Unrestricted Quantifier Scope Disambiguation

We present the first work on applying statistical techniques to unrestricted Quantifier Scope Disambiguation (QSD), where there is no restriction on the type or the number of quantifiers in the sentence. We formulate unrestricted QSD as learning to build a Directed Acyclic Graph (DAG) and define evaluation metrics based on the properties of DAGs. Previous work on statistical scope disambiguation is very limited, only considering sentences with two explicitly quantified noun phrases (NPs). In addition, they only handle a restricted list of quantifiers. In our system, all NPs, explicitly quantified or not (e.g. definites, bare singulars/plurals, etc.), are considered for possible scope interactions. We present early results on applying a simple model to a small corpus. The preliminary results are encouraging, and we hope will motivate further research in this area.