In Teun van Dijk (ed.), Introduction to Discourse Analysis.
London: Sage, 1996, 35-62.
The Story of
Discourse Analysis
Robert de Beaugrande
In the mid-1990s, providing a ‘user-friendly’
introductory survey of the research trends which have contributed to discourse
analysis is a tough problem. My solution is to ‘tell a story‘ highlighting the
main ideas that have helped or hindered discourse analysis, crediting projects
or researchers only for illustration. Many citations and references are found
elsewhere.1
A. ‘Science’
as map-making
In the popular ‘story of science’,
scientists observe things in the real world
and then describe or explain them accurately, steadily piling up more facts and marching on toward the final truth. A more
accurate, though less dramatic, story would tell of scientists drawing a series of maps
for a rugged and periodically shifting terrain
that is not located in the ‘real
world’ of ordinary experience yet is connected
to it in ways the scientists determine. This terrain requires specialised methods for observing, describing, or explaining. Each map fits the
vision of its makers and the intentions of its users, putting some features or
places into sharp perspective and ignoring others, just as ordinary maps look
different if they show roads, climates, altitudes, or mineral deposits. No
scientific map is ever final or complete, just as no ordinary map ever becomes
identical with the terrain.
Again like an ordinary map, a
scientific map gets tested by seeing
how well it helps people find things. The natural
sciences make maps of a sparse terrain — one that can be disconnected from the rich and messy
world of ordinary reality. Looking at the world around us and the society we
live in, we do not see the forces and particles of physics, or the compounds
and polymers of chemistry, or even the cells and neurons of biology. But these
sciences give us reliable maps revealing these things in the underlying order of reality, and give special instruments for observing, such
as microscopes and particle accelerators. Some instruments can even intervene in those orders, e.g., by
splitting atoms, synthesising compounds, or removing unhealthy cells.
The human sciences, in
contrast, face a rich terrain closely connected to human
activities. Here, observing ordinary experience is much easier than deciding
what ‘underlying order’ to look for and what instruments to use for observing,
let alone for intervening, e.g., to render humans more co-operative (Part F).
The human sciences have drawn two
kinds of maps. Outside maps try to record what people are actually observed to do, e.g., how people within a culture
and society behave in the roles of chiefs, priests, warriors, hunters, farmers,
and so on. Inside maps try to infer from observations and
introspections what people think and believe
and how they organise their knowledge of the world
and of their society. Drawing both kinds of maps together might be worthwhile, using the one kind
to better understand the other and trying to show the whole big picture of
human life. But until recently, most scientists and philosophers have
recommended drawing only one kind
or the other to suit the
changing fashions of science. The recent struggle to regain the big picture was
major motive for discourse analysis, as we shall see.
B.
Disconnecting the ‘science of language’ from discourse
This general ‘story of science’ sets
the scene for our more specific ‘story of language science’. Surely, science
should investigate the capacities of human beings for using the natural languages developed by a culture or society for communication in
everyday life. But language is immensely
rich — vastly connected to many
things in many ways. How can science tackle it or keep it from spilling into
the terrain of other human sciences, like history, anthropology, sociology, and
psychology?
The ‘language science’ of the 20th
century, known as modern linguistics,
resolved to disconnect language and study it by itself.2
But in the world of human beings, you won’t find a language by itself — the
Dutch language strolling by the canals, or the English language having a nice
cup of tea, or the German language racing madly along the Autobahn. You only
find discourse, that is, real communicative
events. So ever since, linguists have
been trying to reconstruct language disconnected from discourse, believing that
it should be done to create a proper
‘science’ and not seriously question whether it actually could be done.
Let’s watch what happened in the
‘story of language science’ during the first half of the 20th century. Language
got divided up into various ‘domains’
to be studied separately, starting with the ones that
are easier to disconnect from discourse.
Linguistics made a fine start by describing the simplest language sounds in the domain of phonology. A grand vision of
‘underlying order’: beneath all the practical
sound-units that speakers of a language
actually produce or recognise lies an ideal
system of theoretical sound-units called phonemes we can precisely describe. In physical terms, for pronouncing the sound of ‘d’ in an English word
like ‘daunting’, the front teeth stop the flow of air and the vocal chords
vibrate; just try it. If the vocal chords didn’t vibrate, you’d get ‘t’ in
‘taunting’. So we can classify the sound of ‘d’ as the ‘voiced dental stop’ in
the system of ‘English phonemes’. The system nicely connects back to the real
world and yields a convincing ‘map’ of the anatomy of the mouth, nose, and
vocal tract with the locations and events for producing ‘phonemes’. In mental terms, each ‘phoneme’ must allow
us to tell the difference between
words that are also different in meaning,
say ‘daunting’ versus ‘taunting’. So we get both an outside map of how sounds
are pronounced, and an inside map of what sounds speakers tell apart
in words. Moreover, our ‘phonemes’ match the letters in the alphabet,
which was tidied up and enlarged to make the international phonetic alphabet for recording the ‘phonemes’ of
any language.
The next domain of language to study
was the simplest meaningful forms, called morphemes, described by morphology — for good luck, the two
names rhymed with ‘phonemes’ and ‘phonology’. These theoretical units
correspond to word-parts as practical units. You find ‘morphemes’ by writing down language samples and
‘segmenting’ them into the smallest pieces that still seem to mean something.
For example, the utterance ‘the labels were blurred and unreadable’ would have
6 words but 11 morphemes:
[1]
the+label+s+seem+ed+blur+ed+and+un+read+able
Written this
way, little ‘maps’ show language data as a left-to-right series of smallest
meaningful units in the same order they are spoken or written. Again, the
connected to reality seems clear through the visual image of the segments. But
wait — a visual image is far less ‘real’ and permanent than the human vocal
tract; and written segments can mislead. If we had ‘ineffable’ or ‘impossible’,
not ‘unreadable’, in sample [1], where are the morphemes? English has no verbs
‘to eff’ or ‘to poss’; a native speaker may not know that the stems were
borrowed from Latin via French. Do we count them anyway?
