(born Geneva 1857; died Vufflens-le-Château, 1913)
In language there are only differences.
—Saussure, Cours de linguistique générale
Imagine any kind of catalogue. Each element in the catalogue is distinguished according to whatever properties are recognized as relevant. This is certainly a natural way of organizing a description: one puts together a list of properties that characterize every element in the set. So, for example, if you have ten bicycles, you can describe each one in terms of its color, the type of wheel, the saddle cover, the shape of the handlebars, and so on. It’s likely that the ten bicycles will be quite different from one another in relation to the properties chosen; that they won’t all have the same color, the same type of wheel, the same saddle cover, the same handlebar shape, etc. However, imagine that after taking a longer look at the same ten bikes you notice that in certain ways there are relations among their properties. For example, five of the bikes have racing handlebars and the other five don’t: having handlebars is obviously a property of all the bikes, but having racing handlebars is a property of only five of them. Imagine now that we notice another property: five of the bikes have flat tires and five do not, and maybe this division doesn’t correspond to the one based on the type of handlebars. Carrying on in this way, we reach a descriptive situation completely different from the one we started with: if we’re really lucky, with just four properties defined for each bike we can characterize every one of them uniquely. So if we label the properties A, B, C, D, with, for example, +A meaning the bike has a certain property (e.g., racing handlebars), and–A meaning that the bike in question doesn’t have that property, we can construct 16 (24) different combinations: the bike type [+A, +B, +C, +D], the bike type [-A, +B, +C, +D], and so on until we’ve exhausted all the combinations.
Three points are worth making about this system. First, we’d have to be really lucky, because it’s not obvious that we’d be able to find four properties that divide the bikes into two different groups every time (we might need more properties). Second, given that there are ten bikes and the combinations of our four properties give sixteen classes, that means that there are six types of bike that are not in our group—maybe it would be worth looking for them. Third, in the differential grids that we construct in this way, not all the properties of a member count (assuming that the list of properties can in any case ever be exhaustive); the only thing that counts is how the member of the group is characterized in relation to the properties chosen.
Coming back to our example, any bike of type [+A, +B, +C, +D] is fine; in other words, when a member is characterized in the given way in relation to the chosen properties, any other element having the same characterization in relation to those properties can be substituted (technically, commuted) for that member of the group. Moreover, using grids of this kind, it often turns out that there are subsets of members that can be identified by a smaller number of properties than the number of elements in the subset so defined; these are called “natural classes.” Often these natural classes are what is really interesting, and they can give us a simpler explanation of the facts. This type of classification was developed and used to classify the sounds of language; for example, we speak [+voice] or [-voice] consonants in pairs like p/b, t/d, etc. But in fact the philosophy behind this method, as presented in the introduction to Chomsky’s Pisa Lectures (Chomsky 1981), is for syntax too the real methodological revolution behind generative grammar. The difference between a catalogue and a grid is that in the latter every property defined for one member is also defined for all the others. So in the end every member is defined only in terms of how it differs from all the others. This is one of the most powerful insights ever: it transformed the way we look at the structure not only of language but also of (potentially) any system—so much so that the term associated with this way of looking at things, “structuralism,” is one of the unifying ideas of the twentieth century.
Structuralism fostered and made possible the collection of a huge amount of data, in linguistics and in other fields, revealing unexpected and more perspicuous patterns. Breaking things down into primitive differential elements leads to an original and suprising question: what really exists in the end, the bicycles or the abstract features that generate them?
