or more separate words does have the properties associated with formulaicity or not (see Chapter 8).
While some accounts of formulaic language conservatively only attribute formulaicity to material that already displays the tell-tale signs of it – written as a single word or at least hyphenated, pronounced differently as a whole than as a sequence of component parts, or having a meaning that cannot be easily derived from those parts – other models, including the one used in this book, propose that items become formulaic first, and only later begin to develop the tell-tale signs. The implications of this orientation are immense both for the claims one might make about how much of our language is formulaic, and for how we might identify it – for there could be a huge quantity of formulaic material that does not carry any outward sign of being so. Indeed, any string of words might turn out to be formulaic, and it might take a century or more for that fact to become fully evident.
The situation is exacerbated by the need to distinguish between something that is formulaic ‘in the language’, so to speak, and something that is formulaic for just a particular individual or group; and also to distinguish between what is formulaic for a given speaker and for a given hearer. The notion that the language itself contains formulaic material is inevitable in corpus-driven accounts, where one examines large quantities of text without particular attention to who produced it, and looks for recurrences. If one finds more than a certain number of instances of a particular sequence of words, it is possible to attribute that pattern to ‘the language’.
However, within a psycholinguistically driven account there is less value in attributing formulaicity to strings in ‘the language’, for formulaicity is viewed as the property of a particular string as it is handled by a particular individual. What is formulaic for one person need not be formulaic for another – the reason will become clear as the three claims are discussed below. So a given speaker might produce a word string formulaically, only to have it treated as non-formulaic by the hearer. On the other hand, even an account based on the individual’s knowledge will recognize that many word strings are likely to be formulaic for most native speakers – that is what it means to know the same language.
Definition: the morpheme equivalent unit
As the short account above indicates, it is not really possible to say what formulaic language is without adopting a theoretical position of some sort. Yet any theoretical position sidelines others, making it difficult to relate different people’s accounts, unless everyone is explicit about what they mean. The solution adopted here is to coin a new term and provide a clear and specific definition for it. The term used will be ‘morpheme equivalent unit’ (MEU), the definition of which is imbued with the particular theoretical stance laid out in the rest of this chapter. The MEU is defined as:
a word or word string, whether complete or including gaps for inserted variable items, that is processed like a morpheme, that is, without recourse to any form-meaning matching of any sub-parts it may have.
The definition reflects specific claims about the nature and, by implication, provenance of formulaic material in a language – the claims that are presented in the next sections. It is not, therefore, the kind of definition that can really be used to go out and identify examples of formulaic language in real texts, or to find examples suitable for experimental investigations. The relationship between the MEU definition and that of another term used in this book, the ‘formulaic sequence’ (Wray 2002b: 9), is explored more fully in Chapter 8. A working definition of the terms, sufficient to apply in these earlier chapters, is given in a note in Chapter 1.
Three key conceptual claims about morpheme equivalent units
As noted above, the three key claims presented here summarize features of the model developed in Wray (2002b). It is this model that will provide the context for the interpretations of evidence in the remainder of the book.
The mental lexicon is heteromorphic
The theoretical position taken in this book, on the basis of evidence from multiple sources reviewed extensively in Wray (2002b), is that linguistic material is stored in bundles of different sizes (compare Jackendoff 2002: Chapter 6). That is, the mental lexicon contains not only morphemes and words but also many multiword strings, including some that are partly lexicalized frames with slots for variable material, treated as if they were single morphemes – they are MEUs (see definition above).
Not all strings of words, of course, are prefabricated. As linguistic theory has long recognized, what makes human language special is the huge potential it has for novel expression.3 This potential is only realized because we can both create and understand formulations that we have not encountered before – formulations that, therefore, cannot have been stored in memory in a fully lexicalized form.4
What goes into the lexicon
Because not everything can be stored in the lexicon, any model of how language is composed and understood needs to define criteria to determine which items will have their own entry. Separate consideration must be given to what characterizes a lexically stored word string (for example, regarding its meaning, function, form, and so on) and this matter is extensively discussed in Wray (2002b: Chapter 3). The focal issue here is the impact on the philosophy informing a theoretical model, when notice is taken of psychological processes and sociointeractional preferences as well as the basic patterns and the principles underlying them (see Chapter 6). For some theories it is important that one or another feature of the linguistic system be streamlined (i.e. pared of any unnecessary elements). The desirability of streamlining the system is partly a question of explanatory elegance, and partly, perhaps, a product of implicit or explicit assumptions about the constraints under which human language processing operates. For instance, perhaps there are restrictions on how much lexical information we can remember, though, in fact, there is no evidence of this. Other theorists prefer to keep the ‘rule system’ simple, even at the expense of a streamlined lexicon. That is the position adopted here.
Before the non-streamlined lexical model is explored, it will be helpful to review the main characteristics of the streamlined one. The more that lexical storage is streamlined, the more atomic the lexicon will become. That is, the way to minimize the size of the lexicon is to exclude any items that can be constructed by rule from smaller units. An atomic lexicon will be “like a prison, containing] only the lawless” (Di Sciullo and Williams 1987: 3) – there will be no items in the lexicon that are reducible to smaller semantic units in a somewhat regular way (see Wray 2002b: 265–74 for discussion). An atomic lexicon will still need to admit a few multiword strings, but only those that cannot be generated from smaller parts without a specific rule that has no wider applicability. Examples might include ‘by and large’, ‘very well’ (in the sense of ‘yes’), ‘to boot’ (in the sense of ‘also’), ‘no more’ (in the sense of ‘dead’), ‘at long last’ and ‘as well’, along with – at least for an English speaker with no knowledge of other languages – borrowed expressions such as ‘laissez faire’ and ‘je ne sais quoi’. On the other hand, many word strings could be excluded from an atomic lexicon, even though they carry hallmarks of formulaicity, for example, ‘perfect stranger’ and ‘pack it in’ (in the sense of ‘stop’). Although these items are non-literal in meaning they are regular in form, and so they can, according to such models, be generated using secondary meanings of the component words, or by employing second-order pragmatic mapping to give a holistic meaning to the composed form. Similarly, word strings with an unusual grammatical pattern, for example, ‘believe you me’, might be viewed as
Some versions of Construction Grammar allow for everything to be lexically stored – see Chapter 7– since constructions can be partly or completely lexically unspecified. The point being made here is that the capacity for novelty must be accounted for somehow. In models that accommodate the existence of grammatical rules, albeit sometimes insertional ones that complete lexically stored frames, novelty is achieved by activating the rules to produce word strings that are not independently stored in the lexicon.
