Though a topic of intense discussion in the 1980s, work on the licensing of null arguments has fallen out of fashion of late. The present volume, therefore, represents a welcome revival of interest in the topic. In its introduction, it also suggests a reason why the topic was abandoned by many researchers: initial characterizations of a strong connection between null arguments and rich inflectional agreement proved to be too simplistic and heavily resisted any adequate formal characterization. Rather, over the past twenty-five years it has been demonstrated that a wide variety of factors other than morphological ‘richness’ play a role in determining whether or not a particular argument of a verb can be licensed as null. Some languages, for instance, allow null subjects only with first and second person, but not with third (so-called partial pro-drop); others allow pro-drop only when the subject is post-verbal; others only in certain tenses or only when pro receives certain semantic interpretations. Finally, many languages with no agreement whatsoever freely allow null arguments. All of these discoveries have led to the conclusion that a single characterization of all instances of pro-drop is likely to be impossible. Rather, it may be that null arguments are licensed in a number of independent ways.

The present work reflects this view while still seeking to provide principled explanations for pro-drop phenomena; but while as a whole the book can be praised for looking deeper into the empirical facts to enlighten existing work, it can also be criticized for not looking broader to empirical facts from other languages. Most of the languages explored here are familiar from the existing pro-drop literature (Romance languages, Finnish, Hebrew, for example, though Inge Zwitserlood & Ingeborg van Gijn's chapter on ‘Agreement phenomena in Sign Language of the Netherlands’ is a notable exception). The final two chapters on non-configurational languages might seem to broaden the picture, but as the introduction recognizes, these languages are not understood to be pro-drop languages in the formal sense. One wonders how insights from African and South American pro-drop languages might illuminate the null subject landscape.

Following the introduction, the volume begins with Margaret Speas's influential but previously unpublished paper from the early 1990s, ‘Economy, agreement, and the representation of null arguments’. In this paper, Speas argues that the availability of pro-drop is decided structurally by whether or not a language has an Agreement (Agr) projection as part of its functional structure and, if so, whether agreement inflection is base-generated in Agr or on the verb. If a language has no Agr projection (e.g. Chinese) or has Agr with agreement base-generated in the Agr position (e.g. Italian), null subjects are permitted. Languages where agreement is base-generated on the verb, on the other hand, do not permit null subjects. Assuming that being generated in Agr correlates with a rich agreement paradigm, this provides a formal explanation for Jaeggli & Safir's (Reference Jaeggli, Safir, Osvaldo and Safir1989) observation that null subjects are permitted in languages with very rich agreement or no agreement at all, but not in languages with deficient agreement paradigms.

Speas's proposals are similar to the framework presented by Artemis Alexiadou in ‘Uniform and non-uniform aspects of pro-drop languages’, familiar from the influential Alexiadou & Anagnostopolou (Reference Alexiadou and Anagnostopoulou1998) article. In this framework, pro-drop phenomena are related to the Extended Projection Principle, understood as a feature-checking requirement involving a D(efiniteness)-feature. The latter can be checked either by merging a subject into the relevant specifier or by rich inflection on the verb (resulting in pro-drop). One criticism of this approach is that it leaves open the question of what constitutes ‘rich’ inflection. In the present volume, Alexiadou addresses this question, attempting to account for well-known cases of partial pro-drop in languages like Hebrew and Finnish. Taking the D-feature of Alexiadou & Anagnostopolou (Reference Alexiadou and Anagnostopoulou1998) to be equivalent to the feature [person], Alexiadou argues that cases of partial pro-drop can be accounted for by close examination of the agreement paradigm of the language. She demonstrates that in the Hebrew and Finnish cases (which allow pro-drop in the first and second person but not in the third person), there is evidence that third singular inflection is not specified for [person].

In a similar vein, Hans Bennis argues in ‘Agreement, pro, and imperatives’ that pro itself does not come from the lexicon specified for phi-features, but must acquire phi-feature values from a lexically specified set of uninterpretable features on the verb via an Agree relation. If the morphology of the verbal agreement paradigm is rich enough, pro will be able to identify and acquire a full set of phi-features and will then be able to delete the uninterpretable features of the verb. Without enough morphological information, however, pro cannot be sufficiently specified to delete these features. Thus, languages without fully-specified inflectional paradigms (such as English) cannot license pro while languages with fully-specified paradigms (like Spanish) or no inflectional features at all (such as Chinese) can license pro.

Like Alexiadou and Bennis, Olaf Koeneman argues in ‘Deriving the difference between full and partial pro-drop’ that examining a language's full agreement paradigm provides the best clues for determining the language's pro-drop possibilities. For Koeneman, this is due to the fact that agreement forms must uniquely identify null subjects in order for the latter to be considered pronominal (and therefore legitimate subjects). Koeneman's test for the pronominal status of agreement markers is whether or not the agreement paradigm exhibits syncretism. Only if the paradigm has no syncretic forms can null subjects be licensed. This explains why languages like English and German, though exhibiting agreement affixes that uniquely identify particular arguments (e.g. third person singular in the English present tense) fail to exhibit pro-drop in the presence of those affixes: it is not an identity requirement between affixes and subjects that makes pro-drop possible, but rather an identity requirement that must hold across the entire agreement paradigm. Koeneman notes that there seem to be exceptions to his generalization, mentioning Colloquial Portuguese, which has the same agreement form in the second and third plural. However, it would seem that the number of exceptions must be rather large. For instance, nearly every one of the hundreds of Bantu languages spoken in sub-Saharan Africa has a noun class system in which a single subject agreement marker may be taken to reference nouns from more than one noun class; yet these language exhibit subject pro-drop uniformly.

