Copenhagen-Malmö: Tree Approximations of Semantic Parsing Problems

Abstract

In this shared task paper for SemEval-2014 Task 8, we show that most semantic structures can be approximated by trees through a series of almost bijective graph transformations. We transform input graphs, apply off-the-shelf methods from syntactic parsing on the resulting trees, and retrieve output graphs. Using tree approximations, we obtain good results across three semantic formalisms, with a 15.9% error reduction over a stateof-the-art semantic role labeling system on development data. Our system came in 3/6 in the shared task closed track.

A Transition-Based Directed Acyclic Graph Parser for UCCA

“…Recent interest in SDP has yielded numerous works on graph parsing (Ribeyre et al, 2014;Almeida and Martins, 2015;Du et al, 2015), including tree approximation (Agić and Koller, 2014;Schluter et al, 2014) and joint syntactic/semantic parsing (Henderson et al, 2013;Swayamdipta et al, 2016).…”

Proceedings of the 14th International Conference on Parsing Technologies

“…In the context of the SemEval task 8 on Broad Coverage Semantic Dependency Parsing, Schluter et al (2014) and Agić and Koller (2014) introduce a tree approximation definition for the semantic dependency parsing task, including a number of different approaches. Both present similar pruning pre-processing steps as a potential approach.…”

“…Additionally, Schluter et al (2014) present a packing pre-processing step as a further approach. We discuss the approaches presented in (Schluter et al, 2014), because we adopt the parts of their preprocessing which do not remove any edges. Their approaches also outperformed that of (Agić and Koller, 2014) in the SemEval task.…”

“…The sparsity of digraph representations of some semantic dependency digraph datasets (i.e., the fact that the number of edges is linear in the number of vertices), as well as the well-performing first attempts at such tree approximations from Schluter et al (2014), Agić and Koller (2014), and Agić et al (2015) suggest that tree approximations for digraph semantic dependency structures are a relevant avenue to the development if not the sidestepping of some computationally harder models of directed acyclic graph (DAG) and digraph decoding Martins and Almeida, 2014).…”

“…The key change in the algorithm is to iteratively minimise the error in precision between output spanning subtrees and corresponding training digraph instances, allowing therefore the algorithm to choose best spanning subtree approximations with respect to the dataset, rather than forming tree approximations as a pre-processing step to training as was done by Schluter et al (2014), Agić and Koller (2014), and Agić et al (2015).…”