Multi-Input Multi-Output Sequence Labeling for Joint Extraction of Fact and Condition Tuples from Scientific Text

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 302–312, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 302 Multi-Input Multi-Output Sequence Labeling for Joint Extraction of Fact and Condition Tuples from Scientific Text Tianwen Jiang‡†,1, Tong Zhao†, Bing Qin‡, Ting Liu‡, Nitesh V. Chawla†, Meng Jiang† ‡Harbin Institute of Technology, Harbin, Heilongjiang, China †University of Notre Dame, Notre Dame, Indiana, USA ‡{twjiang,bqin,tliu}@ir.hit.edu.cn †{tzhao2,nchawla,mjiang2}@nd.edu Abstract Condition is essential in scientific statement. Without the conditions (e.g., equipment, en- vironment) that were precisely specified, facts (e.g., observations) in the statements may no longer be valid. Existing ScienceIE methods, which aim at extracting factual tuples from sci- entific text, do not consider the conditions. In this work, we propose a new sequence label- ing framework (as well as a new tag schema) to jointly extract the fact and condition tuples from statement sentences. The framework has (1) a multi-output module to generate one or multiple tuples and (2) a multi-input module to feed in multiple types of signals as sequences. It improves F1 score relatively by 4.2% on BioNLP2013 and by 6.2% on a new bio-text dataset for tuple extraction. 1 Introduction Conditions such as environment and equipment provide validation supports for facts, while the facts focus on scientific observation and hypoth- esis in scientific literature (Miller, 1947). Exist- ing ScienceIE methods, which extract (subject, re- lational phrase, object)-tuples from scientific text, do not distinguish the roles of fact and condition. Simply adding a tuple classification module has two weak points: (1) one tuple may have different roles in different sentences; (2) the tuples in one sentence have high dependencies with each other, for example, given a statement sentence in a bio- chemistry paper (Tomilin et al., 2016): “We observed that ... alkaline pH increases the activity of TRPV5/V6 channels in Jurkat T cells.” an existing system (Stanovsky et al., 2018) would return one tuple as below: 1This work was done when the first author was visiting the University of Notre Dame. Encoder Decoder LSTMd Feed Forward Feed Forward Softmax Softmax {!" #, !% #, … !' #} {!" (, !% (, … !)(} Feed Forward Feed Forward ... Feed Forward Feed Forward ... Softmax Softmax Softmax Softmax LM POS CAP Multi-Input Module + + Statement Sentence relation names Multi-Output Module Multi-Head Multi-Head Relation name tagging layer Tuple completion tagging layer Input Sequences + Input-Gates {*+,-", *+,-% … *+,-'} {./01", ./01% … ./01)} 2[":5] 78 9 , 2[":5] 7: 9 … 2[":5] 7; 9 2[":5] 78< , 2[":5] 7:< … 2[":5] 7= < Input-Gates Input-Gates fact condition Figure 1: Our framework has two modules: (1) a multi- input module (bottom) based on a multi-head encoder- decoder model with multi-input gates; (2) a multi- output module (top) of a relation name tagging layer and a tuple completion tagging layer. (alkaline pH, increases, activity of TRPV5/V6 channels in Jurkat T cells). where (a) the object should just be …

Content truncated to 500 words