Language Learning and Computers
English Grammar in Conversation
DEPARTMENT OF LINGUISTICS AND MODERN ENGLISH LANGUAGE,
LANCASTER UNIVERSITY, LANCASTER LA1 4YT, UK
A thorough description of spoken English grammar is felt to be overdue. After all, spoken language has largely been neglected by grammatical tradition; the word grammar itself descends from the classical Greek word for writing. The grammarian's bias towards the written language remains strong today, and is reflected in such terminology as 'left dislocation' and 'right dislocation' in referring to grammatical transformations. To understand these terms, we have to think of a sentence as immobile and visible on the page, like a laboratory specimen, rather than as an ongoing phenomenon of speech. Morever, discrimination in favour of the written sentence, in grammars, has to some extent been reinforced by the advent of computer corpora, simply because until very recently corpora of spoken language have not been available in sufficient quantity and coverage.
The opportunity now exists, however, for corpus linguistics to redress this balance, by taking advantage of the new availability, in the 1990s, of large and varied spoken English corpora. These include notably the spoken component of the British National Corpus, collected by Longman; the spoken component of the Bank of English corpus, collected by the Collins COBUILD team, and the CANCODE corpus, collected by Cambridge University Press. (To say that these publishers 'collected' these spoken corpora is convenient shorthand for saying that they organized, and paid for, their collection.) It is significant that all these large computer corpora of speech have been brought into existence by major British dictionary publishers: in fact, a fourth such publisher, Oxford University Press, was the lead partner of the whole British National Corpus project.
The project on which I am about to report1 was also sponsored by a publisher: Addison Wesley Longman. It is a project leading to a large and detailed corpus-based grammar of English, to be entitled the Longman Grammar of Spoken and Written English (LGSWE)2. The grammar made use of a large and varied corpus (the Longman Spoken and Written English Corpus) of approximately 40 million words. Within this corpus, a core corpus of some 20 million words subdivided into four 'core registers' (conversational speech, fiction writing, news writing and academic writing) formed the main focus of grammatical study and comparison. In its detailed quantitative comparison of conversation (the prototypical spoken register) with three major written registers (fiction, news and academic prose), the LGSWE is making something of a breakthrough in the comparison of spoken and written grammar. And the final chapter (Chapter 14), entitled 'The Special Grammar of Conversation' shines the spotlight strongly on the spoken corpus data (equally divided between British and American conversation) to bring out what is particularly distinctive or 'special' about spoken grammar, as contrasted in different degrees with the three written registers we also examined in detail.
To the question 'Is there a special grammar of spoken English?', any of the three following answers might be offered:
The first answer does not need to be taken seriously, although it is surprisingly persistent in the mind of the folk grammarian. It is inherited from the age-old tradition associating grammar with the written language, and it is bolstered by examples such as the following, which, like others which follow, is from the Longman spoken corpus:
It is true that often spontaneous spoken language does seem to be grammatically chaotic - mainly because of the well-known phenomena of dysfluency that afflict a speaker trying to cope with the pressures of on-line processing. Here the speaker happens to be faced by a difficult planning task, not just grammatically but cognitively: she is trying to explain how to get to a discount store, and seems to have difficulty both with her own incomplete map of the vicinity, and also with the problem of conveying that to someone who may well have a different, also incomplete, map. But this example can be easily be counterbalanced by others such as the following, which has a clear grammatical structure, and hardly any dysfluency. Here the speaker's on-line planning needs are less taxing, as she recounts a (probably familiar) narrative about the family dog:
The only example of dysfluency here is the repetition of if you're, no doubt caused by the overlap between the first occurrence of those words (in square brackets) and someone else's speech. We do not need to follow the 'speech has no grammar' theory any further, but can instead ask 'What kind of grammar is the grammar of speech?'.
The second answer ('the same as the grammar of writing') is one which is encountered quite often with linguists of a more theoretical turn of mind, and indeed anyone who finds it reasonable to talk about 'the grammar of English' rather than 'the grammar of written English' or 'the grammar of spoken English' - in fact, most of us. The authors of the LGSWE follow this line of thinking through virtually all the chapters of their book. In the main descriptive chapters 2 to 13, the same categories, structures, and rules are used for all four registers, including conversation. An example of the kind of frequency comparison which can be made on the basis of these categories is given in Figure 1, which shows the remarkable contrasts of mean phrase length in the four registers, with academic writing at one extreme and conversation at the other.
We find many such striking differences of frequency between the conversational subcorpus and the three written subcorpora - but these differences of frequency would not have made any sense, had we not also recognized that the same categories occur across the spoken-written divide. Conversation makes use of entities such as prepositions, modals, noun phrases and relative clauses, just as written language does. So - assuming, as many would, that differences of frequency belong to the use of the grammar, rather than to the grammatical system itself - it is quite natural to think in terms of one English grammar, whose use in conversational performance can be contrasted with its use in various kinds of writing. In other words, conversational grammar is seen to be just a rather special implementation of the common grammar of English: a discovery which does not necessarily in any way diminish the interest of studying the grammar (i.e. the grammatical use) of spoken language.
