Computational Etymology and Word Emergence

Proceedings of the 12th Conference on Language Resources and Evaluation (LREC 2020), pages 3252–3259 Marseille, 11–16 May 2020 c⃝ European Language Resources Association (ELRA), licensed under CC-BY-NC 3252 Computational Etymology and Word Emergence Winston Wu, David Yarowsky Center for Language and Speech Processing Johns Hopkins University (wswu, yarowsky)@jhu.edu Abstract We developed an extensible, comprehensive Wiktionary parser that improves over several existing parsers. We predict the etymology of a word across the full range of etymology types and languages in Wiktionary, showing improvements over a strong baseline. We also model word emergence and show the application of etymology in modeling this phenomenon. We release our parser to further research in this understudied field. Keywords: etymology, wiktionary, word emergence 1. Introduction Since antiquity, scholars have been fascinated by etymol- ogy, the study of words’ origins. In modern days, there exist numerous etymological dictionaries for select lan- guages (e.g. English (Skeat, 1884; Partridge, 2006), Al- banian (Orel, 1998), or Old Chinese (Schuessler, 2007)) as well as language families (e.g. Italic (De Vaan, 2018), Slavic (Derksen, 2007), or Altaic (Starostin et al., 2003)). Many of these improve and expand upon existing dictionar- ies as new evidence comes to light about the relationships between languages and their words. However, until very recently, the discovery of these relationships has not been computational driven. In recent years, researchers have developed computational methods for determining relationships between languages (see Nichols and Warnow (2008) and Dunn (2015) for sur- veys of the field of linguistic phylogenetics), but there is lit- tle work on computationally learning the etymological rela- tionships between individual words. Researchers have con- structed a Proto-Indo European lexicon (Pyysalo, 2017), and showed that knowing a word’s etymology can help with text classification tasks (Fang et al., 2009; Nastase and Strapparava, 2013) and reconstructing language phy- logenies (Nouri and Yangarber, 2016). In an era of abundant linguistic data, we seek to address the dearth of computational approaches to modeling etymol- ogy. To this end, we develop a parser that extracts etymol- ogy information and translations from Wiktionary, an open multilingual dictionary. Using this data, we present several approaches to model when (word emergence), from where, and how a word enters a language. We employ RNN-based models to accurately predict a word’s formation mecha- nism, parent language, and year of emergence. For emer- gence, we also experiment with various historical data- driven models. Our methods are language-independent and are applicable for improving existing etymology determi- nations that may be incorrect, as well as providing etymol- ogy for words that may not have an existing etymological entries, both in low- and high-resource languages. en. computer en. compute fr. computer la. computo en. -er enm. -er ang. -ere der bor suffix inh inh Figure 1: Wiktionary etymology graph of the English word computer. Etymological relationships are shown in blue. 2. Wiktionary Etymology Wiktionary1 is a large, free, online multilingual dictionary that is editable by anyone in the world. In addition to con- taining information found in traditional dictionaries (pro- nunciations, part …

Content truncated to 500 words