Described by its author as ‘an exercise in corpus linguistics’ (p. 5), this book surveys the modal auxiliaries and a range of modal-like expressions in three corpora: the British and Australian components of the International Corpus of English (‘ICE-GB’ and ‘ICE-AUS’), and a specially constructed corpus of American English (‘C-US’). The US corpus is smaller, containing just under 200,000 words compared to a million words in each of the ICE corpora: the spoken texts (about 116,000 words) are taken from the Santa Barbara Corpus, and the written part from the Freiburg-Brown Corpus of Written American English (‘Frown’).

After a chapter on the corpora and one called ‘Theoretical preliminaries’, the chapters which present the substantive data are organised in similar ways. For each item we are given a table listing the frequencies of its main meanings in each of the corpora separately and in total, with the figures for C-US and the total also normalised to tokens per million words. The frequency for each sense of an item is further helpfully given as a percentage of the occurrences of that item. The meanings are distinguished as ‘deontic’, ‘epistemic’ ‘dynamic’ and ‘indeterminate’ in almost every case. Examples for each of the meanings are then presented and discussed. Chapter 3 covers necessity and obligation, chapter 4 deals with possibility, permission and ability, and chapter 5 looks at prediction and volition.

For anyone looking for basic information about English modals and related expressions, the book's clarity and internal consistency, and the attention that it pays to quasi-modals make it a major advance on previous corpus-based studies such as Coates (Reference Coates1983) and Mindt (Reference Mindt1995); it also compensates for the lack of frequency information in Palmer (Reference Palmer1990). Studies of regional variation will benefit from knowing, for example, that shall and be to are far more common in British English than the other two varieties. The book will no doubt become the standard starting point for any analysis that takes frequency seriously, and I will recommend it to my students as the first place to go for reliable data about modals. Collins deserves credit for the careful, painstaking work that he has put in.

That said, the book has several limitations. Arguably, the frequency information for the modals and most common quasi-modals does not progress much beyond the statistics in the Longman Grammar (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan1999). Granted, Longman gives bar charts rather than precise figures, and does not distinguish Australian English; but the global relative frequencies of the different modals are easy to discern, with will, would and can by far the most common, and then a big jump down to could, followed by another jump down to may, should and must (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan1999: 486). The more precise figures in Collins (can with 7,663 occurrences and could with 3,557) do not add very much. The situation is similar for the statistics on different uses of individual expressions: for have to, Collins reports 76.1% deontic uses and 22.4% epistemic; Longman gives a comparable picture, with the bonus of separate figures for conversation and academic prose (Biber et al. Reference Biber, Johansson, Leech, Conrad and Finegan1999: 494). Not all the statistics match: Collins has epistemic uses of may accounting for 79% of its tokens, whereas for can the vast majority of instances (81%) are dynamic (that is, ability) with permission uses accounting for 9.9%. Biber et al. agree on may, but for can they have permission uses coming in rather more often (Reference Biber, Johansson, Leech, Conrad and Finegan1999: 491). For geographical data the book is invaluable, but for information about the meaning and use of modals, Collins covers well-trodden ground in slightly more detail – though he covers a wider range of items than Longman.

I was disappointed by the second chapter ‘Theoretical preliminaries’. On the question of whether modals are polysemous or have a single underlying meaning, Collins says ‘In the present study we assume a polysemy position’ (p. 24) but gives no reasons. The distinction between subjectivity and objectivity is said to be ‘of particular importance in this study’ (p. 28), but it is barely elaborated: the brief discussion seems to assume that subjectivity involves ‘speaker involvement’ (again, not elaborated), but does not pay serious attention to more sophisticated work such as Verstraete (Reference Verstraete2001) (though this paper is mentioned in passing on p. 93), or Nuyts (Reference Nuyts2001). The theory chapter suffers from uncritical reliance on Palmer (Reference Palmer1990) and Huddleston & Pullum et al. (Reference Huddleston and Pullum2002): obviously these are important starting points, but the uncritical acceptance, for example, of their rag-bag category of ‘dynamic’ modality (‘to include ability, volition, and also non-deontic root (or circumstantial) meanings’ (p. 23)) is a missed opportunity for more careful analysis.