One of the interesting themes of the volume involves using partial pro-drop languages as a window into the working of null subject licensing, as in Alexiadou's chapter discussed above. Two other authors in the volume, Koeneman and Cecilia Poletto (‘Asymmetrical pro-drop in Northern Italian dialects’), also address this issue and both argue that a historical perspective on a language's agreement inflection can enlighten the discussion. Focusing on the Hebrew and Finnish case, Koeneman argues that while the first and second person agreement affixes in these languages have pronominal origins, the third person affixes do not. In the licensing system he discusses (see above), this means that third person affixes never had the feature [pronominal] that must be underspecified in order for an agreement marker to license a null subject. In a detailed examination of the development of Northern Italian Dialects, Poletto shows that the partial pro-drop systems at work here do not fall along the first and second vs. third person divide observed for Hebrew and Finnish, but rather group first person singular with first and second person plural against the other agreement forms. While the former allow pro-drop, the latter do not. Adopting similar licensing requirements as Speas and Alexiadou, Poletto argues that the asymmetries she observes may be accounted for if the feature [person] is decomposed into the features [speaker] and [hearer].

All of these approaches have in common a view that runs throughout the volume: that the formal licensing of null subjects can be linked to a language's morphological paradigm only in an indirect way. Null subjects are licensed by the feature structure of functional elements in the language. Examining morphological paradigms and their historical development may give clues as to what the relevant features are (there is some consensus in the volume that the feature [person] is central), but how agreement morphemes are manifested phonologically in the language plays no direct role.

While most of the volume addresses the formal licensing of null subjects, two chapters focus on the interpretation of null arguments, or what Rizzi (Reference Rizzi1986) called their contextual licensing. Zwitserlood & van Gijn examine null argument licensing in the sign language of the Netherlands (NGT). NGT shows a high degree of variability in its agreement paradigms, with some verbs agreeing with all arguments, some with only a subset of their arguments, and some not agreeing at all. Rather than locating this variability in the agreement paradigm of the language, Zwitserlood & van Gijn argue that there are various ways of licensing null arguments and that NGT makes use of them all, organizing the various strategies in a hierarchy. They propose that, when interpreting a null argument at Logical Form, the null argument first looks for local agreement to establish its reference. If this is not available, it seeks a local argument. If neither option is available, the null argument assumes the status of a variable and looks for an appropriate sentential topic. There are thus three ways to establish reference for a null argument in NGT and only when all three fail is an overt subject required.

In ‘“Arbitrary” pro and the theory of pro-drop’, Patricia Hofherr revisits Rizzi's (Reference Rizzi1986) proposal that contextual licensing of null arguments is dependent on what sort of theta role the argument receives (full, quasi, or none). Using data from nearly a dozen languages, Hofherr demonstrates that amongst third person referential arguments, languages often allow null subjects when the reference is arbitrary, but not when it is anaphoric. Given this distinction, Hofherr proposes to replace Rizzi's hierarchy with one that distinguishes deictic, anaphoric, and non-anaphoric references for pro. Interestingly, deictic pro is defined as having the feature [+hearer] or [+speaker], the same distinction in person features that Poletto argues is required for the formal licensing of null arguments.

Finally, the last two chapters of the book address the status of null arguments in non-configurational languages. Both Eloise Jelinek (‘The pronominal argument parameter’) and Mark Baker (‘On zero agreement and polysynthesis’) agree that such languages are typologically distinct from the other languages discussed in this volume that exhibit pro-drop. Polysynthetic languages are not, properly speaking, pro-drop languages. They are rather languages in which full Determiner Phrases (DPs) are never true arguments residing in structural A-positions. Instead, the inflectional clitics representing these DPs are the verb's true arguments, while full DPs are always in adjoined positions. Jelinek refers to this type of language as a Pronominal Argument (PA) language, opposed to so-called Lexical Argument (LA) languages, and her chapter is largely concerned with outlining formal differences between the two types of languages. Baker's Polysynthesis Parameter accomplishes much the same typological divide. Baker also argues that argument-licensing morphology on the verb may in some cases be null, observing that typically in polysynthetic languages ditransitive verbs show agreement only with the indirect object. Since this lack of agreement does not affect the syntax of direct objects in these constructions (they can still be either dropped or freely ordered), Baker argues that the verb agrees with the direct object, though this agreement is null. If Baker's hypotheses are correct, then this is strong evidence that macroparameters like the polysynthesis parameter are psychologically real.

If this volume has a single overarching theme, it is that when it comes to the phenomena of null arguments, there is no single overarching theme. Implicitly or explicitly, all of the authors recognize that given all we have discovered about null arguments, one cannot speak of null argument licensing with complete uniformity. This is even reflected in the structure of the volume: nine papers are organized into four thematic sections. But rather than seeing such developments as problematic, the present volume suggests that we see them as an invitation to deeper and more detailed linguistic investigation.