We'll gladly take any feedback on the work, don't hesitate to reach out! 🤗

As always, code and aggregated data openly available 👐
osf.io/n6j8g/

So with such simple mechanisms controlled by just 3 parameters, we're able to quite nicely reproduce the effect we observed: socioeconomic mixing smooths out this dialectal difference between classes.

We show analytically that increased mixing does make different classes converge on their dialect usage. And since it's an ABM, we can simulate agents moving around according to actual mobility patterns, and again we found one of these nice straight lines we all love 🫶

That's why we proposed some basic mechanisms that could explain this, in the form of an agent-based model. Use of a variety depends on how prestigious the standard form is (s), how much the poorer class prefers non-standard (q1), and how much the richer prefer standard (q2).

What this plot shows is that the more different classes mix with one another (low assortativity), the less their use of non-standard language will depend on their class of origin. This is really remarkable (to us at least!), so we wanted to understand further.

This was really surprising to us, so we had to understand where these differences could be coming from! That's when we thought of looking at how much people of different classes mix in each of these cities. And what we found is this very nice line 🔽

This gives a nice proxy for the tendency of Twitter users to use non-standard English. And what we found were smaller correlations than we expected, but more crucially, widely different ones from one metropolitan area of England to the other.

The question that triggered this work was: how inter-dependent are socioeconomic status and deviations from standard language?

One cool aspect of the method is that we used a local version of LanguageTool to compute the frequencies of deviations from standard grammar in tweets.
languagetool.org