The aim of the paper is to provide an analysis of Polish Negative Concord (NC) and Genitive of Negation (GoN) in the context of Verb Clusters (VCs). In Polish, a verb has to be overtly negated in the presence of a (number of) clause-mate n-word(s). This behaviour depends neither on grammatical function nor on linear position of the n-words; in each case a single negation meaning results. However, in VCs this locality of NC breaks down: whenever a dependent of the lowest verb in the cluster is an n-word, one of the verbs in VC has to be negated, not necessarily the lowest one. (In passing, we also show that Polish NC is non-local in the sense that it can cross any number of NP and PP projections; an aspect of NC usually ignored in the literature.) Similarly, whenever a verb subcategorizing for an NP[acc] is negated, this requirement changes to NP[gen]. However, in the context of VCs, negating any of the verbs triggers GoN on the lower verb.
We provide an HPSG analysis of these facts, suggesting that complex predicate formation is happening in Polish VCs. This way we are able to uniformly explain apparent violation to the locality characteristics of both phenomena. We also discuss results of various other tests usually invoked when arguing for complex predicate structures and show that many of them are not applicable to Polish.
