- April 1999, Birmingham
- Michael Toolan
- Department of English
- University of Birmingham
NB A version of this paper also appears in M.Coulthard et al., eds, Dialogue Analysis VII: Working with Dialogue. Selected Papers from the 7th IADA Conference, Birmingham 1999 (Niemeyer, 2000).
Many discourse analysts are deeply sceptical of speech-act or speech-move schemas, regarding them as unreliable, partial, context-insensitive, disproportionately ‘top-down’, oriented to speaker-intention but unable to track revised intentions, and so on (see, as representative of discourse linguists criticisms of speech act theory, those of Eggins and Slade, 1997 and Lesser and Milroy, 1993). My present proposals are at variance with the speech-act tradition which aimed to specify the constitutive rules for producing, say, a promise. At the same time categorizing some things as questions, others as requests, and so on, seems a necessity at some level of processing–and if it could never be done, with any degree of reliability, we could never elicit information from each other, or get others to do things. But we do; so requests and informs surely exist, and can usually reliably be recognized and oriented to by integrated participants in real contexts of engagement. What we may argue about, however, is the degree to which participants’ reliable recognition of requests, informs, etc., draws on, is guided by, or even is ‘governed’ by, the presence of this signifying formal device or element or that confirmatory sequential factor. The Integrationist and reflexivity-attentive tradition (Harris, 1998; Harris and Wolf, 1998) will also draw one to ask: what are we doing when we classify certain moves as questions, undertakings, and so on?
I will give three simple examples why a workable basic categorization of communicational moves, and further to that a workable subcategorization of move-types, is necessary. Though not a random sample, each of these seems representative of recent work which both depends to a degree on act-categorization, and may be undermined by the uncertainty of the very categorization it would hope to be able to invoke.
Example 1: identifying mitigated directives in doctor-patient discourse.
Why the reliable identification of speech-act or speech-move categories remains in my view a pressing requirement in linguistically-minded discourse analysis is I think indirectly highlighted by an interesting recent paper (Skelton and Hobbs, 1999) that appeared in the February 1999 issue of the British Medical Journal. The paper is entitled:
A descriptive study of co-operative language in primary care consultations by male and female doctors.
As it happens, both authors are in the Department of General Practice Medicine of Birmingham University. In passing it is worth noting that a growing number of language-focussed papers like Skelton and Hobbs’ is appearing–papers that analyse and critique the language of one group of specialist users or another and reporting their findings back to those specialists in their own professional journals.
The authors have taped and transcribed something like sixty hours of general practice consultations, involving both male and female general practicioners, and their present paper uses a very simple operational measure of cooperative language to probe whether doctors differ in the cooperative language they use, depending on their gender. They find no significant gender contrast. As an operational quantification of ‘cooperative language’, the authors search in the consultation transcripts for three measures which they postulate may be indexical:
- 1. sheer number of words used by female doctors in comparison to male doctors;
- 2. frequency of use, by the male and by the female doctors, of facilitative tag questions with falling tone such as It’s a bit sore there, isn’t it?. No significant difference found.
- 3. Frequency of use, by male vs. female doctors, of mitigated directives, such as You could try going for a 15-minute walk before going to bed [invented example-MT].
This third diagnostic of cooperative language is arguably the most discoursally interesting of those the authors opted to use, but it also turned out to be the most problematic to identify and isolate:
There was no clear evidence that either male or female doctors used a greater number of mitigated directives, but this was in part because there was no clear distinction between mitigated and other directives. For example, If you don’t see any good improvement with it then come back and I’ll move you up a scale of sort of erm potency really, give you something a bit stronger. Is this a straightforward directive (it has an imperative), or a mitigated directive (the tentative if and sort of soften it), or simply a statement about what might happen? As the authors note, this kind of difficulty in interpretation of language in context makes quantitative claims–such as that male doctors are dogmatic and female doctors cooperative in speech–dubious. (Skelton and Hobbs, 1999: 577)
One might add that such difficulties, if general and unresolvable, render any discourse analysis that invokes and relies on speech act categories like directive etc. unsustainable. And yet, as the paper itself shows, some medical trainers are keen to use such categories. The other obvious thing to note is that, presumably most of the time the doctors who say things like if you don’t see any good improvement with it then come back and I’ll give you something stronger etc, along with the patients who are addressed in this way, do not seem to have the interpretive difficulties that the counting-and-categorizing analyst has. There is nothing wrong with the data, the stimuli to our metalinguistic categorizations, nor indeed any malfunctioning among the integrated users, here doctors and patients, engaged in using the language in the course of meeting particular needs. The data are not impoverished or degenerate; but our descriptive categorizations may be. Nor should it be thought that the researchers’ problem lies only with isolating mitigated directives, as if what they call straightforward directives are unambiguous–because, perhaps, they are coterminous with use of imperative mood. We know the latter notion is false, and that the category of bare directives is as in need of careful characterization as any putative subordinate categories.
