This brief response to Elizabeth Closs Traugott's contribution ‘“Insubordination” in the light of the Uniformitarian Principle’ could begin with John Lennon: ‘Imagine there is no. . . sentence.’

The notion of ‘sentence’ belongs to the oldest and most pervasive ideas in the study of languages, despite all its definitional controversies: is a sentence a unit that expresses a complete thought, or is it a unit with at least a subject and a predicate? Is it what we find between a capital letter and an appropriate punctuation mark, or is it maybe the largest unit to which syntactic rules apply (see Ries Reference Ries1931; Seidel Reference Seidel1935; Matthews Reference Matthews, Jacobs, von Stechow, Sternefeld and Vennemann1993: 91)? In the face of these controversies, Matthews (Reference Matthews, Jacobs, von Stechow, Sternefeld and Vennemann1993: 91) identifies two main criteria for sentences: completeness and correctly fitting words. In this article I would like to focus on the first criterion: completeness. It will be argued, in line with Traugott (this volume) and Miller & Weinert (Reference Miller and Weinert1998: 31), that this notion of completeness has to do with formal written language and schooling. The detailed study of spoken language in interaction might require a different set of criteria and features; here the notion of ‘sentence’ (and especially the implied concept of ‘complete sentence’) can indeed be misleading and should probably be replaced.

This is not to say that we need to reject the idea of sentences altogether. The grammatical unit of the sentence, and with it the criterion of completeness, is of course very valuable and illuminating in the analysis of conceptually written language, as Miller & Weinert point out (Reference Miller and Weinert1998: 71). In spoken informal language, however, the situation seems to be different. Here we find incomplete sentences, including cases of ellipsis, of insubordination, or monoclauses introduced by subordinators (MISs), fragments, chunks. And these seem to be anything but rare, though we are still lacking reliable statistics on the frequency of ‘incomplete constructions’ in spoken language.Footnote ¹ The question then is why many of these utterances are typically classified as ‘incomplete’, i.e. as somehow derived from complete sentences, but lacking. This takes us back to the primacy of spoken language, as advocated by Saussure and many others.Footnote ² Much, if not most of our theorizing today is based on written, planned language. The data we work with are mostly planned and written. Only about one-fifth of the Corpus of Contemporary American English (see http://corpus.byu.edu/coca/) is spoken (but maybe also scripted, planned) data; the NOW corpus (see http://corpus.byu.edu/now/) with 2.8 billion words (and growing) contains only planned, written data. Of course, it is easy to see why spoken data are hard to come by, difficult to present and tricky to work with. And yet we need to constantly remind ourselves that most of the data we work with are written, planned texts. This in turn has consequences for our theory building, our methodology and even our basic assumptions.

Another important point is that spoken, unplanned language (in contrast to written, planned language) usually occurs in actual conversation, in interaction between two or more speakers. It therefore presents completely different problems and thus also requires a major rethinking of methodologies and categories. Halliday (Reference Halliday1989), for example, already pointed out that for spoken language, a much more productive unit of analysis is the clause (which can be combined into various clause complexes). Chafe (Reference Chafe1994) concentrated on intonation units as correlates of cognitive units. In a similar vein, Traugott (this volume) shows that we need to focus our attention on interaction, on the ‘syntax, semantics, pragmatics and prosody of contributions to ongoing discourse in the light of both what precedes and what follows’ (p. 306). In other words, we need to develop a truly interactional grammar if we wish to do justice to real-life spoken interaction.

In concrete terms: what would happen if we reverse our general approach to the study of spoken language and treat spoken, informal language as basic, and formal written styles as secondary and derived? Instead of calling the answer of Cody-1Actor2 in (1) incomplete, elliptical, or even a case of insubordination, we could take a fresh look, as Traugott does.

1. (1) JANICE-1CUSTOMER# What do you want?

   CODY-1ACTOR2-# Like, if you could, just the regular kind.

   (COCA, ABC Primetime 2012)

If we do not assume an underlying, basic, complete sentence, there is no need to treat if in (1) as a subordinator since there is no matrix clause. Consequently, its free use as in (1) does not have to be treated as ‘insubordination’. Rather, one could see this as an interactional chunk, or possibly even a construction in the technical Construction Grammar sense.

Alternatively, one could treat if as a grammatical element which can have a subordinating function (e.g. in complete, written sentences). But this is not necessarily its basic, prototypical function (this is maybe only the case if we treat written language as primary!). And there are a number of elements and constructions that exhibit a similar behavior. In some (mostly written) varieties, they can function as syntactic subordinators and coordinators. But in informal spoken language, they show a very different behavior. Other chunks (or fragments) such as and then what (2), so what then and What about (both in (3)) also seem to operate on a non-sentential, interactional level.

1. (2) But we can share food. We'll share food. Yeah, you guys share food. And then what, you know, these guys don't eat all that much either, I've noticed.

   (COCA, Hot Topics, ABC 2015)

2. (3) The old man held up the bottle of Everclear, raised his eyebrows at her: Well, darlin’? She nodded, took the glass from him, and knocked it back in one swallow. Oh, that's not gonna slow me down, she said. Mero expected her to neigh. So what then, Rollo said, picking at the horse manure under his boot heel. What about Tin Head and his galvanized skull plate? I heard it this way, she said.

