Joint Universal Syntactic and Semantic Parsing

Abstract

While numerous attempts have been made to jointly parse syntax and semantics, high performance in one domain typically comes at the price of performance in the other. This trade-off contradicts the large body of research focusing on the rich interactions at the syntax–semantics interface. We explore multiple model architectures that allow us to exploit the rich syntactic and semantic annotations contained in the Universal Decompositional Semantics (UDS) dataset, jointly parsing Universal Dependencies and UDS to obtain state-of-the-art results in both formalisms. We analyze the behavior of a joint model of syntax and semantics, finding patterns supported by linguistic theory at the syntax–semantics interface. We then investigate to what degree joint modeling generalizes to a multilingual setting, where we find similar trends across 8 languages.

When More Data Hurts: A Troubling Quirk in Developing Broad-Coverage Natural Language Understanding Systems

“…We explore both a sequence-to-sequence (seq2seq) model and a sequence-to-graph (seq2graph) model, using the MISO framework (Zhang et al, 2019b;Stengel-Eskin et al, 2021), which is built on top of AllenNLP (Gardner et al, 2018). The former directly predicts the Lisp string, while the latter produces a DAG as seen at the bottom of Fig.…”

“…Transformer-based transductive A more competitive approach follows the transductive parsing paradigm (Zhang et al, 2019a), which aims to directly produce the underlying DAG instead of the surface form, generating graph nodes as well as edges. We implement a transformer-based transductive model based on the architecture and code from Stengel-Eskin et al (2021). The model directly generates the linearized DAG (cf.…”