In two places, it is possible to compare the treatment in this book with other recent work (in this journal): in both cases, Collins is found wanting. Depraetere & Verhulst (Reference Depraetere and Verhulst2008) give a careful corpus-based breakdown of the different uses of must and have to, with a fine-grained typology of sources of necessity which goes far beyond the simple distinction between ‘subjective’ and ‘objective’ that Collins uses. Declerck (Reference Declerck2010) surveys the different uses of be to, arguing that its modal sense is basic – in accord with Collins, who puts be to in the chapter on ‘Necessity and obligation’ rather than ‘Prediction and volition’. Declerck distinguishes eight modal uses of be to and four ‘futurish’ uses, compared to Collins’ use of ‘epistemic’, ‘deontic’ and ‘dynamic’; Declerck also discusses a range of uses in the protasis of conditionals (of the type if you are to become a professional player, you will have to practise more) in more detail than Collins, who only briefly discusses two examples (p. 85). On the other hand, it may be unfair to expect Collins to devote that degree of attention to each quasi-modal: the book would have been hugely expanded, and serious students can always move on to research articles.

Collins uses the term ‘quasi-modals’ to cover two groups of expressions: ‘semi-modals’ which have an auxiliary as their first element (had better, would rather, be to, have got to) and ‘lexico-modals’ which do not, and which are mostly formed with be and a lexical item (have to, need to, want to, be (un)able to, be about to, be bound to, be going to, be obliged to, be supposed to, be (un)willing to). The semi-modals are said to be like modals in that they lack non-tensed forms (cf. *We will must leave soon and *We will have got to leave soon), whereas the lexico-modals can be non-tensed (cf. We will have to leave soon).

In this messy area of grammar no distinction is likely to be watertight, but this one seems to leak somewhat. If the semi-modals have an auxiliary as their first element, then they can be expected to have the NICE properties (appearing before not, inversion in questions, etc.; cf. Palmer Reference Palmer1990: 4). The lexico-modals formed with be, however, also have all the NICE properties: be in be about to, for instance, is no different in this respect from be to, which has all the NICE properties. Thus the presence or absence of an auxiliary as the first element cannot be the key distinction between semi-modals and lexical modals. Note also that another lexico-modal, have to, behaves like an auxiliary in some varieties of English: Have we to go? would be normal for such speakers, for whom Have you a pencil? would also be acceptable.

The restriction to tensed forms, the other supposed criterial property of semi-modals, is also unreliable. Comparing the lexico-modal be supposed to with the semi-modal be to, we find a similar pattern: for each of them, a search in the BNC yielded only one example containing a non-finite form:

1. (1) You didn't go to dances or anything like that. Yeah. And what about after the funeral day? Did it relax more then Oh yes. or was there a sort of spell afterwards too. Oh yes there would be you'd a year you would be supposed to be not very much. Oh heaven's Oh right. Oh yes, you wouldn't be supposed to go to a dance or corn or anything like that. Oh no. (K6S)

2. (2) Yes we we've adjourned this case for one week other enquiries can be made. So he doesn't have to, on conditional bail today? Well bail will be extended for one week. Well you know what that means Mr, you were told this last week you will be to surrender yourself to bail next Tuesday at er. (F7W)

Both of these are strange, and may perhaps be classed as performance errors, so it is reasonable to say of both these items that they do not have non-finite forms. There are 33 other instances of non-finite be supposed to in the BNC, but they are in formal texts where we are not dealing with the lexico-modal but with the sense ‘be assumed to’, as in:

1. (3) The dynamics of the energy cascade and dissipation may be supposed to be governed by the energy per unit time (J12)

Thus the BNC data do not support a categorial distinction between be supposed to and be to.

The section on will, a particular interest of mine, misses several opportunities to probe a little deeper. Collins follows Huddleston & Pullum et al. (Reference Huddleston and Pullum2002), and the majority of other grammarians, in treating will ‘primarily as an exponent of epistemic modality’ (p. 126); but his figure of 59.2% for instances under this heading includes both the uncontroversial epistemic use and the highly controversial future time use: a pity, as an indication of the relative frequency of these two uses would have been useful. He claims that examples like It'll be Christmas soon include an element of epistemic modality, because ‘the prediction could be prevented from coming true by an unforeseen eventuality’ (p. 128). That may be accurate metaphysics, but it is hardly relevant to everyday usage, in which such predictions are normally treated as certain; and applied consistently, it would mean that no language can be said to have a future tense – a minority position which I doubt that Collins holds.

In summary, this book is an excellent and timely quantitative study, and a solid starting point for further research, but the analysis is rather disappointing.