1. MOTIVES FOR EFFICIENCY
I have argued throughout that the utilization of texts in communication entails
constant management of blocks of knowledge, only some of which are relevant at a
given moment. The sheer volume of this knowledge usually precludes making even a
majority of it explicit in an individual statement. It follows that a language
should provide numerous options for compacting surface expression without
damaging the connectivity of underlying knowledge. In effect, these sets of
options point the participants in communication toward that portion of active
knowledge which is to be currently expanded or modified. The options are clearly
an important contribution to the EFFICIENCY of textuality: processing the
largest amounts with the smallest expenditure of resources. In terms of
CYBERNETICS, the use of formats for restatement responds to the CURRENT CONTROLS
on communication (cf. I.3.4.7), regulating the flow of knowledge up to the
The notion of “cohesion” has been used by some researchers for devices such
as pronominalization, substitution, and ellipsis (see
especially Halliday 1964; Hasan 1968; Halliday & Hasan 1976). Often, no
special consideration is given to the underlying connectivity of text-knowledge
and world-knowledge that makes these devices possible and useful (except in the
discussion of “lexical cohesion” in Halliday & Hasan 1976: ch. 6). Many
factors in linguistic outlooks were responsible for this omission: limitations
to sentences, exclusion of world-knowledge, lack of interest in real
communication, and a general discomfort regarding semantics. The enduring
primacy of syntax in linguistics is revealed by the very terms that were
proposed for the devices we are considering:. “hypersyntax” (Palek 1968),
“macrosyntax” (Gülich 1970), or “suprasyntax” (Dressler 1970a).
Evidently, the notion of ‘syntax” here is not that of syntax proper, but a
hybrid of ‘semantics of syntax” and “syntax of semantics” as envisioned
by the scheme set forth in I.2.8. Bonnie Webber (1980) remarks on the tendency
to treat the cohesive devices as if they served to refer to surface words rather
than to the conceptual-relational content underlying words. Jerry Morgan (1978a:
109f.) notices that tendency even in the writings of Halliday and Hasan (1976:
2), ‘who probably know better.” But Morgan may be too severe: surely we may
say metaphorically that words “refer” to other words, and mean that words
refer to the same referents as other words, provided we do not go on to claim
that we are dealing with nothing but words.
An exception to general trends is the very broad outlook of Roland Harweg
(1968a). His notion of ‘substitution’ subsumes not only the usual devices
such as pronouns and articles, but a diverse range of conceptual relations like
inclusions among classes, superclasses, or metaclasses, part/whole, causality,
and proximity. He is one of the few linguists to make free use of
world-knowledge in defining textuality. In the main, “substitution’ is any
connection between two components of a text or textual world that allows the
second to activate a configuration of knowledge shared with the first. Hence, a
good portion of his examples would be in line with the spreading activation
model of knowledge use (cf. III.3.24).
I will undertake to outline some of the most important devices of cohesion. My
criteria will be their contributions to the processing efficiency. These devices
RECURRENCE is the actual repetition of expressions. The repeated elements may
have the same, different, or overlapping reference, and the extent of conceptual
content they can be used to activate varies accordingly.
DEFINITENESS is the extent to which the text-world entity for an expression at a
given point is assumed to be identifiable and recoverable, as opposed to being
introduced just then.
CO-REFERENCE is the application of different surface expressions to the same
entity in a textual world.
1.4.4 ANAPHORA is the type of co-reference where a lexical expression has is a PRO-FORM (e.g. pronoun) after it in the surface text.
CATAPHORA is the type of co-reference where a lexical expression has is a
pro-form before it in the surface text.
EXOPHORA is the application of a pro-form to an entity not expressed in the text
at all, but identifiable in the situational context.
ELLIPSIS is the omission of surface expressions whose conceptual content is
nonetheless carried forward and expanded or modified by means of noticeably
JUNCTION subsumes the devices for connecting surface sequences together in such
a way that the relations between blocks of conceptual text- world knowledge are
signaled, such as: addition, alternativity, contrast, and causality. Subtypes of
junction are CONJUNCTION, DISJUNCTION, CONTRAJUNCTION, and SUBORDINATION (see
These devices offer a number of contributions to efficiency: (1) the compacting
of surface expression; (2) the omission of surface elements; (3) the carrying
forward of materials to be expanded, developed, modified, or repudiated; (4) the
signaling of knownness, uniqueness, or identity; and (5) a workable balance
between repetition and variation in surface structure as required by the
considerations of informativity.
The dependence of these devices on context emerges from this list of advantages.
At any particular moment during the production and comprehension of a text,
people need cues about what ALTERNATIVES among possible continuations are more
or less probable (cf. IV.1.1). At the same time, it is necessary to keep the
intended alternatives CURRENT without cluttering up the surface text by lengthy
restatement or repudiation.
The STABILITY PRINCIPLE was proposed in I.4.4 as a major factor of systemic
regulation of the kind I envision in the actualization of texts. Such a
principle assigns a high priority to strategies for co-ordinating surface
expressions that share common or contiguous conceptual content. The ECONOMY
PRINCIPLE stipulates that, wherever expedient or doubtful, preference should be
given to re-using already activated content, rather than activating new content.
It follows that cohesive devices like those enumerated in V.1.4 do not make the
text coherent; the prior assumption that the text is coherent makes these
devices useful (cf. Morgan 1978a: I 10).
The recurrence of surface expressions with the same conceptual content and
reference is especially common in spontaneous speaking, as opposed to formal
situations. The eyewitness report of a distraught county supervisor after a
flood in Arizona contained these statements (Gainesville Sun, Dec. 20,
There’s water through many homes.
I would say almost all of them have water in them. It’s just
completely under water.]1 [1. Throughout this chapter,
I use the convention of underlining the elements I wish to address in
rhetorical accumulated effect of this usage has something of the disastrous,
disordered copiousness of the water, an entity normally in short supply in
According to the principles of stability and economy, recurrence would entail
sameness of reference. But this could lead to conflicts in texts where there
seem to be no alternative expressions for different referents (Gainseville Sun,
Dec. 20, 1978):
Weapons and projectile toys have a built-in threat to eyes and cannot be
made child-proof. (74.2) Consumer safety groups have also warned about stuffed
animals with loose eyes and poorly sewn-on accessories. Small children
can pull them off and swallow them. (74.3) “We find eyes all over the
place,” one toy store clerk said.
assume that the clerk was finding toy eyes, not children’s ‘eyes all over
the place’ because in the latter case, the press treatment would be much more
explicit (lack of knowledge inference, III.3.21). Ambiguity is similarly
overcome for this passage of the Ohio Drivers Handbook:
A restricted license may be issued to any person otherwise qualified who
is subject to episodic impairment of consciousness upon a statement from a licensed
of my Ohio informants interpreted the passage such that the physician is
required to have a driver’s license (though some wondered how ‘episodic impairment
of consciousness’ differs from the usual state of Ohio drivers).
Deliberate violations of the stability and economy principles might increase
informativity and interest. For example, a poem allegedly written by the
18-year-old conspirator Chidiock Tichborne just before his execution in 1586
contains the line (Simpson [ed.] 1967: 85L):
(76) My glass is full, and now my glass is run.
discrepancy (1.6.9) arises when the second ‘glass’ cannot be taken as a
drinking vessel and must be processed as ‘hourglass’ instead, reverting to
knowledge about the writer’s personal situation of impending execution.
Psychologically, recurrences should distribute attention away from their
components, except in cases like (76). If the frequency principle of learning
(IV.2.2) applies, the recurrent elements should be impressed on memory.
Processing should be easy, as the point of connection in the ongoing text-world
model should be obvious (cf. Kintsch 1974: 86). Whatever factors may apply,
there must be a difference between the TRIVIAL recurrences required by the
limited repertories of language options and MOTIVATED recurrences where
repetition has some deeper justification (cf. Werth 1976; Beaugrande 1978b,
Consider for instance the Biblical proverb:2 [2. These
examples are taken from a textbook entitled Rhetoric: From Athens to Auburn, ed.
