Morpho-distributional criteria
The languages for which the earliest parts of speech classifications were developed were strongly inflectional in nature. In both Latin and Greek, nouns and verbs took two entirely different sets of suffixes. As a result, the inflections provided a very obvious way of dividing the vocabulary up into different word classes: nouns can be defined as the words which take a certain set of inflectional suffixes; and verbs as words which take a certain different set. The classification into noun and verb, in other words, can unproblematically be read off the morphology, and if we ever had a doubt about what part of speech a given word belongs to, we can quickly resolve it by checking which set of inflections it took. At an earlier stage of its linguistic history, English itself used to be much more like Latin and Greek, as we can see from the paradigms of the Old English nouns stān ‘stone’, giefu ‘gift’ and hunta ‘hunter’, and of the verb fremman ‘do’ (note the character þ, pronounced [θ] and called ‘thorn’):
In order to discover whether hunta is a noun or verb, then, we simply ask what set of inflectional endings it can take.
In Modern English, where the morphology has greatly decayed, we cannot determine word class from inflections: there just isn’t enough inflectional morphology, and what there is isn’t regular enough through any one word class. For example, we couldn’t define nouns as the class of words which take plural markers, since some words which we clearly want to recognize as nouns, like equipment, worth, emptiness, suffrage, marketing or music, typically do not have this possibility. The English category ‘adjective’ is similar. We might suggest that adjectives can be defined as those words which can accept com parative/superlative morphology: think of fast–faster–fastest, hot–hotter–hottest, black–blacker–blackest and kind–kinder–kindest. The problem is that while this criterion applies to some basic adjectives, it will end up excluding many words we accept as adjectives, like prior (*priorer/*more prior), former (*formerer/*more former), total (*totaler/*more total), future (*futurer/*more future) and many others.
QUESTION Is it possible to identify a morphological criterion for verbs in English? What prevents us giving a simple answer to this question?
The difficulty of coming up with satisfactory definitions of these basic parts of speech is somewhat embarrassing, since it is precisely the traditional categorization of verb, noun and adjective that we need if we are to describe English in the way that most syntacticians believe is required.
Another criterion often used to define parts of speech is a distributional one. A distributional approach to parts of speech classifies parts of speech on the basis of the way they pattern in sentences. One way of defining these patterns is to advance sample contexts which can serve as test frames. For example, Radford (2004: 30) suggests that (3) can serve as a test for nouns, since it is only by a noun that it can be completed:
Thus, cardinal nouns like house, dog, patience, recognition and many others can be substituted into the sentence, whereas adjectives/adverbs (sad(ly), correct(ly), canine), verbs (eaten, dwell, conceal) and prepositions (to, with, by) cannot. More examples are given in (4).
However, further reflection soon reveals that not all words which we usually count as nouns can be appropriately put in the frame, as shown in (5):
These sentences might be improved by adding further material after the noun (?They have no possibility of improvement). However, a better test-frame for nouns might be the following one:
This works for all the nouns we have seen so far. But it is not free of problems, since it also admits the following, as any quick search of the Internet will show:
None of the highlighted words is usually considered a noun in the senses which they exemplify here. Consistently, none of these words should be able felicitously to appear in the test-frame in (7).
Even if we were able to devise appropriate test-frames which did isolate the categories in a way that matched our traditional classifications, there is still a problem: the morphological and distributional tests only work if the membership of the different parts of speech categories is already known. That is why we can say that the words in (7a–e) are not nouns in spite of their compatibility with the test-frame, and why we know that the words in (5) are nouns even though they do not fit into the test context in (3). In neither case are we using the test sentences to establish the part of speech classification of a given noun; we are using it to justify a traditional classification which we consider as self-evident. The traditional part of speech categories seem simply to be so entrenched in the way we think of our own language that we cannot avoid using them, even if we have a hard time specifying exactly what criteria govern their application. It seems as though we have to look elsewhere for some insight into this question.
