FastSense: An Efficient Word Sense Disambiguation Classifier

fastSense: An Efficient Word Sense Disambiguation Classifier Tolga Uslu, Alexander Mehler, Daniel Baumartz, Wahed Hemati TTLab, Goethe University Frankfurt Frankfurt, Germany {uslu, mehler, baumartz, hemati}@em.uni-frankfurt.de Abstract The task of Word Sense Disambiguation (WSD) is to determine the meaning of an ambiguous word in a given context. In spite of its importance for most NLP pipelines, WSD can still be seen to be unsolved. The reason is that we currently lack tools for WSD that handle big data – “big” in terms of the number of ambiguous words and in terms of the overall number of senses to be distinguished. This desideratum is exactly the objective of fastSense, an efficient neural network-based tool for word sense disambiguation introduced in this paper. We train and test fastSense by means of the disambiguation pages of the German Wikipedia. In addition, we evaluate fastSense in the context of Senseval and SemEval. By reference to Senseval and SemEval we additionally perform a parameter study. We show that fastSense can process huge amounts of data quickly and also surpasses state-of-the-art tools in terms of F-measure. Keywords: WSD, Big Data, Wikipedia 1. Introduction One of the core tasks in natural language processing is Word Sense Disambiguation. Without disambiguation, we just consider a word as a combination of characters and not the meaning behind it. Without properly disambiguating the lexical constituents of a text, it is almost impossible to process its content automatically. Our goal is to solve this problem and to make it applicable to large amounts of data. To this end, we present a neural network-based classifier for WSD called fastSense. We take sequences of words as input and compute a sense label per ambiguous word in that sequence as output. This approach was mo- tivated by the classifier called fastText (Joulin et al., 2016). As the name suggests, fastText is designed to perform text classifications as quickly as possible. How- ever, fastText is not suitable for disambiguating words. In addition, the neural network used by fastText does not support training of multi-labels. Therefore, we imple- mented our own word embedding-based neural network by analogy to the architecture of fastText. This allows us to apply fastSense to WSD efficiently even on big data. In order to test the time complexity of our approach, we cre- ated a disambiguation corpus from the German Wikipedia with over 50,000,000 training and test sets. We use the disambiguation pages and the link structure of Wikipedia to match words with their corresponding Wikipedia senses. In this paper, we deal with the German Wikipedia. In terms of size or space complexity, its sense model is far beyond what is normally studied, for example, in the framework of Senseval or SemEval. However, in order to show that our approach is language independent, we additionally perform multiple tests related to Senseval and SemEval. These tests show that our model keeps up with state-of-the-art tools by reaching 73.47% at Senseval-2, 73.48% at Senseval-3 and up to 87,57% on SemEval 2007 tasks. The paper is structured …

Content truncated to 500 words