Adjectives, Compounds and Words

English spelling is, as everyone knows, full of peculiarities. Those peculiarities account for the amusement provided by sentences like (1) and Shaw's strange notion that 'fish' could be spelt ghoti, as well as the equally absurd claim that York-Los appears as a 'word' in the larger construction the New York-Los Angeles flight. (1) The dough-faced ploughman coughed and hiccoughed his rough way through Scarborough. Linguists take as given the primacy of the spoken word and the derivative nature of written language. We are aware of absurdities like those mentioned above, but see them as being artifacts of the spelling system and dismiss them as being of marginal relevance to the structure of English. Yet at other times we seem to have great difficulty in discarding the idea that English orthography tells us something important about the language. In this paper 1 I should like to discuss one such instance. However misleading English spelling may be on occasions, there is one place where it seems to match our intuitions perfectly. The description a black bird has black and bird in two orthographic words, whereas the naming function illustrated by a compound form such as a blackbird sees a single orthographic word, and no longer a series of two. We can find these intuitions justified in the literature. Black in blackbird is no longer available for syntactic or morphological modification (we cannot have *a rather blackbird, nor *a blackerbird). This indicates 1 I should like to thank Heinz Giegerich and Winifred Bauer for their helpful comments on an earlier draft of this paper. Errors are my own.

(1) The dough-faced ploughman coughed and hiccoughed his rough way through Scarborough.
Linguists take as given the primacy of the spoken word and the derivative nature of written language. We are aware of absurdities like those mentioned above, but see them as being artifacts of the spelling system and dismiss them as being of marginal relevance to the structure of English. Yet at other times we seem to have great difficulty in discarding the idea that English orthography tells us something important about the language. In this paper 1 I should like to discuss one such instance.
However misleading English spelling may be on occasions, there is one place where it seems to match our intuitions perfectly. The description a black bird has black and bird in two orthographic words, whereas the naming function illustrated by a compound form such as a blackbird sees a single orthographic word, and no longer a series of two. We can find these intuitions justified in the literature. Black in blackbird is no longer available for syntactic or morphological modification (we cannot have *a rather blackbird, nor *a blackerbird). This indicates 1 I should like to thank Heinz Giegerich and Winifred Bauer for their helpful comments on an earlier draft of this paper. Errors are my own.
that it is not a full word in its own right. Blackbird carries stress on the left-hand element of the compound. This is sometimes called 'compound stress' in the literature, but for reasons which will become clear is referred to here as 'firstelement stress'. This points out the difference between the word and the phrase (where nuclear stress tends to fall on the rightmost element in what we can, for present purposes, term 'second-element stress'). The meaning of black bird can be deduced from the meaning of its elements and the meaning of the construction, while the meaning of blackbird cannot be entirely predicted from the meaning of the elements (if that were possible, a sentence such as I saw a brown blackbird this morning would be nonsensical, which is not the case). This means that blackbird must be a dictionary entry, and in that sense is a lexical item (a term used in this paper in preference to the alternatives 'listeme' and 'dictionary word' 2 ). We thus seem to have a very strong set of coincidences, which match our intuitions, as set out in Table 1. Indeed, the whole pattern of Table 1 seems so convincing that it may seem odd to bring up the matter at all in this context.  If all examples were like this particular example there would be no problem; a problem does arise, though, with the notion that this example is in some way typical of English. To show this, we need to see how general or how limited the pattern illustrated in Table 1 is.
We can start with the observation that the number of adjectives that work in the way that black does in our exemple-type seems to be very restricted. If we require exactly parallel conclusions to those laid out in Table 1, we find the kind of adjectives set out in Table 2. Whatever these adjectives may have in common, they are not a random sample of words labeled 'adjective' in our dictionaries. inaccessible to syntactic or morphological modification as it is in the blackbird type of example. The moment we discuss blacker ice or a redder squirrel, we are no longer using these as the names of the entities given in the dictionary definitions.
Rather we are using them in the same way that we might use black bird, as descriptions. The same is true if we talk of a very brown trout or rather white meat.
Thus what we define as lexical items on grammatical criteria need not have a single stress or be written as a single orthographic word.
Consider what would happen if we started with an example like funny business. We would probably say that this is a lexical item because its meaning is not entirely predictable form the meanings of its parts and because if funny is sub-modified in any way the whole no longer retains its idiomatic meaning, but becomes compositional. But in all other respects it meets the criteria for a phrasal construction. If that is the case, we have admitted that orthography and stress are subsidiary criteria which do not need to be met for something to be a lexical item.
