A Typology of English Texts : Douglas Biber

A typology of English texts*
DOUGLAS BIBER
Abstract
This paper develops a typology of texts in English with respect to a five-

dimensional model of variation. Each dimension comprises a set of lexical
and syntactic features that cooccur frequently in texts, reflecting underlying
shared communicative functions. Eight text types are identified with respect
to these dimensions; each type represents a grouping of texts that are
markedly similar to one another with respect to their dimension character-
izations. Although the linguistic variation among texts is continuous, there
are only a few linguistic characterizations that occur frequently, and the text
types identified here represent the prototypical groups of texts having these
characterizations. The types are interpreted by considering their predomi-
nant linguistic features and the general communicative characteristics of the
texts grouped in each type, and by performing microanalyses of particular
texts. Based on these interpretations, functional labels, such as 'Informa-
tional interaction', 'Learned exposition', and 'Involved persuasion', are
proposed for each type.
1. Introduction
Over the last several years, numerous studies have attempted to document
the nature and extent of linguistic similarities and differences among
various kinds of texts. A major goal of such research is to develop an
overall typology of texts, to provide a theoretical and empirical founda-
tion for comparative discourse research. This need has been emphasized
by Tannen (1982: 1):
Linguistic research too often focuses on one or another kind of data, without
specifying its relationship to other kinds. In order to determine which texts are
appropriate for proposed research, and to determine the significance of past and
Linguistics 27 (1989), 3-43 0024-3949/89/0027-0003 $2.00

© Mouton de Gruyter, Amsterdam
Brought to you by | Takming University of
Authenticated | 150.108.161.71
Download Date | 1/7/13 10:17 PM
4 D. Biber
projected research, a perspective is needed on the kinds of language and their

interrelationships ... discourse analysis needs a taxonomy of discourse types, and
ways of distinguishing among them.
It is easy to illustrate the need for a typology of texts; many discourse

studies analyze particular sets of texts without specifying their relations to
other kinds of texts, often making the unwarranted assumption that
findings can be generalized to 'discourse' as a whole. For example, there
have been numerous contradictory conclusions concerning the linguistic
characteristics of speech and writing due to this methodological short-
coming (many studies compare only face-to-face conversation and aca-
demic exposition but assume that their results characterize all speech and
writing; see Tannen 1982; Biber 1986). Similarly, contradictory claims are
common concerning the linguistic characteristics of 'complex' versus
'simple' discourse or 'formal' versus 'informal' discourse (Finegan and
Biber 1986; Besnier 1986). In fact, a typology of texts is a research
prerequisite to any comparative register analysis, whether of speech and
writing, formal and informal texts, restricted and elaborated codes,
literary and colloquial styles, 'good' and 'bad' student compositions, early
and late historical periods, or whatever, to situate particular texts relative
to the range of texts in English.
There have been a number of text typologies proposed within linguis-
tics and related fields. Researchers have typically developed typologies on
a functional basis: first identifying one or two particular functional
dichotomies, and then describing the 'types' defined by the poles of those
distinctions. For example, much of the research on spoken/written
differences can be considered as implicitly typological in this way, where
the mode differences between speech and writing are claimed to distin-
guish between oral and literate text types. More explicit typologies have
been proposed by Longacre (1976) and Chafe (1982). Longacre proposes
a four-way distinction of 'monologic' texts with respect to the parameters
of projected time and temporal succession: narrative, expository, proce-
dural, and hortatory. Chafe proposes a four-way classification of texts
with respect to the parameters of 'involvement-detachment' and 'integra-
tion-fragmentation' .
Within rhetorical theory, four basic 'modes' of discourse are tradition-
ally distinguished: narration, description, exposition, and argumentation.
Although there is wide agreement on the importance of these four discourse
types, there is less agreement on the particular parameters distinguishing
among them; for example, different definitions of exposition have focused
on one or another of the following parameters: content type, organization,
objectivity, purpose, or informational density (Grabe 1984).

A typology of English texts 5
All of these typologies have been proposed on functional rather than