Richard Graves (Auburn: Auburn University Press, 1976), pp. 33, 32, and 19
As in water face reflects face,
So the heart of man reflects man.
two lines are very similar in surface structure, and each contains an element
repeated on either side of the element ‘reflects’. This organization of
expression enacts the content of the textual world: images ‘reflected’ in a
mirror. Less striking is the use of recurrence for signaling repetitious events,
as in Steinbeck’s passage:
They work at it and work
use is similar to that of the county supervisor preoccupied with an
overabundance of water in (73). Speaker outlook can be signaled with recurrences
such as this one from Jeannie Morris:
There are no distractions — and I mean no
time, the surface format iconically nacts the insistence of the speaker on an
attitude as unchanging as the expressions themselves. Possible objections are
accordingly discouraged (cf. Beaugrande & Dressier 1980).
Recurrence can be employed with a shift in the syntactic function of an
expression (Dressier 1979). The recurring element is adapted to its
environments, yet the identity of reference is still obvious. In the American
Declaration of Independence, we find these stretches of text:
to assume among the powers of the earth, the separate and equal station
[...] (80.2) they should declare the causes which impel them to the separation
shift from adjective for an attribute to noun for an action neatly signals the
overall coherence while avoiding the monotony of an exact repetition. Dressier
(1979) notes that this recurrence type offers the text producer the potential to
create new language items, since one occurrence can provide for the
comprehension of the other. Such is the case when Erich Fried entitles a story
‘Turtle-Turning’ and includes this passage (Fried 1975):
Everywhere he finds a helpless turtle fallen on its back, he turns it over
title would be highly non-determinate without this recurrence via word- class
Recurrences of lengthy expressions or whole passages can be disadvantageous,
because they depress informativity unless strong motivation is present. It is
strategically sound to vary expression with paraphrases or synonyms. Yet as in
the case of ‘water’ (73), there may be only one readily available name for
the desired concept. In scientific reports, the use of specially defined terms
must be consistent, despite the repetitiousness entailed. Hearers and readers
presumably adapt their expectations in response to these factors.
The issue of definiteness takes on various dimensions, depending upon whether
one’s outlook is logical or psychological. If meaning is identified directly
with “truth value” (III.I.2), definiteness becomes a property of objects
asserted in a logical world. If meaning is viewed as mental processes, then
definite entities are those that are “uniquely identifiable” to participants
in communication (Clark & Clark 1977: 249f.). Whether the entities are
logical or real, both criteria are too strong. Definiteness applies to many
entities that need not be identified at all with specific objects. Ortony and
Anderson (1977) distinguish the identifiable reference as “extensional
representation” and the reference to entities needed only for conceptual
content as “intensional representation” (cf. I.2.8.2).
The utilization of ARTICLES,in English at least, is revealing, as the terms
“definite and indefinite article” suggest. Usually, the definite article is
claimed to precede the expression of entities already mentioned, and the
indefinite that of newly introduced ones (cf. Firbas 1966). But the following
fragment of a Thurber story (in Thurber 1948: 34), suggests
that the matter is more intricate: 3 [3. James Thurber,
‘The Princess and the Tin Box’, in The
Beast in Me —And Other Animals, published by Harcourt Brace Joyanovich
1948. Reprinted by permission.]
Once upon a time, there lived a king whose
daughter was the prettiest princess in the
whole world. (82.2) On the day the princess was eighteen, the king sent a royal
ambassador to the courts of the five
neighboring kingdoms to announce that he would give his daughter’s hand in
marriage to the prince whose gift she liked
most. (82.3) The first prince to arrive at
the palace […]
classical distinction of new = indefinite vs. previously mentioned = definite
applies here only to ‘a king’ (82. 1) and ‘the
king’(82.2). The beginnings of texts are, of course, likely places for
indefinite articles (Weinrich 1976: 172). Yet the first occurrence of
‘princess’ has the definite article, being a superlative. The usage in ‘the
five neighboring kingdoms’ rests on the postulate of continuity in a textual
world (I.6.4): a geographical region can be expected to have neighbors. ‘The prince’ in (82.2) is a projected entity not yet having any
referent: any prince who meets that description (an “intensional
representation’ in the sense of Ortony & R. Anderson 1977); and ‘the
first prince’ in (82.3) is a member of the candidate class in which there can
be only one for each number in a series. Such varied uses of articles are
essential for the connectivity of the story. De Villiers (1974) found that if
the definite articles in a story text are replaced by indefinite, readers
don’t take the component sentences as parts of a story at all. Loftus and
Zanni (1975) found that eyewitness reports could be influenced by inserting
definite articles in front of strategic items: the articles impelled the
eyewitnesses to accept as factual some items they hadn’t really seen to begin
with. Here, the text surface actually created background knowledge while
pretending to keep it active.
At least the following entities would seem to be eligible for the status of
MENTIONED entities as established in a textual world (e.g. ‘the king);
SPECIFIC entities established by constraining description or definition (e.g.
‘the day the princess was eighteen);
EPISODIC entities stored in the shared knowledge of language users personally
acquainted with each other (e.g. ‘the movie’ in Clark & Marshall 1978:
57; cf. also Goldman 1975: 347);
UNIQUE entities which every sensorially endowed member of a communicative group
is assumed to know about (e.g. ‘the sun’, ‘the earth);
INSTITUTIONALIZED entities that social organization is presumed to require
(‘the president’, ‘the fire department’, ‘the police);
DEFAULT entities created on demand for the continuity of a textual world (e.g.
‘the five neighboring kingdoms’ in [82.2]);
PROTOTYPICAL entities that function as the representative of a class (e.g.
‘the man on the street’, ‘the ugly American) (cf. III.3.27);
SUPERLATIVE entities that occupy the extreme position on some scale of variables
(e.g. ‘the prettiest princess in the whole world’);
RELATIONAL ENTITIES accessible via TYPICAL and DETERMINATE links from already
The criterion of being “uniquely identifiable” fails to cover these various
uses. Often, definite entities have no more identity than is required for the
particular context wherein they appear (Rieger 1975: 204). We can talk about ‘the
police” ‘the ugly American’, or ‘the prettiest princess in
the whole world’ without any commitment to an object, or even to a complete
entity: we are addressing a conceptual configuration whose content may be no
more than the properties we need at the moment. The ‘police’ are people only
in their official capacity, not as private individuals. An ‘ugly American’
need by no means possess a repellent outward appearance. We can easily envision
the man on the street’ not being on any street at all, but sleeping in a
dumpster. And ‘the prettiest princess’ may be decidable in a children’s
tale, but hardly in a reality where beauty is a matter of opinion.
Definiteness might be explicated as the status of entities in a textual world
whose FUNCTION in their respective context is non-controversial. To fix the
status, e.g. with proper names or definite descriptions, is to instruct the
hearer/ reader that the appropriate conceptual content should be easily
suppliable on the basis of already activated knowledge spaces. INDEFINITE
entities, on the other hand, require the activation of further knowledge spaces.
Hence, de Villier’s (1974) test subjects thought that the version with
indefinite articles could not constitute a unified story world. They took the
indefinite articles as instructions to activate new spaces rather than use
already active ones.
No one would have trouble with entities like ‘the sun’ and ‘the moon’.
These entities are not in fact unique, as the exploration of astronomers
attests. But in lack of any wider setting such as a science fiction story,
preference is at once given to the usual referents. Since a textual world is not
committed to exact correspondence with the accepted real world, conventionally
unique entities can be recontextualized into non-unique. In this view,
uniqueness begins to converge with default. Consider this excerpt from a news
article on prostitution (Gainesville Sun, Oct. 8, 1978):
Now that the adult bookstores, formerly the vice squad’s primary
target, have been closed down, the agents are able to devote more time to
definiteness of ‘bookstores’, ‘vice squad’, and ‘agents’ rests on
their typical or institutionalized status in American social organization. They
can be assumed as defaults without any clear notion of where or who they might
be in this particular town. If an unfortunate occasion arises, their uniqueness
can be established. Yet communication would operate very slowly if we had to
establish uniqueness merely in order to talk about these entities.
