The Soul of a New Cliché: Conventions and Meta-Conventions in the Creative Linguistic Variation of Familiar Forms

Creativity – whether in humans or machines – is more than a matter of simple creation. To be “creative” implies an ability to do more than invent, but an ability to recognize and appreciate the inventions of others. After all, the ability to recognize surprising value in the efforts of others is the same ability we use to guide our own creative efforts. Solipsistic creativity is rare indeed, and most creativity relies on an audience that is creative enough to value our efforts. Of what value is an ability to e.g. speak ironically if we cannot also understand or appreciate the irony of others? The goal of imbuing computers with creative abilities must thus include a sub-goal of enabling computers to recognize and respond appropriately to the creativity of others. As computers are increasingly used to analyze the burgeoning texts of the world-wide-web, the ability to automatically detect and analyze the linguistic creativity of speakers has become more important than ever. In this paper we consider how speakers engage creatively with cliché, to achieve creative ends through the novel variation of familiar linguistic forms. Our computational analysis of a large collection of linguistic patterns on the Web shows that speakers are surprisingly conservative in their variation strategies, and novelty alone rarely leads to creativity. This conformity can make it easier for computers to detect when speakers are using familiar language in truly original ways. INTRODUCTION Samuel Goldwyn, the co-founder of MGM studios, famously summed up Hollywood’s attitude to creativity with the line “Let’s have some new clichés”. On the face of it, this seems like just another one of Goldwyn’s many memorable misstatements (like “include me out!”): after all, it’s hard to think of clichés as new, or as something that can be invented on demand. Yet, on closer analysis, one can find real insight in Goldwyn’s remark. Clichés are considered anathema to the creative process because they represent everything that is conventional and jaded about the status quo. However, clichés become tired thru overwork, and are overworked precisely because they prove themselves so useful in so many different contexts. Few creators set out to create a new cliché, but most would like their efforts to become as much a part of the fabric of our culture as the most tenacious of clichés. Nonetheless, cliché is generally derided for its baleful effect not just on language, but on thought. George Orwell, in a much-quoted polemic from 1946 (Politics and the English Language [12]), poured scorn on two particular forms of clichéd language: the expedient use of familiar metaphors that have lost their power to evoke vivid images; and the use of readymade turns of phrase as substitutes for individually crafted expressions. Rather than bend words to their meanings, Orwell worried that clichés entice lazy writers to bend meanings to their words. He derided the over-use of readymade phrases “tacked together like the sections of a prefabricated henhouse”, and fretted that any writer who operates by “mechanically repeating the familiar phrases” is simply “gumming together long strips of words which have already been set in order by someone else”. Orwell offers a typically monochromatic view of clichés that accentuates the negatives and overlooks the positives. In dismissing clichés as flyblown, jaded or over-worked, he himself succumbs to what Christopher Ricks [14] calls the language of cliché-clichés. The critic William Empson admired Orwell’s eye for cliché, but rejected his proscriptive approach, memorably calling Orwell “the eagle-eye with the flat feet” [3]. As Empson demonstrates with this marvelous combination of two old tropes, clichés are simply lexical resources, like words, and their creative value (or lack thereof) lies entirely in how they are used. The linguistic Web is home to all the clichés of language, but it is also a space in which speakers feel free to vary these clichés to suit their own needs. New variations evolve quickly on the Web, and new stereotypes arise to anchor these variations in a shared knowledge of popular culture. In this paper we look at one particular pattern of new clichés, the XYZ construct [22], which allows a speaker to figuratively describe an X from the domain Z in terms of an apt vehicle Y. XYZ metaphors allow speakers to turn a well-known concept into a concise and vivid descriptor, as in “The telegraph was the internet of the 19 century.” We view the linguistic Web as a corpus from which we can harvest a large collection of figurative XYZ instances, and thereby study in microcosm the conventions and metaconventions of everyday linguistic creativity. For easy retrieval from the Web, we focus on XYZ instances in which the Y field is a proper-named individual. Our analysis of these figurative XYZ instances will allow us to explore the ways in which speakers obey, and