In recent years, interest in issues of non-discreteness has grown within both theoretical and descriptive linguistics, as witnessed by such publications as Bod et al. (Reference Bod, Hay and Jannedy2003), Aarts et al. (Reference Aarts, Denison, Keizer and Popova2004) and Fanselow et al. (Reference Fanselow, Féry, Schlesewsky and Vogel2006). The book under review here fits well into this current trend, though it has a different focus of attention.

In its most frequent usage in current linguistics, the term ‘gradience’ refers to gradient data from experimental or corpus studies and the task of developing models that reflect this empirical gradience. Bas Aarts focuses instead on gradient categorization – a topic that has been popular since the 1970s in several branches of theoretical linguistics. Aarts considers the difficulties that arise in syntactic classification and categorization: Which morphosyntactic categories are we to assume, how shall we define them and how can we recognize them? Do they have sharp or fuzzy boundaries; do they overlap or are they inherently gradient?

The book consists of eight chapters, starting with a brief ‘Introduction’. The other seven chapters are organised into three parts. Part I, ‘Theoretical background’; mainly presents a thoroughly accomplished overview of the literature: Chapter 2 considers ‘Categorization in linguistics’, from Aristotle and the early grammarians of the linguistic tradition to twentieth-century approaches to linguistic categorization, focusing heavily on the latter. In this chapter, the centre of attention is not on gradience but on categorization per se, examining the question of which categories are assumed and on what basis they have been defined in various philosophical and linguistic schools and traditions. This historiographical perspective is also prominent in chapter 3, ‘Grammatical gradience’, which reviews a number of proposals on gradient categorization in twentieth-century linguistics. Chapter 4, ‘Gradience and related notions’, differentiates gradience from similar notions like markedness and prototype, among others.

Part I of the book deserves high praise. It will be immensely valuable even to scholars who do not have an active interest in the issues under discussion here but nevertheless want to know what the discussion on gradient categorization is about and what were and are the most influential authors, proposals and ideas. Aarts provides an excellent overview of the debate on gradient categorization, especially for the time period from the late 1950s until today.

Part II, ‘Gradience in English: Case studies’, contains three chapters, whose titles mirror the classification that Aarts develops in the preceding sections: ‘Subsective gradience’ (chapter 5), ‘Intersective gradience’ (chapter 6) and ‘Constructional gradience’ (chapter 7). The case studies in this part are partly taken from the literature, mainly from twentieth-century linguistics, and partly present original work by the author. Part III, ‘Formalization’, consists of only the final chapter, ‘Modelling syntactic gradience’.

Aarts's distinction between subsective gradience, intersective gradience and constructional gradience is useful and helps to classify the findings in chapters 5–7. Intersective gradience (IG) is gradience between categories. Several authors have proposed that categories like noun, adjective and verb are ordered on a continuous scale without sharp boundaries between them. Expressions may have nominal, verbal and adjectival properties at the same time, albeit to different degrees. Aarts convincingly argues against such a conception, claiming instead that categories have sharp boundaries. But he also argues that categories are inherently gradient and located close to each other in some kind of ‘category space’. Hence, it is possible for elements to be peripheral within one category and thereby approximate another category without becoming a member of it. Although I have some doubts about this conception, I will not address this issue here.

Instead, I will concentrate on subsective gradience. Subsective gradience (SG) refers to gradience within one category. As an example of SG, Aarts refers to the relation between common nouns, proper nouns and pronouns within the category of nouns, as widely described in the literature. He discusses in detail Crystal's (Reference Crystal1967) proposal, which makes use of matrices and argues that the best way to define a word class is by listing the phonological, morphological, lexical, semantic and syntactic criteria that pertain to it. Crystal's four criteria for nounhood are as specified in (1).

1. (1) Defining properties of nouns

   1. (a) p[roperty]1: ability to act as subject

   2. (b) p2: ability to take number inflection

   3. (c) p3: ability to co-occur with an article

   4. (d) p4: ability to take a nominal suffix

Aarts's approach towards gradient categorization is largely based on the methodology used by Crystal. But where Crystal sees these criteria as a discovery procedure for categories which have not been set up in advance, Aarts (104f.) adopts the ‘now generally accepted’ view that verbs and nouns exist in all languages and even ‘that the two classes are innate’. Accordingly, he argues that criteria like Crystal's ‘can be regarded as defining criteria for the class of nouns … [and s]etting up such criteria inductively will constitute the partial setting up of the grammar’ (104).

The above quotation reveals a weakness of this study: Aarts makes strong claims about syntactic categorization, for instance, with respect to the categories' innateness and thus universality, but the discussion in the book is exclusively about the syntactic categories of English. The four criteria in (1) might be helpful for English, but what about languages that do not have number inflection, articles or nominal suffixes? Would they not have nouns, or only nouns that belong to the periphery of the noun category as defined above? The four properties listed above may be a reasonable choice as criteria for ‘discovering’ nouns in English, as Crystal put it, but how can they be definitory for an innate category?

