Abstract
As a submission of its own to the open track of Task 8 at SemEval 2014, a subset of the organizers sought to connect the task to pre-existing, ‘in-house’ parsing systems that can output the same types of semantic dependency graphs as used in the task. For each of the three target formats, there is an existing parsing system, which typically was developed in parallel to the creation of the target dependency graphs, viz. (a) for the DM format, the parser using the hand-engineered English Resource Grammar; (b) for the PAS format, the Enju parsing system, with its probabilistic HPSG acquired through linguistic projection of the PTB; and (c) for the PCEDT format, the scenario for English tectogrammatical analysis within the Treex framework, comprising a pipeline of a syntactical data-driven dependency parser and a number of hand-engineered processing blocks that convert trees from the analytical to the tectogrammatical layer.