The spreading activation model of knowledge use, as frequently cited in this
book, is relevant to definiteness. Although it is not decided whether spreading
is consciously controlled or not (cf. M. Posner & Snyder 1975), definiteness
can be one means for channelling it. The appearance of a definite entity not
previously mentioned would then have the effect of singling out a point in
knowledge space to which activation is assumed to have spread. DETERMINATE and
TYPICAL links clearly provide the soundest basis for that assumption. Consider
this news item (Florida Independent Alligator, Oct. 9, 1979):
A seat belt saved a UF student when he fell asleep at the wheel of his
1977 Subaru and turned off into the path of a train.
definiteness of ‘wheel’ arises as a determinate ‘part-of’ a
‘Subaru’, and that of ‘path’ as a typical ‘’location-of-motion-of’
Perhaps the following definition merits consideration: definiteness con
spread to any text-world entity standing in a determinate or typical linkage
(cf. III.3.15) to an entity whose definiteness is already establishedin the
textual world. To see how this principle would work, imagine that (85.1)
were a text beginning; any of the continuations in (85.2) should then’ be
acceptable via the link types (from III.4.7) cited in square brackets:
Never before had we seen such a house.
The plot of land was quite deserted. [location-of]
The rectangular outline looked oddly lopsided. [form-of]
The walls were leaning inward. [part-ofi
The plaster was peeling off. [substance-of]
The furniture was awfully rickety. [containment-of]
The edifice seemed doomed to collapse. [motion-of]
In all of these continuations, the ‘house’is taken as a topic node and thus as a control center to which new material is preferentially connected (cf. III.4.27). This configuration is shown graphically in Figure 22, with all continuations included.
However, if the linkage were accidental, definiteness would not be so likely to spread, e.g. (85.2g) being an odd continuation:
The canary seemed depressed. [containment-of]
oddness of some of my school children’s ‘parts of a house’ (III.3.26) is
due to accidentalness. Definiteness also seems reluctant to spread down longer
pathways, so that (85.2h) is an odd continuation if the house’s inhabitant is
The face was terribly ugly. [part-of-agent-of-possession-of?]
a single accidental instance is taken from an otherwise accessible class, we are
adding an ‘instance-of’ link. Again, definiteness is not clear in such
The nail was rusty. [instance-of-part-of]
The brick hurt my elbow. [instance-of-part-of]
can improve upon these continuations by providing some intermediary entities not
included in numerous classes:3 [3. There may be a constraint
that definiteness cannot spread to an accidental instance of an unordered class
unless the class itself is first evoked.]
The nail holding the name-plate on the front door was rusty.
The brick protruding furthest from the fireplace hurt my elbow.
Linkages to an event can function like these linkages to an object. If a text
begins with (86.1), then the continuations in (86.2a-c) connect up to the whole
The sun was just emerging from behind a cloud.
The day was not yet over. [time-of]
The sudden brightness blinded our eyes.
The improvement in our spirits was
could also link back to ‘sun’ as object:
The golden color was impressive.
The orb blazed down on us. [form-of]
10 Inclusion in classes, superclasses, and metaclasses (III.3.19f.) renders
these matters quite intricate. One entity which usually has no unique or
identifiable referent is the PROTOTYPE (cf. P. Hayes 1977; Fahlman 1977; Rosch
1977; Brachman 1978a; Webber 1978). The prototypical member has a determinate
“instance-of” link to its class, to the extent that the class has a
discoverable identity. In a conversation like the following from The
Importance of Being Earnest (Wilde 1940 [original 1899]: 420):
ALGERNON: In married life, three is company and two is none.
JACK: That is the theory that the corrupt French Drama has been
propounding for the last fifty years.
ALGERNON: Yes, and that the happy English home has proved in half
matters little if speakers have any particular French drama or English
home in mind. The context demands no more than a FUZZY concept which supplies
the needed content.
A class combined with scales of values can yield a SUPERLATIVE as the class
member situated at the extreme end of a scale. Because value scales are in the
main imprecise, superlatives share the fuzziness of prototypes. The usage of
‘the prince whose gift she liked most’ in (82.2) is straightforward enough,
since the princess’s decision will automatically define the referent. But when
Leroy Brown was asserted in the American pop song back around 1977 to have been:
the baddest [i.e. toughest] man in the whole damn town
one would seriously suppose a precise value determination. Leroy is simply being
characterized as an extreme representative of the already extreme class of
‘bad men in south Chicago’. Where competition is so keen, empirical
verification would be absurd (and mighty dangerous). In this one textual world,
Leroy was the superlative, at least until his sudden demise, ‘like a picture
puzzle with some pieces missing’.
As I mentioned in III.1.3, logicians have traditionally been concerned with at
least certain aspects of classes and class inclusions, namely those that fall
under the heading of QUANTIFICATION (cf. Stegmüller 1969: 15f.). As I observed
at a philosopher’s symposium at the University of Bielefeld in June 1979,
logicians generally suppose that definiteness and the use of definite articles
depends on the types of quantification described in III.1.3. My own impression
was that quantification has been introduced not so much for matters of this
kind, but for the special requirements of logic in constructing valid proofs. In
the following famous example, we have a universally characterized class of
‘men’ in (89. 1) and the unique member ‘Socrates’ in (89.2):
All men are mortal.
Socrates is a man.
Therefore, Socrates is mortal.
the proof is clearly valid, I do not see why it should depend particularly on
definiteness of existence or uniqueness. We might replace ‘men’ and
‘Socrates’ with ‘unicorns’ or ‘the King Louis XIII of France’s pet
pageboy’ without making the line of reasoning faulty. The questions of
existence and definiteness, I submit, hinge upon context of occurrence. The
demands of formal logics for precise quantification far exceed the conditions of
many contexts of everyday communication. Whereas logicians have for years
debated the status of the donkey in the (according to the redoubtable Texan
Robert F. Simmons, who vowed prospectors love their donkeys, utterly
Every man who owns a donkey beats it.
language user need merely create a default donkey with whatever further traits
(besides being beaten) are required for the textual world. The demands exerted
by logical quantification are far too strict for natural language communication.
For the text psychologist, the interesting questions are rather how people
recognize objects, and under what conditions they are more or less disposed to
believe statements. People concern themselves with existence and abstract truth
only in special contexts.
INDEFINITENESS, I suggested in V.3.5, is the property of entities for which no
knowledge space is currently active. The beginning of our rocket text:
(35.1.1) A great black and yellow V-2 rocket 46 feet long stood in a
New Mexico desert.
instructs the reader to create active nodes for ‘rocket’ and ‘desert’
and to hang the supplied attributes, locations, etc. onto them. However, the
text could also have begun with ‘The great black and yellow V-2 rocket
[…]’ and still have been perfectly coherent. The effect would be the
writer’s commitment to make further use of the node beyond that one statement.
For example, Through the Looking Glass (Carroll 1960: 175) starts right
out with the statement:
(91) One thing was certain, that the white kitten had had nothing to do
reader justly expects to hear at least enough about the white kitten to make the
statement believable (and will have a long time to wait!). Such usage is very
widespread in texts whose format requires engaging the reader’s interest,
because a knowledge deficit is created. In one collection of short essays (G.
Levin [ed.] 1977), definite articles for not yet established entities at the
beginning of texts is clearly the rule, not the exception (cf. Harweg 1968b):
Each year I watched the field across from the Store turn
caterpillar green. (Maya Angelo, p. 13)
The judging formally begins with the Saturday luncheon at the
Heart of Wilson Motel. (Frank Deford, p. 115)
The train, its metal wheels squealing as they spin along the
silvery tracks, rolls slower now. (Robert Ramirez, p. 127)
Before you even get the cone, you have to do a lot of planning.
(L. Rust Hills, p. 182)
The introduction of entities as definite right at the beginning of the text does
not disprove or undermine the status of the definite/indefinite distinction. We
do see that, given a regularity of natural language communication, people freely
do just the opposite for special effect. It is pointless to argue whether the
essays just cited are “well-formed.” In a linguistics of actual texts, a
rule such as ‘Use the indefinite article for the first mention, and the definite for later mention” can be no more than I
DEFAULT or PREFERENCE (d. 18.104.22.168). Communication takes place gainst a
baekdrop of defaults and preferences, but text users will go their own ways when
it is expedient to do so (cf. 22.214.171.124).
