Unsupervised Does Not Mean Uninterpretable: The Case for Word Sense Induction and Disambiguation

Proceedings of the 15th Conference of the European Chapter of the Association for Computational Linguistics: Volume 1, Long Papers, pages 86–98, Valencia, Spain, April 3-7, 2017. c⃝2017 Association for Computational Linguistics Unsupervised Does Not Mean Uninterpretable: The Case for Word Sense Induction and Disambiguation Alexander Panchenko‡, Eugen Ruppert‡, Stefano Faralli†, Simone Paolo Ponzetto† and Chris Biemann‡ ‡Language Technology Group, Computer Science Dept., University of Hamburg, Germany †Web and Data Science Group, Computer Science Dept., University of Mannheim, Germany {panchenko,ruppert,biemann}@informatik.uni-hamburg.de {faralli,simone}@informatik.uni-mannheim.de Abstract The current trend in NLP is the use of highly opaque models, e.g. works and word embeddings.neural net- While these models yield state-of-the-art results on a range of tasks, their drawback is poor interpretability. On the example of word sense induction and disambigua- tion (WSID), we show that it is possi- ble to develop an interpretable model that matches the state-of-the-art models in ac- curacy. Namely, we present an unsuper- vised, knowledge-free WSID approach, which is interpretable at three levels: word sense inventory, sense feature representa- tions, and disambiguation procedure. Ex- periments show that our model performs on par with state-of-the-art word sense embeddings and other unsupervised sys- tems while offering the possibility to jus- tify its decisions in human-readable form. 1 Introduction A word sense disambiguation (WSD) system takes as input a target word t and its context C. The sys- tem returns an identifier of a word sense si from the word sense inventory {s1, ..., sn} of t, where the senses are typically defined manually in ad- vance. Despite significant progress in methodol- ogy during the two last decades (Ide and V´eronis, 1998; Agirre and Edmonds, 2007; Moro and Nav- igli, 2015), WSD is still not widespread in appli- cations (Navigli, 2009), which indicates the need for further progress. The difficulty of the prob- lem largely stems from the lack of domain-specific training data. A fixed sense inventory, such as the one of WordNet (Miller, 1995), may contain irrel- evant senses for the given application and at the same time lack relevant domain-specific senses. Word sense induction from domain-specific cor- pora is a supposed to solve this problem. How- ever, most approaches to word sense induction and disambiguation, e.g. (Sch¨utze, 1998; Li and Juraf- sky, 2015; Bartunov et al., 2016), rely on cluster- ing methods and dense vector representations that make a WSD model uninterpretable as compared to knowledge-based WSD methods. Interpretability of a statistical model is impor- tant as it lets us understand the reasons behind its predictions (Vellido et al., 2011; Freitas, 2014; Li et al., 2016). Interpretability of WSD models (1) lets a user understand why in the given context one observed a given sense (e.g., for educational appli- cations); (2) performs a comprehensive analysis of correct and erroneous predictions, giving rise to improved disambiguation models. The contribution of this paper is an interpretable unsupervised knowledge-free WSD method. The novelty of our method is in (1) a technique to dis- ambiguation that relies on induced inventories as a pivot for learning sense feature …

Content truncated to 500 words