By Kirsty McHugh, OUP UK
We operate in a world full of continuous variation, using language that is intrinsically vague. How warm is a ‘warm day’? Where should we draw the ‘poverty line’? Are you the same person as you were yesterday? None of these questions can be given a clear-cut answer. Any response is likely to start with ‘That depends…’. In Not Exactly: In Praise of Vagueness, Kees van Deemter, based at the University of Aberdeen, shows us why vagueness is, in various circumstances, both unavoidable and useful. In the post below he writes about the advantages of vagueness.
The recent snow has made us acutely aware of the difficulties of weather forecasting. And with global warming in mind, we know that freak weather events are expected to become more frequent. Forecasting will become more and more important as a way of warning us about dangers ahead.
Weather forecasts were among the first serious applications of computer technology, with early successes in the 1950s: computers churned out huge tables of numbers, predicting temperature, wind speed, and so on. To make sense of these tables was time consuming and prone to subjectivity and error.
In recent years, new computer technology has come to our aid by converting these tables into words. The name of this technology is Natural Language Generation (NLG), an area of research that owes as much to linguistics as to computing science.
Until recently, NLG programs went about in a straightforward manner, associating each weather phenomenon with fixed thresholds. Wind speeds between 32 and 48 knots, for example, counted as “gale force”. Saying “gale force” instead of “33.4 knots” is useful because it allows the weather to be painted with a broad brush stroke, glossing over irrelevant details.
‘Mild for the time of year’
But thresholds are a bit of a nuisance. Take air temperature: in London, 13C is mild in winter, normal in spring, and cold (or “disappointing”) in summer. In Scotland, ‘mild’, ‘normal’ or ‘cold’ would refer to very different temperatures. Consequently, we need a huge number of thresholds, to cater for many places and times. This is a recipe for disaster: on a sleepy afternoon, a programmer could easily enter an incorrect threshold, causing the forecasting program to misbehave in ways that might never be traced.
Moreover, any given threshold is bound to be somewhat arbitrary. Suppose the temperature hovers around a threshold, sometimes just above it, sometimes just below. Do we really want the computer to say “It will be mild at 9AM, warm at 10, and mild again at 11”, and so on? Much better to lift things to a higher level, saying “The temperature will be on the warm side all day”. To produce this sentence, however, based on nothing more than numbers as input, requires a level of artificial intelligence that is well beyond the current state of the art.
In our day-today conversation, we do not give words like “warm” precise black-and-white thresholds – we settle for grey instead. Or to use its technical linguistic term – vagueness. Vagueness has a bad name, but it is much more common than many people think, and for good reasons. To see why, let us look at an example from biology: the tricky notion of a species.
When is a species not a species?
Biologists sometimes draw the line between species where one type of individual is different enough from another that the two cannot interbreed (i.e., they cannot produce fertile offspring). This criterion, however, leads to strange results.
The Ensatina salamander lives in the hills around California’s central valley. In fact there are six types (let’s call them E1 to E6) which are located round a rough circle – so E1 is next to E2, and also next to E6. Each type successfully interbreeds with the two types next to it, but with no others (E2, for example, interbreeds only with E1 and E3). If we now apply the biologists’ interbreeding criterion to find out what the species are, we do not retrieve E1-E6, nor all Ensatinas together, but five different, and strangely overlapping “species”!
If we apply the same idea to our own ancestors, including those long gone such as Neanderthals, the result is that Homo sapiens is no longer the one species that we know and (sometimes) love, but a large and messy gamut of overlapping human-like species.
The ‘simple’ interbreeding criterion causes a mess. If it’s so difficult to separate species in a clear-cut way, then perhaps it ought to be done vaguely! So when the next skull of an ancient humanoid is dug up, and we all ask “Did it belong to a human?”, the right answer might be “It depends where you draw the line.” Just like the weather, a species can defy neat categorization.
The logic of vagueness has bothered philosophers and scientists since Aristotle’s day.