linguistic grounds. That is, in each of these cases researchers have isolated
an important functional difference among texts and have subsequently
attempted to identify the linguistic features associated with that differ-
ence. For this reason, these typologies are not well defined from a strictly
linguistic perspective, considering questions such as (1) do the proposed
sets of defining linguistic features actually cooccur systematically in texts?
(2) are the texts in each type actually similar to each other in their
linguistic form? (3) are the types clearly distinct in their linguistic form?
and (4) does the typology characterise the full range of texts in English?
For example, neither Longacre's typology (see Smith 1985) nor Chafe's
(see Redeker 1984) captures the sets of linguistic features that actually
cooccur in English texts, and therefore these typologies cannot identify
text types that are linguistically well defined. The rhetorical modes of
discourse, on the other hand, are not intended as linguistic types; the
characteristic linguistic features of each mode have not been consistently
defined, and there is considerable linguistic variation among the texts
within each mode. These typologies are important to the extent that they
identify salient functional differences among texts. (The functional dis-
tinctions identified by Chafe have been particularly useful in this regard.)
Functionally based typologies have not been successful, however, in
identifying the salient LINGUISTIC differences among texts in English.
The present study uses a new approach to identify the salient linguistic
text types of English. The typology developed here is based on sets of
syntactic and lexical features that cooccur frequently in texts, rather than
assuming sets of features defined on a priori functional grounds. These
feature sets, the 'dimensions' of variation, are identified empirically by
multivariate quantitative methods, and the linguistic characteristics of
any given text can be specified precisely with respect to each dimension.
The dimensions and associated text characterizations provide the basis
for the present typology: the types are defined such that the texts in each
type are maximally similar in their linguistic characteristics, while the
different types are maximally distinct from one another. These types
represent important functional differences in English, because linguistic
cooccurrence reflects shared function. The order of analysis is reversed
from previous studies, however. The types are first identified on the basis
of their linguistic characteristics and only subsequently interpreted func-
tionally. The resulting text distinctions represent the functional types that
are linguistically well defined in English.
There is one further text typology that should be considered here: the
folk-typology of 'genres'. Genres are the text categories readily distin-
guished by mature speakers pf a language; for example, the genres of

6 D. Biber
English include novels, newspaper articles, editorials, academic articles,

public speeches, radio broadcasts, and everyday conversations. These
categories are defined primarily on the basis of external format. Thus,
newspaper articles are found in the news sections of newspapers; aca-
demic articles are found in academic journals. These distinctions are
related to other differences in purpose and situation, and there are marked
linguistic differences among the genres of English (Biber 1988). Genre
distinctions do not adequately represent the underlying text types of
English, however. Texts within particular genres can differ greatly in
their linguistic characteristics; for example, newspaper articles can
range from extremely narrative and colloquial in linguistic form to
extremely informational and elaborated in form. On the other hand,
different genres can be quite similar linguistically; for example, newspaper
articles and popular magazine articles can be nearly identical in form.
Linguistically distinct texts within a genre represent different text types;
linguistically similar texts from different genres represent a single text
type. In the present typology of texts, genres and text types are clearly
distinguished, and the relations among and between them are identified
and explained.
2. Background: five dimensions of variation
The notion of linguistic cooccurrence is central to linguistic analyses of

style or register. Brown and Fräser (1979: 38-39) emphasize the impor-
tance of this notion, observing that it can be 'misleading to concentrate on
specific, isolated [linguistic] markers without taking into account system-
atic variations which involve the cooccurrence of sets of markers'. Ervin-
Tripp (1972) and Hymes (1972) define 'speech styles' as varieties that are
defined by a shared set of cooccurring linguistic features; the text 'types'
of the present study are the salient varieties of English defined in these
terms. That is, text types are identified quantitatively such that the texts in
a type all share frequent use of the same set of cooccurring linguistic
features. Because cooccurrence reflects shared function, the resulting
types are coherent in their linguistic form and communicative functions.
In the present study, I analyze linguistic cooccurrence in terms of
underlying 'dimensions' of variation. There are three distinctive charac-
teristics of the notion of 'dimension' as I use it. First, no single dimension
is adequate in itself to account for the range of linguistic variation in
a language; rather, a multidimensional analysis is required. Second,
dimensions are continuous scales of variation rather than dichotomous
distinctions. Third, the cooccurrence patterns underlying dimensions are

