Stream-Based Lookup of Context-Sensitive Lexical Information

This book deals with the problem that current algorithms for word sense disambiguation are too slow to use them in a writing assistance environment in real-time. In order to solve this problem, I use concepts of stream processing to enhance existing algorithms in order to analyse a text which is constantly changing. As an example I extend the simplified Lesk and a graph centrality WSD algorithms and test them in a simulation environment which I build specifically for evaluating WSD algorithms regarding run-time and quality for growing texts. The algorithms do not have to start a new analysis each time a sentence changes. It is possible to just consider the recent changes and merge them with the previous results. I was able to prove that the new algorithms are significantly faster than their original variants without changing the outcome. It was possible to speed up the simplified Lesk algorithm by 27.6% and the graph centrality WSD by 44.6% in terms of computation time.