SeCoDa: Sense Complexity Dataset

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 5962–5967 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 5962 SeCoDa: Sense Complexity Dataset David Strohmaier, Sian Gooding, Shiva Taslimipoor, Ekaterina Kochmar University of Cambridge, Department of Computer Science and Technology, ALTA Institute Cambridge, CB3 0FD {ds858, shg36, st797, ek358}@cam.ac.uk Abstract The Sense Complexity Dataset (SeCoDa) provides a corpus that is annotated jointly for complexity and word senses. It thus provides a valuable resource for both word sense disambiguation and the task of complex word identification. The intention is that this dataset will be used to identify complexity at the level of word senses rather than word tokens. For word sense annotation, SeCoDa uses a hierarchical scheme that is based on information available in the Cambridge Advanced Learner’s Dictionary. This way, we can offer more coarse-grained senses than directly available in WordNet. Keywords: word sense disambiguation, lexical complexity, dataset 1. Introduction Both word sense disambiguation (WSD) and complex word identification (CWI) are long established tasks in natural language processing (Navigli, 2009; Iacobacci et al., 2016), with CWI focusing on the detection of words that might be considered complex by readers and therefore in need of simplification (Shardlow, 2013). It has been shown that a number of NLP tasks benefit from WSD, however so far CWI and WSD have been kept distinct. The complexity differences between senses as opposed to the complexity of word tokens is yet to be specifically investigated or de- tected. The Sense Complexity Dataset (SeCoDa) fills this gap by providing both complexity and sense information for word tokens. We release this dataset to open the door for a new combined task of complex sense detection.1 The SeCoDa also makes a contribution to word sense dis- ambiguation research by moving beyond WordNet senses. Instead of WordNet, it uses more coarse-grained sense information found in the Cambridge Advanced Learner’s Dictionary (CALD).2 2. Motivation Our dataset is the first to combine word senses with com- plexity information. We argue that the same word with mul- tiple senses may differ in its complexity when used in each of these senses. For example, the sense that the word driver takes in car driver may be deemed less complex than when it is used in the computer science-related sense. In every- day contexts, explaining what a driver in computer science does is considerably more difficult than what a car driver does. Examples from the dataset of Yimam et al. (2017) annotated with complexity scores support this idea. For in- stance, consider the following sentence: Successive waves of bank sector clean-ups have failed to convince investors 1The dataset can be found at https://github.com/ dstrohmaier/SeCoDa. 2https://dictionary.cambridge.org/ dictionary/english/ Waves in this context is used in the sense of “a larger than usual number of events of a similar, often bad, type, hap- pening within the same period”. This sense is less frequent than “a raised line of water that moves across the surface of an area of water, …

Content truncated to 500 words