Preliminaries:
Possible words, actual words, and the lexicon
A notorious problem in the description of the speakers' morphological competence is that there are quite often unclear restrictions to the possibility of forming (and understanding) new complex words, to the effect that proposed word formation rules may not yield the correct set of complex words.

For instance, word formation rules may predict the existence of forms which are unattested or whose status as well-formed derivatives is more than doubtful. A famous example of this kind is the attachment of the nominalizing suffix -ity to adjectival bases ending in -ous, which is attested with forms such as curious - curiosity, capacious - capacity, monstrous - monstrosity. However, -ity cannot be attached to all bases of this type, as evidenced by the impossibility of *gloriosity or * furiosity. What is responsible for this limitation on the productivity of -ity?

Another typical problem with word formation rules is that they are often formulated in such a way that they prohibit formations that are nevertheless attested. For example, it is generally assumed that person nouns ending in ee (such as employee) can only be formed from transitive verbs and, in informal terms, denote the object of the base verb ('someone who is employed'). However, sometimes even intransitive verbs take -ee (e.g. escapee, standee) or even nouns (festschriftee).¹
 

Furthermore, some affixes occur with a large number of words, whereas others are only attested with a small number of derivatives. What conditions these differences in proliferance?

Intuitively, the notion of productivity must make reference to the speaker's ability to form new words and to the conditions the language system imposes on new words. This brings us to a central distinction in morphology, the one between 'possible' (or 'potential') and 'actual' words. The separation of actually attested words from non-attested but well-formed words is traditionally recognized² and finds a reflection in Coseriu's well-known dichotomy between system and norm (1952).³ A possible, or potential, word can be defined as a word, existing or non-existing, whose morphological or phonological structure is in accordance with the rules of the language. It is obvious that before one can assign the status of 'possible word' to a given form, these rules need to be stated as clearly as possible.⁴
 

The concept of 'actual word', it seems, is harder to define. A loose definition would simply say that actual words are those words that are in use. However, when can we consider a word being 'in use'? Does it mean that some speaker has observed it being used somewhere? Or that the majority of the speech community is familiar with it? Or that it is listed in dictionaries? Rainer defines 'actual word' as a "word that is part of the vocabulary of a specific speaker at a specific point in time" (1987:195f, my translation).⁵ The problem with this definition is that it makes reference to the individual speaker's lexicon, whereas morphological theory needs to make reference to the language system. It should be pointed out, however, that this gap between the vocabulary of the individual native speakers of a language and the abstract lexical-morphological language system should not be overestimated since a large overlap of lexical knowledge among speakers is necessary for language to work.

However, this and other problems have led some theorists to the complete abandonment of the notion of actual word. Kiparsky, for example, has argued that the whole concept of actual word should be disposed of because "it is ill-defined and of no linguistic interest" (Kiparsky 1982b:26). There are essentially three arguments for this position.

The first argument is that speakers may not know whether a given word has already occurred in the language or not. In support of this claim one could cite Aronoff (1983), who has carried out experiments which indicate that, with highly productive rules, "Speakers tend to judge potential words ... as actual words, though they are not" (1983:166). But what does this result really tell us? First, it simply shows that the speakers' intuitions can-not be used to tap this distinction. This does not at all imply that the distinction is non-existent. Second, it tells us that highly productive processes create words that are so similar to many actual words that the two are no longer distinguishable. Again this is not a good argument against the distinction itself, because no one has ever claimed that actual words and potential words should not have many things in common. What has been claimed is that lexically listed regular complex words form a subclass of actual words (e.g. Jackendoff 1975, Aronoff 1976, see below for further discussion). To put it differently, the class of actual words contains both morphologically regular and morphologically idiosyncratic forms. The crucial difference between actual and possible words now is that only actual words may be idiosyncratic (e.g. semantically or phonologically).

The second argument against the 'possible-actual' dichotomy is that this distinction is not clear-cut. However, the fact that a distinction is not clear-cut does not necessarily mean that it does not exist. For example, we know that the distinction between inflection and derivation is more like a continuum and that a strict boundary between the two can often not be found. However, there is still good evidence that this distinction is theoretically useful and psycholinguistically relevant (see, for example, some of the papers in Booij and van Marie 1996).

Although it may seem that with a given speaker a word is either listed or not, the picture is not clear-cut, because many factors, frequency of occurrence chiefly among them, play a role in memorization and retrieval. For example, the memorizability and later accessibility of a word depends crucially on the word's frequency in the speaker's environment. Hence one could claim that some items are 'more listed' than others, in the sense that they are easier for a speaker to access and retrieve than others.

The third, and perhaps strongest, argument against actual words is that "no rule of grammar even depends on whether a word is 'actual' or not" (Kiparsky 1982b:26). Booij (1987:44-51) presents impressive counterevidence to Kiparsky's claim by showing that certain types of morphological change, and processes of 'paradigmatic' word formation must make reference to the notion of 'existing complex word'⁶. The development of affix clusters and the substitution of morphemes in complex words are cases in point, since both phenomena presuppose the prior existence of a complex model form.

To summarize the discussion of the notions of possible and actual word we can say that the distinction is useful and often necessary for the description of morphological processes. Morphology, as conceived here, concerns the study of both actual and possible words. It is one of the aims of this study to find out more about the kinds of mechanisms that are necessary to define the properties of potential words in English, and this aim will be achieved primarily by studying large amounts of actual words.