identified quantitatively (by a statistical procedure known as factor

analysis) rather than on an a priori functional basis.
Dimensions have both linguistic and functional content. The linguistic
content of a dimension comprises a group of linguistic features (such as
passives, nominalizations, prepositional phrases) that cooccur with a
markedly high frequency in texts. Based on the assumption that cooccur-
rence reflects shared function, these cooccurrence patterns are interpreted
in terms of the situational, social, and cognitive functions most widely
shared by the cooccurring linguistic features.
To date, five major dimensions of variation have been identified in
English. Biber (1988) presents a unified description of genre variation in
English in terms of this five-dimensional model. The model is developed
by analyzing the cooccurrence distributions of 67 linguistic features in 481
spoken and written texts of contemporary British English. The texts,
which were taken from the Lancaster-Oslo-Bergen and the London-Lund
Corpora, represent 23 different genres (for example, academic prose, press
reportage, conversation, radio broadcasts). The linguistic features fall
into 16 major grammatical categories: (A) tense and aspect markers, (B)
place and time adverbials, (C) pronouns and pro-verbs, (D) questions, (E)
nominal forms, (F) passives, (G) Stative forms, (H) subordination fea-
tures, (I) prepositional phrases, adjectives, and adverbs, (J) lexical
specificity, (K) lexical classes, (L) modals, (M) specialized verb classes,
(N) reduced forms and dispreferred structures, (O) coordination, and (P)
negation. The features are identified automatically in texts by computer
programs written in PL/1. The cooccurrence patterns among features
(that is, the dimensions) are identified quantitatively by a statistical
procedure known as factor analysis. Biber (1988) includes both a
theoretical analysis of genre variation in terms of the model and a full
discussion of the methodological approach (including situational descrip-
tions of the texts, functional descriptions of the linguistic features, and
technical descriptions of the computational and statistical techniques). I
will begin here with a brief explication of the overall model of variation
and then go on to use this model as the basis for a typology of texts in
English.
The summary that follows on pages 8 and 9 lists the cooccurring
features associated with each of the five dimensions:

8 D. Biber
Summary of the cooccurrence patterns underlying the dimensions:
Dimension 1. 'Involved versus Dimension 2. 'Narrative versu

informational production' nonnarrative concerns'
private verbs past-tense verbs
THAT deletion 3rd person pronouns
contractions perfect-aspect verbs
present-tense verbs public verbs
2nd person pronouns synthetic negation
DO as pro-verb present-participial clauses
analytic negation
demonstrative pronouns present-tense verbs
general emphatics attributive adjectives
1st person pronouns
pronoun IT
BE as main verb Dimension 3. 'Explicit versus
causative subordination situation-dependent referenc'
discourse particles WH relative clauses on objet
indefinite pronouns positions
general hedges pied-piping relative clauses
amplifiers WH relative clauses on subjct
sentence relatives positions
WH questions phrasal coordination
possibility modals nominalizations
nonphrasal coordination
WH clauses time adverbials
final prepositions place adverbials
adverbs adverbs
nouns
word length
prepositions
type/token ratio
attributive adjectives
place adverbials

Dimension 4. Overt expression of Dimension 5. 'Abstract versus

persuasion' nonabstract style'
infinitives conjuncts
prediction modals agentless passives
suasive verbs past-participial clauses
conditional subordination BY passives
necessity modals past-participial WHIZ deletions
split auxiliaries other adverbial subordinators
possibility modals
no complementary features
no complementary features
Most of the dimensions consist of two groupings of features, which

represent sets of features that occur in a complementary pattern. That is,
when the features in one group occur together frequently in a text, the
features in the other group are markedly less frequent in that text, and
vice versa. To interpret the dimensions, it is important to consider likely
reasons for the complementary distribution of these two groups of
features as well as the reasons for the cooccurrence pattern within each
group.
For example, consider dimension 2. The features in the top group
(above the line) are past-tense verbs, perfect-aspect verbs, third-person
pronouns, and public verbs (primarily speech-act verbs), while the
features in the bottom group are present-tense verbs and adjectives.
Considering all of the features on dimension 2, this dimension is
interpreted as distinguishing narrative discourse from other types of
discourse, suggesting the interpretive label 'Narrative versus nonnarrative
concerns'. Narrative concerns are marked by considerable reference to

10 D. Biber
past time, third-person animate referents, and reported speech (public

verbs); nonnarrative concerns, whether expository, descriptive, or other,
are marked by immediate time and attributive nominal elaboration. The
complementary groupings on the other factors reflect similar functional
relations.
To represent the communicative function(s) underlying each cooccur-
rence pattern, the dimensions are labeled as follows:
1. Involved versus informational production
2. Narrative versus nonnarrative concerns
3. Elaborated versus situation-dependent reference
4. Overt expression of persuasion
5. Abstract versus nonabstract style
Dimension 1 (see summary, page 8) represents a dimension marking
high informational density and exact informational content (the bottom
grouping of features), versus affective, interactional, and generalized
content (the top group of features). Two communicative parameters seem
to be involved here: (1) the primary purpose of the writer/speaker:
informational versus involved; and (2) the production circumstances:
those circumstances providing careful editing possibilities, enabling preci-
sion in lexical choice and an integrated textual structure, versus circum-
stances that are characterized by real-time constraints, resulting in
generalized lexical choice and a generally fragmented presentation of
information. To reflect these parameters, the interpretive label 'Involved
versus informational production' is used for this dimension.
Considering both groups of defining features, dimension 3 is inter-
preted as characterizing highly explicit, context-independent reference
versus nonspecific, situation-dependent reference. WH relative clauses
(including pied-piping constructions) are used to specify the identity of
referents within a text in an explicit and elaborated manner. Time and
place adverbials, on the other hand, are usually used for text-external
references, where the addressee must identify the intended place and time
referents in the actual physical context of the discourse. Overall, the label
'Elaborated versus situation-dependent reference' captures the character
of this dimension.
The interpretive label Overt expression of persuasion' is used for
dimension 4. This dimension marks the degree to which persuasion is
marked overtly, whether marking the speaker's point of view, or the
speaker's attempt to persuade the addressee.
Finally, the cooccurrence of conjuncts, passive constructions, and past-
participial clauses on dimension 5 marks informational discourse that is
abstract, technical, and formal in style versus other types of discourse.

