Chasing Hypernyms in Vector Spaces with Entropy

Abstract

In this paper, we introduce SLQS, a new entropy-based measure for the unsupervised identification of hypernymy and its directionality in Distributional Semantic Models (DSMs). SLQS is assessed through two tasks: (i.) identifying the hypernym in hyponym-hypernym pairs, and (ii.) discriminating hypernymy among various semantic relations. In both tasks, SLQS outperforms other state-of-the-art measures.

A Short Survey on Taxonomy Learning from Text Corpora: Issues, Resources and Recent Advances

“…More recently, several studies suggest that DIH is not correct for all the cases (Santus et al, 2014;Rimell, 2014). For example, "American" is a hypernym of "Barack Obama" but the (politicsrelated) contexts of "Barack Obama" cannot be covered by those of "American".…”

SemEval-2018 Task 9: Hypernym Discovery

“…Earlier work on hypernym modeling was unsupervised, and leveraged various interpretations of the distributional hypothesis. 1 Most of the recent work on the subject is however supervised, and in the main based on using word embeddings as input for classification or prediction (e.g Baroni et al, 2012;Santus et al, 2014;Fu et al, 2014;Weeds et al, 2014;Sanchez Carmona and Riedel, 2017;Nguyen et al, 2017). As shown by Shwartz et al (2016), pattern-based and distributional evidences can be effectively combined within a neural architecture.…”

Dual Tensor Model for Detecting Asymmetric Lexico-Semantic Relations

“…Distributional models come in both unsupervised and supervised flavors. Unsupervised metrics for hypernymy detection assume either that the hyponym's contexts are included in the hypernym's contexts (Weeds and Weir, 2003;Kotlerman et al, 2010) or that the linguistics contexts of a hyponym are more informative than the contexts of its hypernyms (Rimell, 2014;Santus et al, 2014). Supervised hypernymy classifiers represent the pair of words by combining their distributional vectors in different ways -concatenating them (Baroni et al, 2012) or subtracting them (Roller et al, 2014) and feeding the resulting vector to a supervised classifier like logistic regression.…”

“…Distributional models predict the hypernymy relations by combining raw distributional vectors of concepts in a pair (Baroni et al, 2012;Roller et al, 2014;Santus et al, 2014), whereas path-based models base predictions on lexico-syntactic paths from co-occurrence contexts obtained from a large corpus (Snow et al, 2004;Nakashole et al, 2012;. combine the path-based and distributional models to reach state-of-the-art performance in hypernymy detection.…”