Larger-Context Tagging: When and Why Does It Work?

Proceedings of the 2021 Conference of the North American Chapter of the Association for Computational Linguistics: Human Language Technologies, pages 1463–1475 June 6–11, 2021. ©2021 Association for Computational Linguistics 1463 Larger-Context Tagging: When and Why Does It Work? Jinlan Fu†, Liangjing Feng†, Qi Zhang†, Xuanjing Huang†, Pengfei Liu♯ ∗ † School of Computer Science, Shanghai Key Laboratory of Intelligent Information Processing, Fudan University, ♯Carnegie Mellon University {fujl16,qz,xjhuang}@fudan.edu.cn, pliu3@cs.cmu.edu Abstract The development of neural networks and pretraining techniques has spawned many sentence-level tagging systems that achieved superior performance on typical benchmarks. However, a relatively less discussed topic is what if more context information is introduced into current top-scoring tagging systems. Al- though several existing works have attempted to shift tagging systems from sentence-level to document-level, there is still no consensus con- clusion about when and why it works, which limits the applicability of the larger-context ap- proach in tagging tasks. In this paper, instead of pursuing a state-of-the-art tagging system by architectural exploration, we focus on inves- tigating when and why the larger-context train- ing, as a general strategy, can work. To this end, we conduct a thorough com- parative study on four proposed aggregators for context information collecting and present an attribute-aided evaluation method to in- terpret the improvement brought by larger- context training. Experimentally, we set up a testbed based on four tagging tasks and thirteen datasets. Hopefully, our preliminary observations can deepen the understanding of larger-context training and enlighten more follow-up works on the use of contextual in- formation. 1 Introduction The rapid development of deep neural models has shown impressive performances on sequence tag- ging tasks that aim to assign labels to each token of an input sequence (Sang and De Meulder, 2003; Lample et al., 2016; Ma and Hovy, 2016). More recently, the use of unsupervised pre-trained mod- els (Akbik et al., 2018, 2019; Peters et al., 2018; Devlin et al., 2018) (especially contextualized ver- sion) has driven state-of-the-art performance to a ∗Corresponding author new level. Among these works, researchers fre- quently choose the boundary with the granularity of sentences for tagging tasks (i.e., sentence-level tagging) (Huang et al., 2015; Chiu and Nichols, 2015; Ma and Hovy, 2016; Lample et al., 2016). Undoubtedly, as a transient, sentence-level setting enables us to develop numerous successful tagging systems, nevertheless the task itself should have not be defined as sentence-level but for simplifying the learning process for machine learning models. Naturally, it would be interesting to see what if larger-context information (e.g., taking informa- tion of neighbor sentences into account) is intro- duced to modern top-scoring systems, which have shown superior performance under the sentence- level setting. A small number of works have made seminal exploration in this direction, in which part of works show significant improvement of larger- context (Luo et al., 2020; Xu et al., 2019) while oth- ers don’t (Hu et al., 2020, 2019; Luo et al., 2018). Therefore, it’s still unclear when and why larger- context training is beneficial for tagging tasks. In this paper, …

Content truncated to 500 words