Other problems soon arise.
How do we make the whole big map of all
the ‘morphemes’ in a language, like the phonologists’ map of all the phonemes?
Listing all word-parts plus all undividable words is an enormous job. Andidentifying the ‘morphemes’ by
segmenting transcribed utterances doesn’t tell us just how to classifying the pieces? In [1], we
already see different types of meanings. Some units
like ‘‑ed’ have sparse meanings
(e.g. ‘past tense’), while others like ‘read’ have rich meanings (e.g. ‘inspect and understand writing’). Thesparse ones come in little sets, such as the
set of present and past endings on verbs, whereas the rich ones come in big sets, such as the set
of verbs like ‘blur’ and ‘read’. Not
surprisingly, the sparse little ones were picked to be the main ‘morphemes’,
while the big rich sets were called lexemes,
the theoretical units corresponding to words
as practical units, and handed over to the domain of lexicology.
We readily recognise the pattern in
the science of language called ‘modern linguistics’. The resounding success in
describing sounds established the idea that you can indeed disconnect language
from real discourse and discover an underlying order. After that, language
science had some standard ‘rules’:
(1) Study one domain of language
at a time.
(2) Describe each domain as asystem of theoretical units corresponding to the practical units in the data.
(3) Describe each unit by the features that clearly identify it from
the rest (e.g.,, being a ‘voiced dental stop’ or being ‘past tense’).
(4) Investigate by carefully transcribing the native speaker’s utterances, segmenting them into units, and classifying
the units.
These ‘rules’ worked well in
‘phonology’ and got transferred into ‘morphology’, where they worked under more
special conditions. In ‘lexicology’, they started to break down because the
number of units is so vast and many have no features to clearly identify them.
So lexicology was often left along the margins of linguistics or assigned to
‘lexicographers’, who explore how to make dictionaries.
The next domain had an old
name waiting for it: syntax, a name
for ‘tying things together’. The units to be tied would presumably be
‘morphemes’ and ‘lexemes’. But what does the tying’? We can only inspect the pieces; we have toinfer the ties. And the usual
tactics of segmenting data and then classifying the pieces involvesuntying the pieces, which doesn’t
help much.
To be consistent, the theoretical units would be syntagmemes,3
corresponding tophrases andclauses as practical units. But how many
units would the whole system have,
and how should we find and classify
them? Compare sample alternative wordings [1a-1d]: do we give each sentence its
own separate description or do we try to show how they are all related?
[1a] the labels were blurred and unreadable
[1b] the labels were blurred and
unread
[1c] the labels were too blurred to
be readable
[1d] the blurred labels weren’t readable
Linguists decided to the show the
relations among similar sentences. The ‘outside maps’ of the units as we find them
‘in’ language data were traded for ‘inside maps’ of underlying patterns and rules
that explain how units get tied together to produce the data we find.
This shift raised another
important issue for our story: not which
data you study butwhere you go searching for your data. The project of studying ‘language
by itself’ doesn’t say where because language isn’t found by itself. The search
methods that most firmly established linguistics as a science were developed by
fieldwork linguistics You draw outside maps by going out to ‘work’ in the ‘field’ of cultural and social
activities and carefully recording what native speakers of previously undescribed languages are actually observed to say. Your work is difficult and obliges you
to live for extended periods in remote areas. In exchange, you have the
privilege of working at the cutting edge of your science and describing a
language that was unknown to other scientists.
Also, you have the huge
practical advantage of meeting authentic
data of natural language in actual contexts of situation,
instead oflanguage by itself.
Gradually, you join the social practices of interaction and conversation, which
supply continual tests: if your conclusions are wrong, you’ll get corrected, misunderstood,
teased, or ignored. So you have constant opportunities to check your results
before you present them as a ‘map’, And you can justly claim to have made the
map the hard way, without relying on personal intuitions the way you could do
for your own native language.
The other search method has
no special name, so I invented the term homework
linguistics. You draw inside maps of language by staying at home or
in your office and using introspection
and intuition to determine what
native speakers know about their language. Being a
native speaker yourself, you might decide that fieldwork and authentic data aren’t necessary for doing linguistics. You can do your homework
with your own invented data and state the ‘rules’ you were
presumably following.
At this point in our story
the main definition of a ‘language’ also got shifted: not a set of
systems of theoretical units corresponding to the practical
units you find in the
data, but rather a system of underlying patterns and rules that arrange
and transform the data. Looking back at samples [1a],
[1b], [1c], and [1d], this system would explain
a complicated pattern by transforming
it into simpler patterns, e.g., changing [1a] into [1e].
[1a] the labels were blurred and unreadable
[1e] something blurred the labels + someone couldn’t read the labels
The system did not have to explainwhy the more complicated pattern might be actually uttered. So
linguistics could continue officially studying ‘language by itself’,
represented by handfuls of invented sentences and still disconnected from
ordinary discourse.
But here our story begins to
look bleak. After three decades of research on syntax, no such system of
underlying patterns and rules has yet been produced for any natural language.
All we have is a pile of fragments such a system might contain, but no idea how
they fit together and how we can supply the rest. The problem is simple and, I
am convinced, unsolvable: the arrangement of words in phrases and sentences is decided only partly by syntax,
and partly by speakers’ knowledge of the world and of their society. To explain the arrangement, we must reconnect language with that knowledge and shelve the project of describing language by itself. And we must
quit working with invented data and start working with authentic data. These prospects are precisely what discourse analysis intends
to achieve.
