Topic Levels 
Joseph E. Grimes 
Cornell University and 
Summer Institute of Linguistics 
Now that the sentence is no longer 
the edge of our world, we see more clearly 
than ever how totally responsive speech is 
to the situation that calls it forth and 
to the people involved in it. Bare content 
is shaped and packaged to meet many 
requirements at once. 
I have tried to sort out two broad 
categories of these requirements. The 
first, cohesion, is hearer oriented. It is 
the influence on form of the speaker's own 
assumptions about what the hearer knows at 
each instant of the communication process. 
The second, staging, is speaker oriented. 
It reflects how the speaker calibrates the 
importance of different parts of what he 
himself intends to say. I find it helpful 
to tie down the discussion of both 
cohesion and staging to differences in 
linguistic form; there may be other 
psychological or philosophical overtones 
to cohesion and staging that have no such 
direct repercussions, but getting at those 
overtones is another matter. 
One of the areas where we are making 
progress in the linguistic study of 
discourse is in seeing how speaker and 
hearer always seek a common ground of 
reference. This area, however, is hidden 
in a terminological thicket. Charles 
Hockett (1959.201) originally identified 
it as topic, in which, in his terms, 'the 
speaker announces a topic and then says 
something about it.' Gundel (1974) has 
followed this usage, and I think it the 
best label for now even though it gets 
confused with topicalization, which may or 
may not be part of the same package. Much 
earlier some of the Prague School 
theorists (summarized in Danes 1974), and 
later Halliday (1967), used theme for a 
similar concept, and now Grosz (1977) has 
used focus for something not very 
different. Each of these three terms, 
topic, theme, and focus, has also been 
used for at least two other kinds of 
phenomena by reputable lingu{sts, so 
eventually we are going to have to put 
together a road map to all the 
alternatives; but just because the terms 
are confused is no reason to conclude that 
no headway is being made. 
The idea that I am going to continue 
to refer to as ~ is this: for 
communication to succeea, speaker and 
hearer have to establish common ground. 
This common ground is usually a presumed 
agreement about the identity of certain 
objects in the real world. It may also be 
agreement about certain events or about 
certain relations that hold. As far as its 
linguistic expression goes, I think it 
significant that its formal makeup appears 
to revolve around nominals most of the 
time, treating things that are not 
necessarily objects as if they were. 
One reason the common ground 
phenomenon seems so important is that 
without its narrowing effect, the hearer 
might not be able to manage the numerous 
semantic alternatives that could be 
developed from each expression in the text 
constructed by the speaker. Gundel has 
pointed out the utility in this regard of 
a formulation attempted by Searle 
(1969.126): 
For any speaker S, any object X and 
any predicate P, it is a necessary 
condition of S's having predicated p 
of X in the utterance of a sentence 
containing P, that X should have 
been successfully referred to in 
that utterance and all the 
presuppositions of P should be true 
of X. 
Searle's X is very much like what I 
am calling the topic, in that unless the 
hearer can relate to it referentially, he 
can neither agree nor disagree with 
whatever else may be said about it. 
Gundel illustrates how even an 
isolated sentence like George ate a plate 
o__ff shrimp cannot be assimilated as part of 
a real communication unless the hearer has 
some way of knowing which of the people 
named George is being referred to; once he 
knows that, he can react with yes he did 
or no he didn't or oh, the sign of a new 
constel-latl--~of -fnformation in memory 
(Winograd 1972). 
Grosz has made a useful distinction 
104 
between the hearer's memory for concepts, 
which tends to be long term and global to 
a text, and the hearer's memory for 
linguistic form, whfch tends to be short 
term and local to a segment of text. Her 
distinction interlocks with the one 
Halliday and Hasan (1976) make between 
reference and substitution. What they 
call reference identifies concepts, 
objects outside of language, and even 
pieces of language itself, as they are 
mentioned once they have been introduced. 
What they call substitution includes 
ellipsis; it refers to the reactivation of 
stretches of speech from earlier in the 
text in order to talk about situations 
that have not already been referred to, 
but which have enough in common with 
others that have been referred to that the 
same linguistic expression can apply to 
the new situation. 
It is the first of these, memory for 
concept rather than form, from which the 
speaker appears to take what he hopes is 
common ground between himself and his 
hearer. This selection from the field 
available for global reference is what is 
behind Gundel's observation that the topic 
has to be accessible to the hearer. (The 
most accessible things are 
characteristically the standard elements 
of the communication situation: I, you, 
here, and now.) 
In the course of a temt, be it 
monologue or dialogue, the referential 
common ground that is used as the core of 
communication may change. This is true of 
the global topic and apparently of local 
topics as well. The initial core of 
reference may be designated very simply, 
for example by a single noun phrase, with 
no differentiation of parts or functions 
at the beginning. Gundel even shows that 
the topic of some sentences may be 
implicit, not mentioned in that sentence, 
but nevertheless to be taken into account 
if the sentence is to make sense. 
