Barlow_Blends

Usage, Blends, and Grammar MICHAEL BARLOW Rice University 1. Introduction The extract in (1) 1 contains a short sample of language in use, an orthographic representation of part of an exchange between two American speakers set in an academic environment. (1) a. Why is it chapter 3? b . Well, that's the thing to talk about. While the representation in (1) clearly omits a considerable amount of information, extending from the phonetic realisation of the utterances to their discourse context, it does nevertheless provide empirical data relating to an actual speech event, and as an illustration of language usage it will serve as a useful starting point for the discussion of usage-based grammar presented here. We can take some comfort from the fact that example (1) represents an actual exchange, if small one, between two speakers. However, if we look to the example for insights into usage-based grammar, we see the difficulty of bridging the gap between the particular, the utterance itself, and the general, the grammatical representation that ultimately underpins the 315 3 16 I USAGE, BLENDS , AND GRAMMAR production and comprehension of the utterance-differences between production and comprehension notwithstanding. The strategy adopted here is simply to examine multiple utterances (or sentences) in a corpus, looking for regularities in usage. The relationships drawn between multiple instances of utterances and linguistic generalisations have some parallels in the connection between usage data and the form of a usage-based grammar in that, as we will see, both relationships involve data-driven pattern extraction. Even a cursory inspection of a corpus reveals recurrent regularities in the data, the most obvious of which are collocations or recurrent word associations (Firth 1957, Sinclair 1991).3 We find, for example, that the word thing, which appears in (1b), is used very frequently (in both spoken and written discourses) and occurs in a number of collocations such as the thing is and the .. .thing to do. With appropriate usage data (i.e., corpus data) at hand and keeping in mind the parallels alluded to above between corpus-based generalisations and usage-based grammar, we can make some progress towards answering a question posed (somewhat rhetorically) in Bolinger (1961: 381): Is grammar something where speakers produce (i.e. originate) construct- ions, or where they reach for them from a pre-established inventory, when the occasion presents itself? Obviously, the answer to this question is not going to simply appear out of thin air once we have the requisite corpus data. Consequently, we have to look for clues to the relationship between frequent collocational patterns and their representation in a usage-based grammar, and then consider the implications for less frequent collocational patterns such as the thing to talk about in (1b), which is not a particularly strong collocation. The prevailing view of syntax is that at its core it comprises a set of rules or constraints which account for the creative use of language, and that secondarily or peripherally, the grammar (or lexicon) contains a list of multi-word idioms and other set phrases. I claim that this is exactly backwards: The main component of grammar instead comprises a large set of redundantly specified schemata, both abstract and lexically-specified, and the role of rules or constraints (or highly abstract schemata) is to provide the glue or mortar to combine these prefabricated chunks." Exploring this alternative view, I investigate the relation between instances of usage illustrated by the fragment in (1) and a usage-based grammar, paying particular attention to the role of blending as described in Turner and Fauconnier (1995), Fauconnier and Turner (1996), and Turner (1996) inter alia. I aim to show that what looks like creativity in language is in many MICHAEL BARLOW / 317 instances the result of blending of prefabricated units rather than the output of a set of generative rules. The structure of the paper is as follows: I first examine the consequences of the pervasiveness of collocations for the structure of grammar, turning in Section 3 to an exploration of the nature of syntax in a collocation-rich grammar. The main part of the paper, Section 4, demonstrates the use of corpus data to investigate the extent of blending of prefabricated forms in language production. 2. Collocations and Usage-based Grammar We can describe how a usage-based grammar might differ from a standard grammatical description by considering the following contrasts. Adopting a generic X-bar syntax framework, one could view the utterances in (1) as bracketed structures or tree-structures in which the main category represents a sentence containing phrasal categories (such as NP, N', etc.) which, in turn , contain or dominate lexical categories (such as N). Only the latter categories are linked to individual lexical items, and therefore the control of connections among words is achieved only indirectly through restrictions on the combination of syntactic categories. From this perspective, syntax is a set of constraints governing the combination of phrasal and lexical categories.' And, as is well-known, the typical data used in this tradition is grounded not in actual usage but rather in intuitions about sentences and specifically judgements of the grammaticality of sentences. A model of syntax that aims to explain the distribution of syntactic categories, say AP and N' (rather than lexical items themselves), clearly has value as a way of capturing the range of possibilities of association (as famously illustrated by colorless green ideas), but it is bound to be unsuccessful in accounting for frequent collocations such as those involving the cooccurrence of particular adjectives (e.g., broad) with particular nouns (e.g., daylight). In other words, an X-bar framework can only account for relations between syntactic categories and is unable to represent the lexical or collocational dimension of syntagmatic structure. The ubiquity of dependencies among words (as opposed to dependencies among syntactic categories) is self-evident once corpus analyses are undertaken, suggesting the need for some kind of collocational grammar. And while no such grammar has been produced, the extent of collocations is extensively documented in dictionaries and other reference works such as Benson, Benson and Ilson (1997) , Kjellmer (1994), and Moon (1998), which provide extensive lists of collocations. Moreover, the degree of attraction (colloc- 3 18 / USAGE, BLENDS, AND GRAMMAR ational strength) between