C.
Reconnecting the ‘science of language’ to discourse: Large-corpus linguistics
My story so far was naturally
too simple. Many distinguished linguists never proposed to disconnect language
from discourse, such as American tagmemics, the Prague school, and British
systemic functional linguistics, whose projects are reviewed in Part E. And
fieldwork linguists, as I said, maintained the connection in practice regardless
of their official theories.
Nonetheless, the majority view in
linguistics has usually been that discourse is too rich and diversified, too
intimately tied to the ordinary world of human activities, such as casual
conversations among friends or family, to be a proper object for science. In
the vision of one leading discourse analyst, Joseph Grimes (1975: 2),
‘linguists’ might feel ‘like the Dutch boy with his finger in the dike’,
fearfully imagining ‘the whole wild sea out there’ — ‘business letters, conversations,
restaurant menus, novels, laws’, ‘movie scripts, editorials, without end’. So
we can appreciate why ‘discourse analysis’ has, until recently, rarely been a
title for academic courses, or for chapters in introductory textbooks, or for
sections in congresses of linguistics
Today, the scientific trend carrying
the label of discourse analysis is
increasingly conceived to be a programmatic counter-current to the
self-conscious ‘disconnection programme’ sketched in part B. We emphatically
define language as a system integrated with speakers’ knowledge of the world and society. This system
should be described inlinguistic, cognitive, and social terms, along with the conditions
under which speakers use it.
Our precepts might seem to make the job of describing language messier and less
disciplined than the old programme. In fact, I shall claim just the opposite.
For the old programme language, was
a uniform and stable system defined in its own
terms — it had one ideal underlying order beneath the mass of particular data. If this idea were
correct, then linguistics should plainly reveal three kinds of success:
(1) the coverage
of a language should keep getting wider;
(2) the various descriptions should converge; and
(3) linguists or linguistic schools should reach a
firm consensus about how to proceed.
But we actually see a mixed picture: very high
success in phonology and fairly high in morphology, middling in lexicology, and
low in syntax. In my story, the mix has to happen because these domains are not
equally easy to disconnect from real discourse, especially after you’ve booted
out fieldwork with authentic data in favour of homework with invented data. To
make real headway on coverage, convergence; and consensus, we must regard
language as a diverse and dynamic system designed to provide the means for
human communication. The system does not have or need one underlying theoretical order because it is constantly
creating multiple modes of practical
order wherever discourse takes place. When you try to disconnect the system from discourse, those
modes of order dissolve, and you start imagining all sorts of complicated rules
for ‘assigning structures’, ‘disambiguating meanings’, and so on — for solving
the very problems you’ve created by disconnecting. But if you insist on
connections between language discourse, these problems don’t arise.
To see my point, let’s examine one
recent linguistic approach, which has by far the most real data ever assembled.
Large-corpus linguistics, as
pioneered by John McHardy Sinclair and his team, exploits the advanced
technology of powerful computers with spacious memories to gain ‘access to a
quality of evidence that has not been available before’ (Sinclair 1991: 4).
Intriguingly, the Sinclair team, whose ‘functional linguistics’ I cited at the
start of Part C, were among the earliest linguists to call their research
‘discourse analysis’ (e.g. Sinclair & Coulthard 1975; see Part E). Today,
Sinclair compares his computer to the instruments and technologies that enabled
swift advances in other sciences, such as the microscopes and particle
accelerators I mentioned way back in Part A. Of course, the computer differs by
displaying evidence that was easy to observe before, but too plentiful and
diverse to manage. The computer enables us to see patterns that don’t emerge
either from modest sets of samples or from the introspection and intuition — a
very different ‘underlying order’ than sets of ‘theoretical units’ and
‘syntactic rules’. Instead of just making general statements about ‘the English
language’, we can use the data base to explore how general or specific our
statements ought to be.
My data here were taken from
the ‘Bank of English’ corpus at the University of Birmingham, storing around
200 million words of authentic spoken and written English discourse from books,
newspapers, radio broadcasts, telephone conversations, letter-box mailings,
radio broadcasts, and so on. The primary data display consists of the key words
we pick to search for, plus their collocations,
i.e., the ‘company they usually keep’.4 For the English verb
‘warrant’ as key word, some of the corpus data I found are shown here:5
< their circumstances
simply do not warrant charitable assistance. <t> For >
< bark disease. Degenerating trees warrant
specialist attention. Felling or >
< Costa Rica’s economic conditions warrant the cut
in aid, which the country >
< insists there is enough evidence to warrant an
investigation / One suggestion >
< the national objectives at stake warrant the
deaths of U.S. troops / Oil, >
< these old homes are chilly enough to warrant
guests wearing thermal long johns >
< revelations of an affair did not warrant my
leaving the Government. <t> I >
Before looking at the data, I had no clear idea of the
meaning and uses of the verb ‘warrant’, and I hardly use it myself. Now, I can
define the meaning as: ‘provide an occasion where a reaction would be
appropriate or expected’. I can also tell which people are likely to utter such
data: one who might be expected to react, or somebody reporting what they said.
The contexts imply public exposure, where the discourse would raise large
issues like ‘economic conditions’ or ‘national objectives’. So the typical
speaker represents some institution or authority, and the data also tell what
kind: government, judiciary, military, sports, business, science, and medicine.
To use the word is to engage in a subtle
gesture of power — maybe why I
don’t use it.