This example is just one of the numerous legal disputes we can find which hinge on the court’s assessment of the nature or intensity or scope of a particular communicative act. This case I will mention is the early medical negligence case of Clarke v Adams (1950) 94 Sol Jo 599, Judge Slade presiding. A physiotherapist was treating a patient who suffered from a fibrositic heel, and the treatment involved subjecting the heel to heat provided by a machine; unfortunately the heel was subjected to too much heat, was further damaged, and an amputation was necessary. At trial the defendant physiotherapist and his employers were held liable for the resultant damage to the patient. The trained and experienced physiotherapist had warned the patient, using the wording he had been taught to use, as follows:
When I turn on the machine I want you to experience a comfortable warmth and nothing more: if you do, I want you to tell me.
Clarke v Adams (1950) 94 Sol Jo 599
But the judge held that this warning was inadequate in the circumstances, in which what he called a ‘warning of danger’ was required: “The warning must be couched in terms which make it abundantly clear that it was a warning of danger.” In short, the courts were of the view that there are such things as warnings, and also a more particular category of ‘warnings of danger’. If anyone ought to be able to characterize and explain such ‘warnings of danger’, discourse analysts should.
The third example is offered simply as a recently published representative of the many articles published over the years, that report ‘difficulty’ with applying extant move or act schemas, and which they then feel little embarrassment about adopting selectively, to devise their own customized system. The article is by Deirdre Bishop and others, appeared in 1998, and is titled ‘When a nod is as good as a word: form-function relationships between questions and their responses’. Applied Psycholinguistics. 19, 415-432. A crucial stage in Bishop et al’s study is the ‘Coding of communicative functions’ (419ff.). They write:
The system for coding communicative functions was developed over a period of five years and was influenced by a range of sources [including Dore, Coulthard and Fey]. No existing system proved entirely suitable for our purposes. Most of the coding schemes developed for handling child language were targeted at relatively young children engaged in toy play, and, as Chapman (1981) pointed out, little information is available on the reliability of existing coding systems.
In the literature, many researchers applying discourse analytical categories can be found making declarations of this kind. The level of divergent understandings and classifications that this seems to indicate is I suggest regrettable and damaging.
A Simple Schema of Speech Moves.
I briefly outline my speech-move schema below. But it is a moot point whether the modifier speech should be retained, in the phrase ‘speech move’, since one underlying goal is to reconnect spoken and non-speech moves, in the way that these are subtly interwoven in many interactions. Ultimately the fact that particular acts are moves in an interactional encounter should trump whether they are verbal or not. In the longer run what I think we should be characterizing are better called communicational moves rather than speech moves. My warrant for this line of reasoning is the Integrational Linguistic principle that verbal and non-verbal behaviour should be re-integrated in our analyses to match their interwovenness in real life (see, e.g., Harris, 1990: 43). Clearly, not all the moves (gestures, actions, etc.) in a spoken encounter need be speech moves; nevertheless, since many are (particularly in plays), and to reflect the interest here in the linguistic exponents of moves, the complex phrase is retained here.
Devising the scheme began with the general points made in chapter 3 of Halliday (1994): when individuals talk to each other, they are enacting exchanges (although below I will substitute the term engage(ment) for that of exchange), and these exchanged phenomena can be thought of as predominantly either mental or physical, and the grammar of English reflects this. If the enacted exchange is chiefly mental, the conversational contribution amounts to a giving of information or a seeking of information; if the exchange is chiefly physical, the contribution amounts to a giving or seeking of goods and services. Four core conversational moves, or acts, thus amount to the giving, or seeking, of either information or goods and services. These can be represented as in the grid below, with example utterances likely to perform each of the acts:
goods & services information
speaker is giving Can I give you a hand? I mustn’t do any heavy lifting
speaker is seeking Will you give me a hand? Have you got a good hold?