Once the topic of a text is put into 
play, that topic may be developed in at 
least three different ways that have been 
described so far in the literature: it may 
be expanded, shifted, or split. 
Expansion adds things to the core of 
reference. Shift adds new referents to 
the core and leaves others off, so that 
what is taken as common ground at one 
point in the text differs from what was 
taken as common ground earlier. Splitting 
the core results in local topics being 
brought into play in relation to global 
topics; or rather, higher level topics are 
split into a higher level part and a lower 
level part, a process which if repeated 
may yield more than two levels in the same 
text. 
Topic expansion is illustrated in a 
story from Time magazine (June 21, 1976, 
p. 56). The da--te-of the issue needs to be 
taken as the initial topic; there is no 
other common ground to begin with between 
the writer and the reader. This is normal 
in news stories: consider how impossible 
it is to agree or disagree with anything 
like The Giants beat the Dodgers until one 
knows the occasion. The title of this 
piece is Teton: Eyewitness to Disaster. 
For a reader who knows his geography, 
Teton identifies a place, while for one 
w-~ does not, there is at least a good 
chance that it is a place name or the name 
of a person. The idea of referential 
common ground gives us for starters a 
reasonable guess at an event that happened 
the week before the appearance of the 
magazine, and possibly a place, as a 
limiting field within which to place the 
interpretation of the rest of the message. 
The text begins with a paragraph set 
off in italic type and quotation marks, in 
which the speaker is not identified: "This 
wet spot on the side of the dam started 
spurting a little water ..." The noun 
phrase that begins the sentence is 
definite, as is the dam contained within 
it. The definiteness here suggests that 
the writer expects the reader to be able 
to find the reference because it is 
accessible within the limits already set. 
If he follows that suggestion and takes 
Teton as the name of the dam and accepts 
this as identifying something new within 
that field, his reference succeeds. (The 
side is legitimately definite once we 
identify the dam by what Halliday and 
Hasan call lexical cohesion; dams have 
sides.) As far as pinning down a core of 
reference is concerned, the text so far 
has.its topic built up as clearly as if 
the article had begun much more dully, as 
for example Last week at a place called 
Teton where there is a dam, a wet spot 
appeared on its side. 
The text goes on "... and I asked my 
mother, 'D_oo you think we shou-~ Hotl--i-~-y the 
authorities?' She said: 'I don't 
think ...'" Here the person who is making 
the report is mentioned explicitly for the 
first time, as is the mother to whom the 
question is addressed. This complex of the 
observers, the wet spot on the dam, the 
location, and the time persists as the 
referential core or topic through the 
course of over a column. It is built up by 
small references to give an expanded 
topic. 
Topic shift differs from expansion in 
that some referential elements appear to 
be dropped from the topic as the text 
progresses; some things that were treated 
as part of the common ground in the 
earlier part of the text are not so 
treated later. Schank (1977) focuses on 
the intersection of the referential field 
of one sentence and that of its successor, 
and tries to define some (but not all) 
ways in which that intersection relates 
to the referential field of the next 
i05 
sentence. His initial definition of topic 
as 'any object, person, location, action, 
state, or time that is mentioned in the 
sentence to be responded to' is probably 
too inclusive, because it does not take 
into account Searle's factor of successful 
reference; but once a text is begun, 
Searle's boundary condition is no longer 
needed, because the reference has been 
established by the text itself. 
For Schank a new topic is 'derived 
from the original input but is not 
identical to it', in that reference may 
shift from a specific element mentioned in 
the earlier sentence to the class of which 
it is a part in the later sentence, or 
vice versa. The element in the later 
sentence may also be a different 
conceptualization that is like the first 
in kind; Schank calls this supertopic. It 
may also be 'a comment that can be 
inferred from the interaction of two 
conceptualizations', or metatopic. Schank 
suggests more specific rules that he hopes 
will characterize the way topics shift in 
the course of a text. 
The key concept, however, is his 
observation that a sentence out of context 
cannot be said to have a topic, because 
for him the topic arises only out of the 
interaction of adjacent sentences by the 
process of intersection. If he is right, 
or close to right, it is reasonable that 
some of the things that are treated as 
topic earlier in the text be given 
different treatment later in the same 
text, because they are no longer taken as 
part of the referential common ground 
between the speaker and hearer. 