The data also showed me something
else my intuition couldn’t supply: the occasions doing the ‘warranting’ are
usually bad ones, so that the
reaction would control the ill effects or punish the persons responsible. The
most common collocation was whether ‘evidence warrants an investigation’ or a
‘trial’, recalling the legal discourse with the noun ‘warrant’ in collocations
like ‘search warrant’ or ‘warrant for arrest’. So if we read ‘there wasn’t a
single incident to warrant any action from me’ we can assume it was a bad
‘incident’ calling for retaliatory ‘action’. Often, the situation is even more
specific: when something bad happens or gets discovered, excuses are made why
the expected reaction won’t happen,
as when the ‘revelations of an affair did not warrant my leaving the
Government’, or when somebody else’s ‘circumstances simply do not warrant
charitable assistance’ from me. Such uses are gestures for showing power while denying responsibility.
Authentic discourse samples also show
that infrequent data may be no harder to manage than frequent data. We easily
tell what is meant when ‘degenerating trees warrant specialist attention’ or
when ‘old homes are chilly enough to warrant thermal long johns’. Even odd or
vague data appear simple, e.g.:
[2] < shampoos are effective enough to warrant only
one shampoo per wash.>
[3] < the White House says these air leaks do not
warrant military interception >
[2] was probably intended to praise a commodity, the
shampoos’ (substances) that ‘warrant only one shampoo’ (one act of use). In
military discourse like [3], an ‘air leak’ can have the unusual meaning of
aeroplanes or missiles rather than air passing through.
As you see, describing authentic
language data does not get messier
and less disciplined when we retain the connections to human knowledge of world
and society. On the contrary, the data only get that way when we try to
disconnect them by segmenting them into theoretical units or writing
‘underlying rules’ or ‘features’.
Large corpuses offer valuable
support for the project of discourse analysis to return to authentic data. This
support is all the more vital now that the label has caught on, and many
studies claim to be ‘discourse analysis’ while still writing abstract ‘rules’
for invented data, rather like a textbook called ‘modern astronomy’ but still
teaching astrology.
Large-corpus linguistics also
promises unprecedented advances in coverage
as big teams of linguists with sophisticated software collaborate in the
description of corpus data. Descriptions would tend to converge, thanks to maintaining rich connections. And we should
reach a consensus because we are
using our shared linguistic, cognitive, and social skills, which are also
shared by the people who produced the data in the fist place.
Now, Joe Grimes was right about that
‘whole wild sea’ that a science of discourse must confront, as quoted near the
start of Part C. But having so much data has the advantage of forcing us to
consider and explain the human relevance that justifies us doing one
project in discourse analysis rather than some other. Even my brief
demonstration with ‘warrant’ raised the issues of power and responsibility you
soon encounter when you quit segmenting or analysing ‘sentence structures’ and
start asking who says what and why.
D.
Language and discourse, science and power
The popular ‘story of science’
(invoked at the very beginning) observing things in the ‘real world’ and
describing or explaining them accurately handily conceals the prospect that science has power — power to say
what the ‘truths’ shall be, who gets to state them and decide what they mean,
and who gets to learn them and where. Even better concealed is the prospect
that science also has responsibility — to consider and influence
how its results will be used to make life better or worse, safer or riskier,
more humane or inhumane. This responsibility is greatest when the object of
investigation happens to be discourse: the main human channel for organising
life and deciding who knows or does what — whether knowledge and power will be
shared or hoarded, whether people accept or deny responsibility for what they
do or say, and so forth.
Now, if a science of discourse
analysis neglects its responsibility, we are not just being cynical or lazy; we
are also bypassing the most humanly relevant issues within our own domain. If
participating in discourse carries social responsibility, participating in a
science discourse carries far more, as we get a steadily clearer and larger picture
of how some people are much better than others at using discourse to reach
their goals.
To prevent misunderstandings, we
must stress the contrast between our programme versus the campaign of traditional grammar. Since ancient times, language
guardians have been rallying to
‘preserve good language’ and ‘rescue it from destruction by vulgar speech’.
They invoked such lofty ideals as ‘logic’ and ‘purity’, but their real motive
was to legitimise the language
variety of powerful people as the only ‘correct’
or ‘proper usage’. In this
way, other people could be disempowered
by excluding them from public discourse or else forcing them to participate
with feelings of anxiety or humiliation. So most of the ‘incorrect’ or
‘improper’ usages were picked out from the normal language of the social groups
who were to be disempowered.
We must understand that the issues
of usage over which the campaign was usually fought were unrelated to successful communication, and why. They were technical or finicky fine-points
worked out or just made up by the ‘language guardians’, typically schoolmasters
or clerics who were naturally self-conscious about their speech and keen to
supplement their meagre incomes by giving lessons or writing books and
treatises about ‘correct usage’. Having no scientific methods of gathering and
sorting data, the guardians just followed their own intuitions and wishful
thinking.
Over the years, long disorganised
lists have been compiled and passed along, in handbooks or schoolbooks, of all
the things you should or should not say. In school, pupils are regaled with
bales of confusing and impractical advice purporting to ‘improve’ their
language. Most of the pupils who prosper come from social backgrounds whose
language varieties already resemble the ‘proper’ variety. A few others succeed
in switching to the ‘proper variety’ though great diligence, but risk sounding
a bit awkward in ‘proper society’ and very awkward among friends and family.
The rest just leave school convinced they still don’t speak or write ‘good
English’, now regarded as a mysterious game whose rules change from teacher to
teacher and never get clearly explained. Indeed, it’s a game everybody is told
to play if they want a good life, but the rules are made so confusing and
complicated that people need special lessons at exclusive places fittingly
called ‘grammar schools’.
In this game, prissy and irrelevant
rules have a good market value to mystify ordinary people If you proclaim it
‘wrong’ to ‘put a preposition at the end of sentence’, people struggle to say
fussy thing things like [4] and hesitate to say things like [5] and [6], where
the alternatives are plain awful.
[4] to whom did you give it? [not: whom did you give
it to?]
