A look at structural ambiguity in language model based parsers

Modern language model based parsers don't address the problem of structural ambiguity, where a given sentence may have multiple interpretations. Exploring the idea that such sentences may be disambiguated in the context of a prior sentence, we set up an experiment in Dutch involving relative clauses. Experiments show that including disambiguating context helps to mitigate existing structural biases, but not fully solve the issue. If time I present this work in the context of an NWO project in which more resources and evaluation methods for Dutch have been developed.