Learning Event Expressions via Bilingual Structure Projection

Proceedings of COLING 2016, the 26th International Conference on Computational Linguistics: Technical Papers, pages 1441–1450, Osaka, Japan, December 11-17 2016. Learning Event Expressions via Bilingual Structure Projection Fangyuan Li1, Ruihong Huang2, Deyi Xiong1∗, Min Zhang1 Soochow University, Suzhou, China1 Texas A&M University, College Station, USA2 fyli@stu.suda.edu.cn, huangrh@cse.tamu.edu {dyxiong, minzhang}@suda.edu.cn Abstract Identifying events of a specific type is a challenging task as events in texts are described in numer- ous and diverse ways. Aiming to resolve high complexities of event descriptions, previous work (Huang and Riloff, 2013) proposes multi-faceted event recognition and a bootstrapping method to automatically acquire both event facet phrases and event expressions from unannotated texts. However, to ensure high quality of learned phrases, this method is constrained to only learn phrases that match certain syntactic structures. In this paper, we propose a bilingual structure projection algorithm that explores linguistic divergences between two languages (Chinese and English) and mines new phrases with new syntactic structures, which have been ignored in the previous work. Experiments show that our approach can successfully find novel event phrases and structures, e.g., phrases headed by nouns. Furthermore, the newly mined phrases are capable of recognizing additional event descriptions and increasing the recall of event recognition. 1 Introduction Event recognition aims to identify documents that describe a specific type of event. Accurate event recognition is challenging due to ambiguities of event keywords. In the previous work, Huang and Riloff (2013) (hereafter H&R) proposed multi-faceted event recognition method that uses event expressions as well as event defining characteristics (aka “event facets”, such as “agents” and “purpose”) to achieve high accuracy in identifying civil unrest events. They also presented a bootstrapping solution that can learn event expressions and event facet phrases from unannotated texts. However, to achieve high quality phrases, strict syntactic constraints have been enforced and their bootstrapping algorithm can only learn two particular types of V-O (Verb-Object) Structure for both event expressions and facet phrases. Obvi- ously, diverse forms of other verb phrases and non-verb phrases exist to describe events and are ignored by the proposed algorithm. For instance, a verb phrase where two verbs are connected with a particular dependency relation “xcomp”1, (e.g., “came out to demonstrate”) is one of these structures. Civil unrest events can also be invoked by some noun structure phrases, such as just a noun word phrase (e.g., “sit- ins”) or phrases starting with a noun (e.g., “disobedience of order”), even a passive form phrase structure like “rallies held (in)”. In order to address this issue, we propose a simple yet effective bilingual structure projection method that explores syntactic divergences (Georgi et al., 2012) between two languages and mines new syntactic structures for event expressions and event facet phrases effectively using parallel corpora. This is inspired by many recent cross-lingual research that utilize the second language to provide a different view (Balcan and Blum, 2005; Burkett et al., 2010; Ganchev et al., 2012) and complementary cues (Che et al., 2013; Wang et al., 2013) in improving Natural language Processing (NLP) …

Content truncated to 500 words