[5] what did she look like? [awful:
like what did she look?]
[6] he’s worth listening to [awful: to him is worth
listening]
Such ‘rules’ are in effect unworkable solutions for imaginary
problems, and reveal an insensitivity
for English grammar: an expression such as ‘look like’, ‘listen to’, or ‘pay
for’ is not verb + preposition but rather a phrasal verb that can no more be
taken apart than can ‘resemble’ or ‘hear’.
I am not suggesting some grand
conspiracy among teachers of English; many have sincerely aspired to improve
the English language and to help ordinary people. Instead, educational
practices have evolved and survived to suit the fundamental contradiction
in the ideology of Western
societies between inclusive theory versus exclusive practice (Beaugrande 1996). In theory,
all citizens have the same basic human rights to free speech, public education,
scientific training, and so on; in practice, the great majority are
systematically excluded. We can find this contradiction in all our democratic
institutions, above all in our public schools, which in theory offer everyone
equal chances to succeed by merit and in practice programme many young people
for a lifetime of failure and poverty (Lemke 1990). Applied to language, the
same contradiction favours approaches that in theory teach everyone to speak
‘properly’ and in practice select pupils from privileged backgrounds — quite
apart from what the teachers might intend.
The first organised resistance
against the campaign of language guardians was mounted when ‘modern
linguistics’ declared its resolve to describe
actual usage: what people do say and not what language guardians think
they should say. ‘The non-standard speaker’ was sensibly counselled to ‘take
pride in simplicity of speech and view it as an advantage’ (Bloomfield 1933:
499). But this resolve was stunted by the project, sketched in Part B, of
orienting the description toward some ideal
underlying order of ‘units’ and ‘rules’, which was too near the old ideals of
‘logic’ and ‘purity’ invoked by grammarians. Providing accurate registers of
the ‘phonemes’ and ‘morphemes’ of many languages and language varieties does
not yet account for the effective discourse strategies genuinely essential to
successful communication. Nor does it supply practical alternatives for
language teachers who would gladly relinquish the old campaign against
‘incorrect grammar’. So the prospects for concerted interaction between English
teachers and linguists have not moved very far beyond the programmatic stage of
hopeful talk about a ‘revolution in teaching’ (Postman & Weingartner 1966).
I raise these issues because we can
expect the programme of discourse analysis for supporting effective
communication to be both accidentally misunderstood and deliberately
misrepresented in contexts of institutional power, where language is currently
a major pretext for denying human rights (Phillipson & Skuttnabb-Kangas
[eds.] in press). Language guardians will resent us for demystifying their
campaign not to ‘improve language’ but to protect privilege and legitimise
social and ethnic discrimination. Homework linguists with fancy ‘theories of
syntax’ will call us ‘unscientific’ and absurdly lump us together with the
language guardians. Both of our opponents parties are high in power and low in
responsibility, and bent on maintaining the status quo.
Yet another danger looms, as Teun
van Dijk (19__) has foreseen. If we do discover and describe the strategies for
effective communication, our results can be co-opted by the same manipulators,
indoctrinators, and exploiters who have long been using discourse to seize and
secure their power — politicians, demagogues, bureaucrats, profiteers, and
advertisers. We cannot prevent them, but we can help ordinary people use
discourse to resist their schemes and scams, provided we can make our results
widely available in plain language and in readable (or listenable) sources.
Either we take our own advice and communicate effectively — or we don’t deserve
to be believed any more than our opponents, the snobbish guardians and glib
homeworkers.
E.
The programme of discourse analysis: Looking back
So far, I have told you why
discourse analysis was resisted so long (Part B) and why I think its time has
come (Part C). I have also suggested why large-scale projects for helping
people to participate in discourse more effectively run counter to powerful
interests in society and science (Part D). Next, we can look back the resources
and outlooks for a programme of discourse analysis along the lines I have been
pursuing.
It’s essential to consider which disciplines might be contributors. One obvious candidate we
have already met in Part B: the linguists who developed methods for fieldwork on previously undescribed languages. Instead of ‘language by itself’, you analyse discourse
encountered in human interaction, a principle emphasised by American tagmemics (e.g. Pike 1967). Also, you
benefit from being ‘defamiliarised’ away from your own culture and meeting
‘strange’ ways of saying and doing things. Evelyn Pike told a story about one
culture whose language she was studying, where locations are expressed by the
points of the compass rather than the sides of the body. At the approach of a
poisonous snake, somebody yelled ‘jump to the east!’
Fieldwork also reveals differing
notions about what a language needs to express. For example, Mumiye, a
Niger-Congo language, has special forms to indicate when an action is
‘progressive’ (is continuing) or ‘durative’ (goes on for a longer time), such
as ‘-yi’ in [7-8] and ‘naa’ in [9] (data reported by Danjuma Gambo, in Longacre
et al. 1990: 151f). The piece-by-piece translations indicate what the morphemes
contribute, while the idiomatic translations suggest what an English speaker
might say:
[7] kpanti nwang kn sha-yi
chief sat food eating-durative
‘the chief sat eating and eating’
[8] sombo da-yi di ya bii ka jaa
gbaa
squirrel go-durative
go take focus child hoe
‘the squirrel was going to go take a small hoe’
[9] kura gbãa yuu naa
tortoise returning road progressive
‘the tortoise
was returning on the road’
English has no special form for
‘durative’; and its ‘progressive’ form with ‘‑ing’ need not indicate that
something went on for long time. Often, it indicates an action going on when
something else happened (e.g., ‘the tortoise was returning when he met the
squirrel’). For some actions, we can use repetition (e.g., ‘eating and eating’)
but for others it would sound odd (e.g., ‘returning and returning’ might
suggest several different returns rather than one long one). Yet we can rely on
our world-knowledge, e.g., that tortoises take a long time to travel — just as
the world-knowledge of Mumiye speakers grasps a small hoe as a ‘child hoe’
although nobody saw hoes bear children.