The label 'Abstract versus nonabstract style' can thus be proposed for
dimension 5.
3. Multidimensional characterizations of texts
In the same way that the frequency of nouns in a text might be called the
'noun score' of that text, 'dimension scores' can be computed to
characterize each text with respect to each dimension. First, the frequen-
cies of all linguistic features are normalized to a text length of 1,000 words
and standardized to a mean of 0.0 and a standard deviation of 1.0. On
such a scale, a score of 1.0 marks a value that is one standard deviation
higher than the overall mean score; a score of — 1.0 marks a value that is
one standard deviation below the mean.1 Standardized scores are used
because they set frequency counts to a single scale, making the frequencies
directly comparable across features.
After standardization, dimension scores are computed by summing, for
each text, the frequencies of the salient defining features of the dimension.
To illustrate, consider dimension 3 as given in the summary on page 8.
The dimension score representing dimension 3 is computed by adding
together the frequencies of WH relative clauses on object and subject
positions, pied-piping relative clauses, phrasal coordination, and nomi-
nalizations (the features with positive loadings), and subtracting the
frequencies of time adverbials, place adverbials, and general adverbs (the
features with negative loadings) — for each text
The linguistic relations among texts can be considered by comparing
their dimension scores, and the relations among text varieties can be
considered by comparing the mean dimension score of each variety. For
example, Figure 1 plots the mean dimension scores for nine English
genres with respect to dimension 1, 'Involved versus informational
production'. Face-to-face conversation has the highest value, marking it
as extremely involved and interactive; this high score reflects high
frequencies of present-tense verbs, private verbs, first- and second-
person pronouns, contractions, etc., together with markedly low fre-
quencies of nouns, prepositional phrases, long words, etc. Personal
letters and interviews have moderately high scores on dimension 1, while
general fiction and prepared speeches have intermediate values. Genres
like press reportage, academic prose, and official documents have the
lowest values on dimension 1, marking them as quite informational and
noninvolved; these low scores reflect very high frequencies of nouns,
prepositional phrases, etc., plus very low frequencies of private verbs,
contractions, etc.

12 D. Biber
35 + Face-to-face conversations
30
25
20 + Personal letters
Interviews
15 +
10 +
Prepared speeches
General fiction
-5 +
-10 + Editorials
-15 + Academic prose; Press reportage
O f f i c i a l documents
-20
Figure 1. Mean scores of dimension 1 ('Involved versus informational production') for nine
genres
The overall relations between any two texts or varieties can be analyzed
by consideration of their relative scores on all five dimensions. In previous
studies, I have used the dimensions to examine the relations among
various genre classes (see for example 1987, 1988; Biber and Finegan
1988); the present paper develops a typology of English texts with respect
to the five dimensions.

4. A typology of texts in English
4.1. Identification of formally distinct text types
Groupings of texts that are similar in their linguistic form can be

identified empirically by using a statistical procedure known as cluster
analysis. In the present case, the groupings are identified on the basis of
similarities with respect to the five dimensions outlined above. Cluster
analysis groups texts such that the texts within each cluster are maximally
similar to each other in their exploitation of the textual dimensions, while
each cluster is maximally distinct from the others. That is, those texts with
the most similar dimension scores are grouped in each cluster.
To identify the major text types in English, it was necessary to analyze
the similarities and differences among a large number of texts representing
many different spoken and written genres. In all, 481 texts taken from 23
major genre categories were analyzed, as summarized in Table I. 2 Some
of these genre categories represent several distinct subgenres. For exam-
ple, press reportage includes cultural, sports, and financial news reports;
academic prose includes humanities, social sciences, and engineering
expositions; broadcasts include sports reporting and reportage of less-
exciting events, such as a funeral and a scientific demonstration. Taken
together, these texts represent a large range of the communicative
situations and purposes found in English.
A cluster analysis produces different solutions for different numbers of
clusters (that is, a one-cluster solution, a two-cluster solution, etc.).
Therefore, the first task for the researcher is to determine which solution
provides the best 'fit' to the data, that is, in which solution the texts within
each cluster are maximally similar while the clusters themselves are
maximally distinct. The 'fit' of a solution can be assessed quantitatively,
and in the present analysis, the solution for eight clusters provides the best
fit to the data.3
The cluster analysis assigns every text in the study to some cluster. If
each text is labeled with the number of its cluster, all 481 texts can be
plotted in a way that illustrates the differences among clusters. Figure 2,
for example, shows the distribution of texts with respect to dimension 1
(Involved versus informational production) and dimension 3 (Explicit
versus situated reference). This plot represents the distribution of texts
according to their exploitation of the linguistic features on these two
dimensions; the horizontal axis plots the dimension score of each text for
dimension 1; the vertical axis plots the scores for dimension 3. The
numbers in the plot represent the cluster number of the texts having the
given scores on these two dimensions. For example, the position held by

