Generating Timelines by Modeling Semantic Change

Proceedings of the 23rd Conference on Computational Natural Language Learning, pages 186–195 Hong Kong, China, November 3-4, 2019. c⃝2019 Association for Computational Linguistics 186 Generating Timelines by Modeling Semantic Change Guy D. Rosin1, Kira Radinsky1,2 1Technion – Israel Institute of Technology, Haifa, Israel 2eBay Research, Israel {guyrosin,kirar}@cs.technion.ac.il Abstract Though languages can evolve slowly, they can also react strongly to dramatic world events. By studying the connection between words and events, it is possible to identify which events change our vocabulary and in what way. In this work, we tackle the task of cre- ating timelines—records of historical ‘turning points’, represented by either words or events, to understand the dynamics of a target word. Our approach identifies these points by lever- aging both static and time-varying word em- beddings to measure the influence of words and events. In addition to quantifying changes, we show how our technique can help isolate semantic changes. Our qualitative and quanti- tative evaluations show that we are able to cap- ture this semantic change and event influence. 1 Introduction Languages respond to world events in many ways. New words, phrases, and named entities are cre- ated, new senses may develop, and valences may change. Various approaches support the study of historical linguistics (e.g., comparative linguistics, etymology, etc.). In this work, we focus on a spe- cific process for tracking the progression of mean- ing over time in sense, semantics, and in relation to other words and concepts. By leveraging changing relationships in temporal corpora, we demonstrate a way of ‘embedding’ words and world events. Observing changes in this embedding allows us to construct timelines that support the study of evolv- ing languages. The timeline of scientific and technical discov- eries, for example, can drive the emergence of new word senses as these discoveries are ‘named’. Take the word “cell” which evolved from its 12th century meaning (a small room or chamber) to a new sense in the 17th century (a basic unit of an organism) to the 19th century meaning (an elec- tric battery) and most recently to a shorthand for a mobile phone1. Critically, the dominant senses of a word vary over time as some meanings become less commonly used while others gain in popular- ity. This dynamic need not be driven only by the addition of certain senses. The prevalence of a hy- ponym, for example, may also drive a change in the ranking of senses. The word ‘disaster’ may call to mind very different things depending on the latest type of disaster. Thus, the word may evoke ‘nuclear disaster’ in a reader in 2011 (e.g., driven by the Fukushima incident). However, in 2012 the ‘storm’ sense may be more salient (e.g., driven by Superstorm Sandy). Evolution of senses is but one way a language can evolve. Broader semantic changes can also occur. For example, the valence of the word may move or even flip (e.g., terrific or bully). Of partic- ular interest to us are those changes that are more immediate and precipitated …

Content truncated to 500 words