The idea of splitting up the 
referential field embraced by the topic 
into higher and lower level topics has 
been treated in two different ways, each 
of which may be valid in its place. Grosz 
recognizes the phenomenon, and gives an 
intriguing example (1977.23) of a pronoun 
it which refers back to a global topic 
last mentioned half an hour earlier, even 
though a whole series of local topics has 
come between the pronoun and its 
antecedent. In her discussion of global 
and local topics, however, she tends to 
equate the first with memory for concepts 
and the second with memory for forms, 
chiefly because she finds the domain of 
ellipsis to be restricted to a local 
segment of text, and to always involve 
memory for forms. 
Meyer (1974) and Clements (1976), 
however, find the global-local phenomenon 
operating independently of ellipsis or 
other substitute-like memory for form. 
They construct a topical hierarchy 
consisting of a global topic, whatever 
local topics are talked about as part of 
the discussion of the global topic, and 
whatever lower level topics are talked 
about as part of the discussion of those 
local topics, in what apparently gives a 
recursively definable topic tree of 
unrestricted depth. In this model, 
psychological tests of recall show that 
subordinate position in the topic tree 
regularly gives worse recall than 
superordinate position. 
The definition of topic used by Meyer 
and Clements is earlier than Gundel's, so 
that one could expect the variance in 
their results to be reduced by attention 
to her principles for recognizing topics. 
Topic for Clements is more like Halliday's 
theme, ordinarily the first thing in the 
sentence. His topic hierarchy comes from 
three rules: 
(i) Topic rule: 
Identify the topic of each 
clause and simple sentence. 
(ii) Old/new rule: 
Decide whether the topic is new 
(never previously mentioned) or 
old (mentioned in an earlier 
topic or comment). If new -- 
assign it one level below the 
previous topic. If old -- 
assign it the same level as its 
first mention. 
(iii) Coordination rule: 
If a topic is coordinated with 
an earlier topic or comment, 
assign it the same level as 
that earlier topic or comment. 
The work of Clements and Meyer lends 
credence to the idea that there may be a 
hierarchy of topics in a text, all 
referential in Halliday and Hasan's sense, 
rather than dependent only on short term 
memory for form. Some observations from 
Koine Greek, the vernacular Greek of the 
first century before Christ to the third 
century after, appear to bear this out. 
Word order is used much less in Koine 
than it is in English to specify 
grammatical relations, because the case 
system of nouns carries that load. One of 
the functions which word order expresses 
seems to be that of identifying shifts in 
topic (Grimes 1975). Noun phrases in the 
nominative case that precede the main verb 
of a clause regularly make that nominative 
the topic; that is, they signal the reader 
to take it as part of the referential core 
that is to be the common ground between 
him and the writer. 
The conjunctions of Koine Greek also 
play a part in the topical structure. 
There are three kinds of conjunctions: 
coordinating, subordinating, and 
resumptive. The coordinating conjunctions 
include most of the ones translatable as 
'and' or 'but' The subordinating 
conj unctions include the ' because ' and 
'if' varieties. The resumptive ara, dio, 
and occasionally idou mean something like 
106 
'now, back to the main point'. They reset 
the topic level from wherever it was to 
the global topic. 
Working along with the conjunction 
system is the system of grammatical 
subordination. It uses relativizers, 
complementizers, participles, and verbal 
nouns to signal that certain propositions 
are peripheral in the author's perspective 
on what he has to say. All these 
subordinating grammatical mechanisms are 
used constantly in the ancient Koine 
documents, even though it would be 
perfectly possible to express what they 
express in strings of independent clauses 
with no subordination. 
The result is texts with a richly 
elaborated grammatical structure, with 
some clauses at four or even five levels 
of subordination. The general distinction 
between global and local applies, but 
recursively: at any level of 
subordination, it appears possible to have 
yet another level of subordination 
attached. 
When we turn, however, to languages 
that make distinctions of topic level 
explicit in their pronominal systems, we 
find no greater elaboration than a 
distinction between global and local 
topics. Bacairi of Brazil (Wheatley 1973) 
is such a language. The terms Wheatley 
uses in his description are 'thematic - 
athematic' and 'focal - nonfocal'; but it 
seems clear from perusal of his paper that 
we would now want to call the thematic 
category the topic, and the focal category 
global. Someone who has been identified as 
part of the common ground between speaker 
and hearer for a text as a whole is 
referred to by the pronoun maca 'thematic 
focal animate'; someone who Ts--6opic for a 
local segment but is not the global topic 
at the same time is referred to by auaca 
'thematic nonfocal animate'; someone who 
is global topic, but is referred to within 
a stretch that has a local topic active, 
is mauanca or maunca 'athematic focal 
animate', and those who are topic neither 
at the global level nor at the local level 
are referred to by uanca 'athematic 
nonfocal animate' There are inanimate 
counterparts for all four pronouns. 