14 D. Biber
Table 1. Distribution of texts across 23 genres
Genre Number
of texts
Written — genres 1-15 from the LOB corpus

1. Press reportage 44
2. Editorials 27
3. Press reviews 17
4. Religion 17
5. Skills and hobbies 14
6. Popular lore 14
7. Biographies 14
8. Official documents 14
9. Academic prose 80
10. General fiction 29
1 1 . Mystery fiction 13
12. Science fiction 6
13. Adventure fiction 13
14. Romantic fiction 13
15. Humor 9
16. Personal letters 6
17. Professional letters 10
Spoken — from the London-Lund corpus
18. Face-to-face conversation 44
19. Telephone conversation 27
20. Public conversations, debates, and interviews 22
21. Broadcast 18
22. Spontaneous speeches 16
23. Planned speeches 14
Total 481
Approximate number of words 960,000
the circled number 5 on Figure 2 locates the text that has a dimension
score of 13 on dimension 1 (the horizontal axis) and a score of -5 on
dimension 3 (the vertical axis), and that belongs to cluster 5; these
dimension scores mark this text as moderately involved in focus and
moderately situated in reference.
Figure 2 shows relatively distinct groupings for clusters 1, 2, 5, 7, and 8,
while the remaining three clusters (3, 4, and 6) are less well distinguished
in terms of dimensions 1 and 3. The texts in cluster 1 (marked in the plot
by the numeral 1) are characterized by quite high scores on dimension 1
and relatively low scores on dimension 3; cluster 2 is similar except it has
lower scores on dimension 1. Clusters 5, 7, and 8, all have unmarked
scores on dimension 1; they differ from one another along dimension 3:

N CM «
.m ©
eo
eo
in
oo
in
in
m
ιηιη
in I
in oo in in
in in « m in .! 1
I I
o « oo in in in m
oo in oo oo in
in in in
in in
in m
m m
ω ο ω
")
ω to
(Ο (Ο (D
α
« ω ω (ο ω
m α) Φ
3ω ο
ω to α> ο α>
ν « «t to 5
•J5
*
o-eecw-oc n
'ε;

16 D. Biber
cluster 8 texts have markedly high scores, cluster 7 texts have extremely
low scores, cluster 5 texts have unmarked scores. Clusters 3, 4, and 6 all
have low scores on dimension 1; on dimension 3, cluster 4 texts tend to
have high scores, cluster 6 texts tend to have relatively low scores, and
cluster 3 texts have unmarked scores.
The asterisks on Figure 2 plot the 'centroids' (the central characteriza-
tions) for each cluster with respect to the two dimensions. An overall
summary of each cluster is given in Table 2, including the number of texts
in the cluster, the nearest cluster, and the 'distance' to the nearest cluster.
The 'distance' measures the cumulative difference between the cluster
centroids with respect to the five dimensions. Table 2 confirms the
impression given by Figure 2 that the types are not equally distinct in their
linguistic characterizations. In particular, this table shows that clusters 3,
4, and 6 are relatively nondistinct: both cluster 3 and cluster 6 have cluster
4 as the nearest cluster, with a distance of only 8.3 between cluster 3 and
cluster 4.
The cluster analysis identifies the 'core' text types in English: the
groupings that contain very high concentrations of texts. There is a group
of core texts and a group of peripheral texts associated with each cluster.
Core texts are very similar to the central linguistic characterization of a
cluster; peripheral texts are relatively dissimilar to the central cluster
characterization, but even more dissimilar to other clusters.4 Out of the
481 texts in this study, 345 are grouped into one of the core text types by
the cluster analysis; Figure 2 plots only these core texts. Peripheral texts,
however, are not aberrant; their existence rather reflects the fact that
textual variation is continuous. Texts do not divide into sharply distinct
'types' — instead there is a continuous range of variation in linguistic
form and use. The notion of 'text type' developed here is based on the
Table 2. General summary of the clusters
Cluster number Frequency of Frequency of Nearest cluster Centroid

