HyperBox: A Supervised Approach for Hypernym Discovery using Box Embeddings

Proceedings of the 13th Conference on Language Resources and Evaluation (LREC 2022), pages 6069–6076 Marseille, 20-25 June 2022 © European Language Resources Association (ELRA), licensed under CC-BY-NC-4.0 6069 HyperBox: A Supervised Approach for Hypernym Discovery using Box Embeddings Maulik Parmar, Dr. Apurva Narayan Independent Researcher, University of British Columbia British Columbia, Canada, British Columbia, Canada maulikres@gmail.com, apurva.narayan@ubc.ca Abstract Hypernymy plays a fundamental role in many AI tasks like taxonomy learning, ontology learning, etc. This has motivated the development of many automatic identification methods for extracting this relation, most of which rely on word distribution. We present a novel model HyperBox to learn box embeddings for hypernym discovery. Given an input term, HyperBox retrieves its suitable hypernym from a target corpus. For this task, we use the dataset published for SemEval 2018 Shared Task on Hypernym Discovery. We compare the performance of our model on two specific domains of knowledge: medical and music. Experimentally, we show that our model outperforms existing methods on the majority of the evaluation metrics. Moreover, our model generalize well over unseen hypernymy pairs using only a small set of training data. Keywords: Box Embeddings, Hypernym Discovery, Hypernym relation 1. Introduction In linguistics, hypernymy is a semantic relation be- tween a hypernym denoting a superordinate and a hy- ponym denoting a subordinate. Hypernymy is a major semantic relation and a vital organization principle of semantic memory (Miller and Fellbaum, 1991). It is an asymmetric relation between a hypernym (supertype) and a hyponym (subtype), as in animal-dog and sport- tennis. Figure 1 shows some examples of a hypernym class along with some of their hyponyms. As we can see from Figure 1, hypernyms are a more general class of hyponym terms. It plays a crucial role in language understanding because it enables generalization, which lies at the core of human cognition. Therefore, it has been an active area of research in NLP for decades. Automatic hypernym discovery is useful in many tasks like taxonomy creation (Snow et al., 2006; Navigli et al., 2011), recognizing textual entailment (Dagan et al., 2013), and text generation (Biran and McKeown, 2013). In this paper, we tackle the problem of hypernym dis- covery (Espinosa-Anke et al., 2016) instead of hyper- nym detection (Shwartz et al., 2017). Generally, eval- uation benchmarks for modeling hypernymy are such that they are reduced to a binary classification task where one tries to predict if a hypernymy relation ex- ists between candidate pairs. Hypernym detection uses this experimental setting and tends to suffer from lexi- cal memorization phenomena (Levy et al., 2015) due to the inherent modeling of the datasets by supervised sys- tems. Thus, to alleviate this problem (Espinosa-Anke et al., 2016) proposed to frame the problem as Hypernym Discovery i.e given a query term and search space of domain vocabulary, discover the best candidate hyper- nyms of input hyponym. This reformulation not only helps in alleviating the issues discussed above but also Figure 1: Examples of Hypernym-Hyponym pairs helps to use it with other downstream applications …

Content truncated to 500 words