636LANGUAGE, VOLUME 67. NUMBER 3 (1991) dividual basis, as W has noted. This choice would then be more dependent on economic prestige. In her final chapter, W discusses the differences between first- and secondgeneration immigrants to Catalonia with regard to the learning of Catalan. For those immigrants who do not live in a predominantly Catalan neighborhood, W hypothesizes that those who were born outside Catalonia are more likely to learn Catalan than those who are native-born of immigrant parents. She says that immigrants who have moved to Catalonia are not part of a social group when they first arrive, so it is not a great risk for them to learn Catalan and thus define themselves as members of the Catalan social group. Those who are born in Catalonia, however, grow up with a social identity, which for immigrants is usually Castilian. These speakers must accept the risk of rejection by their own social group if they wish to learn Catalan. W points out that other researchers have keyed on the cognitive-linguistic background of immigrant children, but that in view of her findings in Catalonia it might be worthwhile to examine the social risks as well. In spite of shortcomings in the design of the matched-guise test, Woolard's book is a valuable study. She explains very clearly the complexity of bilingualism in Catalonia, and her ethnographic portrait elucidates the feelings and conflicts between Catalans and Castilian immigrants. This book will be of great interest to researchers in the areas of bilingualism, language prestige, ethnic identity, and language planning, and to anyone who wants to understand fully the unusual status of the Catalan language in Spain. REFERENCES Mier, Jeanne Zang. 1986. Estudi sociolinguistic de certs aspectes de la llengua catalana. Treballs de sociolinguística catalana 6.33-112. Vallverdú. Francesc. 1973. El fet linguistic com a fet social. Barcelona: Edicions 62. 5706 Vandegrift Avenue[Received 11 September 1990; Rockville, MD 20851revision received 11 January 1991.] The syntactic recoverability of null arguments. By Yves Roberge. Kingston & Montreal: McGill-Queen's University Press, 1990. Pp. x, 217. $34.95. Reviewed by Donna Jo Napoli, Swarthmore College Roberge begins his book with an introduction to the theoretical framework adopted, that of Government and Binding (GB) Theory. A survey of previous accounts of null arguments, particularly in Romance languages (the primary data base for the book), follows. The prose is clear and easy to understand, despite the fact that the works being summarized are full of complex arguments. The reader can be deceived into thinking this is a rather simple book; but the remaining chapters disabuse one of this initial false impression. There are four chapters in all: 'Syntactic theory and null arguments' (10-34), 'Null arguments in Romance languages' (35-84), 'On clitic doubling' (85-151), and 'Clitics and agreement markers' (152-80). The main thesis of the book has two parts. First, all empty categories must be licensed, a familiar concept in GB and one that R calls the Recoverability REVIEWS637 Principle (8). Second, two mechanisms for licensing null arguments are a rich enough inflection (AGR) and clitics. The notion of recoverability itself is not explored explicitly. But from the very fact that R claims that AGR and clitics can license because they have the relevant features of person, number, and, sometimes, gender and Case, we are led to the idea that recoverability here means not just predictability, but an actual subset relation between sets of assigned roles or features (as in Larson 1990;613, n. 14). For example, an act ofeating predictably involves something eaten (or to eat). But we cannot deduce from that pragmatic fact that a verb like eat has a syntactically present null object in a sentence like John ate already. With R's approach a verb could have a null argument only if there is some syntactically realized element that will allow us access to all the relevant information about the null argument. Since lexical pronouns in Romance languages are feature bundles for person, number, gender, and Case, only if this set of features can be recovered about a null argument will the null argument be licensed. Pro, the null pronoun, is allowed in subject...