core texts peripheral texts distance to
nearest cluster
1 22 1 2 15.3
2 49 24 1 15.3
3 28 15 4 8.3
4 53 18 3 8.3
5 47 13 8 11.0
6 117 33 4 10.2
7 7 5 5 15.3
8 22 27 5 11.0

frequent and therefore typical clusterings of texts, which account for the
majority of texts in English. In a sense, these can be considered the text
'prototypes' of English. There are, however, other texts that fall in
between clusters, grading from one type to the next. I return to the
continuous nature of variation among texts in section 5.
The grouping of texts into clusters is determined on the basis of their
characterization with respect to all five dimensions. That is, texts that are
similar with respect to one dimension but very different with respect to
other dimensions are likely to be grouped into different clusters. Figure 2
shows the distribution of texts with respect to only two dimensions, but it
can be used as an illustration of the way texts are grouped into clusters.
For example, texts in clusters 1, 2, and 5 are very similar with respect to
their dimension 3 scores (the vertical axis); texts in all three clusters
generally have scores between 0 and — 8. With respect to their dimension
1 scores (the horizontal axis), however, the texts in these three clusters are
distinct: texts in cluster 1 have scores ranging from 40 to 54; texts in
cluster 2 range from 22 to 40; texts in cluster 5 range from — 3 to 15.
Similar comparisons can be made for clusters 8, 5, and 7: texts in these
clusters are quite similar with respect to their dimension 1 scores (ranging
generally between — 3 and 12), but quite distinct with respect to their
dimension 3 scores (cluster 8 ranging from 8 to — 1; cluster 5 ranging from
2 to — 8; cluster 7 ranging from — 10 to — 16). The picture given by Figure
2 is incomplete because only two dimensions are considered. When all five
dimensions are considered, it is possible to identify the salient distinguish-
ing characteristics of all eight clusters.
Figures 3 and 4 summarize the distinguishing characteristics of the
eight clusters, plotting the centroid score of each cluster with respect to
each dimension. These two figures present the same information: Figure 3
highlights clusters 1-4, while Figure 4 highlights clusters 5-8. The
information presented in these figures overlaps the information presented
in Figure 2; the centroid values for dimensions 1 and 3, which are given by
asterisks on Figure 2, are repeated on the respective scales of Figures 3
and 4.
On the basis of Figures 3 and 4, it is possible to describe the
distinguishing linguistic characteristics of each of the eight text types.
Using the interpretive dimension labels, cluster 1 is situated, nonabstract,
and extremely involved, but not marked for narrative concerns or
persuasion; cluster 2 is similar to cluster 1, except it is less involved.
Clusters 3 and 4 are also similar to each other: both are extremely
informational, highly elaborated, nonnarrative, and nonpersuasive. These
two clusters differ primarily with respect to dimension 5, where cluster 3 is
extremely abstract in style while cluster 4 is only moderately abstract.

18 D. Biber
-20 * CA
-30
Involved Narrative Elaborated Overt Abstract

vs. concerns vs. persuasion style
informational situated
production reference
Figure 3. Dimensional characterization of the eight text types, highlighting clusters 1,2,3, and
4: '*' marks the centroid score of each cluster on each of the dimensions; dimensions 2-5 use the
same scale (from —13 to +9), while dimension 1 uses a compressed scale (from —30 to +48)
Cluster 5 is extremely narrative, moderately involved, situated, and

nonabstract, and not marked for persuasion. Cluster 6 combines the
distinctive characteristics of clusters 4 and 5: it is informational as well as
narrative, while it is not marked on dimensions 3, 4, and 5. Cluster 7 is

40
30
20
10
0 -*^
-10
-20 * C4
-30
Involved Narrative Elaborated Overt Abstract

vs. concerns vs. persuasion style
informational situated
production reference
Figure 4. Dimensional characterization of the eight text types, highlighting clusters 5,6, 7, and
8: '*' marks the centroid score of each cluster on each of the dimensions; dimensions 2-5 use the
same scale (from —13 to +9), while dimension 1 uses a compressed scale (from —30 to + 48)
extremely situated in its reference, in addition to being markedly nonnar-

rative, nonpersuasive, and nonabstract. Finally, cluster 8 is distinctive in
being extremely, persuasive, in addition to being moderately involved,
nonnarrative, and elaborated.5