Figure 1 : Mean phrase length across registers
The third answer is one which has recently become popular - for example, through David Brazil's book The Grammar of Speech, and indirectly through a number of thought-provoking publications by Ronald Carter and Michael McCarthy. In handling spoken language, Brazil argues for a totally different approach to grammar from the approach which has become familiar through conventional focus on the written language. He argues for a linear model moving dynamically through time, and puts aside the more traditional architectural model in terms of hierarchies of units. Although Carter and McCarthy do not take this thorough-going approach, they do throw the spotlight on grammatical features of spoken language which they feel have been largely neglected by standard grammars entrenched in the 'written tradition'. They argue that structures which are inherent to speech have not been properly studied until the advent of the spoken computer corpus, and are consequently absent from canonised written grammar familiar to learners of English throughout the world: structures such as the 'dislocated topic' of This little shop ... it's lovely or the 'wagging tail' of Oh I reckon they're lovely. I really do whippets. These tend to find their raison d'être in the fact that conversation constructs itself in a dynamic fashion, giving the speaker only a small look-ahead window for planning what to say, and often inducing retrospective add-ons. Carter and McCarthy (1995) put forward a structural model for the clause in conversation, containing in addition to the core clause itself a pre-clause topic and a post-clause tail. With their refreshing emphasis on the dynamic modelling of grammar in action, Carter and McCarthy seem to be taking a line similar to Brazil's advocacy of a new grammar of speech.
My own view on answers 2 ('same grammar') and 3 ('different grammar') was somewhat ambivalent when I came to work on Chapter 14 of LGSWE. But the first task I set myself was to read through the drafts of all the preceding chapters, noting grammatical phenomena which were strongly biased in frequency towards the spoken medium. The result of this was a rich profile of conversational grammar as it distinguishes itself from written grammar in all its variety. This profile included some features, such as the disjunctive prefaces and tags illustrated above, which had already found their way into the 'mainstream' presentation of early chapters. But in frequency terms, I noted a scale of conversational features going from those which are well represented also in the written medium to those which are virtually absent from it - such as dysfluency phenomena, which in written language are restricted to writing modelled on speech, as in fictional dialogue. There were on the other hand features which appeared not to find a place in conversation - such as for as a conjunction, and the substitute pronoun those of those which..., although even here there was the occasional exception. It never seemed realistic, on reflection, to argue that certain features would never occur in speech, or would never occur in writing, because even if they were not detectable in several million words of conversation or written language (as the case might be), they could crop up if more data were considered. I therefore found myself adopting the 'same grammar' point of view, seeing both speech and writing as making use of the same overall grammatical repertoire, but allowing always for cases where the feature in question might be overwhelmingly commoner in one than the other.
To illustrate this 'same grammar - different use' point of view, here is a list of grammatical phenomena which, collectively, go a considerable way towards characterising the grammar of conversation in terms of frequency of use. The important thing is that these can be grouped together (although there is also considerable overlap) in terms of the functional constraints or influences which condition the spoken medium. Without going into precise frequencies, I will mark these features as
In my list there are seven functional headings, representing the conditions operating in conversation which help explain why the characteristics of conversational grammar are as they are: 1. shared context, 2. lack of elaboration, 3. interactiveness, 4. personal expressiveness, 5. real-time constraints, 6. restricted repertoire, and 7. vernacular range.
HDF = high differential frequency; LDF = low differential frequency.
1. Conversation takes place in a shared context
2. Conversation avoids elaboration or specification of meaning
That was sort of a special deal
Kind of a touchy subject
3. Conversation is interactive
4. Conversation is expressive of personal politeness, emotion and attitude
Common adjectives in conversation are mostly evaluative (e.g. good, lovely, nice); these characteristically occur in intensifying coordination good and..., nice and ..., etc.
5. Conversation takes place in real time
6. Conversation has a restricted and repetitive repertoire
7. Conversation employs a vernacular range of expression
The diagram in Figure 2 shows the seven arranged in a ring which shows their overlapping interconnections with one another. I suggest that the connections between the satellite circles can be explained as follows. At the top, shared context links to, and overlaps with, interactiveness in the sense that interactive dialogue can make grammatical short cuts on the basis of an ongoing shared context. On the right-hand side, shared context links to non-elaboration: if we share context, we tend to rely on implicit meanings which require little or no elaboration. Non-elaboration, in turn, links to restricted repertoire because the lack of need to elaborate and specify means that the speaker can rely upon a repetitive repertoire of much-used words and phrases. Restricted repertoire also ties up with real-time processing, because on-line pressures encourage reliance on a limited repertoire of items readily retrievable from memory. If we move again to the left-hand side of the diagram, interactiveness clearly associates itself with politeness, emotion and attitude, both involving interpersonal and affective communication; and politeness, emotion and attitude have a connection to vernacular range of expression, since the use of vernacular or non-standard grammar frequently conveys speakers' solidarity as members of a group. The missing link at the bottom of the diagram defeated me - until I thought of one characteristic which unites them: they both comprise aspects of speech which attract a low evaluation, from the standpoint of the grammatical correctness associated with the written language. But I felt even less convinced of this, and so only tentatively filled in the missing line in the circle!
Figure 2: Seven conditions operating in conversation
Of the seven conversational conditions, some can be thought of as constraining and some are liberalising. For example, 2 'lack of elaboration' constrains a speaker to use few complex referring expressions; whereas 5 and 7 are liberalising, in giving speakers the licence to use a disjunctive range of structures or a vernacular range of expressions that would normally be excluded from edited written language.
It is well known that 'all grammars leak', but thinking of the 'same grammar' / 'different grammar' controversy, I prefer to think of English grammar as made of a rubbery substance that enables it to be squashed or inflated in one part or another according to circumstances. The circumstances of conversation lead to a reduction of the repertoire in certain areas and an enlargement of the repertoire in others - but this is in terms of likelihood rather than in terms of all-or-nothing rules.. So, in the end, this image enables me to keep to the view that English grammar is common to both written and spoken language -- but its shape can be moulded to the constraints and freedoms of each. In this sense, there is a special grammar of conversation.
Brazil, D. (1995). A grammar of speech. Oxford University Press, Oxford.
Carter, R. and M. McCarthy. (1995). Grammar and the spoken language. Applied Linguistics 16(2): 141-158.
Download this paper as a doc-file.
Download this paper as a rtf-file.