories that deal with the relationship between verbs and their arguments. It addresses the problem of argument realization in its many aspects such as semantic roles, lexical semantic representation, conceptualizations of events, thematic hierarchies, and verb alternations. This is a linguistics book and therefore geared toward researchers in syntax and semantics. Nevertheless , this extensive and clearly presented body of work by Levin and Rappaport Hovav deserves a prominent place in computational linguistics libraries. The book is organized into three broad parts. The first part reviews current theories of semantic roles, lexical semantic representation, and event structures. The second part focuses on the mapping from lexical semantics to syntax and on thematic hierarchies. The last part examines perspectives on multiple argument realization. Chapter 1 presents a good survey of linguistic theories and their different ways of explaining syntactic similarities in language. It discusses proposed strategies of isolating semantic components of verbs and the fact that these meaning components cannot be taken in isolation to explain syntactic behavior. The chapter concentrates on the problems that must be taken into account when developing a theory of argument realization and a related theory of lexical semantic representation. It discusses the importance of viewing verb meanings as construals of happenings and explores the variety and uniformity in argument realization through cross-linguistic examples showing that argument realization involves more than just the agent and patient roles. Chapter 2 introduces work on semantic role lists and examines in detail their properties. This chapter includes a discussion on the limitations of semantic role list representations, in particular: (1) the difficulty of determining the right granularity for defining semantic roles; (2) the lack of internal structure of the typical semantic role inventory and the fact that cross-linguistic examples attest similarities between certain roles (e.g., patient and recipients, goals and benefactives); and (3) the problem of assigning a single instance of each role per clause and, if this constraint is relaxed, why only some pairings of roles are possible. Even with these limitations, Levin and Rappaport Hovav still maintain the notion that semantic roles are an appropriate basis for a lexical semantic representation. This chapter also reviews proposed efforts to solve these limitations through decomposition of semantic roles into binary features, and by allowing arguments to be assigned more than one semantic role. Chapter 3 reviews other current notions of semantic …