Transductive Non-linear Learning for Chinese Hypernym Prediction

Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1394–1404 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1128 Proceedings of the 55th Annual Meeting of the Association for Computational Linguistics, pages 1394–1404 Vancouver, Canada, July 30 - August 4, 2017. c⃝2017 Association for Computational Linguistics https://doi.org/10.18653/v1/P17-1128 Transductive Non-linear Learning for Chinese Hypernym Prediction Chengyu Wang1, Junchi Yan2,1, Aoying Zhou1, Xiaofeng He1∗ 1 Shanghai Key Laboratory of Trustworthy Computing, East China Normal University 2 IBM Research – China chywang2013@gmail.com, yanesta13@163.com {ayzhou,xfhe}@sei.ecnu.edu.cn Abstract Finding the correct hypernyms for entities is essential for taxonomy learning, fine- grained entity categorization, knowledge base construction, etc. Due to the flexi- bility of the Chinese language, it is chal- lenging to identify hypernyms in Chinese accurately. Rather than extracting hyper- nyms from texts, in this paper, we present a transductive learning approach to es- tablish mappings from entities to hyper- nyms in the embedding space directly. It combines linear and non-linear embedding projection models, with the capacity of encoding arbitrary language-specific rules. Experiments on real-world datasets illus- trate that our approach outperforms pre- vious methods for Chinese hypernym pre- diction. 1 Introduction A hypernym of an entity characterizes the type or the class of the entity. For example, the word country is the hypernym of the entity Canada. The accurate prediction of hypernyms benefits a variety of NLP tasks, such as taxonomy learn- ing (Wu et al., 2012; Fu et al., 2014), fine-grained entity categorization (Ren et al., 2016), knowledge base construction (Suchanek et al., 2007), etc. In previous work, the detection of hypernyms requires lexical, syntactic and/or semantic analy- sis of relations between entities and their respec- tive hypernyms from a language-specific knowl- edge source. For example, Hearst (1992) is the pi- oneer work to extract is-a relations from a text cor- pus based on handcraft patterns. The following- up work mostly focuses on is-a relation extrac- tion using automatically generated patterns (Snow ∗Corresponding author. et al., 2004; Ritter et al., 2009; Sang and Hof- mann, 2009; Kozareva and Hovy, 2010) and re- lation inference based on distributional similar- ity measures (Kotlerman et al., 2010; Lenci and Benotto, 2012; Shwartz et al., 2016). While these approaches have relatively high precision over English corpora, extracting hy- pernyms for entities is still challenging for Chi- nese. From the linguistic perspective, Chinese is a lower-resourced language with very flexible expressions and grammatical rules (Wang et al., 2015). For instance, there are no word spaces, ex- plicit tenses and voices, and distinctions between singular and plural forms in Chinese. The order of words can be changed flexibly in sentences. Hence, as previous research indicates, hypernym extraction methods for English are not necessarily suitable for the Chinese language (Fu et al., 2014; Wang et al., 2015; Wang and He, 2016). Based on such conditions, several classification methods are proposed to distinguish is-a and not- is-a relations based on Chinese encyclopedias (Lu et al., 2015; Li et al., 2015). Similar to Prince- …

Content truncated to 500 words