The treatment of PROTOTYPES illustrates another facet of the definite/ indefinite distinction. Either of the following utterances
could be produced
in a situation of receiving unqualified advice:
(96a) A layman shouldn’t give advice to an expert.
(96b) The layman shouldn’t give
advice to the expert.
one utters (96a), bearers are instructed to look immediately into the
situational context for referents, so that the indefiniteness is removed. One
can use (97b) with a less obvious directness, because the tendency is to
envision prototypes for the class of laymen and experts. Once more, the
question of “well-formedness” would miss the main point.
3.16 Indefiniteness could also be applied unconventionally. If we had the utterance (traditional saying):
man who never loses his head doesn’t have a
had to lose.5 [5.
A similar saying in German goes back to Gotthold Ephraim Lessing.]
the usage ‘a head’ presents as indefinite something that is a determinate “part-of” ‘man’. The effect is to weaken the determinateness of that link by suggesting that there could be men without heads after all.
The definiteness of text-world entities, as we can see, is complex. The usual
criteria (cf. V.3.1f.) for exploring the issues are too narrow. If people could
assign definiteness only to uniquely identifiable objects in the world, or
objects whose existence (either singly or as a class) has been explicitly
asserted, communication as we now find it would scarcely be feasible. We might
make better headway by treating definiteness as something that arises out of the
connectivity of stored knowledge being used in a real situation, where
efficiency is referring
CO-REFERENCE VIA PRO-FORMS
If REFERENCE is the relationship between expressions and the objects, events,
and situations in a world those expressions designate (III.1.3). The
use of alternative expressions in a text for the same text-world entity could be
termed CO-REFERENCE. Although there are many types of co-reference, (e.g.
synonyms, paraphrase), I shall explore only co-reference via PRO-FORMS.
Pro-forms are derivative in their actuãIized content from their co-referring
expressions. As such, pro-forms differ from their co-referring expressions in
systematic ways (cf. Padučeva 1970; Dressler 1972a: 26f.):
Pro-forms have a wider range of potential application.
Pro-forms are comparatively empty of inherent content.
4.1.3 Pro-forms are usually shorter-a fact which Dressier (1972a: 26f.) sees in agreement with Zipf’s (1 935) “law”. the more frequently a word is used, the shorter it tends to be or become.
Pro-forms obey constraints upon their occurrences, such that comprehension is
not rendered unduly problematic.
Pro-forms need a distinctive surface appearance. In English, PRONOUNS are the
only word class in the nominal system that maintains different forms for gender
(masculine, feminine, neuter) and case (subject vs. object)-nouns distinguish at
most possessive, singular, and plural. DEICTICS (pointing words) generally begin
with ‘th-’and are the only word class in which initial ‘th- ’is voiced
in pronunciation (except the article ‘the’ and the pro-forms
4.2 PRONOUNS are the best known type of pro-forms. In general, they have as
co-referent expressions nouns appearing in the text (cf. Postal 1969). Yet some
uses of pronouns do not follow this application, for instance, in the popular
may be required to recover some referents, as in the well-known slogan of the
Bell Telephone Company:
Calling long distance is the next best thing to being there.
where ‘there’ must be co-referent with an inferrable location. Pronouns may apply to entities whose previous introduction did not occur via nouns, as in a recent statement by a U.S. newscaster:
The Congressional privilege of giving consent to treaties is one they
seem unwilling to sacrifice.
where the co-referent must be derived from the adjective ‘Congressional’.
4.3 If other expressions sharing referents are used together with pronouns, the natural order would seem to be from most specific to least. Lakoff (1968a) foresees an order of. (1) proper name; (2) specific description; (3) a general class name; and (4) pronoun. An invented example might be:
Napoleon entered the room. (104.2) The
famous general made some announcement. (104.3) The
man was very excited. (104.4) He spoke
at top speed.
this order is far from obligatory. A text producer might use just the reverse in
order to create a knowledge deficit (like the deficit evoked by introducing new
entities as definite, cf. V. 3.13). We find that tactic used for suspense in
this passage by the marvellous Russian story-teller Nikolai Leskov (1961: 55).
The door to the cell of the Archbishop mysteriously opens:
Who should walk in but a venerable old
man in whom his Grace immediately recognized one
of the saints of the church, no other than the Right
order of ‘who-man-saint-Sergius’ is a complete reversal of that foreseen by
Lakoff,6 [This example is not meant as a refutation of Lakoff,
who was dealing with sequels where each element was in a invented separate
sentence. Rather, it illustrates how flexible language regularities are in
general (cf. note 10 to Chapter 1).] and the gradual emergence of the mysterious
figure’s identity is perfectly matched to the gradual increase of specificity
in the co-referring expressions. The usage is both effective and appropriate
The replacement of surface expressions also brings up the problem of class
inclusions such as we saw in V.3.10ff. The pro-forms can refer to the same set
of entities as their co-referent expressions (examples here from Webber 1978:
45): (106a) Several linguists attended the masquerade. They were dressed up as
cyclic transformations. But distinctions can be found between a COLLECTIVE
inclusion, as in (106b), and a DISTRIBUTIVE inclusion, as in (106c):
Several linguists attended the masquerade. They all came as parse
Several linguists attended the Yorktown Strutters’ Ball. They each
came dressed as a different transderivational constraint. [Insider joke: The
original hit tune was the 'Darktown Strutters’ Ball', but IBM Labs, known at
the time for airy, boastful claims about its language programs, are in Yorktown
distinction has important effects upon the text-world model, as these examples
(from Webber 1978: 44) reveal:
The three men who tried to lift a piano dropped it.
The three men who tried to lift a piano dropped them.
pronoun ‘it’ creates a textual world with the men lifting one piano
together, while ‘them’ leaves us with the three lifting one piano each.
The efficiency of pro-forms is especially evident when they apply to large
stretches of discourse that activate sizeable knowledge spaces:
“Give your evidence,” said the King, “and don’t be nervous, or I’ll
have you executed on the spot.” This did not encourage the witness at
all. (Carroll 1960: 148) 6
(108), ‘this’ stands for the entire content of what the King of Hearts has
said, and places the entirety in a ‘reason-of’ relation to the state of the
‘witness’. A pro-form can even stand for a block of content whose limits are
left open by remaining unexpressed:
(109) “My father and mother were honest, though poor—”
‘Skip all that!” cried the Beliman in haste.
I skip forty years,” said the Baker in tears
(The Hunting of the Snark, [Carroll 1973: 63])
depiction of forty years would have constituted a vast expanse of content.
Pro-forms also serve in the REPUDIATION of some portion of previously expressed
content (cf. IV.3.12), as in (Belloc 1940: 177f.):
(110) I shoot the hippopotamus
With bullets made of platinum
Because if I use leaden ones
His hide is sure to flatten ‘em.
class of ‘bullets’ is divided into the subclass of ‘platinum’ versus
‘leaden’, and the expectation that the latter subclass should be used is
repudiated. In the following remark of the White King, ‘one’ designates a
currently present member of the class of ‘Pencils” while a still indefinite
‘thinner’ member is envisioned:
(111) My dear, I must get a thinner pencil. I can’t manage this one a bit. (Carroll 1960: 190)
referents in a textual world can be similar in every respect but one, and the
pro-form need only attach that respect to keep them distinct, as in the case of
Tweedledee and Tweedledum (Carroll 1960: 229, e.a.):
She was just going around to see if the word “TWEEDLE” was written on the
back of each collar, when she as startled by a voice coming from the one
The pro-form ‘one’ is useful also if the entity in question is to be kept indefinite (Carroll 1960: 100):
(113) The March Hare said: “I vote the young lady tells us a story.” “I’m afraid I don’t know one,” said Alice.