Studying discourse data from
unfamiliar languages makes you more sensitive to data from familiar ones.
People’s sensitivities to a language like English have long been dulled and
distorted by unreliable schoolroom ‘grammars’ (Part D). Moreover, English has
been dominated by written culture, encouraging the belief that the
order of language only emerges when written down in neat sentences — a belief
shared by many homework linguists (Part B). So we have not properly appreciated
the different and highly elaborate order of everyday spoken language, as
uncovered by ‘conversational analysis’ (see below).
In contrast, the more remote
languages of Africa, Asia, Oceania, and South America have been centred on oral cultures.
They have been spared from campaigns against ‘incorrect’ usage, and from the
bookish equation of orderly language with written language — some never devised
writing systems at all. High values were placed on speaking skills in communal
activities such as story-telling, which vitally supported cultural traditions
against the ravages and dislocations of slavery and colonialism. Whole systems
of spoken discourse signals were developed to organise the
story-line with its individual events and their participants, as discovered by
Longacre and his group (1990) in some 40 languages of East and West Africa. In
Gimira, an Omotic language of Ethiopia (data reported by Mary Breeze, in
Longacre et al. 1990: 27f), a ‘switch reference marker’ (shown as S/R) is used
at the high point of a story when the main characters alternate major actions.
We see the marker in the nasal sound ‘-n/m’ affixed to verb-forms in the story
[10] about a man named Gartn who was up in a tree collecting honey when he was
spotted by a hungry leopard (FUT = future; LOC = location; SUB = subject; STAT
= static verb-forms indicating the setting):6
[10] panc’i yi tok’an yisti ič gapmk’an wognsi hammsagyis maki
leopard-SUB
his foot-LOC being tree branch-LOC having-sat go-FUT-STAT-he saying
bak’u. Taci peški
hami šidni iču Gartn nasi esa myac’a kabnsi
waited
thinking not-to-go refused Gartn man-SUB honey bees having-wrapped
yi apm hazn
m’msi myac’am dusti woc’i koškan wot’i
his face-LOC throwing-S/R having-eaten
by-bees being-stung running valley
šičamm yiam gurt’i yisti dodn wornti hank’u.
leaving-descending-S/R
by-it trembling being ground-LOC descending went
‘The leopard being at the foot of the tree, he had been sitting on the
tree branch and waiting, saying and thinking it would go. It refused to go.
Gartn wrapped the honey and the bees and threw them in its face. The bees stung
it and it ran away down into the valley. After it had gone, trembling he
descended to the ground and left.’
Both the piece-by-piece translation and the more
idiomatic one show rapid switches of the agent (the one who does the action)
between ‘leopard’ and ‘Gartn’, who are named only once each. We can use
world-knowledge to infer from context who is meant, though I have cleared up
the English translation by switching between the pronouns ‘he’ versus ‘it’. The
switch reference marker only appears for the two decisive actions of Gartn’s
‘throwing’ (‘haz-n’) and the leopard’s finally ‘leaving and descending’ (‘iam-m’) from the valley, and not for less decisive
actions like ‘sitting’ and ‘thinking’. We see here how forms and patterns
relate to the total discourse and are not merely ‘morphemes’
or ‘sentence structures’ to segment and classify.
We can still find a few discourse
signals in traditional English stories. In an 1878 rendering of the familiar
folktale ‘Tom Tit Tot’7 (a demonic cousin of Rumpelstiltskin), ‘well’
systematically appears at important turning points in the story, often with a
shift of time:
[11] Well, once there were a woman and she baked five
pies [through a misunderstanding, the king proposes marriages to her daughter
if she will spin for him] Well, so they was married [king orders her to spin or
die] Well, she were that frightened [she bargains with a demon to do the
spinning; she must guess his name or be ‘his’] Well, the next day her husband
took her into the room and there were the flax [demon appears and spins] Well,
when her husband he come in: there was the five skeins [a cycle of spinning and
name-guessing begins] Well, every day the flax and the vittles was brought
[king unwittingly reveals name] Well, when the gal heerd this, she fared as if
she could have jumped outer her skin for joy [she tells the demon his name]
Well, when that heerd her, that shrieked awful and away that flew into the dark
We hardly notice such uses of ‘well’ because the word
has many other functions, e.g., as a conversational signal indicating you are
about to give a spontaneous opinion.8 Also,
these functions do not figure in traditional grammar-books; and ‘well’ is not
deemed ‘proper’ for written English.
Data like those from Mumiye and
Gimira indicate why fieldwork naturally leads into discourse analysis. But only
around the mid-1970s did such work adopt the term ‘discourse analysis’ (led by Grimes 1975; Longacre 1976;
Grimes [ed.] 1978). By then, linguistics was turning away from the project to
disconnect language from discourse and favouring alternative projects
The most important alternative was functional linguistics, which has had
several branches. A Czechoslovakian branch, sometimes called the ‘Prague School’ and founded by Vilém Mathesius
and his pupils, exploited their knowledge of Slavic languages like Czech and
Slovak, where the order of words in a sentence is more flexible than in English
and depends crucially on degrees of ‘knownness’ and ‘focus’. In comparison to
the more ordinary version [12], the English order [12a] is emphatically focuses
on the ‘problems’ by fronting the expression ahead of the sentence subject. In
contrast, the Czech version in [12b] (REFL PRON = reflexive pronoun) is not
emphatic, showing that ‘what is regarded as unusual in one language need not
appear so in another’ (Firbas 1992: 125f).
