Multi-Label Transfer Learning for Multi-Relational Semantic Similarity

Proceedings of the Eighth Joint Conference on Lexical and Computational Semantics (*SEM), pages 44–50 Minneapolis, June 6–7, 2019. c⃝2019 Association for Computational Linguistics 44 Multi-Label Transfer Learning for Multi-Relational Semantic Similarity Li Zhang and Steven R. Wilson and Rada Mihalcea Computer Science and Engineering University of Michigan {zharry,steverw,mihalcea}@umich.edu Abstract Multi-relational semantic similarity datasets define the semantic relations between two short texts in multiple ways, e.g., similarity, relatedness, and so on. Yet, all the systems to date designed to capture such relations tar- get one relation at a time. We propose a multi-label transfer learning approach based on LSTM to make predictions for several rela- tions simultaneously and aggregate the losses to update the parameters. This multi-label re- gression approach jointly learns the informa- tion provided by the multiple relations, rather than treating them as separate tasks. Not only does this approach outperform the single-task approach and the traditional multi-task learn- ing approach, it also achieves state-of-the-art performance on all but one relation of the Hu- man Activity Phrase dataset. 1 Introduction Semantic similarity, or relating short texts or sen- tences1 in a semantic space – be those phrases, sentences or short paragraphs – is a task that re- quires systems to determine the degree of equiv- alence between the underlying semantics of the two sentences. Although relatively easy for hu- mans, this task remains one of the most difficult natural language understanding problems. The task has been receiving significant interest from the research community. For instance, from 2012 to 2017, the International Workshop on Seman- tic Evaluation (SemEval) has been holding the Semantic Textual Similarity (STS) shared tasks (Agirre et al., 2012, 2013b, 2015, 2016; Cer et al., 2017), dedicated to tackling this problem, with close to 100 team submissions each year. In some semantic similarity datasets, an exam- ple consists of a sentence pair and a single anno- tated similarity score, while in others, each pair 1In this work, we do not consider word level similarity. comes with multiple annotations. We refer to the latter as multi-relational semantic similarity tasks. The inclusion of multiple annotations per exam- ple is motivated by the fact that there can be dif- ferent relations, namely different types of simi- larity between two sentences. So far, these rela- tions have been treated as separate tasks, where a model trains and tests on one relation at a time while ignoring the rest. However, we hypothesize that each relation may contain useful information about the others, and training on only one rela- tion inevitably neglects some relevant information. Thus, training jointly on multiple relations may improve performance on one or more relations. We propose a joint multi-label transfer learning setting based on LSTM, and show that it can be an effective solution for the multi-relational semantic similarity tasks. Due to the small size of multi- relational semantic similarity datasets and the re- cent success of LSTM-based sentence representa- tions (Wieting and Gimpel, 2018; Conneau et al., 2017), the model is pre-trained on a large corpus and …

Content truncated to 500 words