   (COCA, Annie Proulx. 1997. The half-skinned steer. The Atlantic Monthly 280(5), 116)

What would be the consequences for historical linguistics from a cognitive point of view if we abandoned the idea of ‘complete sentences’ as the basic unit of syntactic analyses? The first problem that we encounter has to do with historical data, as Traugott points out. The further we go back in time, the fewer informal or interactional genres we find. This means that for earlier Englishes, it is very difficult to investigate language in its natural interactional context. But even here there are some genres, or at least some texts (like Ælfric's Colloquy) that lend themselves to such an approach. Note, by the way, that example (3) above is not from a spoken, interactional genre, but rather from literature representing spoken interaction. This is a methodological aspect closely related to historical pragmatics (see Kytö Reference Kytö, Jucker and Taavitsainen2010), which could lend itself to fruitful collaboration between literature, linguistics, pragmatics and syntax in the future. At the same time, it is also clear that the surviving manuscripts of, for example, Old English or even Middle English can in no way be representative of the Old or Middle English language as a whole, mainly because there is only a certain number of text types available. Even such an impressive collection as the Helsinki Corpus (see http://clu.uni.no/icame/manuals/HC/INDEX.HTM), which explicitly tries to be representative, finds its limits. In the category philosophy, for example, we find only very few texts, and none in the Old English III subperiod. Similarly, the category Law is represented by only one text for every subperiod. When we take the corpus behind the Old English Dictionary as our basis (http://tapor.library.utoronto.ca/doecorpus/wordcount.html), we only get fewer than 4 million words. Even the ‘smaller corpora’ of Present-day English operate with at least 100 million words.

Moreover, the further we go back in time, the less we know about the concrete text production settings (dictation, editing etc.), and the way texts were received. All this does not preclude a cognitive perspective on historical language processing – on the contrary. If we assume Traugott's Uniformitarian Process Principle (UPP) to be operative, then we can use our present-day findings and hypotheses to investigate historical language stages, and maybe we can even shed some more light on how texts were, produced, processed and perceived. Maybe this brings us a bit closer to a fuller understanding of the relationship of conceptually oral and conceptually written language in pre-standardized times. But we need to keep in mind that this all is still hypothetical. It might well be the case that former times fostered, or even required, different cognitive skills in speaker/hearers or reader/writers. Chafe (Reference Chafe1994), for example, suggests that it is writing and written language that led to new forms of representing consciousness. This could involve different cognitive abilities – but the jury is still out on this. What seems plausible at this point, though, is that when writers were forced to decontextualize what they previously had available as orally conceptualized and contextualized language, they were also forced to conjure up the ‘distant consciousness’ of the reader and the cognitive rigging behind it. To what extent this means that spoken language structures such as adjacency pairs and incomplete sentence are distilled into written structures remains a question for future research.

It would probably be wrong to claim that MISs and other ‘syntactic fragments’ are developments of Early and Late Modern English. First, given the UPP, we should assume that the processes involved in the production and perception of language were the same then as they are now.Footnote ³ Since the negotiation of meaning today can be shown to happen in interaction, as ‘the expansion of prior utterances by constituent increments’, or ‘“chunks” that can be used for moving discourse forward’ (p. 306 above), it seems plausible to assume that the same kind of processes also took place in the past – until we find evidence to the contrary. Meanwhile, given the UPP, we can try to reconstruct some of the interactional patterns for earlier Englishes, even with sketchy data. This finds further support in the fact that ‘the text-sentence is not a stable unit of syntax but a discourse unit whose composition and complexity is subject to cultural variation and rhetorical fashion. Clauses (and phrases) are not subject to such variation and fashion’ (Miller & Weinert Reference Miller and Weinert1998: 41). Miller & Weinert show that we can find ‘fragmented’ or ‘incomplete’ sentences (even in writing) across many different cultures and languages, and also across time. They argue that in the development of some languages – like Latin or French – we see the gradual, and often even consciously organized movement from simpler, paratactic structures to more hypotactic, complex constructions, which are then treated as the hallmark of elegant, elaborate style. For English, and many other modern, standardized languages this elegant elaborate style, with independent, complex and complete sentences is taught at schools, and is expected in many genres of formal, written language. And it informed many of the architects of syntactic theory. But when we look at the phenomenon from the viewpoint of language and cognition, it is certainly plausible to assume that earlier spoken interactional language was not that much different (process-wise) from contemporary interaction. As a related question, we'd then have to ask ourselves how written styles developed over time. Mair (Reference Mair, Hickey and Puppel1997) describes the ‘colloquialization’ of written language under the influence of spoken styles. But an equally valid question looks at how formalization of written language took place before that (and maybe still is taking place in some genres).

Traugott in her article argues that it could be worthwhile to shift our focus. When we study natural, spoken interactional language we need to take into account syntax, semantics, pragmatics and prosody in order to uncover the joint meaning making processes. But this means that we might have to reconsider some long-standing, dear traditions, such as the complete sentence. This is not to say that the complete sentence is useless and should be discarded altogether. As has been argued before: it is still an important and valuable concept for the analysis of written, planned language. But perhaps spoken, unplanned, interactional language functions differently. This new, usage-based, cognitive-interactional perspective is something that every now and then has already been addressed in the literature (e.g. in Miller & Weinert Reference Miller and Weinert1998 or Chafe Reference Chafe1994; note that the latter offers an interesting discussion of the impact of writing on consciousness and cognition). But it has never taken center stage. And in doing so, it is also important to look into language use in the past. This enterprise is surely complicated by the available data; at the same time, understanding the present is often easier when we consider its development.

This short reply began with John Lennon, so it may as well end with him: ‘Imagine there's no sentence. . . it's easy if you try’.