sometimes transcend, semantic and pragmatic conventions when seeking to use language creatively. COMPUTERS AND CREATIVE LINGUSITIC VARIATION There is a popular misconception that creativity is a matter of breaking the rules. If this were the case, there could be no “fair” creativity in chess, in sports, or in any rule-defined context, yet we see many admirable instances of creativity in each of these contexts. It is more accurate to say that creativity is matter of exploiting or subverting conventions [4], and because many of our conventions are so deeply entrenched, we unquestioningly view them as rules [22]. For a computer to be creative, it must have a knowledge of these conventions, not at a hard-wired level where they are coded as rules, but at a knowledge-representation level where they can be modified and manipulated. To attain this level of knowledge, it is best if a computer acquires an understanding of conventions (and their limitations) for itself, rather than have them hard-coded by a programmer. Creative people often exploit an expectation gap between themselves and their audiences: they see orthodoxy as a set of conventional norms that can be challenged [4, 5], while their audiences expect these norms to be obeyed as rules. Creativity in language is largely a matter of playing with conventions [4], to establish expectations in the minds of an audience that are then dashed, in ways that surprisingly add rather than subtract meaning. Our analysis here reveals that while creative variation is pervasive in language, speakers tend to be semantically conservative in how they combine their clichés and stereotypes, so that very few expectations are undermined. Such linguistic variation can yield novelty, but yields little in the way of lasting value or creativity. These findings thus suggest several reasons to be optimistic about the ability of computers to detect and appreciate the creative language of others (and, in turn, to generate language of comparable creativity themselves). The first reason is that creativity is surprisingly dependent on a robust knowledge of cliché and stereotype. Before humans or computers can be creative with language, they first require a firm grasp of the conventional ways in which speakers are uncreative with language. The necessary knowledge can be easily harvested from everyday language, as e.g., found in abundance on the Web [20,21,23,24]. The second reason is that the ways in which people playfully combine stereotypes or vary the contents of familiar forms can also be learned from observing how speakers achieve these variations in well-defined contexts, such as e.g. in similes and in figurative XYZ constructions. These are meta-conventions for being creative with language, and these too can be learned from careful observation of human speakers. The third reason is that speakers often mark their attempts at creativity through their use of linguistic support structures. Some structures are as subtle as the addition of a hedge marker like “about” or “not exactly”, while others are as overt as the addition of an explanation. These structures help to convey the meaning of a linguistic novelty even when the underlying conceit (such as a humorous metaphor or ironic viewpoint) falls flat. RELATED WORK AND IDEAS The linguistic Web is a vast reservoir of opinions and beliefs, as well as a fertile breeding ground for new stereotypes and for new variations on familiar forms. The Web can thus be used as a corpus [7] from which general observations about language can be derived, which may support specific observations about linguistic creativity. When used as a language corpus, the Web has some noteworthy qualities, beyond the obvious benefits of scale. For one, it is a dynamic corpus, always changing to reflect new additions and new users. When one uses the Web as a corpus, one is using the most up-to-date corpus available. For another, the linguistic Web has a variable resolution. If one finds a phenomenon of interest, one can always go back for more examples, or wait for more to occur naturally. Thus, one can work with a sample of the Web, downloaded locally, and extend this sample dynamically as the need arises. One can also use the linguistic Web as a bridging corpus, to dynamically fill in the gaps in a more conventional corpus. Of course, care must be taken when using the linguistic Web as a corpus, not least because this is not the principal purpose of the Web, and those who create it and provide access to it have no responsibility to provide the balance that linguists usually require from a corpus. As noted in [8], the hit counts provided by search engines for a given query do not have the same authoritativeness as statistics derived from a corpus compiled by linguists. Generally speaking, one should be skeptical of absolute page counts, and work instead with relative measures where possible. Despite these caveats, the Web is a marvelous source of language users [9] and of language data [7]. When the folklorist Archer Taylor compi