A Unified Syntax-aware Framework for Semantic Role Labeling

Abstract

Semantic role labeling (SRL) aims to recognize the predicate-argument structure of a sentence. Syntactic information has been paid a great attention over the role of enhancing SRL. However, the latest advance shows that syntax would not be so important for SRL with the emerging much smaller gap between syntax-aware and syntax-agnostic SRL. To comprehensively explore the role of syntax for SRL task, we extend existing models and propose a unified framework to investigate more effective and more diverse ways of incorporating syntax into sequential neural networks. Exploring the effect of syntactic input quality on SRL performance, we confirm that high-quality syntactic parse could still effectively enhance syntactically-driven SRL. Using empirically optimized integration strategy, we even enlarge the gap between syntax-aware and syntax-agnostic SRL. Our framework achieves state-of-the-art results on CoNLL-2009 benchmarks both for English and Chinese, substantially outperforming all previous models.

Effective Subword Segmentation for Text Comprehension

“…To model texts into vector space, the input tokens are represented as embeddings in deep learning models [28,29,30,45,46,55,57]. Previous work has shown that word representations in NLP tasks can benefit from character-level models, which aim at learning language representations directly from characters.…”

Adversarial Domain Adaptation Network for Semantic Role Classification

“…Roth and Lapata [27] embed dependency path as syntactic information into a model and exhibited a notable success. Li et al [28] extended existing models and proposed a unified framework to investigate more effective and more diverse ways of incorporating syntax into sequential neural networks. Different from the above works, our work is constructed based on a Bi-LSTM neural model and an adversarial domain classifier is incorporated into the model to enhance the domain-invariant features to emerge.…”

“…The problem can be tackled if we can construct a new vector space where the divergence between the two distributions are very small. Some works [28], [29] suggest that the adversarial loss can measure the H-divergence between two distributions. Thus, this paper incorporates adversarial learning to narrow down the divergence between the source distributions and the target.…”

“…Following the setup of the adversarial training [28], we design a min-max game to optimize the whole network. The train process consists of the following two parts.…”

“…Here, we compare our method with SourceLabeler system. We divide all sentences of Brown test set into six groups [1][2][3][4][5], [6][7][8][9][10], [11][12][13][14][15][16][17][18][19][20], [21][22][23][24][25][26][27][28][29][30], [31][32][33][34][35][36][37][38][39][40][41][42][43][44][45][46][47][48][49][50] and [51-]. The statistics about different groups are shown in Table 2.…”