[12] a computer could take in its stride most of these
problems
[12a] most of these problems a
computer could take in its stride
[12b] většinou problémů by si počítač
hravě poradil.
with most problems it-would REFL PRON
computer with-great-ease it-cope
The functional sentence perspective, as this approach has been called
(bibliography in Firbas & Golková 1976), revealed previously unnoticed ways
for text and context to influence the arrangement of English sentences. In
Katherine Mansfield’s short story ‘At the Bay’ [13] (analysed by Firbas 1992:
26ff, 74ff), the setting is made the opening theme for Linda Burnell’s appearance on the scene. As in many
discourse beginnings, the communicative
dynamism — how ‘informative’ the
content is — starts out high for the opening sentence [13.1] presenting first
the ‘setting’ and then the main person in the story as the subject of the
sentence and her ‘dreaming’ action as the main verb. Putting the setting in a
long phrase ahead of the subject suggests that the story will highlight the
setting.
[13.1] In a steamer chair under a manuka tree that
grew in the middle of the front grass patch, Linda Burnell dreamed the morning
away. [13.2] She did nothing. [13.3] She looked up at the dark, close, dry
leaves
Though we cannot anticipate a ‘steamer chair’ or an
exotic ‘manuka tree’, it is normal for a chair to be placed ‘under a tree’ and
for a tree to ‘grow’ in a ‘grass patch’; in such a pastoral setting, ‘dreaming’
is a typical action too. In contrast, the dynamism of [13.2] is uniformly low:
the pronoun subject ‘she’ is already identified, and the activity of ‘doing
nothing’ is expected from a ‘dreaming’ person. [13.3] is a bit higher, with the
same pronoun subject ‘she’; and someone in a reclining chair under a tree
easily looks up and sees ‘leaves’, though we might not anticipate them being
‘dark’ or ‘close’.
Starting the story line for the
whole discourse thus way indicates that the main character is Linda Burnell and
that we’ll learn what she ‘dreams’ about. Also, prominently placing the setting
at the start suggests that the dream will be associated with the tree; a
lengthy reverie about ‘flowers’ indeed ensues, and about Linda ‘feeling like a
leaf’. As a British writer, Mansfield probably hoped the ‘manuka’ species of
tree would make the story more informative than a typical British tree. In the
Gimira story [10], in contrast, the ‘tree’ setting remains unspecified, the
‘thematic’ focus of attention being the plight of a man who, if ‘dreaming’
about anything, then about being safe at home away from leopards.
The British branch of functionalism,
led by linguists like J.R. Firth, Michael Halliday, and John Sinclair, also
rejected the disconnection of ‘language by itself’ and studied what speakers
actually say. Sinclair’s group pioneered ‘discourse analysis’ through fieldwork
on classroom discourse. Instead of ‘linguistic units’ and ‘rules’, the main terms
highlighted discourse moves like initiation, nomination,
andfollow-up by the teacher, and bid and response by a learner, e.g. in [14]:9
[14]
Initiation T Give me a sentence using
an animal’s name as food, please.
Response L1 We
shall have a beef for supper tonight.
Follow-up T Good.
That’s almost right, but ‘beef’ is uncountable so it’s ‘we shall have beef’,
not ‘we shall have a beef’.
Initiation Try again, someone else.
Bid L2 Sir
Nomination T Yes
Freddie
Response L2 We
shall have a plate of sheep for supper tonight.
Follow-up T No,
we don’t eat ‘sheep’, we eat ‘mutton’, or ‘lamb’.
Initiation Say it correctly.
Response L2 We
shall have a plate of mutton for supper tonight.
Follow-up T Good. We shall have mutton for supper. Don’t use ‘a plate’ when there’s more than one of you.
Such discourse plainly occurs only in classrooms,
pursuing the old campaign for ‘correct’ usage (Part D). The pupils are not to
tell what they like to eat and why, or how to cook it. The task is far more
artificial: saying ‘an animal’s name as food’, which easily traps pupils with
the tricky English usage of French loan-words for the foods (e.g. ‘mutton’,
‘beef’, ‘veal’) instead of the animals’ usual names. Communication is
subordinated to fine points of usage that the teacher illustrates without
giving useful explanations.
Several British functionalists were
guided by knowing Oriental languages like Chinese, rather than Slavic ones; and
they too brought new insights into English. They developed a view of language
being a network of options that are
assigned their functions when language is used in discourse. Instead of
‘correctness’, the key criterion is markedness,
e.g., to emphasise ‘these problems’ in [12a]. This ‘network’ view carries the
British brand name of systemic
functional linguistics’ and assumes that the organisation of a language is
expressly designed to support its use.
One classic demonstration, also a
model for stylistics, was given by
Halliday for William Golding’s The Inheritors. To evoke a ‘Neanderthal
tribe’s point of view’, Golding uses clause patterns whose ‘subjects are not
people’ but ‘parts of the body or inanimate objects’; the effect is ‘an
atmosphere of ineffectual activity’ and ‘helplessness’, and a ‘reluctance to
envisage the “whole man”’ ‘participating in a process’ (1973: 123, 125). When
the Neanderthal Lok watches a person from a more advanced tribe shooting
an arrow at him, the event is expressed as a series of natural processes
performed by a ‘stick’ and a ‘twig’:
[15] The bushes twitched again [...] The man turned
sideways in the bushes and looked at Lok along his shoulder. A stick rose
upright and there was a lump of bone in the middle [...] The stick began to
grow shorter at both ends. Then it shot out to full length again. The dead tree
by Lok’s ear acquired a voice. ‘Clop!’ His ear twitched and he turned to the
tree. By his face there had grown a twig. (Golding 1955: 106f)
These choices deliberately omit the
connection between ‘stick’ and ‘twig’ in a single weapon of bow and arrow, plus
the causes and effects involved, e.g., bending and releasing the bow, seen
head-on as a stick ‘growing shorter at both ends’ and then ‘shooting out to
full length’ and propelling the ‘lump of bone’ and its shaft to ‘the tree by
his face’. Lok’s notion of a ‘dead tree’ suddenly ‘growing a twig’ symbolises
the Neanderthals’ archaic and mystified world-view, dooming them to a
destruction they can neither understand nor resist, at the hands of a more
evolved people.
