ABDN at SemEval-2018 Task 10: Recognising Discriminative Attributes using Context Embeddings and WordNet

Proceedings of the 12th International Workshop on Semantic Evaluation (SemEval-2018), pages 1017–1021 New Orleans, Louisiana, June 5–6, 2018. ©2018 Association for Computational Linguistics ABDN at SemEval-2018 Task 10: Recognising Discriminative Attributes using Context Embeddings and WordNet Rui Mao, Guanyi Chen, Ruizhe Li and Chenghua Lin Computing Science Department University of Aberdeen Aberdeen, Scotland, UK {r03rm16, g.chen, ruizhe.li, chenghua.lin}@abdn.ac.uk Abstract This paper describes the system that we sub- mitted for SemEval-2018 task 10: capturing discriminative attributes. Our system is built upon a simple idea of measuring the attribute word’s similarity with each of the two seman- tically similar words, based on an extended word embedding method and WordNet. In- stead of computing the similarities between the attribute and semantically similar words by using standard word embeddings, we pro- pose a novel method that combines word and context embeddings which can better measure similarities. Our model is simple and effective, which achieves an average F1 score of 0.62 on the test set. 1 Introduction Capturing discriminative attributes is a novel task, which is very different from classical semantic tasks that model similarities in semantics. The task aims to recognise semantic differences be- tween words. Traditional semantic similarity eval- uation tasks were designed for evaluating the qual- ity of word representations based on the fact that words with similar semantics will be close to each other in vector space. Recent state-of-the-art dis- tributed semantic models (Ling et al., 2015; Bo- janowski et al., 2017) inspired by the success of word2vec (Mikolov et al., 2013) gave good per- formance in these similarity measure tasks. Nev- ertheless, how to capture discriminative attributes between semantically similar words is still a chal- lenge for traditional word embedding methods, be- cause these methods are designed to capture simi- lar semantics. We have two observations for the nature of the task and the provided data: (1) only limited data is available for model training; (2) the in- puts of the model are merely isolated words them- selves, which lack context information for apply- ing complex models. Therefore, we propose a novel framework that differentiates two semanti- cally similar words with the attribute word by us- ing their word and context embeddings. We ex- perimented with both Continuous Bag of Words (CBOW) and Skip-gram, demonstrating that using the combination of word and context embeddings outperforms using word embeddings alone. The contribution of this work can be sum- marised as follows. We examine word and con- text embeddings in CBOW and Skip-gram, show- ing that using both word and context embeddings can better measure the co-occurrence of two words in sentences than simply using word embeddings. Hence our similarity measure can recognise the discriminative attributes of two semantically simi- lar words more accurately. 2 System Description Our system is trained based on word and context embedding features as well as WordNet features (Fellbaum, 1998). Before introducing our frame- work in detail, we first introduce the two key tech- nical parts of our framework, i.e., context embed- ding and WordNet. 2.1 Context …

Content truncated to 500 words