Familiar labels for the typical kind of conversational act performed in these four core categories can be proposed:
goods & services information
(typically, non-verbal action) (typically, non-verbal action)
giving Undertaking Inform
seeking Request Question
To a very considerable extent these are Halliday’s categories, if not quite his labels (he opts for Offer, Command, Statement, and Question, respectively). For reasons I cannot elaborate here, given space constraints, I much prefer the move labels Request and Inform to those of Command and Statement; and the Undertaking category is arguably more comprehensive than that suggested by Offer (offers are one type of Undertaking in my schema).
We can also group Undertakings and Requests together, and Informs and Questions: since Undertakings and Requests both concern future proposed action by one interactant or the other, they are called Proposals; since Informs and Questions provide or seek information, they are called Propositions. The future action that a Proposal specifies is normally nonverbal (washing the dishes, closing the door, paying for a purchase item, etc.) although occasionally it can involve a verbal performance (where the teacher requests: Billy, recite the present tense conjugation of ‘donner’, please). The information sought or given in response to a Proposition is normally verbal, but replies to Propositions can be performed nonverbally (A: Where’s the oilcan? B: [points to far corner of garage]).
Undertakings seem to be the least extensively used. But they are a relatively well-defined group: they are proposed future actions or services on the part of the speaker, ostensibly to the benefit of the addressee, the undertaking of which is, significantly, made contingent upon the addressee’s express or implied consent. This last point is arguably crucial, and distinguishes Undertakings from announcements (I’m going to reorganize your bookshelves into some sort of order), which are a kind of Inform. Grammatically, Undertakings often involve a first person pronominal and one of the modal verbs shall, can, or may, where these can be interpreted as contributing to a proposal meaning ‘Do you consent to me doing x for you?’. Alternatively they may use a second person pronominal subject in a projecting structure using a verb such as want or like: Do you want me to take the dog for a walk? Would you like some tea? Semantically, Undertakings seem united in at least implicitly foregrounding the addressee’s postulated modality of willingness or inclination: what they wish or want. A typical and expectable response to an Undertaking is a reply expressing consent–OK; alright–or a declination with a reason for so declining–No thanks, I’ve just had one.
Each of these category types includes various subtypes. Thus Requests range in force from orders and directives to mild suggestions. One of the most extensive recent discussions of requests in English comes in Tsui (1994: 90-115). But there are divergences between Tsui’s category of Requestives, which covers five rather distinct subclasses, and the Request category adopted here. Tsui summarizes Requestives in the following chart:
speaker action speaker benefit request for permission
addressee benefit offer
Requestives addressee action speaker benefit request for action
addressee benefit invitation
speaker and speaker benefit or
addressee action speaker + addressee proposal
Relative to my scheme, Tsui’s requests for action and invitations (such as a host’s saying to a guest “Please sit down”) are unambiguously Requests, while request for permission (granted by addressee for speaker’s benefit–e.g., Can I get myself a drink from your cabinet?), is essentially a request, getting the addressee to consent to a course of action which is to the speaker’s benefit. But Requestives involving speaker action for addressee’s benefit–Can I get you a drink?–are a kind of Undertaking. It is interestingly doubtful whether proposals can be established as a distinct category, since they are quite intermediate: Shall we share a bottle of Friuli? implies shared action and mutual benefit, and by those considerations is both a Request and an Undertaking. Such borderline cases, linguistically non-congruent in Hallidayan terms, with politeness and acknowledge uncertainty of sequel subtly grammaticalized, may well be the most interesting to study in the long run.
Undertakings include promises, vows, some invitations, and offers of help. Informs include claims, warnings and compliments; they entail the imparting–at one level or another–of verbalizable information. Whether or not the addressee finds an Inform informative is a separate matter. And Questions are acts designed to obtain the kind of information that Informs supply.
As is well recognized, the matching of types of speech move with particular linguistic forms is nothing if not problematic. Nevertheless, situated functionalist interpretation can be usefully underpinned, often with reference to formal and grammatical evidence. Confirmatory criteria are ordinarily available, which help ensure that disputes where analyst A says “That’s an Inform” and analyst B says “No it’s not, it’s an Offer” are actually quite rare (and interesting). Similarly, in actual interaction, we are infrequently forced to pause and wonder “Is he asking me or telling me?”, “Was that an offer or a request?”; and similarly the occasions when we do hesitate in this metapragmatic way are worthy of attention for that reason alone.
