Strycharczuk et al.’s Frontiers paper sparks controversy

Posted on July 16, 2020 by

LEL’s Pat Strycharczuk and colleagues: Manuel López-Ibáñez, Georgina Brown and Adrian Leemann have a new paper out in Frontiers in Artificial Intelligence, titled “General Northern English. Investigating regional variation in the North of England with machine learning”. The paper’s barely been out for 24h, but it has already generated a lot of controversy, if 1.1k angry comments on Mail Online are anything to go by. It is clear that machines were used to analyse northern accents here, but what can we actually conclude from this? The featured photo presents some of the more daring conclusions that have been proposed (not by Pat or her team). To clear things up a bit, Manchet has interviewed Pat to establish what’s what (we have a direct line of communication).

What is the machine you’ve used to study speech?

They’re machine learners, a type of algorithm or computer programme. What they do is “listen” to samples of northern accents, while being told which accents are from Manchester or not. Well, they don’t listen as such, since they don’t have ears, but they get some information about vowel sounds. Then they are presented with a new speaker, and they have to decide: “Is this speaker from Manchester or not”?

Why use a machine for something like that?

Because the alternative would’ve been to ask people, which sounds like a simpler solution, but it’s not. If we ask people, we have no control over what they believe someone from Manchester sounds like. Anyone may give you a different judgement, depending on whom they know that’s from Manchester. People are also really bad at explaining how they make this type of judgement – you want them to say “I could tell this person is from Manchester by the way they pronounced the word letter“, but they rarely say such things. They’re more likely to say “Sounds kind of Manchester, dunno why, just does”.

So this machine, can it tell if someone’s from Manchester or not?

A lot of the time, it can. However, it also makes mistakes, and the mistakes reveal an interesting pattern. For example, it often thinks speakers are from Manchester when they are in fact from Leeds or Sheffield.

Does it mean all northern accents are becoming the same?

No, it means that some of them are quite similar.

Maybe your machine is just bad at telling accents. I know someone who can tell which street you’re from in Manchester by the way you talk.

The thought has occurred to us, so we also analysed the average pronunciations of various vowel sounds in different cities. They really are remarkably similar, especially when we compare Manchester, Leeds and Sheffield, so no wonder the learners make mistakes.

Are we losing our northern accents? Our identity? Our soul?

No, this is simply some people in the North speaking a new type of accent. Or maybe even not all that new. Linguists have been talking about General Northern English for at least the past forty years, so it’s not really news that this accent is out there. What’s new is actual data that confirm there is a good number of people speaking GNE.

Are these people ashamed of their own accent?

No, it’s the normal way to talk in their community, among their peers. GNE is their own accent.