Another contributor to discourse
analysis would be the discipline of sociolinguistics,
which reconnects language with society by studying the language varieties
corresponding to differences in social, regional, and economic status. These
varieties differ not just in sound patterns, but also in discourse patterns,
depending especially on whether the participants come from a more ‘written’ or
more ‘oral’ culture. When shown a series of pictures and asked to tell the
story, middle-class children from written cultures specified nouns for things
like ‘boys’ and ‘window’ in [16], whereas the working-class children from oral
cultures, assuming anybody can see what’s meant, used pronouns like ‘they’ and
‘he’ and pointing expressions like ‘there’, e.g. [16a].10
[16] three boys are playing football
and one boy kicks the ball and it goes through the window
[16a] they’re playing football and he kicks it and it
goes through there
In Basil Bernstein’s (1964) unwisely-named ‘deficit hypothesis’, working-class people with a more ‘restricted code’ are
also more limited in their mental capacities than middle-class people with a
more ‘elaborated code’. This hypothesis triggered a storm of controversy
because both common sense and science wrongly assume that ‘intelligence’ is a
fixed and innate capacity, which would lead to the offensive hypothesis that
the working class is genetically inferior; Bernstein was claiming instead that
social conditions create disparities in mental capacities, including ones for
using language. The idea that intelligence is a social construct is repugnant
to many Western scientists and educators, because it demystifies our
fundamental contradiction between inclusive theory versus exclusive practice
(Part D), and the alibi of education that failure is caused by the biological
and psychological limitations of individuals. Moreover, Bernstein’s work
suggested that the ‘remedial programmes’ tacked onto ordinary schooling to
bring pupils’ language varieties into line with the ‘standard’ would be
ineffective — as we now know. Improvement demands transforming the social
conditions under which intelligence and discourse competence are constructed,
away from producing and legitimising inequalities over toward supporting
equality in practice as well as in theory (see Part F).
The most detailed picture of real
talk in society has been supplied by the analysis
of conversation in ethnomethodology.
Its home discipline was sociology,
which developed its own methods to study language rather than borrowing them
from linguistics. Harold Garfinkel (1974) reported coining the term ethnomethodology after such terms as
‘ethnoscience’ or ‘ethnomedicine’ for people’s commonsense knowledge of what
‘science’ or ‘medicine’ do. His method proposed studying real conversational
data and uncovering the participants’ commonsense ‘methodology’ for ordinary
social interactions.
This method focused for more on real speakers
than linguistics, and upon oral culture, this time in familiar languages
like English (Part B). Conversation is usually managed by its participants
quite tightly and fluently, with few conspicuous breaks or disturbances. The
significance of utterances is clearly a function of the ongoing interaction as
a whole rather than just the meanings or words or phrases, witness this bit of
taped conversation collected by Emmanuel Schegloff (1987: 208f) (small capitals
show emphasis; brackets show overlap; colons indicate lengthened sounds):
[17.1] B. Well, honey?
I’ll probl’y see yuh one a’ these
days
[17.2] A. Oh::: God yeah
[17.3] B: éUhh huh!
[17.4] A: ëWe—
[17.5]
A: But I c— I jis’ écouldn’ git down éthere
[17.6]
B: ëOh— ëOh I know I’m not askin
éyuh tuh écome down
[17.7] A: ëJesus ëI mean I just didn’t have five minutes yesterday
Two middle-aged sisters who haven’t visited each other
for some time are conversing on the telephone. Sister B probably intends to
signal a closing with the usual reference to a future seeing [17.1], as in
English ‘see ya’, French ‘au revoir’, German ‘Auf Wiedersehen’, etc. But sister
A understands a complaint about not having visited, and makes excuses for why
she ‘jis’ couldn’ git down there’ [17.5]. Sister B displays that she
appreciates A’s problems and signals that she was not pressing her claims to a
visit, overlapping with A’s excuse of ‘not having five minutes yesterday’
[17.6-7].
Ethnomethodologists like Schegloff
emphasise that the conversational analysis can document its own interpretations
with those made by the actual participants, in this case, A’s misunderstanding
and B’s venture to amend it. The theory thus stays far closer to the practice
than in ‘homework linguistics’ (Part B). Though ‘often unnoticed or
underappreciated in casual observation or even effortful recollection of how
talk goes’, the ‘detailed practices and features of the conduct of talk —
hesitations, anticipations, apparent disfluencies, or inconsequential choices’
— ‘are strikingly accessible to empirical inquiry’ (Schegloff 1992). This
lesson should be noted by linguists and philosophers.
The empirical lesson came more
easily to the discipline called text
linguistics (early survey in Dressler 1972). Predictably, some early ‘text
linguists’ tried constructing a disconnected theoretical unit of ‘language by
itself’ called (what else?) the ‘texteme’
and represented by complicated and impenetrable formulas. But a text
disconnected from context resembles a miscellaneous array like a pattern on
wallpaper.
Most text linguists gradually reconnected texts to discourse participants’ knowledge of world and society. We developed the concept of textuality: not just a set of ‘theoretical units’ or ‘rules’, but a human achievement in making connections wherever communicative events occur (cf. Beaugrande & Dressler 1981). The connections among linguistic forms like words or word-endings make up Cohesion, and those among the ‘meanings’