Assume that ‘noun’ is an innate category, call it INOUN. Assume further that the English learner has learned to categorize the parts of speech in her linguistic input in such a way that there is one category ENOUN which is defined in the intended way by the four criteria in (1). How does the speaker know that ENOUN is an instantiation of INOUN? If there were any substantial criteria for deciding this, they would have to be innate, too, and they would provide an innate definition of INOUN that may or may not fit ENOUN. Assume now that this definition of INOUN does not fit ENOUN. Would the language learner then feel that her language is missing something? Presumably not. Hence, I would suggest that we either use innate categories with innate definitions, or language-particular categories with language-particular definitions. Confounding innate and language-particular categories and criteria in the proposed way does not make sense to me.

With respect to the proposed gradient ‘nounhood’ of pronouns, proper nouns and common nouns, one might wonder whether this is really an instance of gradience rather than subcategorization. Indeed, I do not see any need for assuming gradience in this particular case. Pronouns, proper nouns and common nouns are different subcategories of the category ‘noun’, and none of the three ought to be considered more central than either of the others.

Aarts further claims that definitorial criteria must be morphosyntactic in nature (105). In his section on the category ‘adjective’, he nevertheless uses criteria that do not quite fit this requirement.

1. (2) Defining properties of adjectives

   1. (a) p1: attributive position  a happy woman

   2. (b) p2: predicative position she is happy

   3. (c) p3: intensification very happy

   4. (d) p4: gradedness happy/happier/happiest

   5. (e) p5: un-prefixation unhappy

The criteria in (2) might appear morphosyntactic since they refer to morphological and syntactic properties. However, it seems to me that the explanation for an adjective lacking one or the other property is often of a semantic nature. For instance, colour adjectives lack properties p4 and p5. This has clear semantic reasons. Does that make them less central adjectives? What we have here (as well as in property p3) are morphosyntactic properties that result from particular semantic properties of adjectives. It seems to me that a gradient syntactic category ‘adjective’ based on these criteria would in the end group adjectives with respect to particular semantic properties. Whether one considers these properties to be essential for the category or not – or indeed more essential than other semantic properties – is an arbitrary decision. Why should happy be a more central adjective than blue? I fail to see how this could deepen our understanding of syntactic categories.

Criterion p2 is independently problematic. While it is correct that adjectives can occur in predicative position, this is also true of other categories, like noun phrases, particles, infinitival or finite subordinate clauses.

1. (3)

   1. (a) Mary is an author.

   2. (b) The TV is on.

   3. (c) He is to leave his job.

   4. (d) She is what I would call ignorant.

The question then arises why p2 should be definitorial for adjectives if it does not discriminate between categories. However, if we exclude p2 in addition to p3–p5 (the latter for being semantically-based), we end up with a single definitorial property (p1). An adjective would then be defined as an element that stands in attributive position, which means that ‘adjective’ would be a particular grammatical function rather than a word class.

A troublesome consequence of the book's restriction of the discussion to English is that morphology is largely ignored when determining word class membership. For example, German adjectives in attributive position appear with adjectival agreement morphology, which provides a clear indicator of word class. English, too, gives a few morphological hints in the case of adjectives: only adjectives form comparatives and superlatives (cf. (2d)), and only adjectives are suffixed with -ly to form an adverb. We can simply assume that elements that have one of these properties are without doubt adjectives.

Not all adjectives have one of these properties, though. Colour words, for example, have neither. However, given that they have the same syntactic distribution as clear cases of adjectives, i.e. they occur in attributive and predicative positions (cf. (2a, b)), colour words may be unambiguously classified as adjectives. What I describe here is a heuristic, a discovery procedure for English adjectives. Such procedures are well-established practice in descriptive linguistics but do not lead to gradience within a category. As mentioned above, in languages with agreement morphology on adjectives, such a procedure will be even simpler and likewise not impose any gradience.

Neither from Aarts's review of earlier work, nor from his own original contribution, does a strong argument follow that we need to sort words into categories in a gradient fashion. The facts that are correlated with gradient category membership are often observations of the same kind as those that are used as definitorial properties for a category. It appears that in principle, any observation (related to the morphosyntactic domain) might count as a defining property.

Aarts's formalization of syntactic gradience in the final chapter of this book provides a model in terms of sets of properties based on a set-theoretical treatment. As this model follows rather straightforwardly from the discussion, I will not go into its details here.

All in all, the book leaves the reviewer with mixed feelings. On the one hand, I strongly agree that gradient phenomena are a central issue in contemporary theoretical linguistics. On the other hand, the gradient approach is much less motivated in the area of categories than in other areas, and the strong bias towards English may have led Aarts to misconceive the importance of semantically-based criteria. This is a shortcoming of the book which could have been avoided under a more comparative approach.