Encoding Relation Requirements for Relation Extraction via Joint Inference

Proceedings of the 52nd Annual Meeting of the Association for Computational Linguistics, pages 818–827, Baltimore, Maryland, USA, June 23-25 2014. c⃝2014 Association for Computational Linguistics Encoding Relation Requirements for Relation Extraction via Joint Inference Liwei Chen1, Yansong Feng∗1, Songfang Huang2, Yong Qin2 and Dongyan Zhao1 1ICST, Peking University, Beijing, China 2IBM China Research Lab, Beijing, China chenliwei,fengyansong,zhaodongyan@pku.edu.cn huangsf,qinyong@cn.ibm.com Abstract Most existing relation extraction models make predictions for each entity pair lo- cally and individually, while ignoring im- plicit global clues available in the knowl- edge base, sometimes leading to conflicts among local predictions from different en- tity pairs. In this paper, we propose a joint inference framework that utilizes these global clues to resolve disagree- ments among local predictions. We ex- ploit two kinds of clues to generate con- straints which can capture the implicit type and cardinality requirements of a relation. Experimental results on three datasets, in both English and Chinese, show that our framework outperforms the state-of-the- art relation extraction models when such clues are applicable to the datasets. And, we find that the clues learnt automatically from existing knowledge bases perform comparably to those refined by human. 1 Introduction Identifying predefined kinds of relationship be- tween pairs of entities is crucial for many knowl- edge base related applications(Suchanek et al., 2013). In the literature, relation extraction (RE) is usually investigated in a classification style, where relations are simply treated as isolated class labels, while their definitions or background information are sometimes ignored. Take the relation Capital as an example, we can imagine that this relation will expect a country as its subject and a city as object, and in most cases, a city can be the capital of only one country. All these clues are no doubt helpful, for instance, Yao et al. (2010) explicitly modeled the expected types of a relation’s argu- ments with the help of Freebase’s type taxonomy and obtained promising results for RE. ∗Yansong Feng is the corresponding author. However, properly capturing and utilizing such typing clues are not trivial. One of the hurdles here is the lack of off-the-shelf resources and such clues often have to be coded by human experts. Many knowledge bases do not have a well-defined typing system, let alone fine-grained typing taxonomies with corresponding type recognizers, which are crucial to explicitly model the typing requirements for arguments of a relation, but rather expensive and time-consuming to collect. Similarly, the car- dinality requirements of arguments, e.g., a person can have only one birthdate and a city can only be labeled as capital of one country, should be con- sidered as a strong indicator to eliminate wrong predictions, but has to be coded manually as well. On the other hand, most previous relation ex- tractors process each entity pair (we will use en- tity pair and entity tuple exchangeably in the rest of the paper) locally and individually, i.e., the ex- tractor makes decisions solely based on the sen- tences containing the current entity pair and ig- nores other related …

Content truncated to 500 words