On the thematic side -- that is, in 
the pronouns used for topics -- there is a 
situational or deictic use that confuses 
the picture. The pronoun inara 'thematic 
nondeictic animate' is only textually 
defined, with an anaphoric antecedent that 
is taken always from the preceding 
sentence, not from the topical structure 
of the text as a whole. On the other hand 
there is a pronoun mira 'thematic deictic 
animate' which applies only to animate 
things that can be seen and are near the 
speaker. In between the nondeictic and 
the deictic are the situational uses of 
maca and auaca. The first denotes someone 
far away but in sight, and the second 
denotes someone nearer to the speaker but 
not as near as mira. 
There are situations in which these 
two pronouns appear to flip their 
reference; what they actually do is change 
the basis of the reference from the 
situation to the text. Thus I can begin a 
text by identifying a boy over there as 
maca because he is relatively far away, 
and a woman standing closer as auaca 
because she is not so far away. But i-f 
what I have to say revolves around the 
woman rather than the boy, I will switch 
after a few sentences to the textual 
definition of the pronouns and use maca 
for the woman because she is more central 
to what I have to say, and auaca to the 
boy when I treat him as a loc--~ topic. 
However, we have no information about the 
use of these pronouns to topicalize at 
more than two levels at a time. 
Longuda of Nigeria has a less 
elaborate pronoun system with respect to 
topics than that of Bacairi (Newman, in 
press), but nevertheless it distinguishes 
topic from nontopic. There is actually a 
series of pronoun-aspect particles that 
appear in sequence in texts to identify 
actions of the central character and 
distinguish them from the actions of 
others. In some parts of a text the 
character singled out by the pronouns is 
the one the text is about; he is treated 
as the global topic. Where local topics 
are introduced, however, the topic pronoun 
set switches over to the local topic, and 
the referent of the global topic is 
referred to by the same nontopic pronoun 
as all the other characters. 
For example, in a story about a 
rabbit, most of the story uses the topic 
pronoun series for the rabbit and a 
nontopic pronoun for everybody else, 
including an elephant who interacts with 
the rabbit. One section, however, is 
about what the elephant did. In that 
section it is the elephant who gets the 
topic pronoun series, and the rabbit gets 
the nontopic series, even if it is the 
rabbit whom the story is about globally. 
When the section about the elephant ends, 
the topic pronoun series reverts to the 
global topic, the rabbit. 
Evidence from languages like these 
proves the linguistic realism of a 
distinction between one kind of topic and 
another. It also agrees in the main with 
the kinds of analysis we have been making 
for English; apparently we are fairly 
close to the right track, in terms of 
those languages that put this kind of 
thing right out on the surface. What is 
not yet clear is the number of levels of 
topic we may deal with in a text: 
Clements' s analysis of English and my 
analysis of Greek point toward more than 
two simultaneous levels of topic, while 
107 
the languages that distinguish levels of 
topic in their pronoun systems seem to 
support two levels. 

Bibliography 
Clements, Paul. 1976. The effects of 
staging on recall from prose. Cornell 
University Ph.D. thesis. 
Danes, Frantisek, ed. 1974. Papers on 
functional sentence perspective 
(Janua Linguarum series minor no. 
147). The Hague: Mouton. 
Grimes, Joseph E. 1975. 'Signals of 
discourse structure in Koine'. In 
George MacRae, ed. Society for 
Biblical Literature 1975 Seminar 
Papers, Vol. i. Missoula, Montana: 
Scholars Press, 151-164. 
Grosz, Barbara. 1977. The representation 
and use of focus in dialogue 
understanding (Technical Note 151). 
Menlo Park: Stanford Research 
Institute. 
Gundel, Jeannette. 1974. The role of topic 
and comment in linguistic theory. 
University of Texas Ph.D. thesis. 
Halliday, Michael A. K. 1967. 'Notes on 
transitivity and theme in English, 
Part 2' Journal of Linguistics 
3:199-244. 
and Ruqaiya Hasan. 1976. Cohesion 
in English. London: Longman Group 
Limited. 
Hockett, Charles F. 1959. A course in 
modern linguistics. New York: The 
Macmillan Company. 
Meyer, Bonnie J. F. 1974. The organization 
of prose and its effects on recall. 
Cornell University Ph.D. thesis. 
Newman, John F. in press. 'Participant 
orientation in Longuda folktales'. In 
Joseph E. Grimes, ed. Papers on 
Discourse. Dallas: Summer Institute 
of Linguistics. 
Schank, Roger. 1977. 'Rules and topics in 
conversation'. Cognitive Science 
1:4.421-442. 
Searle, John. 1969. Speech acts. 
Cambridge: Cambridge University 
Press. 
Wheatley, James. 1973. 'Pronouns and 
nominal elements in Bacairi 
discourse'. Linguistics 104:105-115. 
Winograd, Terry. 1972. Understanding 
natural language. New York: Academic 
Press. 
