Book Review

Wray, A. (2008) Formulaic Language: Pushing the Boundaries. Oxford: Oxford University Press.

Reviewed by Caroline Tagg

In Formulaic Language: Pushing the Boundaries, Wray extends the theoretical model put forward in Formulaic Language and the Lexicon (Wray 2002), the main claims of which are usefully summarised in the newer book. Wray’s concern is with the psychological and social causes of formulaic language (FL) and, as she herself points out, her definition of FL as strings of words which appear to be processed holistically (that is, without analysis of subcomponents of the string) will not resonate with, for example, a corpus linguist’s identification of FL as frequently-occurring patterns. Nonetheless, the breadth of Wray’s scope and the way in which she ‘pushes’ at a number of assumed boundaries are of relevance to any researcher interested in FL.

Part one establishes the initial boundaries between formulaic language and novel constructions by laying out Wray’s (2002) ‘heteromorphic’ model, which itself challenges traditional boundaries between words and longer strings, and between novel and formulaic language. In the model, linguistic items of various sizes (morphemes, words, phrases, sentences and texts, as well as partly-fixed frames) are stored whole in the mental lexicon. Linguistic input, according to Wray, is interpreted according to a Needs Only Analysis (NOA) so that larger items are only analysed into sub-components where meaning cannot otherwise be assigned. This dual-systems model reflects Sinclair’s distinction between the Idiom and the Open Choice Principles. Wray, however, goes on to suggest that her heteromorphic model in fact “neutralises the distinction between holistic and analytic processing” (p33), in that combination rules still apply regardless of the size of the units: a view which, in my opinion, represents a logical progression in the argument against the special status of the ‘word’ as the building block of language. However, the problem with conflating formulaic and non-formulaic language in this way is, as Wray points out, that it is hardly helpful in exploring FL, and so the distinction between holistically-processed and analysed units paradoxically stands.

Part Two covers theoretical and practical issues which serve to locate further boundaries. In exploring which language models are compatible with her NOA analysis, Wray returns to some of the models discussed in Wray (2002), but while her analysis then focused on ascertaining the extent to which models support a view of complex, holistically-processed units, Wray’s aim this time is to situate her view of FL within a comprehensive model which accounts not only for the psychological or social causes of language use, but also observable language patterns and underlying principles. Her evaluation of Jackendoff’s Simpler Syntax, Halliday’s Systemic Functional Grammar, Pattern Grammar and Frame Semantics, and Construction Grammar is a wide-ranging and interesting, if necessarily selective, analysis of how different models incorporate FL (as defined by Wray). Wray’s conclusion is that her NOA model can be accommodated most neatly into a model of Construction Grammar which, like Pattern Grammar, attributes meaning to units larger than the word but also seeks to explain the causes and underlying principles in a way consistent with Wray’s perspective (p89).

The boundary between formulaic and non-formulaic language emerges again as the definition and identification of FL is realistically problematised and only partially resolved. The first problem is that of circularity: in order to define FL, one needs to find examples of it; but in order to identify FL, one needs a definition. Wray’s solution is to distinguish between a ‘formulaic sequence’ (Wray 2002) as language “which is, or appears to be, prefabricated” and the ‘morpheme equivalent unit’ (MEU) which she introduces in this book as a subset of formulaic sequences which are actually found to be prefabricated: thus giving a step between an initial identification of possible MEUs and a subsequent analysis and definition, whilst providing a broad definition which can be narrowed down. As Wray points out, the problem is not with sequences which are definitely formulaic or those which are definitely not, but with those, on the boundaries, that “may or may not be”. Retracing her steps over ground which she covered in 2002, Wray reviews various criteria adopted by other researchers, including frequency, phonology, form, spelling and intuition, and she makes the useful points that not only does a researcher need to clearly define what they mean by FL, but that their definition will depend on their purpose in exploring FL. The particular problem for Wray is that her definition of FL means that a unit may not have any identifying textual characteristics at all (language processed holistically may not, for example, be very frequent). Her ‘partial solution’ is to establish justifications for the intuitive judgement of a string as formulaic (p. 116-121). These diagnostics include “By my judgement there is something grammatically unusual about this word string” and “By my judgement, this precise formulation is the one most commonly used by this speaker/writer when conveying this idea”. To a corpus linguist, Wray’s reliance on intuition is hardly satisfactory but, given our understanding of both intuition and FL, and acknowledging Wray’s approach, the diagnostics offer a working solution to the potentially unresolvable problem of identifying (at least through data analysis) what in language is being holistically processed, whilst raising researchers’ awareness of the intuitive judgements they may implicitly be making. One final limitation of Wray’s book, however, for those who are interested in the identification of FL (or MEUs) as they occur in attested data, is that the case studies which follow do not fully engage with the issue of identification and are anyway too specialised or ‘local’ to produce definitions for exploring FL more generally in language use.

