Learning Relatedness between Types with Prototypes for Relation Extraction

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 2011–2016 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 2011 Learning Relatedness between Types with Prototypes for Relation Extraction Lisheng Fu∗ Ralph Grishman Computer Science Department New York University New York, NY 10003, USA {lisheng, grishman}@cs.nyu.edu Abstract Relation schemas are often pre-defined for each relation dataset. Relation types can be related from different datasets and have over- lapping semantics. We hypothesize we can combine these datasets according to the se- mantic relatedness between the relation types to overcome the problem of lack of training data. It is often easy to discover the con- nection between relation types based on re- lation names or annotation guides, but hard to measure the exact similarity and take ad- vantage of the connection between the rela- tion types from different datasets. We pro- pose to use prototypical examples to represent each relation type and use these examples to augment related types from a different dataset. We obtain further improvement (ACE05) with this type augmentation over a strong base- line which uses multi-task learning between datasets to obtain better feature representa- tion for relations. We make our implementa- tion publicly available: https://github. com/fufrank5/relatedness 1 Introduction Relation extraction identifies specific semantic re- lationships between two entities within a single sentence. For example, there is a Physical.Located relationship between George Bush and France in the sentence: George Bush traveled to France on Thursday for a summit. Relation extraction is a crucial task for many applications such as knowl- edge base population. Relation schemas are mostly pre-defined in ex- isting datasets. The definition of the relation type depends on the annotation guide. There is no clear intrinsic Ontology for relation types. In practice, relation types can be created based on interests. ∗This is the work that the author has done before joining Amazon Alexa AI. This leaves datasets with similar, related or over- lapping schemas. For example, the annotation guides for Automatic Content Extraction (ACE) 03-05 changed from year to year. The later cre- ated Entities, Relations and Events (ERE) dataset was similar in the schema, but differs in details. Because of the difficulty of annotating relations, these datasets are all small individually and hard to be utilized together. It is not an easy task to learn relatedness be- tween relation schemas across different datasets since there is no instance-level labels available for the relatedness. However, we can observe the con- nections between the relation types from differ- ent datasets based on relation names or annota- tion guides. We propose to simplify the related- ness as binary (related or not) and to use manual review of relation names to decide the relatedness labels. This would give the prior knowledge that one relation type in one dataset may have closer relationships to some types than the others in an- other dataset. Then we design a model to rec- ognize this similarity. We propose to use proto- typical examples to represent each …

Content truncated to 500 words