This is a big book. There have been many theories of language since Plato, but this is the first to be based on evidence. (That is, evidence of a straightforward variety recognizable to any scientist: substantial samples of the phenomenon under investigation, gathered systematically, in which all sorts of variables can be measured.) Hanks has spent a long and rich career exploring how words do their work. He has been responsible for no less than three landmark dictionaries: Collins English Dictionary in the nineteen seventies, COBUILD in the eighties, and the Oxford Dictionary of English in the nineties. Alongside writing dictionaries, he has been in the vanguard of the use of corpora – in lexicography and everywhere else. (He is most cited in the language technology literature for Church and Hanks (Reference Church and Hanks1989/1990), his work with Ken Church in which they introduced Mutual Information to the field, inaugurating the specialism of lexical statistics.) When Hanks looks at corpus data, he looks at it long and hard, in a way that Natural Language Processing (NLP) people scarcely ever do: he looks at the evidence for a word, to find the patterns and structures which are the substrate of how a word means what it means. Hanks has been doing this as long as anyone, with insights that have inspired many others. He is a, perhaps the, guru of how to do it. The book is rooted in that long and deep experience.

The curse of linguistic theorizing is untamed evidence. Popper tells us that scientific theories must be falsifiable, and can be falsified by a critical experiment giving counter-evidence. In linguistics, counter-evidence is all too easy to come by. Humpty Dumpty said ‘When I use a word, it means just what I choose it to mean – neither more nor less.’ (Carroll Reference Carroll1872.) This is the problem: we can use just about any word to mean just about anything. But that then means we can knock down any linguistic theory by inventing a scenario in which words relate to things in ways, which are at odds with the theory. This is a big problem for linguistics.

Chomsky addressed it by giving special status to native speaker introspective judgements, which convinced many people over a long period, but leaves unaddressed a wide range of problems like differences of introspection between individuals, the constructs of competence and performance, and the strange status of introspective judgements.

Alice's response to Humpty Dumpty is ‘The question is, whether you can use words to mean so many different things.’ Hanks gives this quotation (along with one from Firth) a full page at the start of the book. Hanks is with Alice, and the book aims to show how, and in what ways, words can mean different things.

The subtitle of the book is Norms and Exploitations. A word has its norms of use and meaning, and it is by virtue of these that it can bear meaning, when used in a speech event within a linguistic community. They are known to and shared by the community and we can find evidence of them in corpora – indeed the task of the lexicographer is to identify and present these norms. A word may also be exploited to mean something other than what it usually means, or to play some role other than the role it usually plays. When a speaker does this, they usually take some aspects of the word's norms, and other aspects, linguistic or non-linguistic, of the speech situation, to build some meaning that is not normal for the word. This is an exploitation.

Where an exploitation becomes familiar to a speech community, it might become the basis of a new norm. To give an example: in 1990 Saddam Hussein said, in the widely reported English translation of the Arabic original, there would be ‘the mother of all battles.’ This was not a usual use of ‘mother’ in English, even though it did build on various aspects of the norm for mother, such as deep feeling and significance, and related quite closely to exclamations and expletives such as ‘Holy Mother of God’ (particularly in Irish English) and ‘motherfucker’ (particularly in American English). It was an exploitation. The phrase struck a chord, and before long we started hearing ‘the mother of’ as an emphasizer in a range of contexts. It was showing signs of becoming a new norm.

The distinction between norms and exploitations is central to what should go into a dictionary entry. The dictionary should state the norms, not the exploitations. The norms belong to the language; exploitations belong to specific speech situations. A key skill for a lexicographer is to work out where a particular corpus instance of a word is an exploitation. It can then be set aside when giving a full account of a word's norms. Of course, things change: yesterday's exploitation may be today's norm, and there are delicate judgements to be made.

Given a corpus, we have the prospect of making a statistical distinction between norms (which account for most of the evidence) and exploitations (which are one-offs). This is ambitious: there are many confounding factors, and the chapters, both case studies and theoretical ones, unearth many of them.

The book has thirteen chapters. The introduction starts by asking ‘how do people use words to make meanings?’ and gives a first account of the theory of norms and exploitations, Hanks's account of language. Chapter two, ‘What is a word,’ explores the complexities of that question.

The following chapters, three–six, were familiar to me as they had already been presented at various meetings and published in their proceedings. But each is a masterclass in lexical analysis, of the verbs check, climb, urge, file and the nouns enthusiasm and condescension (and related items), and there can be no complaints that they are brought to a larger audience. They present the method of lexical analysis: the practice of working from the concordances for a word to find its meanings, and the collocations and grammatical structures, which go together to signal and create those meanings.

In some sentences, let us call them ‘cat-sat-on-the-mat’ sentences, all words are used with their normal meanings and patterns of use. But in other sentences they are not. As a lexical analyst (that is, a person doing the core of what a lexicographer does: finding and cataloguing a word's norms of meaning and behaviour), it is important to distinguish the cases. The norms will only be apparent when overlays of other processes are peeled away. Those processes include alternations, and Chapter seven, ‘Three types of alternations,’ discusses lexical, semantic and syntactic alternations. On syntactic alternations, Hanks critiques in detail Levin's ‘English Verb Classes and Alternations’ (Reference Levin1993), finding it interesting but with the kinds of failings in the specifics of analysis that are inevitable when an enterprise of this kind is not firmly based on corpus data.

Another reason why sentences are not all cat-sat-on-the-mat sentences is exploitations, the topic of Chapter eight, which sets itself the ambitious task of building a taxonomy of exploitations. This taxonomy includes many of the bugbears of semantic analysis: ellipsis, semantic-type coercion, metaphor, simile, metonymy, creative uses of constructions, and a bundle of devices for ‘not saying what you mean’: understatement, euphemism, hyperbole, irony and sarcasm. We are also given guidance on distinguishing exploitations from mistakes.

Chapter nine takes us further from NLP than other chapters as its topic is literature and intertextuality, with discussions of the role of folk tales, the Bible and Shakespeare on the lexical fabric of English, and stars James Joyce as the ‘ultimate exploiter’: ‘Finnegan's Wake was Joyce's revenge on the English: he took their language and threw it back at them, scrambled’ (Hanks quotes an Irish critic, p. 276).

Chapter ten, on the ‘complex linguistic gestalt’ of word and pattern meaning, contains a couple of surprising errors. We are told that, in contrast to brambles, cauliflower, cabbage, apple and gooseberry, the words rice, parsnip, radish, leek, apricot, pear and raspberry are examples that are not exploited figuratively and have not developed conventional secondary senses (p. 285). But what about the pear-shaped man blowing raspberries?

The next two chapters pit Hanks's account of language against others, from philosophy, anthropology and psychology (Chapter eleven) and linguistics, of a range of flavours (Chapter twelve). While I can see that these chapters are needed so that Hanks's theory is placed in the intellectual tradition and compared and contrasted with others, these chapters only become relevant if one is arguing against Hanks from one of those other positions. The glory of the book is in the positive account of language it gives, not its contrasts with others.

The book concludes with chapter thirteen, ‘the broader picture.’