any of the words in a particular corpus can be estimated by statistical measures such as mutual information, t-score, cost and entropy." Collocations are not only a feature of corpora; speakers know about collocations, and collocations come in different strengths in terms of how well routinised or entrenched they are in speakers' linguistic systems. Does it matter that X-bar syntax fails to capture certain aspects of what it means to know a language? In one sense, no, it does not. If syntax is taken to be an idealised view of those aspects of language structure related to conjectural learnability issues rather than a model of day-to-day language use, then one might say that the details associated with cooccurrence patterns and frequency of occurrence are peripheral and of little or no importance. On the other hand, if we focus on usage and usage-based grammar, we find that it is the idealised syntax which is of secondary importance and plays a background role in speaker's knowledge, as explained in the discussion below. Using corpora to examine patterns of usage reveals the important role that lexical units larger than a word perform in language production and shows that much of language in use is not creative in the Chomskyan sense, but is based mainly on the use of prefabricated or semi-prefabricated chunks. I argue that the creativity or the expressive dimension to language comes in large part from the modification of prefab structures, rather than the novel combination of lexical categories. Well-entrenched collocations like broad daylight and other examples of fixed and semi-fixed structures such as those shown in (2) are instances of syntagmatic units that can be described in terms of schemata or constructions.' (2) a. one thing b. the thing is c. the right thing to do d. sort of thing e. it is one thing to ... it is another thing to ... The building blocks of a usage-based grammar are not lexical items dominated by lexical categories, but are form-meaning pairings of differing degrees of complexity and different degrees of specificity. Adopting this view of the units of grammar, we see that, in fact, the building block metaphor is not the best for this approach. (See also Langacker, this volume.) For one thing, the "blocks" have both a formal and a meaning side; and some of the blocks come already partially assembled. Nevertheless, MICHAEL BARLOW / 3 19 some sort of "building block model" is necessary to capture the compositional nature of syntax/semantics. The sort of model required, however, must allow for the composition and blending of units large and small. The connection between usage and usage-based grammar is not a simple one, but we can attempt a basic description as follows: repeated exposure to collocations leads to the entrenchment of collocational patterns and their associated meanings in the grammar." This process applies not only to strong lexical collocations such as broad daylight, but also to looser collocational links between words as in [it is one thing to X, it is another thing to Y]. The existence of a cline from strong to weak collocational links among words and the presence of links between collocations (collocations of collocations) are amenable to treatment in a usage-based grammar that, by definition, is structured in a way that reflects input. Additionally, appropriate conditions of language usage are clearly a part of a speaker's knowledge of language and the connections between grammatical units and register, genre, and other types of situational information must also be part of the grammatical representation, but such associations are not pursued here. 3. Combining and Modifying Collocations If we take the position that chunks represented by schemata are the fundamental units of syntax, then we must face up to a couple of problems. One concerns the modification of chunks. How are fixed or semi-fixed expressions such as close, but no cigar modified? If they are unanalysed chunks, then it would not be possible for them to (ever) take modifiers or to allow word substitutions. And it is in solving this problem that something akin to X-bar syntax" may be useful as a guide to the internal structure of chunks, which is a necessary step in the creation of modified structures. A phrase like close , but no cigar may be represented in the grammar with a category label for the whole phrase, but with little identification of internal structure. However, a syntactic template may be imposed on the structure if a speaker wants to create a modified structure, as in close, but no banking cigar''' or close, but no goal. The characterisation of this procedure in terms of an imposition of a syntactic template is, perhaps, an overly dramatic description. Chunks in the grammar have differing degrees of internal structure and sometimes the internal structure will be quite transparent, as in broad daylight and the thing is but in other cases, rather opaque, as in one fell swoop and easy does it. Focussing on usage-based grammar at this level 320/ USAGE, BLENDS, AND GRAMMAR of detail, we also face the fact that individual speakers will vary, perhaps quite markedly, in the internal structure assigned to different fixed and semi- fixed expressions. Individual variation aside, some kind of analysis of chunks can always be made, but given the fundamental role of collocational units, then inevitably higher order syntactic categories playa secondary role. Further issues relating to the modification of chunks will be explored in detail in Section 4. The second problem can be stated as follows. We have made the claim that collocations are an important aspect of syntagmatic structure and yet do not fit within traditional syntactic approaches. However, if collocational units are taken to be fundamental units of syntax, then we need a mechanism to account for the combination of these units. Here once again we can turn to abstract schemata to provide the glue allowing combination of syntagmatic units, although the structures involved will have to reflect the organisation of spoken discourse and hence may differ from traditional tree structures, which are perhaps better suited to written discourse. In a similar vein, we should note that collocational units do not necessarily correspond to traditional syntactic categories. And without promising a complete resolution of these issues , we can make some progress in understanding form/meaning composition by examining evidence from corpus data concerning usage and cognitive representations, focussing in particular on the role of blending. To illustrate these notions, let us look at