20 D. Biber
As discussed above, the dimensions represent frequency scales for sets

of linguistic features, and the clusters of texts are determined on the basis
of their similarities in the use of the features grouped on each dimension
(summarized on pages 8 and 9). Clusters are marked with respect to
particular dimensions to the extent that they have large positive or
negative scores for those dimensions. For example, cluster 1 is marked for
dimensions 1, 3, and 5. Ranking extremely high on dimension 1, the texts
in this cluster tend to have very frequent occurrences of private verbs,
contractions, first- and second-person pronouns, etc. (the upper features
on dimension 1), together with shorter words and lesser lexical variety
(the bottom features on dimension 1). Ranking low on dimension 3, the
texts in this cluster are characterized by frequent time and place adver-
bials (the bottom features on dimension 3) and markedly few relative
clauses, phrasal coordinators, and nominalizations (the upper features on
dimension 3). Finally, ranking very low on dimension 5, these texts tend
to have very few conjuncts, agentless passives, past-participial clauses, etc.
The following is a breakdown of the texts in each cluster by genre. For
each cluster, the total number of texts in the cluster (both core texts and
peripheral texts), an interpretive label, the number of texts from each
genre occurring in the cluster, and (in parentheses) the percentage of texts
from each genre that occurs in the cluster are given. The asterisks identify
cases where a majority of the texts from a single genre occur in a
particular cluster (for example, cluster 1 contains 62% of the telephone
conversations between personal friends).
Composition of the text types by genre, giving the number of texts in the
core cluster, the number of texts peripheral to the cluster, and the
percentage of texts from each genre occurring in that cluster ('*' marks
the case where more than 50% of a genre occurs in a single cluster):
Text typel. Intimate interpersonal interaction (22 core-f 1 peripheral
texts)
12+ 1 Face-to-face conversations (29%)
8 + 0 Telephone conversations — personal friends (62%)*
1 + 0 Telephone conversations — disparates (17%)
1 + 0 Telephone conversations — business associates (13%)
Text type 2. Informational interaction (49 core+ 24 peripheral texts)
26 + 5 Face-to-face conversations (70%)*
7 + 4 Interviews (50%)*
5 + 2 Telephone conversations — business associates (88%)*
4+ 1 Telephone conversations — personal friends (39%)
3 + 1 Telephone conversations — disparates (67%)*
2 + 7 Spontaneous speeches (56%)*

2 + 1 Personal letters (50%)*

0 + 1 Nonsports broadcasts (13%)
0 + 1 Professional letters (10%)
0+ 1 General fiction (3%)
Text type 3. 'Scientific' exposition (28 core +15 peripheral texts)
23+12 Academic prose (44%)
3 + 1 Official documents (29%)
1 + 0 Biographies (7%)
1 + 0 Press reviews (6%)
0+ 1 Hobbies (7%)
0+ 1 Press reportage (2%)
Text type 4. Learned exposition (53 core+ 18 peripheral texts)
17 + 8 Academic prose (31 %)
10 + 1 Press reportage (25%)
6 + 2 Official documents (57%)*
6 + 2 Press reviews (47%)
5 + 0 Popular lore (36%)
4 + 0 Biographies (29%)
2+1 Hobbies (21%)
2+1 Religion (18%)
1 + 0 Press editorials (4%)
Text type 5. Imaginative narrative (47 core+ 13 peripheral texts)
12 + 0 Romance fiction (92%)*
12 + 3 General fiction (51 %)*
7 + 2 Mystery fiction (70%)*
7 + 2 Adventure fiction (70%)*
4 + 3 Prepared speeches (50%)*
2 + 0 Interviews (9%)
1 + 1 Science fiction (33%)
1+0 Popular lore (7%)
1+0 Biography (7%)
0+ 1 Personal letters (17%)
0+1 Religion (6%)
Text type 6. General narrative exposition (117 core + 33 peripheral texts)
30 + 2 Press reportage (73%)*
15 + 8 Press editorials (86%)*
12 + 0 General fiction (41 %)
8 + 0 Biographies (57%)*
7+1 Humor (89%)*
7+ 1 Press reviews (47%)
7 + 6 Academic prose (17%)