For reusing event-based knowledge, PRO-VERBS can be employed, such as ‘do’
(Carroll 1960: 47):
“I don’t know the meaning of half those long words, and what’s more, I
don’t believe you do, either.”
pro-form ‘so’ could can be added on to ‘do’in order to cover material
attached to the original verb:
To this day I am ashamed that I did not spring up and pinion him then and there.
Had I possessed one ounce of physical courage, I should have done
so. (Beerbohm 1958: 57).
‘do so’ carries forward the content of an entire phrase of two actions with
direction and time. Alternately, ‘do it’ can perform such a function:
“Smoothe her hair — lend her your nightcap — and sing her a soothing
lullaby.” “I haven’t got a nightcap with me,” said Alice, as she tried
to obey the first direction; “and I don’t know any soothing lullabies.”
“I must do it myself then,” said the Red
Queen. (Carroll 1960: 326)
pro-forms pick up the content of two out of three mentioned actions.
By selecting pro-forms of various word-classes, speakers can allow hearers to.
re-utilize their mapping strategies between grammatical and conceptual
dependencies. For ‘spring up/pinion [...] then and there’, the ‘do so’
repeats the “head-to-modifier” dependency (II.2.15.7). For ‘lend [...]
nightcap/sing [...] lullaby’, the ‘do it’ repeats the “verb-to-direct
object” dependency (II.2.15.2). One might consider setting up a scheme with
designations like “pro-modifier,’ “pro-direct object,’ etc. However,
pro-forms can have diverse applications in the same occurrence. In:
Yond Cassius has a lean and hungry look. He thinks too much. Such men are
dangerous. (Julius Caesar, 11. ii, 194-95)
pro-modifier ‘such’ carries forward both the modifiers ‘lean and hungry’
and the “cognition”-verb ‘think too much’. In the following sample, the
text receiver is left to infer, as co-referent for ‘such’, an attribute not
expressed in a surface modifier at all: (Carroll 1960: 279):
“I see nobody on the road,: said Alice. “I only wish I had such
eyes,” the King remarked in a fretful tone. “To be able to see Nobody! And
at that distance too!”
King’s remarks suggest that the implied modifier might be ‘good’ or
Pro-forms are far more likely to occur ANAPHORICALLY, i.e. after their
co-referring expressions, than CATAPHORICALLY, i.e. before them. The anaphoric
use would provide a control center to which the material attached to the
pro-form can be readily added on (cf. III.4.27). It is harder to envision how
cataphoric use can be managed. The pro-form might be placed on a HOLD STACK
until its co-referring expression occurs (cf. II.2. 10); or it might be left as
an unlabeled state in FUZZY PARSING until labelling becomes feasible (cf.
IV.1.10). In either case, it would not be advisable to create substantial
distance between the pro-form and its co-referring expression. Cataphora is most
common inside the single sentence, e.g. in this student paper from the
University of Florida:
I don’t know if he’s serious, but my roommate wants to walk a
tightrope over Niagra Falls.
can also announce a large block of content that spans a series of utterances:
(120) That you have wronged me doth
appear in this:
You have condemned and noted Lucius Pella
For taking bribes here of the Sardians,
Wherein my letters, praying on his side
Because I knew the man, were slighted off.
(Julius Caesar, IV, iii, 1-5)
Like definiteness, cataphora can be used to create a knowledge deficit that will be later filled (cf. V.3.13). Warwick Deeping (1930: 720) deliberately begins a story with a cataphoric pro-noun for which the co-referring expression is postponed to the end of a long sentence:
Her father was a snuffy little man, who, after living for fifteen years
as. a widower in the white house at the end of Prospect Terrace, had developed
mannerisms and peculiarities that were neither criticized nor questioned by his
The constraints upon cataphora are part of the conditions of language processing
at large. It is hard to maintain connectivity between elements which are either
placed far apart or whose identity is uncertain because of alternative
candidates. Our ‘rocket’ sample (35) in III.4.20, however, shows that these
difficulties can be offset, for example, by attaching co-references across wide
spaces to a TOPIC node; or by considering what concepts are preferentially
compatible in the sense of Wilks (1978) (e.g. ‘rocket-plunge’).
Ambiguous pro-forms have received considerable attention in linguistics, such as
the classic example:
(122) I love my wife. So does Harry.
the social implications of the possible textual worlds might be intriguing for
bored but lascivious linguistics professors in the American Midwest back before
cable TV and 901-numbers. If English differentiated reflexivity in the way that
Russian does, this kind of ambiguity would be precluded (Dressler 1972a: 24).
All the same, such ambiguities are seldom really unresolvable. Wallace Chafe
(1976:47)suggests that in:
(123) Ted saw Harry yesterday. He told him about the meeting.
co-referents might easily be sorted by keeping the subject and direct object
slots constant, hence ‘Ted = he’ and ‘Harry = him’. That account would
agree with the principle of STABILITY (V.1.7), though it may be agents rather
than subjects that are decisive here. But world-knowledge is surely an
overriding factor. For a sample such as:
(124) Billy told Johnny’s mother that he hit him.
might not rely on stability of subject or agent (making ‘Billy’ do the
‘hitting) so much as on the knowledge that children tell parents about
others’ misdeeds much more often than about their own (hence ‘Billy’ got
hit, and no doubt deserved it). Still more constraining is this passage about
the death of a solicitor (Ipswich Journal, Jan. 12, 1878):
He was going to the Court, when he staggered as if in a fit, and fell against
the wall close to the watchman’s room in the central hall. The watchman and a
policeman, running to his assistance, took him into a room. Some brandy was
administered to no effect, and Mr. Bond, the surgeon of Parliament Street,
arriving, he pronounced him dead.
language user with “autonomous syntax” would spend a long time computing
alternatives about who pronounced whom dead (solicitor? watchman? policeman?
surgeon?); most real readers notice only one possibility. Anything else would
call for some odd elaboration, e.g.:
So much brandy was administered that when Mr. Bond, the surgeon of Parliament
Street, arrived, the solicitor drunkenly pronounced him dead. And
then died as if by mistake.
world-knowledge will find referents even where a wrong pro-form is used
(headline from the Midnight Globe, July 4, 1978):
Sophia Loren reveals love scandals that haunt my marriage.7
[The effect is like “erlebte Rede’— Ms. Loren seems to be speaking
herself, though the predicate has a third-person verb.]
With so many supporting factors to use, the recovery of a wrong co-referent is
unlikely and might signal a refusal or inability to communicate. Gracie’s
mistake in applying ‘it’ to her ‘car’ rather than the ‘pile of
trash’ back in (2) (II.1.8) would not be made by any reasonable person. To
create a non-text without ‘an atom of meaning in it’, Lewis Carroll (1960:
158) needs merely to provide no co-referents for pro-forms:
(127) They told me you had been to her,
And mentioned me to him […]
this poem, no cues regarding identity are forthcoming. A language model using
world-knowledge is justly stumped here, though a reader with a strong enough
attitude on acceptability can wring some meaning even from these verses, as the
King of Hearts demonstrates.
The TRADE-OFF between compactness and rapid access already mentioned in regard
to the storage of knowledge in memory (III.3.18) is also applicable to the use
of co-reference via pro-forms. The pro-forms allow an enormous savings in the
creation and utilization of surface structure. But the gain would be lost again
if there were problematic ambiguities in the identification of the co-referring
expressions. I have argued above that people use all kinds of cues to preclude
ambiguities which the pro-forms themselves, due to their inherent indeterminacy,
might allow. The fact that actual misunderstandings are so seldom in human
communication is an impressive indicator of the co-operative nature of
textuality (especially intentionality and acceptability), and of the regulatory
controls upon systemic actualization (cf. 1.4.5. 1).
In exophoric reference, the pro-forms apply directly to entities recoverable in
the situation, rather than via co-referent expressions in the same text or
discourse. Such a device strongly argues against a division between language and
its settings (cf. III.3.18). Exophora is particularly efficient in that it
bypasss the intermediary step of concept-naming. Even more than anaphora and
cataphora, exophora depends crucially on context. If a concept’s meaning is
its place in a textual world, the meaning of an exophoric expression’s
referent is its place in a text-world with focus on the situational world of
communicating. For example, (128) was used as the very first utterance of a
conversation by someone opening a door and finding a familiar person outside:
(128) She’s not here.
speaker was aware of the visitor’s usual intention to pay a call on a certain
person, and the visitor in turn knew about that awareness.