a couple of examples. In the sentence taken from a spoken corpus shown in (3a), we see the common collocation the thing to do forming a part of the utterance. Based on an inductive view of language learning, the frequency of this construction in corpora might lead one to suggest that the internal representation of this string will have a cognitive reality as a set phrase or grammatical unit. On the other hand, we might ask what is to be made of the occurrence of a worthwhile thing to footnote in (3b). This latter example seems, on the face of it, to be a good illustration of the need for a generative grammar in which syntactic processes control the combination of lexical categories. Certainly a worthwhile thing to footnote is going to be so rare in usage as to count for all intents and purposes as a unique utterance. (3) a. And I think probably the thing to do is have a group write that .... b . I think at some point that might be a worthwhile thing to footnote, ... There are a variety of ways in which this pair of utterances could be handled. For instance, it might be argued that the thing to do is a more or MICHAEL BARLOW / 321 less prefabricated chunk, whereas a worthwhile thing to footnote is created according to a rule-like compositional system. Adopting this approach is equivalent to saying that there are two main ways of constructing sentences: adopting a bottom-up metaphor, we can say that one way is from the word up, and that the other is from the word up, except for those parts for which a prefabricated chunk exists. Little, if any, discussion is given within general syntax to the manner in which the two constructionmethods can co-exist so as to ensure that a prefabricated chunk such as the thing to do meshes with a syntax based on the combination of syntactic categories. I 1 Disregarding the collocational connections between the chunk and other words in the sentence, we still have the problem of how a grammar in which phrasal nodes dominate either other phrasal nodes or lexical categories can cope with a complex four-word expression like the thing to do or with variants such as the best thing to do. Alternatively, some might argue that if a rule-like syntactic system is needed in any case for (3b), then in the interests of parsimony the string in (3a) should also be accounted for by the same rule-like system. To the extent that the utterance in (3a) can be generated by the same kinds of rules as (3b), this seems reasonable. However, this approach is implausible because we are then left with the gulf between, on the one hand, a fully compositional syntax and, on the other hand, usage data that indicates the extensive presence of prefabricated chunks which have unit status on syntactic or semantic/pragmatic grounds. Let us examine a third possibility, which is that a prefabricated chunk is involved in some way in both (3a) and (3b). This third option, explored in detail below, rests on the notion that some of the apparent creativity in language is in fact the result of merger or modification of lexical prefabricated chunks-a partial creativity based on the re-use or re-purposing of prefabricated structures, rather than complete from-the-bottom-up assembly. That is not to say that a bottom-up assembly based on the combination of lexical items never occurs, only that it is the exception rather than the rule. In other words, I argue for the view that strings such as a worthwhile thing to footnote are the result of blending of stored cognitive representations. Thus we can explore the idea that in meeting the needs of a particular communicative situation, there is a merger, or mixing, or blending of form-meaning pairs which leads to an output that may be much more variable than even the full range of underlying or stored cognitive representations, including all entrenched collocations. In the following section we examine the mechanisms of this blending process. 322 / USAGE, BLENDS, AND GRAMMAR 4. The Role of Blending Blending is a general cognitive process involving the merger of formal and conceptual structures to produce new structures that contain partial projections from the input domains, along with new emergent properties specific to the blend (Turner and Fauconnier 1995, Fauconnier and Turner 1996, and Turner 1996). Metaphors and figurative language can be taken to be one type of blending, but there are many other kinds of blends in literary and everyday language, and in non-linguistic domains such as advertising images. A series of Absolut Vodka advertisements, for example, have for several years been based on visual blends which entice the viewer to marvel at the integration of the shape of the Absolut bottle with a view of a famous landmark or scene, and which, at the same time, invite the retrieval or isolation of one input to the blend: the vodka bottle itself. Absolut London, for instance, appears to be a common photograph of the Prime Minister's official residence, 10 Downing Street. Looking closely, however, you see that the outline of the famous door combined with a strategically placed lantern represents the vodka bottle. And in another example, Absolut Amsterdam, the scene is of three tall, narrow Dutch houses next to a canal, but the middle house, which has the same texture and same facing as its neighbouring houses, has taken on the shape of the vodka bottle. These particular examples illustrate a situation in which the formal blend does not correspond to a conceptual blend. In the Absolut ads, there is no conceptual blending that accompanies the visual blending of the vodka bottle and the Prime Minister's residence; there is no blending that goes beyond the blend in the visual image itself. The ad does not suggest, for example, that the Prime Minister drinks vodka at home by the bottleful. Turner and Fauconnier (1995: 202) discuss the potential disconnection of conceptual and formal blends, noting that formal blends can occur in the absence of conceptual blends (as illustrated above). We will return to this disengagement of concept and form below. An important motivation for blending is the push to consolidate several events into a single unit. Noting the pressure to integrate conceptual structure, Fauconnier and Turner (1996: 117) give an example of a non- integrated sequence of actions: 'Jack sneezed. The napkin moved. It was on the table. Now it is off the table.' They point out that English allows the same content to be expressed with the form Jack sneezed the napkin off the table, which represents an integrated conceptual structure. This integration occurs by conceptual blending in which one input