Learning Antonyms with Paraphrases and a Morphology-Aware Neural Network

Proceedings of the 6th Joint Conference on Lexical and Computational Semantics (*SEM 2017), pages 12–21, Vancouver, Canada, August 3-4, 2017. c⃝2017 Association for Computational Linguistics Learning Antonyms with Paraphrases and a Morphology-Aware Neural Network Sneha Rajana∗ Chris Callison-Burch∗ Marianna Apidianaki∗Ψ Vered ShwartzΦ ∗Computer and Information Science Department, University of Pennsylvania, USA ΨLIMSI, CNRS, Université Paris-Saclay, 91403 Orsay ΦComputer Science Department, Bar-Ilan University, Israel {srajana,ccb,marapi}@seas.upenn.edu vered1986@gmail.com Abstract Recognizing and distinguishing antonyms from other types of semantic relations is an essential part of language understand- ing systems. In this paper, we present a novel method for deriving antonym pairs using paraphrase pairs containing negation markers. We further propose a neural net- work model, AntNET, that integrates mor- phological features indicative of antonymy into a path-based relation detection algo- rithm. We demonstrate that our model outperforms state-of-the-art models in dis- tinguishing antonyms from other semantic relations and is capable of efficiently han- dling multi-word expressions. 1 Introduction Identifying antonymy and expressions with con- trasting meanings is valuable for NLP systems which go beyond recognizing semantic related- ness and require to identify specific semantic re- lations. While manually created semantic tax- onomies, like WordNet (Fellbaum, 1998), define antonymy relations between some word pairs that native speakers consider antonyms, they have lim- ited coverage. Further, as each term of an antony- mous pair can have many semantically close terms, the contrasting word pairs far outnum- ber those that are commonly considered antonym pairs, and they remain unrecorded. Therefore, automated methods have been proposed to deter- mine for a given term-pair (x, y), whether x and y are antonyms of each other, based on their occur- rences in a large corpus. Charles and Miller (1989) put forward the co- occurrence hypothesis that antonyms occur to- gether in a sentence more often than chance. How- ever, non-antonymous semantically related words Paraphrase Pair Antonym Pair not sufficient/insufficient sufficient/insufficient insignificant/negligible significant/negligible dishonest/lying honest/lying unusual/pretty strange usual/pretty strange Table 1: Examples of antonyms derived from PPDB paraphrases. The antonym pairs in column 2 were derived from the corresponding paraphrase pairs in column 1. such as hypernyms, holonyms, meronyms, and near-synonyms also tend to occur together more often than chance. Thus, separating antonyms from pairs linked by other relationships has proven to be difficult. Approaches to antonym detec- tion have exploited distributional vector represen- tations relying on the distributional hypothesis of semantic similarity (Harris, 1954; Firth, 1957) that words co-occurring in similar contexts tend to be semantically close. Two main information sources are used to recognize semantic relations: path- based and distributional. Path-based methods con- sider the joint occurrences of the two terms in a given sentence and use the dependency paths that connect the terms as features (Hearst, 1992; Roth and Schulte im Walde, 2014; Schwartz et al., 2015). For distinguishing antonyms from other re- lations, Lin et al. (2003) proposed to use antonym patterns (such as either X or Y and from X to Y ). Distributional methods are based on the dis- joint occurrences of each term and …

Content truncated to 500 words