In some situations, the pro-form can be applied to entities that may not be
given conceptual classification:
What on earth is that?
(130) I can’t believe this!
uses can also signal that the speaker’s expectations — provided they are
presupposed to be known to the hearer-are being disappointed, so that an
explanation or change is in order.
Exophora is handy for SITUATION MANAGING, where participants might have
conflicting views about what is going on. Some robbers recently confronted the
drivers of an armored car with (Gainesville Sun, Dec. 20, 1978):
This is a holdup. We’re not kidding.
description of the situation was, as might be expected, reinforced with the
authority of firearms (cf. Goffman 1974: 447). When an engineer said about the
Arizona floods (V.2.1) (Gainesville Sun, Dec. 20, 1978):
(132) It’s going to get worse before it gets better.
was no clear definition of what the ‘it’ designated; presumably the entire
situation brought about by the events being depicted in previous utterances.
Halliday and Hasan (1976: 53) cite several types of “institutionalized
exophoric reference” in which pro-forms are conventionally used without
commitment to specific conceptual content:
First and second person pronouns are inherently exophoric, and their use
presupposes the mutual identifiability of the communicative participants, though
more directly for speaking than for writing. Conceptual content enters
prominently when the referents are assigned to METACLASSES (III.3.20):
(133) O, what a rogue and peasant slave am I! (Hamlet, 11, ii,
(134) You blocks, you stones, you worse than senseless things! (Julius Caesar, 1, i, 40)
A conceptually empty dummy (II.2.15.6) is employed in subject slots for
describing the state of the weather:
(135) It’s raining/ snowing/ hailing/ etc.
preferences for mapping continuous events onto verbs and for having at least one
subject and verb in a clause (III.4.26) creates a need for a dummy subject
lacking content, agency or instrumentality. This usage might be designated by
the French term “servitude grammaticale.”
A frequently unspecified ‘they’ points toward anonymous agents whose status
is not currently relevant, let alone ‘uniquely identifiable” (V.3.1), as in
this opening from a student’s paper:
They told me when I came here I would have to work hard.
vague agents are DEFAULT entities created for the connectivity of events (cf.
V.3.3.6), and they are processed no further than they need to be.
A partly non-determinate ‘we’ permits the speaker to include himself or
herself into a broad class of undetermined size, e.g. in another fragment from
that student’s paper:
(137) In Florida, we don’t see things like other Southerners.
writer probably had no intention of including the entire class of Florida
inhabitants, but only a group of PROTOTYPES (V.3.3.7). Another vague use of
‘we’ is enlisted to “identify the writer and reader as involved in a joint
enterprise” (Quirk et al. 1972: 208), as in this passage from the Atomic
Energy Commission (cited and discussed in Beaugrande 1977b: 329):
(138) Now we are hearing more and more about another kind of radiation [
the expert writer and the lay readers are hardly ‘hearing’ about ‘atornic
energy in any comparable way. But the hope of obtaining readers’ support for
atomic power plants makes it desirable to evoke solidarity, however phoney.
In general, ‘you’ serves as agent for actions that are considered typical,
whoever may be doing them. We find this element also in students’ papers, for
(139) You never know what the teacher wants on these
are fond of suggesting a personal address with ‘you’, even though they are
talking to an anonymous group (cf. Marcuse 1964:92). One brand of car, for
instance, claims to be (TIME, Nov. 13, 1978):
(140) The difference between a car you like and a car you love.
‘You’ is a cross between an impersonal pro-form and the kind of direct
address we see in this ad (TIME, Nov. 13, 1978):
(141) Could the car you’re driving pass this test?
A tendency to rely on exophora without clear conceptualization is noted by
Halliday and Hasan (1976: 34), who supply this dialogue between Hasan and their
three-year old son (stressed words in italics):
CHILD: Why does that one come out?
That o n e!
there that you push to let the water out.
child was reluctant to provide a conceptual description, assuming that the adult
must have the same focus of attention as he himself. The shifting of stress
among pro-forms indicates his hope that more emphatic pointing will do the job.
The ‘you’ in his final remark is that cited in V.5.4.5.
In a highly publicized study, Peter Hawkins (1969) noted exophora in
descriptions of pictured scenes to be more prevalent in the speech of working-
class children than in that of middle-class. His illustrations were like this.
(143a) Three boys are playing football and one kicks the ball and it goes
through the window
(143b) They’re playing football and he kicks it
and it goes through there […]
Influenced by Basil Bernstein (cf. Bernstein 1964), Hawkins
took this material as a demonstration of the divergence between the
“elaborated code” of the middle class (i.e., having many options) and the
“restricted code” of the working class (i.e., having few options). Aside
from the dubiousness of these notions (cf. Oevermann 1970), they seem to miss
the point here. The working-class children probably saw no reason for conceptual
naming of events and objects which they could see and point out in front of
them. In contrast, the middle-class children probably had much richer experience
with the WRITTEN code, whose ‘elaborateness” is due to its frequent
removedness from an apperceivable situation. Also, the middle-class children
would identify an interview with a school situation where the written code is
favored. Still, (142) shows that even a child of two celebrated university
professors tends naturally to rely on exophora in relaxed situations. Perhaps
Hawkins would have us view Shakespeare as a user of the “restricted code”
because of exophora in this famous scene from Hamlet (III, iv, 131-34)?
QUEEN: To whom do you speak this?
Do you see nothing there?
Nothing at all; yet all that is I see.
Nor did you nothing hear?
No thing but ourselves.
HAMLET: Why, look you there! Look how it steals away!8 [8. For an Elizabethan audience, the spectral referent might well have been real enough. See now Zeffirelli's and Branagh’s magnificent film versions.)
Halliday and Hasan (1976: 36) point out the influence of PEER GROUPS on
exophoric reference. They quote Bernstein’s listing of “close-knit social
groups” compiled without apparent realization of the biting sarcasm in this
juxtaposition: “prison inmates, combat units of the armed forces, criminal
subcultures, the peer group of children and adolescents, and married couples of
long standing.” The close-knitness of working-class children, as suggested by
the Hawkins data, might have much to do with their limited social mobility.
Conceivably, the superfluous pronouns taken as social markers in phrases like:
My sister she plays the piano. (student data)
quasi-exophoric. Due to predominant experience with the spoken mode, text
producers might use a two-stage means of referring (1) present a concept name;
and (2) co-refer to the concept via a pro-form. Significantly, this construction
appears, as far as I can judge, virtually always in subject slots — we would
not, for instance, encounter:
(146) They gave my sister her a piano it.
this account is justified, the initial naming of the concept (‘my sister) in
(145) would function as an announcement of TOPIC that is not deemed a part of
the subject-verb dependency. Apparently, the creation of a control center for
the conceptual-relational network is detached from that for the grammatical
dependency network—a departure from a standard preference (cf. III.4.14).
In principle, exophora can be applied to whatever is evident in the
communicative situation. Entering a room where food was spilled on the floor, a
Florida mother was recently heard to say to her child:
You did that?
the entire utterance consists of pro-forms. However, as the late Adrian Akmajian
(personal communication) notes, there are some constraints upon the compacting
of utterances. The mother could have said (148a), but hardly (148b), and almost
certainly not (148c) or (148d):
Exophora demonstrates the reciprocity of the interaction between language use
and situation. The situation strongly effects the actualization of strategies,
but certain conventions will nonetheless be upheld. In the samples (148a-d), the
terseness of exophora is constrained by the sequencing operations of English. We
shall note further illustrations of the limits upon terseness of expression in
the next section on ellipsis.