Thus, implicitly, we admit that orthography and stress are, if not irrelevant, then no more than supporting material in the discussion of blackbird. And at that point we should acknowledge that blackbird (and other words like it) just happen to have various criteria align, but that this is not crucial, and that stress and orthography are not ways of defining lexical items.
At this point, though, we need to cast our net wider, because there are also first-element stressed adjective-noun constructions which are not covered by the discussion above, largely because they are never written as a single orthographic word. Some examples are given In Table 3. presumably normal in the sense that it provides a norm for new teachers.

«floral arrangement
The items in Table 3 differ from the adjective-noun compounds illustrated in Table 2 in that they cannot be glossed as 'an N which is (stereotypically) A'. That is, while a blackbird is 'a bird which is stereotypically black', the romantic period is not 'a period which is stereotypically romantic'. While that factor does appear to distinguish some of the examples in Table 3  Similarly easy-chair and happy hour seem, in some ways, to fit better in Table 3 than in Table 2. We might hesitate about how to gloss silly-season in Table 2: is it 'a season which is silly' or 'a season in which silly things get reported' or 'season in which the silly is done/reported'?
So we find first-element stress doing at least two different things. With the adjectives listed in Table 2 it indicates that the adjective is to be interpreted as non-gradable (as a classifier, in one terminology), while with the adjectives illustrated in Table 3 it indicates that the adjective is to be interpreted in its nonpredicate meaning. But, and this is crucial, in neither case is the stress pattern a reliable marker of the function shown in the relevant table. In both instances, phrasal stress can have precisely the same reading, sometimes with precisely the same adjectives (recall black bear and primary education).
What, then, is the function of first-element stress? We must now admit that it is starting to look as though its function is not to delimit a compound in any structural sense, which is why the label 'compound stress' has been avoided here.
To consider the type of construction illustrated in Table 3 in more detail, the patterns in which school appears will be considered more closely. What seems likely to be relevant in assigning stress to the first element of the constructions illustrated in Table 3 is a set of factors including the frequency of the particular collocations involved, contrasting patterns of premodification, and the collocations in which the particular adjectives are used. In order to elucidate these factors, the collocates of school in the one million words of the Wellington Corpus of Written New Zealand English (Bauer 1993) were listed, and these are set out in Table 4 Kingdon's (1958: 151) notion that teacup (for instance) is stressed on the first element because of 'an implied sense of contrast' with items such as breakfast cup and coffee cup. But here we do seem to have some evidence which would point to just such a conclusion: school appears so often with a modifier that it is the modifier which is more important than the head noun. This becomes even clearer when we look at some of the modifiers involved. For example, in WCWNZE primary is used ten times in connection with health care or health services, ten times in connection with produce/product/production/producer, 48 times in connection with education, schools, teachers etc. and only 30 times in all other uses. Intermediate is used 12 times with reference to education, and only seven times in any other connection (one of which is an examination!). While we do not have to consult a corpus to tell us that high, for example, has a much wider range of uses, there is a sense in which the occurrence of primary already predisposes us to expect the word school, and the word school is insufficiently distinct without the modifier, so that stress on primary can be excused, if not explained.
Furthermore, it is interesting to note that school itself is used attributively in 25% of its occurrences (or rather more if instances like primary school teacher, listed in Table 4 under primary, are taken into account). That is, in a quarter of its uses, school is actually not defining a class of school, but is being used to define another class. Examples such as school teacher, School Certificate might be deemed irrelevant in making the general point about the way in which school is premodified contrastively, in which case we might want to claim that in relevant instances, the preponderance of classifying premodification is even higher than is shown in Table 4.
Another survey with a different pair of words provides broadly similar conclusions. In Table 5, the uses of society in the WCWNZE are broken down into various patterns. If we add to this the word operatic, which occurs only six times in the corpus, with just one of these modifying society, we can see that we end up with a similar, although not identical pattern. Here we might claim to have two lexemes society. The one meaning 'club' occurs most frequently with some kind of modifier (operatic, Royal, building, etc.). The other society occurs most frequently without a modifier; while the modifiers tend to be different, they can overlap: in principle royal society or New Zealand society could belong to either meaning of society until disambiguated by the context. The modifiers of society (in both senses, as it happens) are all relatively rare, and thus become particularly important in context. But when they are describing 'people living together' they are not the main focus of the communication, whereas when they are naming the 'club' they are of crucial importance.
