[Paper Review] Mapping the geographical diffusion of new words

Jacob Eisenstein, Brendan O’Connor+2 more|arXiv (Cornell University)|Oct 18, 2012

Digital Communication and Language27 references36 citations

TL;DR

This paper proposes an autoregressive model to map the geographical diffusion of new words in U.S. social media, revealing that lexical innovations spread through city-to-city linguistic influence networks rather than globally. The key contribution is identifying that geographic proximity and demographic similarity significantly drive the spread of neologisms, even when social media enables global connectivity.

ABSTRACT

Language in social media is rich with linguistic innovations, most strikingly in the new words and spellings that constantly enter the lexicon. Despite assertions about the power of social media to connect people across the world, we find that many of these neologisms are restricted to geographically compact areas. Even for words that become ubiquituous, their growth in popularity is often geographical, spreading from city to city. Thus, social media text offers a unique opportunity to study the diffusion of lexical change. In this paper, we show how an autoregres-sive model of word frequencies in social media can be used to induce a network of linguistic influence between American cities. By comparing the induced net-work with the geographical and demographic characteristics of each city, we can measure the factors that drive the spread of lexical innovation. 1

Motivation & Objective

• To investigate how new words spread across American cities through social media.
• To model the geographical diffusion of lexical innovations using social media text.
• To identify the geographic and demographic factors influencing linguistic influence between cities.
• To construct a network of linguistic influence based on word frequency dynamics in social media.

Proposed method

• An autoregressive model is trained on time-series word frequencies from geotagged social media posts across U.S. cities.
• The model estimates the influence of each city’s word usage on others, capturing directional spread patterns.
• Linguistic influence is inferred from the temporal correlation of word frequency changes across cities.
• The resulting network of influence is compared with geographical and demographic data to identify driving factors.
• Statistical analysis measures the correlation between network structure and city-level characteristics such as population size and distance.

Experimental results

Research questions

• RQ1 How do new words diffuse across American cities through social media?
• RQ2 Which cities act as primary sources or hubs for lexical innovation?
• RQ3 To what extent do geographic proximity and demographic similarity predict linguistic influence between cities?
• RQ4 How does the structure of linguistic influence compare to physical or social distance?

Key findings

• Neologisms often spread from city to city rather than achieving global reach simultaneously.
• Geographic proximity is a significant predictor of linguistic influence between cities.
• Demographic similarity, such as shared cultural or socioeconomic traits, enhances the likelihood of linguistic diffusion.
• Even with global social media connectivity, lexical innovations remain geographically constrained in their early adoption.