Construindo um domínio não-finito: a sintaxe de orações de gerúndio em português brasileiro

In this dissertation, I investigate gerund clauses (GC) in Brazilian Portuguese, nonfinite clauses the verb of which is in gerund form, {-ndo}. GCs can show up in a series of different syntactic environments. They can be selected by different predicates, both as subjects and as complements. When they are not thematic, GCs can be adjoined both to nominal and to verbal elements. When they are verbal adjuncts, they can be attached to different portions of the structure. Despite this diversity, I try to analyze GCs in unified fashion. The main proposal I put forth is that there are three classes of GCs, such that they differ in structural make-up. On one end of the scale are GCs that project only the bare minimum to build a GC at all, namely, an AspP. I claim that the gerund morpheme heads this phrase because it seems to contribute aspectual meaning to the overall sentence. On the other end of the scale are the GCs that project the most complex structure, a CP that dominates a TP, which in turn dominates the minimal AspP. Between these two ends are GCs of intermediate size, a TP. For convenience, the most structurally complex GCs are dubbed class 1, the intermediate GCs, class 2 and the smallest GCs, class 3. All GC subtypes will undergo this tripartite analysis. It will allow us to capture their individual behavior, as well as the way the classes are grouped together along a series of syntactic properties. Some of these properties, I analyze as TP-dependent, namely, sentential negation and sentential adverbial licensing. As far as these properties are concerned, class 1 and class 2 GCs behave alike. This is captured by the presence of TP in their structure. Class 3 GCs are different from both and this is captured by the absence of TP in their structure. The remaining properties are the ones I propose to be phase-based. The proposal is that this concept plays a unifying role in syntax, allowing it to be simpler. Specifically, I propose that both syntactic operations and relations are phase-dependent. Thus, phenomena as diverse as feature valuation, binding and quantifier raising are unified in that their domain is a phase. Now it is class 2 and class 3 GCs that behave alike, to the exclusion of class 1 GCs. The topmost projection of class 2 GCs is TP, while that of class 3 GCs, an AspP. Neither of them is a phase. On the other hand, the topmost projection of class 1 GCs is CP, a phase. I try to derive the characteristic behavior of each GC subtype and the way the classes proposed are clustered together from their syntactic computation. To this end, based on minimalist assumptions, I outline the way I take the computational system to work. The same system will be an important asset in accounting for the control properties (or lack thereof) in certain GC subtypes and for the temporal properties of these nonfinite constructions.