Measuring productivity: the significance of neologisms

So far, I have discussed various aspects of the productivity of derivational processes, such as -ity and -ness suffixation, without the help of any objectively quantifiable measurements. I have not introduced any scale from 0 to 1, say, in terms of which -ness might score 0.9 and -ity 0.5. This may look like a serious defect. Any conclusions about formal regularity and generality, in particular, must be more or less subjective unless they are based on figures, one may think. What we need is a comparison between the actual frequency of a process and its potential frequency, appropriately defined. The more closely the ‘actual’ figure approaches the ‘potential’ figure, the more productive the process is, in some sense.

In practice, however, devising such a measure has turned out to be extremely tricky. For example, I suggested earlier that -ity is formally regular when applied to adjectives with certain suffixes such as -ous, -ive and -able, but otherwise irregular, for example, when it is attached to suffixless adjectives. This has the advantage that a non-existent noun such as ‘richity’ can be classed as formally irregular, but the disadvantage that it entails that actual nouns such as purity, sanity, oddity and severity must be irregular too. If we dislike this outcome, we must extend the range of adjectives that count as potential bases for -ity suffixation so as to include at least pure, sane, odd and severe. But how far should this extension go? If the new potential bases are taken to be just those un-suffixed adjectives for which corresponding nouns in -ity exist, then the ratio of actual to potential -ity nouns will remain high – but only because we have contrived that it should be so. If, at the other extreme, we let pure, sane, odd and severe persuade us that any adjective whatever can be a potential base, then the actual-to-potential ratio dwindles to almost zero, and our measure fails to capture the difference in ‘feel’ between ‘gloriosity’ (plausible, but blocked by glory) and *richity (highly implausible). What is the appropriate intermediate position between these extremes, then? It is hard to answer that question except subjectively. Thus the objectivity that a numerical ratio would supply turns out to be frustratingly elusive.

Since about 1990, however, a new set of numerical measures have been devised that avoid subjective bias and yet seem to correspond well to what we feel we mean when we talk about ‘productivity’. These measures exploit the extremely large corpora, or bodies of linguistic material, that have been assembled on computer by linguists and dictionary-makers for the purpose of studying both the frequency with which words (lexemes and word forms) occur and the contexts in which they occur. For a process to be productive, in one sense, it should be a process that can be used to form brand new lexemes, or neologisms. So can we identify neologisms in one of these large corpora? Unfortunately, we cannot do so directly; all we can tell for certain is that a lexeme with an earlier dated occurrence in the corpus is not brand new. However, we can rely on the fact that most neologisms within the corpus will be rare. In fact, all will be rare except those that quickly become fashionable. So, even if we cannot directly identify neologisms, an alternative that is both appropriate and feasible is to identify words that are extremely rare, especially those that appear only once in the whole corpus: so-called hapax legomena (singular hapax legomenon), a Greek expression borrowed from classical studies, meaning ‘said (only) once’. We can now focus on the morphological processes that are used in hapax legomena (and other very rare words), and compare them with processes that are used in more frequently occurring words.

Such studies shed interesting new light on the relationship between the familar pair -ity and -ness. In the Cobuild corpus of about eighteen million English word-tokens (based at Birmingham University and used by the dictionary publisher Collins), the number of word-types exhibiting -ity (roughly 400) is not greatly less than the number exhibiting -ness (roughly 500). However, most of the -ity words are of common occurrence (more technically, their token-frequency is high), while many more of the types exhibiting -ness have low token frequency, including hapax legomena. That is, although by one measure -ness seems to be not much more productive than –ity is, it is far more likely than –ity to be used in the creation of neologisms.

The suffix -ness rates high both in the number of words that contain it (words as types, that is, not tokens), and in its availability for neologisms. The suffix -ity ranks high by the first measure but low by the second. Could an affix rank low by the first measure and high by the second? The answer is yes. The Cobuild corpus contains relatively few word-types with the suffix -ian (as in Canadian, Wagnerian), yet a very high proportion of these are of low token-frequency. Rather surprisingly, therefore, for an affix to be suitable for use in a brand new word, it does not have to appear in a large number of existing words.

There is far more that could be said about the ways in which studies of very large corpora can shed light on word formation in English. To understand it in greater depth presupposes some knowledge of statistical techniques, however. For present purposes, it is enough to be aware that such statistical studies are being carried out, and that they go a considerable way towards firming up the notions of generality and formal regularity which I defined in an unquantified fashion.