Two long-recognized confirmatory criteria of Request function are (1) ‘please-insertability’ and (2) prospection. With Requests, unlike Informs and Questions, you can usually insert the word please (or the slightly old-fashioned kindly) before the verb denoting the action to be performed, even if the Request is indirect as in I want you to kindly eat your spinach this minute! Typical prospections of the four moves are as follows:
move type possible prospection:
Undertaking ‘Thanks’ / ‘No thanks’
Request ‘OK’ + action/ ‘No!’ +action
Question ‘Yes’/ ‘I don’t know’
The items listed here as prospections are not cited as necessarily the normal or most usual response to the given act, but rather as entirely possible responses for that act which are in addition highly implausible as responses for any of the other acts. Consider Oh, a legitimate response to Informs. While this would be an awkward response to some kinds of Informs it remains a coherent one; but Oh as a complete and freestanding response to an Undertaking (I’ll put the garbage out) or a Request (Put the garbage out, would you?) or a Question (Did you put the garbage out?) would be decidedly odd.
While Undertakings, Requests, Informs and Questions are argued here to be central to discourse, they are not the only acts involved. But they are treated here as the core ones on the hypothesized grounds that just one among these four can occur as the nucleus of a first move in an interaction. But interlocutors’ responses also merit classification. And in fact each of the canonical initiating acts strongly specifies a particular kind of response:
Undertake – Acknowledge (accept/decline)
Request – (Acknowledgement +) Non-verbal Performance
Inform – Acknowledgement
Question – Inform
This picture introduces just one broad new kind of move, Acknowledgements. Typical examples of verbal Acknowledgements are: thanks, ok, very well, oh, and really?; and nonverbal equivalents are often used instead. Acceptances and acknowledgements are secondary in that they are semantically attenuated, as the above examples suggest, and in that they are contingent upon some prior, exchange-driving act from among the set of four described above: an Acknowledgement cannot normally initiate a speech exchange. Relatedly, while talk is often characterized as kinds of exchange, very little is given (back) when an acceptance or acknowledgement alone is made; and a bare positive acceptance of an Undertaking is not very different from a bare negative one, e.g., a declination of an offer, even though the interactional implications and consequences may be great. Nor, taken as whole groups, are Acknowledgements profoundly different from each other in form or function. For these various reasons, all verbal Acknowledgements are treated here as members of a single secondary class of move.
One revision here of my previous accounts (Toolan, 1997; 1998) is the adoption of the term Undertaking to denote the fourth core move, and the demotion of offer, here treated as just one kind of Undertaking. Offers, too, can be informally characterized as ‘proposals by addresser to do something, for or with the consent of the addressee’, but Undertaking seems the more appropriate general term, confirming that the class includes promises as well as offers (that is, undertakings that are typically ‘binding’ upon acceptance, as well as those that may remain indefinite). Undertakings comprise all speaker-proposals where the addressee’s acceptance or consent is at least implied, if not express. Where the declared speaker undertaking is detrimental to the addressee–I’m going to have you fired–then it will typically constitute a threat, with implied if not overt lack of addressee consent; it thus falls out of the Undertaking category and into that of Informs. The alternative approach, which I have rejected, would be to say that Undertakings are a yet broader category, covering all addresser proposals to or for the addressee, with the latter’s express or implied consent irrelevant. Intriguingly, that approach would somewhat parallel recent judicial reasoning concerning the crime of theft, where it has been controversially argued that the victim’s consent–or lack of it–cannot be determinative (see, e.g., the House of Lords ruling in R v Gomez).
As implied earlier, any proposal (Request or Undertaking) involves a speaker intent on some act of doing, while a proposition (Inform or Question) involves a speaker intent on some act of knowing. And the relations of dependence or obligation between speaker and addressee are often quite different in the four basic cases. A rather different dependence-relation is implied when a speaker Undertakes:
Can I give you a hand with that? Do you want a hand with that?
than when she Requests:
Will you give me a hand with this?
