Understanding misunderstanding: How quick-fix solutions undermine explanation

Communicative alignment is the capacity that, even when talking about novel things and without a fully shared vocabulary, people can come to understand each other. State-of-the-art models so far fail to provide a sufficient explanation for this phenomenon. When such models fail, researchers are often quick to suggest a solution to quickly fix the model and make it work. We propose a metatheoretical approach using four criteria to systematically assess explanatory quality of computational explanations. We analyse the efficacy of different types of such suggested quick fixes following guidelines that a good computational explanation must be transparent, unambiguous, sufficient, and possible. Using these criteria we show that quick-fixes cannot be sufficient to achieve an explanation for communicative alignment. We show that even if there is a strong body of work on a topic, that does not guarantee such work is sufficient to quickly fix a computational model. It may be that a quick fix suggestion has the right idea, but often the methodology is under-specified or ill-formed. In some cases these quick fixes may even undermine finding explanations of communicative alignment, by obscuring the capacity or adding more theoretical problems to an existing explanation. We conclude it is not easy to fix an explanation for communicative alignment. With these analyses as starting point, we then classify the pitfalls within these different quick fixes in a meta-theoretic typology using maps as an analogy. It abstracts away from communicative alignment, and can thus be used to analyse possible quick-fixes in other domains.