Generalizing Natural Language Analysis through Span-relation Representations

Proceedings of the 58th Annual Meeting of the Association for Computational Linguistics, pages 2120–2133 July 5 - 10, 2020. c⃝2020 Association for Computational Linguistics 2120 Generalizing Natural Language Analysis through Span-relation Representations Zhengbao Jiang1, Wei Xu2, Jun Araki3, Graham Neubig1 Language Technologies Institute, Carnegie Mellon University1 Department of Computer Science and Engineering, Ohio State University2 Bosch Research North America3 {zhengbaj,gneubig}@cs.cmu.edu1 xu.1265@osu.edu2, jun.araki@us.bosch.com3 Abstract Natural language processing covers a wide va- riety of tasks predicting syntax, semantics, and information content, and usually each type of output is generated with specially designed architectures. In this paper, we provide the simple insight that a great variety of tasks can be represented in a single unified format consisting of labeling spans and relations be- tween spans, thus a single task-independent model can be used across different tasks. We perform extensive experiments to test this in- sight on 10 disparate tasks spanning depen- dency parsing (syntax), semantic role label- ing (semantics), relation extraction (informa- tion content), aspect based sentiment analysis (sentiment), and many others, achieving per- formance comparable to state-of-the-art spe- cialized models. We further demonstrate ben- efits of multi-task learning, and also show that the proposed method makes it easy to analyze differences and similarities in how the model handles different tasks. Finally, we convert these datasets into a unified format to build a benchmark, which provides a holistic testbed for evaluating future models for generalized natural language analysis. 1 Introduction A large number of natural language processing (NLP) tasks exist to analyze various aspects of hu- man language, including syntax (e.g., constituency and dependency parsing), semantics (e.g., seman- tic role labeling), information content (e.g., named entity recognition and relation extraction), or sen- timent (e.g., sentiment analysis). At first glance, these tasks are seemingly very different in both the structure of their output and the variety of infor- mation that they try to capture. To handle these different characteristics, researchers usually use specially designed neural network architectures. In this paper we ask the simple questions: are the Figure 1: An example from BRAT, consisting of POS, NER, and RE. task-specific architectures really necessary? Or with the appropriate representational methodology, can we devise a single model that can perform — and achieve state-of-the-art performance on — a large number of natural language analysis tasks? Interestingly, in the domain of efficient human annotation interfaces, it is already standard to use unified representations for a wide variety of NLP tasks. Figure 1 shows one example of the BRAT (Stenetorp et al., 2012) annotation interface, which has been used for annotating data for tasks as broad as part-of-speech tagging, named entity recogni- tion, relation extraction, and many others. Notably, this interface has a single unified format that con- sists of spans (e.g., the span of an entity), labels on the spans (e.g., the variety of entity such as “per- son” or “location”), and labeled relations between the spans (e.g., “born-in”). These labeled relations can form a tree or a graph structure, expressing the linguistic structure of sentences (e.g., …

Content truncated to 500 words