The six case studies, five of which have previously been published by Wray and co-writers, comprise Part Three of the book. They make for fascinating reading and provide useful data although, as Wray points out, the studies remain descriptive and are not analysed until Part Four. They illustrate Wray’s interest in FL ‘at the boundaries’, where there are unusual constraints on communication and where people are forced or choose to use previously assembled output; that is, where FL meets creative or novel language use. TESSA, for example, is a translation system designed for basic counter transactions at the Post Office (Chapter 10); while TALK is a software programme which uses a voice synthesiser to relay text typed by individuals who cannot physically articulate language (Chapter 11). What these case studies illustrate is, in Wray’s words, the “advantage of matching large, internally complex items rather than their components” (p135), and similar advantages are described for the memorisation of phrases in language learning, either by beginners rote-learning entire sentences which they do not necessarily break down to analyse, or advanced learners memorising ‘native-like’ expressions (Chapters 12 and 13). The final two case studies represent two extremes in terms of the size of lexical unit: the memorisation by actors of an entire script, illustrated by a French and Saunders sketch (Chapter 14); and (in Chapter 15) how differing interpretations of ‘coonass’ as referring to (white) Louisiana Cajuns by a white employee and as the racially-offensive ‘Coon ass’ by a black employee were settled in court (showing how familiarity with a formulaic string can blind users to its component parts). As the above overview shows, Wray’s case studies are (deliberately) extreme ones, both in terms of how holistically processing is defined and the situation in which it occurs and, as such, for those interested in language description, they shed light on, but do not directly explore, the use and identification of FL in everyday language.

The case studies help to address five questions arising from the literature on FL and discussed by Wray in the fourth and final part of the book:

  1. Do we use FL by default?
  2. What determines the level of formulaicity?
  3. How central is FL to language learning for humans?
  4. How central should it be when modelling for computers?
  5. Does FL constrain what we say and think?

Wray’s argument that we use FL by default (which resonates Sinclair’s argument that the Idiom Principle is the default interpretative device) is, for me, the most convincing. In support of the argument, Wray draws on our tolerance for grammatical irregularity within formulaic sequences, our increased reliance on FL under greater processing pressure such as that experienced by sufferers of autism or Alzheimer’s, and the fact that our intuition overlooks the subcomponents of formulaic sequences. This last argument, that we do not analyse formulaic strings, is of relevance to corpus linguists who observe a mismatch between intuition and corpus data regarding collocation and frequency. If, as Wray suggests, our intuition involves consideration of lexical units (words, phrases, or whole texts) rather than their subcomponents, this would explain why we overlook the use of words within larger units and therefore why our answer to questions regarding the use of a word tend to focus on how it is used as an individual unit rather than its phrasal use: why, for example, we define order primarily as, for example, a command or a sequence, but not in the larger lexical unit: in order to . This is, however, an argument also made in Wray (2002).

In considering how the level of formulaicity is determined, Wray puts forward her argument (as outlined in her other publications) that a holistic approach to communication predated, in human history, that of a combinatorial one; and that the adoption of the latter is determined in part by the fact that esoteric or closed societies utilise more FL than exoteric or outward-facing societies. The fact that different societies (or individuals) can have different (holistic or componential) interpretations of the same string is illustrated in Wray’s ‘coonass’ case study.

Wray’s argument that FL should be more central to human language learning builds on her 2002 discussion of second language learning, and constitutes a useful overview of the advantages and limitations of whole-phrase learning. It rides, however, a little roughshod over the question as to how in practical terms MEUs should be incorporated into language learning. Her argument, in the following chapter, that the modelling of language for computers would be facilitated through a focus on large unanalysed units rather than on combinatorial rules seems sensible, although (as with adult language learning) this remains problematic given our inability to define holistically-processed units other than by intuition, and the vast number of non-generative phrases the computer would need to ‘know’.

Wray’s final argument is that FL compromises our creative expression, although it is less likely to constrain what we think. Drawing on Maoist China, Wray suggests that although citizens were frightened into repeating verbatim words from Mao’s Little Red Book, their ability to apply what they were being made to think to real-life situations depended on their having the courage to think outside the book, so to speak. The constraints imposed by FL on what we say, however, are illustrated not only by the TALK software and language learning experiences documented in the case studies, but by the communication systems used in motor racing and in battle. Again, however, these extreme and very constrained communication systems can only speculatively and indirectly shed light on the impact of FL in everyday language; nor is there discussion as to the implications of the possible constraint on creative expression.

Parts of this book serve to reiterate arguments made in Wray (2002) but new contributions include the term MEU and the diagnostics for intuitively identifying them; and the collection of case studies which inform the five questions addressed. The 22 chapters which make up this book are short and thus easy to read and there are a number of clear and useful diagrams, but (for this reader at least) there are too many footnotes, some of which include references to other works which could be included in the main text. The main impression that the reader is left with, however, is the scope of the book and the extensive research that informs it: Wray’s arguments and observations are supported at all times by succinct interpretations of findings of relevant research.


Caroline Tagg (2008) “Book Review of Alison Wray (2008) "Formulaic Language"”, ELR Journal, 2 (1).