Negative Focus Detection via Contextual Attention Mechanism

Proceedings of the 2019 Conference on Empirical Methods in Natural Language Processing and the 9th International Joint Conference on Natural Language Processing, pages 2251–2261, Hong Kong, China, November 3–7, 2019. c⃝2019 Association for Computational Linguistics 2251 Negative Focus Detection via Contextual Attention Mechanism Longxiang Shen1, Bowei Zou12∗, Yu Hong1, Qiaoming Zhu1, Guodong Zhou1, Ai Ti Aw2 1School of Computer Scienceand Technology, Soochow University, China 2Aural & Language Intelligence Department, Institute for Infocomm Research, Singapore lxshen.scu@gmail.com, zoubowei@suda.edu.cn, {yhong, qmzhu, gdzhou}@suda.edu.cn, aaiti@i2r.a-star.edu.sg Abstract Negation is a universal but complicated lin- guistic phenomenon, which has received con- siderable attention from the NLP community over the last decade, since a negated state- ment often carries both an explicit negative fo- cus and implicit positive meanings. For the sake of understanding a negated statement, it is critical to precisely detect the negative fo- cus in context. However, how to capture con- textual information for negative focus detec- tion is still an open challenge. To well ad- dress this, we come up with an attention-based neural network to model contextual informa- tion. In particular, we introduce a framework which consists of a Bidirectional Long Short- Term Memory (BiLSTM) neural network and a Conditional Random Fields (CRF) layer to effectively encode the order information and the long-range context dependency in a sen- tence. Moreover, we design two types of attention mechanisms, word-level contextual attention and topic-level contextual attention, to take advantage of contextual information across sentences from both the word perspec- tive and the topic perspective, respectively. Experimental results on the SEM’12 shared task corpus show that our approach achieves the best performance on negative focus de- tection, yielding an absolute improvement of 2.11% over the state-of-the-art. This demon- strates the great effectiveness of the two types of contextual attention mechanisms. 1 Introduction As a linguistic phenomenon that reverses the po- larity of a statement or its property into opposite, negation is ubiquitous in human languages. Con- sider following sentence S1. A negative affix -n’t negates the statement that mutual fund trades take effect until the market close. Negation process- ing has been shown critical for a number of NLP ∗ corresponding auther applications, such as biomedical information ex- traction (Morante, 2010; Mehrabi et al., 2015) and sentiment analysis (Wiegand et al., 2010; Jimenez- Zafra et al., 2017). (S1) Mutual fund trades don’t take effect {until the market close}.1 In general, negation is marked by a word or an affix and interact with other parts in sen- tence. Blanco and Moldovan (2011) demonstrated that over 65% of negated statements in Prop- Bank (Palmer et al., 2005) convey implicit positive meanings. For instance, underneath sentence S1, the statement that mutual fund trades take effect is true, when removing the propositional clause until the market close from the original sentence. As the definition of negation focus in SEM’12 (Morante and Blanco, 2012), a negative focus is the most prominently negated part of sentence (Huddleston and Pullum, 2002). In sentence S1, the negative focus is the propositional clause until the …

Content truncated to 500 words