Discussions of ellipsis, sometimes arcanely called “substitution by zero,”
have been marked by controversy (compare Karlsen 1959; Gunter 1963; Isačenko
1965; Crymes 1968; Dressler 1970b; Halliday & Hasan 1976; Grosz 1977). The
dispute might be stated as follows. The surface structures in texts are often
not so complete as they might be in the judgment of the investigator. Language
theories with clearly drawn boundaries of grammatical or logical well-formedness
necessarily proliferate the treatment of utterances as elliptical, according to
the explicitness of the well-formed idealizations. An extreme view is suggested
by Clark and Clark (1977: 16) when they assert that (149a) is really an
“elliptical” version of (149b):
Napoleon conquered Italy, Prussia, and Austria.
Napoleon conquered Italy, Napoleon conquered Prussia, and Napoleon conquered
is hard to see the psychological justification for such a claim. It would seem
either to exhume the old notion that people have to work with kernel sentences
in communication; or to imply an overly literal interpretation of the notion of
the PROPOSITION LIST (cf. VII.3.6). A processor would scarcely create three
separate entries for ‘Napoleon’ and ‘conquered’ for either (149a) or
The standpoint apparently advanced by the Clarks would make it necessary to view
most utterances as elliptical, and to bloat procedures enormously with redundant
entries requiring subsequent removal. Even less extreme examples seem difficult
to classify as elliptical. In the following samples from an essay by star
athlete Jim Brown (in Levin [ed.]
1977:42ff.), I have added in square brackets items that might conceivably be
Manhasset was going to be just as playful as St. Simons Island [was].
She was, no doubt, a good woman, but [she was] quite [a] stern [woman].
I loved my mother as much as any son would [love his mother].
is still questionable whether the production and comprehension of these
fragments as they stand would be improved or impeded by filling in the bracketed
additions. I argued in II.2.36f. and III.4.40 that people could not plausibly be
converting everything they say or understand into complete sentences. If they
did that, they ought to prefer talking in complete sentences much more than they
do. Grammatical completing turns out many pointless, undecidable structures.
Similar dilemmas attend upon such a literal interpretation of the notion of
proposition lists as the Clarks seem to accept. Walter Kintsch (1977a: 312)
reports that (153a) is indeed easier to perceive than (153b):
(153a) Fred runs faster than the girl.
(153b) Fred runs faster than the girl runs.
model of completion-then-deletion predicts the opposite findings. A model using
conceptual-relational networks, on the other hand, is in agreement, since (153b)
requires testing to see if a second node is needed in addition to an earlier
one, as opposed to direct reutilization of one node in (153a).
If the surface unity is taken to be the GRAMMATICAL DEPENDENCY between two
elements, one of which at least cannot stand alone, then ellipsis ought to be
identifiable via a dangling structural component. We can use empirical tests to
probe people’s judgments on missing components, analogous to the studies of
grammatical expectations (e.g. Stevens & Rumelhart 1975), although bearing
in mind their expectations about text types (cf. my findings given in 11.2.36).
We would be able to settle the dispute on the basis of what language users,
rather than abstract sentence grammarians, consider to be elliptical.
The phenomenon of GAPPING (Ross 1970b) can safely qualify as ellipsis: a
follow-up utterance without a verb, but with a structure otherwise similar to
its predecessor’s, as in this ungainly and obtuse synopsis of a Brecht play (Ohio
State Lantern, Sept. 30, 1970):
(154) It is the story of someone trying to achieve something (Mother Courage
sequence ‘Mother Courage survival’ is noticeably discontinuous (even by Ohio
standards), and must be given connectivity via transfer from the preceding
structure to yield ‘Mother Courage trying to achieve survival’. The transfer
is eased by the fact that the preceding structure contains placeholders
‘someone’, ‘something), so that a reader would be on the lookout for
integrating further knowledge. A preceding structure can supply various
quantities of material to fill a gap. In:
(155) PASTOR: Do you promise to have, hold, love, cherish, and respect this man?
BRIDE: Me him!?
whole series of verbs supplies content addressed by the bride’s response. In a
series with diverse direct objects, only applicable ones could be addressed,
PASTOR: Do you promise to have a fit, hold your tongue, love your neighbor,
cherish this ring, and respect this man?
time, only ‘respect’ carries over into the “gap.”
The term SLUICING (Dressler 1972a: 35) signifies a device in which the verb in a
subordinate clause is elided:
(157) John is busy staring at the girls. I think at the blondes.
Again, a sequence like ’think at the blondes’ is
noticeably discontinuous as it stands, and the content of sexist ‘John is
staring’ must be carried over.
Ellipsis is most noticeable for verbs, because English clauses can dispense with
other elements more readily. For example, utterances without subjects are more
common than those without verbs. However, as Leech and Svartvik (1975: 168)
remark, the ellipsis of subjects in subordinate clauses is not usual., we would
not be likely to encounter such an utterance as:
He was so tired that went to sleep.
constraint is similar to the requirement of dummy subjects for verbs, even where
no agency is to be conceived (cf. V.5.4.2).
Ellipsis, like co-reference, is helpful for REPUDIATING content that hearers
might expect (cf. V.4.6):
(159) And tell them that I will not come to-day.
Cannot, is false; and that I dare not, falser. (Julius Caesar, 11,
quarrels, ellipsis can be used, with proper intonation, to signal a repudiation
of content expressed by someone else (cf. Brazil 1975):
BRUTUS: Let me tell you, Cassius, you yourself Are much condemned to have an
CASSIUS: I an itching palm? (Julius Caesar, IV, iii, 9-10, 13)
Under normal conditions, people tolerate substantial ellipsis, depending on the
extent to which SITUATIONALITY is mediated (cf. 126.96.36.199). SPREADING ACTIVATION
alone would allow for ellipsis of determinate and typical linkage, and
INFERENCING could be applied where needed. Even the notoriously disjointed texts
of Dickens’ (1899 : 25) crafty Alfred Jingle are quite intelligible:
Negus too strong here—liberal landlord—very foolish—very—lemonade much
better—hot rooms—elderly gentlemen—suffer for it in the morning—cruel.
is taken to be missing can be supplied by inferencing as PROBLEM-SOLVING (cf
1.6.4ff.). However, ellipsis as extensive as Mr. Jingle’s is not convenient to
hearers who have to perform inferencing in many directions at once in limited
time. Having the text preserved in print for readers makes matters easier.
Uncooperative hearers might, of course, impede communication by supplying
inappropriate content for elliptical utterances. Uncle Henry’s response (3.3)
cited way back in II.I.8 makes it plain that he does not wish to be sociable.
For ellipsis to be unresolvably ambiguous, we would have to find very unusual
settings. Imagine the still greater confusion that would make (162) ambiguous,
as opposed to (120) with pro-form:
(162) I love my wife. Harry too.
Ellipsis is a further illustration of the TRADE-OFF between compactness and
rapid access (cf. III.3.18). Heavy ellipsis, while cutting drastically back on
surface structure, would demand increased effort for connecting the underlying
text-world model. The presence of ellipsis in varying degrees, each APPROPRIATE
to a type of text and situation (cf. I.4.14), is another demonstration of the
regulatory controls on actualization.
Whereas recurrence, co-reference, and ellipsis keep knowledge spaces current,
junction serves to signal the relations between spaces or between entities
within spaces. The configurations joined via conjunction, disjunction, and
contrajunction are preferentially taken as possessing an analogous surface
organization. Previously successful model-building strategies can accordingly be
applied to the mapping phase for the following structures (cf. III.4.16.11).
This constitutes PATTERN-MATCHING between occurrences of the same text, so that
one stretch of input acts as a model for another (cf. IV.4.5; VII.2.36).
Junction also signals the comparability and relatedness of elements and
configurations in the textual world (cf. VIII.I.24). I shall look into four
types of junction:
CONJUNCTION links two or more knowledge configurations which, in regard to their
environment, are additively the same or similar.
DISJUNCTION links two or more knowledge configurations which, in regard to their
environment, are alternatively the same or similar. While in conjunction, all
content is taken as valid for the textual world, only one configuration in
disjunction need be valid.
CONTRAJUNCTION links two knowledge configurations which, in regard to their
environment, are antagonistically the same or similar, i.e. that deal with
related topics, but via combinations not foreseen in spreading activation. Both
configurations may be true for the textual world, but their inherent relatedness
is not obvious.
