Multi-facet Universal Schema

Proceedings of the 16th Conference of the European Chapter of the Association for Computational Linguistics, pages 909–919 April 19 - 23, 2021. ©2021 Association for Computational Linguistics 909 Multi-facet Universal Schema Rohan Paul∗ Haw-Shiuan Chang∗ Andrew McCallum CICS, University of Massachusetts Amherst {rohpaul,hschang,mccallum}@cs.umass.edu Abstract Universal schema (USchema) assumes that two sentence patterns that share the same en- tity pairs are similar to each other. This as- sumption is widely adopted for solving various types of relation extraction (RE) tasks. Nev- ertheless, each sentence pattern could contain multiple facets, and not every facet is similar to all the facets of another sentence pattern co- occurring with the same entity pair. To address the violation of the USchema assumption, we propose multi-facet universal schema that uses a neural model to represent each sentence pat- tern as multiple facet embeddings and encour- age one of these facet embeddings to be close to that of another sentence pattern if they co- occur with the same entity pair. In our ex- periments, we demonstrate that multi-facet em- beddings significantly outperform their single- facet embedding counterpart, compositional universal schema (CUSchema) (Verga et al., 2016), in distantly supervised relation extrac- tion tasks. Moreover, we can also use multiple embeddings to detect the entailment relation between two sentence patterns when no man- ual label is available. 1 Introduction Relation extraction (RE) is a crucial step in au- tomatic knowledge base construction (AKBC). A major challenge of RE is that the frequency of rela- tions in the real world is a long-tail distribution but collecting sufficient human annotations for every relation is infeasible (Han et al., 2020). Distant supervision is proposed to alleviate the issue (Mintz et al., 2009). Distant supervision as- sumes that a sentence pattern expresses a relation if the sentence pattern co-occurs with an entity pair and the entity pair has the relation. For example, we assume the sentence pattern “$ARG1, the partner of fellow $ARG2” is likely to express the spouse ∗* indicates equal contribution Training Data Neural Encoder Neural Encoder Baseline: Compositional Universal Schema Close Close Neural Encoder and Decoder Neural Encoder and Decoder Pattern Embedding 5 Entity Pair Embedding Close Close Multi-facet Universal Schema Table Lookup Table Lookup Pattern Embedding 4 Pattern Embedding 3 Pattern Embedding 2 Pattern Embedding 1 Pattern Embedding 5 Pattern Embedding 4 Pattern Embedding 3 Pattern Embedding 2 Pattern Embedding 1 Entity Pair Embedding Pattern Embedding Pattern Embedding co-occur Sentence Pattern $ARG1 moved in with $ARG2 Sentence Pattern $ARG1, the partner of fellow $ARG2 Entity Pair $ARG1 = Angelina Jolie $ARG2 = Brad Pitt co-occur Figure 1: Comparison between the multi-facet and compositional universal schema. In our training loss, we encourage one of the facet embeddings from a sen- tence pattern to be similar to its co-occurred entity pair. relation if we observe a text clip “... Angelina Jolie, the partner of fellow Brad Pitt ...” in our training corpus and a knowledge base tells us that Angelina Jolie and Brad Pitt has the spouse relation. Ac- cordingly, we …

Content truncated to 500 words