An alternative, and perhaps preferable way of looking at this is provided by Ladd (1984). Ladd suggests that heads get destressed (and that we therefore get first-element stress) when the modifier is not merely 'descriptive'. This approach seems promising, though we need rather more idea of what it means not to be descriptive. Providing a naming function seems to be important here. Now let us return to the monosyllabic adjective + noun constructions we started with. If the principle of contrast holds for them as it might be thought to hold for these constructions with longer adjectives, we would expect to find that a noun like bird is relatively frequently premodified (because we find blackbird where the modifier gets stress) whereas one like bear is usually not premodified (because we find black bear where the head noun carries the stress). More accurately, we would expect this to have been the case when the lexical items blackbird and black bear received their current stress patterns. Note that it is not clear that WCWNZE is a relevant corpus for such a comparison, first because of the period it covers, and secondly because nearly all the bears mentioned are of the stuffed variety rather than animate. Nevertheless, it is disconcerting to find precisely the wrong distribution of modification, as shown in Table 6. Bird meaning 'young woman' (3 occurrences), bird used as a premodifier (12 occurrences) and one instance of ladybird are omitted from Table 6. Let us sum up. We have, apparently, two adjective + noun constructions in English, one of which is a single word, the other of which is a phrase. Yet the stress criterion does not match the semantic criterion by which wordhood might be expected to be determined. It turns out, and this is the main observation of this paper, that an apparently parallel use of stress is found with a disjunct set of adjectives, and that constructions involving this second set of adjectives have not been traditionally viewed as words at all. For this new set, it seems that there is some sense in which the first element stress can be correlated with pragmatic contrast. The same does not appear to be true for the original set of adjective + noun 'compounds' (although the data that has been used here is not necessarily as relevant as we could wish).
If, instead of looking at implicit contrast, we consider the non-descriptive destressing account provided by Ladd (1984), we seem to be on firmer ground, in that the two types of adjective + noun constructions can be seen as acting rather more in the same way. But then we have the problem that so many apparently relevant constructions end up not being destressed at all. While Ladd has further requirements on heads that become destressed (for example, that they should be fairly generic, though that is not his terminology) the difference between blackbird and black bear might be covered, but not, I suspect, the difference between a «social worker and a manual «worker.
Yet another possible solution, which has not so far been discussed here, is that first-element stress is simply a matter of lexicalisation. It is hard to know how to measure this, since lexicalisation does not necessarily correlate with absolute frequency in any given corpus. For example, in WCWNZE, because of one particular text in the corpus, fossil bird, with 4 occurrences, is more frequent than blackbird with 2. Here it seems that a larger corpus might be more revealing. The results are shown in Table 7. 5 Green belt appears as though it should fit in the second column of Table 7, but is given first-element stress by the Macmillan English Dictionary for Advanced Learners (2002). It gets a correspondingly high score with 222 occurrences.
be seen as lexicalised in comparison with, for example, manual worker (31 occurrences), floral arrangement (7 occurrences) seems scarcely different from financial arrangement (6 occurrences). A table corresponding to Table 7 is hard to construct here, given the difficulty in finding and in confirming examples of the appropriate types. For example, a search for feudal system finds not only «feudal system, but also examples of feudal «system, and we also find a wide range of frequencies from the clearer members of the set such as nervous system (567 occurrences) and cardiovascular system (12 occurrences). It may be that a similar kind of result would emerge on average, but probably not to the same extent.
The introduction of a new set of adjective + noun constructions with firstelement stress into the discussion of the status of constructions with first-element stress at first looks as if it might be helpful in resolving a problem of some standing. While this new body of data raises a number of interesting questions and suggests some possible solutions, it still seems that first-element stress is doing more than one thing in English. While this does not in itself disprove the notion that there might be two discrete classes of construction here, it makes it a lot more difficult to sort out the facts and to provide the kind of description which will be useful to language teachers and lexicographers, such as our honoree. More disturbingly, it raises questions about how lexicographers are supposed to identify lexical items (dictionary words). While stress and orthography have often been taken as contributory criteria, consistent patterns of mismatch between the two, and regular mismatches between either of these and a naming function suggest that the lexicographer needs new strategies for identifying relevant material.