SUBORDINATION signals that the relationship between two knowledge configurations
is hierarchical, i.e., that the determinacy of the one is contingent upon access
to the other. Roland Posner (1972) observes that the subordinated material has a
lower position on a gradation of relevance (cf. IV.3.8). Unlike other kinds of
junction, subordination need not signal any analogous organization of surface
structure; indeed, in many languages (e.g. German), subordinate clauses have a
markedly dissimilar structure from that of main clauses.
These various relationships among knowledge configurations can often obtain
without the explicit use of junction, simply because people have predictable
ways of organizing knowledge. It seems reasonable to use the term “junction”
only where there are junctive expressions (‘and’, ‘or’’ ‘but’,
‘because’, and so on). The behavior of natural language junctives is in many
ways different and much more diversified than that of logical connectives (van
Dijk 1977a, 1977b) whose main function is to decide the truth values of complex
statements (cf. III.1.1).
The stories extracted by Hawkins (1969) from children’s protocols illustrate
an extreme of conjunction:
Three boys are playing football and one boy kicks the ball and it goes through
the window and the ball breaks the window and the boys are looking at it and a
man comes out [etc.]
joined configurations are similar to each other: actions, motions, and
apperceptions; and their surface structuring is comparable. The conjunction
signals simple addition of events in a temporal and causal sequence. Because
those relations are recoverable from content, the junctive expression ‘and’
is dispensable, or replaceable with subordinatives:
Three boys are playing football. One boy kicks the ball. It goes through the
Three boys are playing football when one boy kicks the ball so that it goes
through the window. [etc.]
non-committal nature of conjunction makes it the default junction (11.2.24).
Also, ‘and’ might be used as a signal of incompleteness (VII.1.18) so as not
to lose a speaking turn; or as a filler during whose utterance a continuation of
the discourse can be planned out.
Disjunction, in contrast, requires express signaling and cannot be replaced by
subordinative junctives (cf. Leech & Svartvik 1975: 160). Perhaps the
processing of disjunctive configurations is difficult, because the exclusivity
between alternatives is a threat to connectivity and continuity. To keep a
text-world integrated, a processor would want to select the valid alternative
and attach it, discarding the others. Disjunction thus functions as an even
stronger opposition than contrajunction. Consider the watchman’s refrain from
Gilbert and Sullivan’s Iolanthe.:
Every man that’s born into this world alive is either a little bit Liberal or
else a little Conservative.
was no room here for compromise between alternatives (though by the year 2000
there was far too much compromise). In logic, disjunction also figures as the
“law of the excluded middle” (‘either the sun is shining or else it
Contrajunction is usually thought of in terms of opposition; but it is, I have
suggested, weaker than disjunction in that dimension. Two situations, events, or
whatever are treated as inherently not very compatible, yet nonetheless
coexisting in a textual world. Accordingly, spreading activation would not be
expected to connect the two, and hearers must be alerted. A football player
commented on an infamous incident during a game when the tyrannical Woody Hayes
actually punched out an opposing player for spoiling his scheme (Gainesville
Sun, Dec. 3 1, 1978), since one normally expects that people at a location
notice the events there.
(166) I was on the field, but I didn’t see what happened.
Many contrajunctions link longer stretches of material (van Dijk 1977a: 87), and the opposition is accordingly more elaborate. Paragraph 7.4 begins with a phrase containing ‘in contrast’ to announce the differences between conjunction and disjunction regarding replaceability by subordination. The reader may have entertained no particular expectations on the matter. Yet if systemic actualization depends on continuity (I.4.4), contrajunction eases transitions between antagonistic knowledge blocks and hence supports stability.
7.6 Subordination signals more detailed and diffuse dependencies than do conjunction, disjunction, and contrajunction. Subordinative junctives can be treated as TAGS on conceptual relations of the types proposed in III.4.7. Their distribution is strikingly unequal across the set of relations, with causal and temporal linkage being much favored over others. The tags for cause, enablement, and reason overlap somewhat: ‘because’. ‘since’, ‘as’, ‘so’, ‘accordingly’, ‘hence’, ‘thus’, etc. Time relations indicate order, such as previous (‘before’), subsequent (‘after’, ‘next’), and concurrent (‘as’, ‘while’); proximity is often entailed. Many relations have junctives made from preposition plus relative pronoun, such as location (‘near which’, ‘under which’), and so forth. The density of tags for causality and time shows the prominence of those relations for organizing textual worlds, at least for the English-speaking cultures, especially in narratives (VIII.2). The one-word junctives for these relations and the several-word junctives for others would illustrate Zipf’s (1935) law of correlation between frequency of use and shortness.9 [9, Indeed, in many dialects of spoken English, ‘because’ has been shortened down to ‘cause’, as we see in sample (230.2) in VIII.I.14.]
7.7 If causality and time relations are indeed so prominent for coherence, they should naturally be favored in spreading activation and inferencing, whether there are surface junctives used or not. The junctives might increase processing case and yet be dispensable. Consider the old nursery rhyme:
(167) The king was in his counting house, counting all his money;
The queen was in the parlor, eating bread and honey;
The maid was in the garden, hanging out the clothes;
Along came a blackbird and pecked off her nose.
There are no subordinating junctives here; the surface text consists of main-clause sentences. Yet the mere juxtaposition of the statements, reinforced by their parallel structure, gives rise to strong inferences: that the ‘counting-house’, ‘parlor’, and ‘garden’ are proximate in their location; that the actions expressed in the first three lines are proximate in time, while that of the last line interrupts the others; and that it was the maid’s nose, not the queen’s, that was ‘pecked off, because the location of ‘garden’ would enable the bird’s action more readily than ‘parlor’. This inferrability of relations which can dispense with junction is a significant difference between the junctives of natural languages and those employed in logic.
7.8 A text producer might deliberately omit a statement of causality relations, lowering processing case but increasing depth (cf. IV. 1.6). The Bell Telephone Company issues this warning to people doing excavations:
(168) Call us before you dig. You may not be able to afterwards.
leaving the reader to recover the disenablement relation when your phone, or maybe even you, gets terminated by electric shock. People also can infer the reason or purpose of utterances on their own. The following sign reputed to be displayed in Swedish youth hostels offers only the advice which the management of the hostel has reason to believe important for the respective groups of addressees:
(169) Germans: don’t get up before 6 A.M. Americans: don’t come home after 2 A.M. Italians: don’t sing after 10 P.M. Swedes: don’t bring girls into the hostel.
7.9 Hearers might even infer causality relations that the text producer presumably did not intend. This classified advertisement (Gainesville Sun, Sept. 24, 1978):
(170) For sale: office safe. Owner out of town. Call after 6 P.M.
is probably not calculated to encourage the inference that the owner’s absence enables the safe to be stolen after business hours and then quietly sold off. And a psychologist discussing and evaluating the work of Neal E. Miller on motor learning of rats, where curare is used as a local anaesthetic to eliminate any intentional physical interference,10 [10. On the experiments involved and the role of curare, see Gerald Jonas, “Visceral Learning,” New Yorker, Aug. 26, 1972, pp. 49ff. Of course, darts with curare would ‘silence critics’ (or anybody else ) who claimed that Miller’s animals were intentionally producing such effects as altered rate of heartbeat by some trick of the muscles.] was surely not envisioning a causality based on the use of curare by certain South American tribes to manufacture deadly poisonous darts when he said:
(171) Over the years, Miller’s use of curare has silenced many critics.
7.10 I have argued here that explicit subordination provides surface signalling
of underlying conceptual relations that might, in some cases, be inferable via
world knowledge. The subordinative junctives contribute to the efficiency of
processing as long as their use does not become unduly frequent; one would
certainly not want to signal every relation with a junctive. The preference
strategy is probably to use junctives for relations that cannot be readily
inferred because they are variable or non-expected. We saw, however, in (168)
that effectiveness can be increased by not employing junction. I would conclude
that the use of natural language junctives in communication—quite in contrast
to that of logical junctives in proof—should be accounted for in terms of such
design criteria as I proposed in I.4.14.
Click here to go to Chapter VI
Click here to go to the Table of Contents