Friday 10 January 2020

The computer isn't always right


There is something vaguely comforting about the research carried out at Canterbury University in New Zealand which suggested that almost everyone in Wales will be able to speak Welsh in a mere 300 years from now.  It implies that everything required to ensure such a transformation is in place, and all we have to do is carry on as we are – for three centuries.  The computer says so, so it must be right.  The problem, of course, is that whilst the study is an interesting exercise in statistical modelling, and in identifying the factors which guide the outcome, it is, ultimately, based on a whole series of assumptions.  They’ve obviously put a lot of thought into those assumptions, used whatever data is available to arrive at them, and produced a model which will be pored over for many years to come.  It’s a sound basis for research and study – but none of that makes it a good basis for language policy.
The first question which struck me was about empirical validation – are there any examples, anywhere in the world, against which the predicted outcome can be validated?  If a model predicts an outcome which has never been seen before, and which runs counter to the experience of languages with similar numbers of speakers in similar circumstances, it is time to apply what my old mathematics teacher used to call the ‘reasonableness test’.  If it doesn’t look right, then it probably isn’t.  No matter how interesting the model is, and the way in which it has been constructed, a predicted outcome which is at odds with real-world experience needs to be treated with a healthy dose of salt.
The danger of the way in which this study has been reported is that it encourages complacency – an attitude that we’re doing all the right things, and just need to carry on as we are.  I am aware of only one real success story in terms of reviving a declining language, and that is Hebrew in Israel – and that revival wasn’t achieved by depending on the factors considered in this study.

No comments: