Transition-Based Deep Input Linearization

TL;DR

This paper introduces a transition-based deep input linearization model that jointly performs multiple NLG tasks, significantly reducing error propagation and improving accuracy over traditional pipeline methods.

Contribution

The paper presents a novel transition-based approach that integrates linearization, function word prediction, and morphological generation in a single model, outperforming previous pipelined systems.

Findings

Achieved the best results on a standard deep input linearization shared task.

Significantly improved accuracy over pipelined baseline systems.

Demonstrated the effectiveness of joint modeling in NLG tasks.

Abstract

Traditional methods for deep NLG adopt pipeline approaches comprising stages such as constructing syntactic input, predicting function words, linearizing the syntactic input and generating the surface forms. Though easier to visualize, pipeline approaches suffer from error propagation. In addition, information available across modules cannot be leveraged by all modules. We construct a transition-based model to jointly perform linearization, function word prediction and morphological generation, which considerably improves upon the accuracy compared to a pipelined baseline system. On a standard deep input linearization shared task, our system achieves the best results reported so far.