Bilingual Active Learning for Relation Classification via Pseudo Parallel Corpora

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 582–592, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Bilingual Active Learning for Relation Classification via Pseudo Paral- lel Corpora Longhua Qian Haotian Hui Ya’nan Hu Guodong Zhou* Qiaoming Zhu Natural Language Processing Lab School of Computer Science and Technology, Soochow University 1 Shizi Street, Suzhou, China 215006 {qianlonghua,20134227019,20114227025,gdzhou,qmzhu}@suda.edu.cn Abstract Active learning (AL) has been proven ef- fective to reduce human annotation ef- forts in NLP. However, previous studies on AL are limited to applications in a single language. This paper proposes a bilingual active learning paradigm for re- lation classification, where the unlabeled instances are first jointly chosen in terms of their prediction uncertainty scores in two languages and then manually labeled by an oracle. Instead of using a parallel corpus, labeled and unlabeled instances in one language are translated into ones in the other language and all instances in both languages are then fed into a bilin- gual active learning engine as pseudo parallel corpora. Experimental results on the ACE RDC 2005 Chinese and English corpora show that bilingual active learn- ing for relation classification signifi- cantly outperforms monolingual active learning. 1 Introduction Semantic relation extraction between named en- tities (aka. entity relation extraction or more con- cisely relation extraction) is an important subtask of Information Extraction (IE) as well as Natural Language Processing (NLP). With its aim to identify and classify the semantic relationship between two entities (ACE 2002-2007), relation extraction is of great significance to many NLP applications, such as question answering, infor- mation fusion, social network construction, and knowledge mining and population etc. * Corresponding author In the literature, the mainstream research on relation extraction adopts statistical machine learning methods, which can be grouped into supervised learning (Zelenko et al., 2003; Culotta and Soresen, 2004; Zhou et al., 2005; Zhang et al., 2006; Qian et al., 2008; Chan and Roth, 2011), semi-supervised learning (Zhang et al., 2004; Chen et al., 2006; Zhou et al., 2008; Qian et al., 2010) and unsupervised learning (Hase- gawa et al., 2004; Zhang et al., 2005) in terms of the amount of labeled training data they need. Usually the extraction performance depends heavily on the quality and quantity of the labeled data, however, the manual annotation of a large- scale corpus is labor-intensive and time- consuming. In the last decade researchers have turned to another effective learning paradigm-- active learning (AL), which, given a small num- ber of labeled instances and a large number of unlabeled instances, selects the most informative unlabeled instances to be manually annotated and add them into the training data in an iterative fashion. Essentially active learning attempts to decrease the quantity of labeled instances by en- hancing their quality, gauged by their informa- tiveness to the learner. Since its emergence, ac- tive learning has been successfully applied to many tasks in NLP (Engelson and Dagan, 1996; Hwa, 2004; Tomanek et al., 2007; Settles and Craven, 2008). It is trivial …

Content truncated to 500 words