Book Reviews: Parsing with Principles and Classes of Information

After Chomsky's (1981) introduction of the Government and Binding (GB) theory of grammar, a research area called GB parsing developed in the mid-eighties to explore parsing architectures based on that framework. In this area, parsing is viewed as the characterization of a mental process rather than a crude mapping from strings to syntactic structures. Therefore in GB parsing there is a need to develop a motivated mapping between the postulated model of humans' knowledge of language (the grammar) and the parsing architecture, an enterprise in which psychological as well as computational issues are at stake. Both practical experimentation and theoretical investigation revealed early on that imposing a direct correspondence between the assumed grammar and the parsing architecture causes serious problems of efficiency (and in some cases the parsing process might not even terminate). This is in strong contrast with the empirical fact that humans make use of their knowledge of language in a very effective way. For these reasons, research on GB parsing has to face the strong tension between the desiderata of keeping the parser architecture close enough to the abstract linguistic formulation of the grammar, in order to inherit its explanatory power and its generalizations, and the clear need for grammar covering and compiling techniques to reduce the inefficiencies of the abstract linguistic formulation of the theory. This research monograph, which is a revised version of the author's doctoral dissertation, is an original contribution to this line of research. (A shorter version of this research was published as Merlo [1995].) Below, I will provide a general presentation of the book and of the author's ideas. Following that, I will discuss in greater detail the content of each chapter, adding specific comments and observations. The leitmotif in this book is the distinction between linguistic modularity and computational modularity. As is well known, GB theory conceives of grammar as a set of modules, each consisting of a set of parameterized principles concerning some specific linguistic abstraction, as for instance phrase structure, abstract case assignment, binding, movement, and so on. In other terms, linguistic modularity accounts for the linguistic data by means of the interaction of abstract devices, in a way that achieves a high degree of explanation and generalization. From the computational perspective,