The debate on the status of actual words in morphological theory is a direct reflection of the controversy on the nature of the lexicon and its role in the grammar. I will not attempt to review the numerous approaches that can be found in the literature but will only give a very much simplified version of some of the issues pertinent to our discussion.

In syntactic approaches to morphology there is a tendency to see the lexicon "like a prison - it contains only the lawless" (Di Sciullo and Williams 1987:3)⁷, i.e. only simplex words, roots and affixes (e.g. Lieber 1981, 1992). What non-syntactic morphological theories have in common is that they assign morphological processes either to a separate component, or see them as part of the lexicon. Their concept of the lexicon may, however, be very similar to the one just mentioned. For example, in a lexicon à la Kiparsky (1982), only simplex words, roots, and affixes have a place, but no regular complex words. Others (e.g. Jackendoff 1975, Aronoff 1976, Booij 1977) assume that both simplex and complex words, regular and idiosyncratic, can be listed in the lexicon together with redundancy rules that re late them to one another.⁸

But why would so many researchers want to bar complex words from being listed in the lexicon? The main reason for excluding these forms from the lexicon seems to be the widely shared assumption that the lexicon should be "minimally redundant" (Kiparsky 1982:25). As already dis cussed, there is little independent justification for this assumption. Hence, the major argument for a non-redundant lexicon seems to be the elegance and parsimony of the theory itself which proposes this kind of lexicon. This elegance is, however, achieved at the cost of empirical adequacy and, as we will shortly see, under neglect of psycholinguistic evidence.

Many psycholinguistic studies have demonstrated that the economy of storage, i.e. the elimination of redundancy, must be counter-balanced by the economy of processing, i.e. an increase in storage. Simplifying a bit, there are two conflicting models of the processing and storage of morphologically complex forms, full listing and direct access on the one hand (e.g. Butterworth 1983, Manelis and Tharp 1977), and decomposed lexical storage and morphological parsing on the other (e.g. Taft 1985, Taft and Forster 1975). The full listing model claims that all words, complex and simplex, are stored as single units and accessed via the direct retrieval of the whole word. The decomposition model says that all complex words are obligatorily decomposed into their smallest morphological elements. The full listing model rests on the assumption that direct access involves less cognitive costs than parsing. In contrast to that, the decomposition model assumes that storage is more costly than processing. In short, there seems to be an irresolvable conflict between storage and processing in the sense that less storage involves more processing and vice versa. For example, by increasing the storage costs the processing costs can be minimized, since direct look-up may involve less processing than decomposition or parsing. Among others, Frauenfelder and Schreuder (1992) and, most recently, Baayen, Dijkstra and Schreuder (1997) have argued that the conflict of the two models can be resolved by assuming a morphological race model, in which both ways of accessing a complex form in the lexicon are in competition.⁹ The winner of the race is determined mainly by the frequency and the phonological and semantic transparency of the word to be accessed:
"The probability that the parsing route wins the race is highest for transparent low frequency words. ... The direct route will win the race for high frequency word forms and those word forms that are problematic for the parser, for example, opaque word forms containing unproductive affixes." (Frauenfelder and Schreuder 1992:182). Notably, the dual route mechanism implies that complex words can be stored as a whole and not only in a decomposed fashion.
Additional evidence for this view comes from a series of most recent experimental studies by Schreuder and Baayen (1997), Baayen, Lieber and Schreuder (1997), Baayen, Dijkstra, and Schreuder (1997). These authors demonstrate that the processing of a simplex word depends both on the token frequency of complex words containing the word as a stem and surprisingly - on the number of different complex words (types) that contain the simplex word as a stem. In other words the size of the so-called morphological family co-determines the processing of the base word. Whatever the explanation for these results is, it must be based on the assumption that a non-negligible amount of regular complex words is stored in the lexicon.

To summarize, both psycholinguistic and structural linguistic arguments support a view of the lexicon, according to which regular complex words can also be stored. Hence the distinction between possible and actual words should be upheld and the notion of a non- or minimally redundant lexicon should be rejected. Having thereby illuminated some of the basic ideas on which notions of productivity are based, we may now turn to the explication of what productivity actually is.

¹See Barker (1995) for discussion.
²See, for example, van Marie (1985:38-42) for a review of the structuralist literature.
³See Burgschmidt (1977) for an elaboration of these concepts with respect to word formation processes.
⁴It should be noted that there is also a school of thought that completely re fuses to consider non-attested language material, but instead exclusively de votes itself to the study and classification of corpora data (e.g. Harris 1960). As noted by many previous authors, such an approach fails to meet a central concern of linguistics, namely the description of the speaker's generative capacity.
⁵See Aronoff (1976:19) for a similar view.
⁶ See also Aronoff (1976:18) for a similar argument.
⁷ Note that Di Sciullo and Williams (1987) do not argue for a purely syntactic model of morphology but rather take a special kind of lexicalist position.
⁸ See also Segui and Zubizarreta (1985) for a similar discussion.
⁹ The so-called Augmented Addressed Morphology model developed in Laudanna and Burani (1985) and Caramazza et al. (1988) is similar to Frauen-felder and Schreuder's in that it also assumes dual routes of access. It differs from the race model in that the decomposition route is only seen as a back-up procedure. A more detailed discussion of these psycholinguistic works is beyond the scope of this study.