In an Undertaking, as the grammar reflects (…I give you…; â€¦you wantâ€¦), the speech act is focused on the needs of the Other, the addressee–or at least on what the speaker thinks are the Other’s needs. Other is cast as beneficiary, speaker-Self is presented as the giver who, as in any genuine act of giving, is likely to incur some costs. Additionally, although less overtly, the speaker-Self who makes an Undertaking may be adopting a stance of deference or subordination to whoever the addressed Other is (but not in all cases: cf. threats). In a Request, on the other hand, the relations are broadly reversed: the utterance is Self-oriented, imposes some cost on the addressee, and may cast that Other in a subordinate stance. Similar characterizations of Informs and Questions are possible. On the assumption that knowledge is desirable, Informs involve a knowing Self going to the trouble or cost of informing an Other, primarily for the latter’s benefit; and Questions typically involve a not-knowing Self imposing on an Other-addressee, to Self’s benefit. So, unlike the cases in Proposals, in which the subordinate-superior roles match up with those of cost-incurrer and beneficiary, in Propositions the two pairs of roles diverge:
Undertaking: speaker cast as subordinate and incurs cost, addressee cast as superior and beneficiary
Request: speaker cast as superior and beneficiary, addressee cast as subordinate and incurs cost
Inform: speaker cast as superior (the Knower) but incurs cost, addressee cast as subordinate but beneficiary
Question: speaker cast as subordinate but beneficiary, addressee cast as superior (Knower) but incurs cost
But it must be emphasized again that these characterizations are tendencies, fitting canonical instances fairly well and other instances more loosely or not at all. Thus in a canonical inform, the speaker is a cost-incurring Knower, and the addressee is informationally subordinate but the beneficiary. But in an atypical Inform such as a compliment, these characterizations are obviously modulated or tweaked: the superior/subordinate contrast is minimized, and speaker’s cost is slight, and the extent to which the speaker is telling the addressee something they do not already know may range from great to small. Propositions and Proposals contrast in other interesting ways besides. Undertakings and Requests specify actions scheduled to occur within a timespan that extends from the speaker’s present into the future; their temporal reference is delimited to the non-past. Informs and Questions, by contrast, are quite unrestricted in their potential temporal reference: a Proposition can refer to a state of affairs sited in the distant past as easily as in the distant future.
Critics, with reason, sometimes ask ‘why just four primary moves?’; the answer is simply that this level of classification seems to be the one that strikes the best balance, at least as a starting-point for linguistically-supported analysis of communicational moves, between generality and specificity. Four moves is only one category fewer than speech act theory’s five types. And these primary distinctions are the crucial ones to get right. These are the categories that need to be robust and enduring; in this they contrast with the more controversial or opaque subclassifications that have been proposed, notwithstanding the need to develop such more delicate descriptions (e.g., that of Tsui, 1994, or that of Eggins and Slade, 1997). Interestingly, Eggins and Slade’s elaboration of move types is almost entirely focussed upon the second or non-initial moves in exchanges (Eggins and Slade, 1997: 169-226). Their inventory of kinds of ‘sustaining’ moves, if I have counted correctly, numbers 38 sub-types. But their opening moves (and for them all moves are either opening or sustaining), following Halliday as I have done, essentially number just four, with quite secondary distinctions between fact and opinion and openness vs. closedness (Eggins and Slade: 193-4). Thus, for example Is Al Gore ‘Presidential’? would be classed as ‘question:closed:opinion’.
By way of encapsulation, all four core moves can be represented, not so much as kinds of Exchange, but as kinds of interactional Engagement:
EngagementProposition ProposalInform Question Undertaking Request
How valid and valuable can such a scheme be, with its abrupt categorization of initiating moves into four broad types–Inform, Question, Undertaking and Request? Numerous caveats need to be entered. To begin with, the scheme is notional and functional: it suggests that in functional terms, these four moves constitute the four most fundamental, and fundamentally contrasting, possible first steps in interactions. And while I seek to underwrite the validity of the types by making reference to formal and sequential and interactive criteria, ultimately these can never be anything more absolute than an appeal to typical or normal codified practices. Consider please-insertability, prior to the lexical verb, in Requests. Noting such a tendency, in congruent or non-indirect requests, does not amount to saying that please-insertability is defining of or possible in all requests, as any number of utterances, justifiably received as indirect requests will confirm: e.g., It’s awfully stuffy in here with your window rolled all the way up like that. Please – insertability is only a useful (not definitive) diagnostic with interrogative mood utterances that might otherwise appear quite Question-like or Undertaking-like: compare Would you please phone me when you get there with ?Would you please like to live there? and Would you please like me to do that for you?
We cannot establish any set of criteria which will either a) guarantee production of a Request move or b) guarantee production of moves which are not Requests; and mutatis mutandis, the same applies for the other three moves equally; only the situated language user, not any grammatical system, controls these matters. In particular unfolding and unpredictable circumstances, any kind of imaginable verbal or nonverbal material, appropriately deployed, may be interpreted as intended to convey a Request. Nevertheless there is copious evidence to support the contention that users of English routinely perform each of the core moves in patterned and usually contrasting linguistic ways, to the point that those standard formats serve as cues or ’tilts’ in the course of situated sense-making. Accordingly it is useful to pay a good deal of attention to such language patterns, proceeding in a similar spirit to Aijmer, who elects to focus on “the set of strategies which have turned out to be functionally most appropriate and therefore [have] been codified linguistically” (Aijmer (1996): 131), or what Geis (1997: 23) terms “the conventionalization of form for function” which is “a general characteristic of colloquial language”. Such attention to language patterns does not compromise the recognition that no linguistic or semiological feature is a necessary or sufficient condition for any of the core engagement moves.
Finally, it should be emphasized that the core schema of four moves proposed above is intended to be at a remove from speech-act models as conventionally designed. There is no assumption or requirement here that language, as traditionally understood, need be involved in any of these–all four moves can be performed by gesture or gaze, suitably integrated with circumstances. It follows that the moves are not language-specific (they may not even be species-specific), but universals. Ultimately, the moves are to be conceived of as core resources of people, and not of ‘a language’.
- Adams, J. and Bishop, D. 1992. Conversational characteristics of children with semantic-pragmatic disorder. I. Exchange structure, and cohesion. British Journal of Disorders of Communication, 24, 211-39.
- Aijmer, K. 1996. Conversational Routines in English: Convention and Creativity. London. Longman.
- Bishop, D., J. Chan, J. Hartley, and F. Weir. 1998. When a nod is as good as a word: form-function relationships between questions and their responses. Applied Psycholinguistics. 19, 415-432.
- Chapman, R. 1981. Exploring children’s communicative intents. In J. Miller, ed. Assessing language production in children, 111-36. Baltimore: University Park Press.
- Clarke v Adams (1950) 94 Sol Jo 599.
- Culpeper, J., M. Short, and P. Verdonk, eds. 1998. Exploring the Language of Drama: from text to context, London. Routledge.
- Coulthard, M. ed. 1992. Advances in Spoken Discourse Analysis. London. Routledge.
- Eggins, S. and D. Slade. 1997. Analysing Casual Conversation. London. Cassell.
- Fey, M. 1986. Language intervention with young children. Boston: College-Hill Press.
- Geis, M. 1998. Speech Acts and Conversational Interaction. Cambridge. CUP.
- Halliday, M.A.K. 1994. An Introduction to Functional Grammar. 2nd edition. London. Arnold.
- Harris, R. 1990. On redefining linguistics. In H. Davis and T. Taylor, eds., Redefining Linguistics. London. Routledge. pp.18-52.
- Harris, R. 1996. Signs, Language & Communication. London, Routledge.
- Harris, R. 1998. An Introduction to Integrational Linguistics. Oxford, Pergamon.
- Harris, R. and G. Wolf, eds. 1998. A First Reader in Integrational Linguistics. Oxford, Pergamon.
- Lesser, R. and L. Milroy. 1993. Linguistics and Aphasia: Psycholinguistic and Pragmatic Aspects of Intervention. London. Longman.
- R. v Gomez. House of Lords  3 W.L.R. 1067.
- Skelton, J and J. Hobbs. 1999. A descriptive study of co-operative language in primary care consultations by male and female doctors. British Medical Journal. February 26.
- Toolan, M. 1997. Language in Literature: An Introduction to Stylistics. London: Arnold.
- Toolan, M. 1998. The give and take of talk, and Caryl Churchill’s Cloud Nine. In J. Culpeper et al., eds, Exploring the Language of Drama: from text to context, London, Routledge, 142-160.
- Tsui, Amy B. M. 1994. English Conversation. Oxford: Oxford University Press.