22 D. Biber
6 + 4 Religion (59%)*
4-h 2 Hobbies (43%)
34-2 Nonsports broadcasts (63%)*
3 + 0 Science fiction (50%)*
3 -l-1 Adventure fiction (31 %)
3 + 0 Mystery fiction (23%)
3-l· 0 Popular lore (21%)
2 + 3 Prepared speeches (35%)
2 + 0 Official documents (14%)
1 + 0 Romance fiction (8%)
0 + 2 Sports broadcasts (20%)
Text type 7. Situated reportage (7 core+ 5 peripheral texts)
7 + 1 Sports broadcasts (80%)*
0 + 1 Nonsports broadcasts (13%)
0 + 1 Science fiction (17%)
0 +1 Mystery fiction (8%)
0+1 Hobbies (7%)
Text type 8. Involved persuasion (22 core+ 27 peripheral texts)
5 + 4 Interviews (41%)
4 + 3 Spontaneous speeches (44%)
4 +1 Popular lore (36%)
2+1 Religion (18%)
2 + 0 Prepared speeches (14%)
1 +0 Telephone conversations — disparates (17%)
1+0 Humor (11%)
1 + 2 Editorial letters (11 %)
0 + 7 Academic prose (9%)
0 + 3 Hobbies (21%)
0 + 2 Personal letters (33%)
0+ 1 Nonsports broadcasts (13%)
0 + 1 General fiction (3%)
The labels in this breakdown summarize the interpretations of the

clusters as text types, based on consideration of the predominant linguis-
tic features in each cluster (summarized by the centroid dimension score
of each cluster given in Figures 3 and 4), the communicative character-
istics of the texts grouped in each cluster, and microanalyses of individual
texts. I turn now to a detailed consideration of each type.

4.2. Interpretation of the clusters as text types
4.2.1. Clusters 1 and 2: 'Intimate interpersonal interaction' and 'Informa-

tional interaction'. I noted above that clusters 1 and 2 are quite similar to
each other (see Figure 3). Both are characterized by 'situated reference'
(dimension 3), 'a nonabstract style' (dimension 5), and relatively un-
marked scores on dimensions 2 and 4. Even on dimension 1, they are both
characterized as 'involved production'; the major difference between these
two clusters is that cluster 1 has an extreme characterization on this
dimension, while cluster 2 has a more moderate characterization. The
linguistic features associated with these dimension scores are summarized
on pages 8-9: for example, very frequent private verbs, contractions etc.,
plus markedly infrequent nouns, prepositions, etc., associated with the
high score on dimension 1; frequent time and place adverbials and
infrequent WH relative clauses associated with the low score on dimen-
sion 3; and markedly infrequent conjuncts and past-participial forms
associated with the low score on dimension 5.
The texts grouped into clusters 1 and 2 in the breakdown at the end of
the last section reflect the shared linguistic characteristics of the two
clusters as well as the major difference between them on dimension 1.
Cluster 1 comprises strictly 'involved', interpersonal conversations, for
the most part face-to-face conversations and telephone conversations
between personal friends. Cluster 2, on the other hand, comprises person-
to-person interactions that have an informational concern, such as
interviews, business telephone conversations, and face-to-face conversa-
tions in professional contexts. In both of these text types, the interaction
is primary — speakers address individual listeners who are immediately
present and personal. The main difference between these types relates to
the primary purpose of interaction: to convey information in cluster 2 and
to maintain the interpersonal relationship in cluster 1. This difference can
be illustrated by comparison of text sample 1, representing cluster 1, with
text samples 2, 3, and 4, representing cluster 2.6
Cluster 1 texts:
Text sample 1 (LL:1.8.; informal face-to-face conversation between
friends)
A: there are cups # [pause] Nescafe # [pause]
B: shall we have a cup of coffee # [pause]
A: yes certainly # yes certainly # [pause] yes #
B: I see # they're all [???] [long pause]
A: some of them are rather large # [pause]
B: mm #

24 D. Biber
A: some of them are rather large # [pause]
B: want any sugar # [pause]

A: yes please Brenda # [pause]
B: one # [pause]
A: that's about right # yes that's enough thank you # [long pause] not
yours is it #
B: oh no # those are my scripts # I just saw the note # and I know
that's all right # [long pause]
A: just put my glasses on § I can't see a thing without them # [long
pause] well after all they're too dark to be inspiriting # aren't they #
[pause]
B: I don't want one # I'm afraid #
A: I think I'd rather substitute #
B: yes # I haven't space # I don't want [pause] portraits #
A: no # [pause]
Cluster 2 texts:
Text sample 2 (LL:5.5; panel discussion)
Question: Do you think that there is any chance that the Labour
Party will provide an effective opposition in the forseeable future?
A: Christopher Chataway #
C: I've seldom heard a string of sentences # that I really do believe #
to [pause] contain quite so many # [pause] faulty analyses # of the
present situation # [long pause] I don't believe # that this country is
swinging to unilateralism #
A: Lord Boothby #
B: well # [pause] I don't think you know # that Tony Wedgwood
Benn can seriously say that personalities [pause] don't matter # [long
pause] because I think they do matter tremendously # in [pause] politics
today # and especially in the politics of the Left # [long pause] what has
happened is...
Text sample 3 (LL:1.1; face-to-face conversation between academic
colleagues, concerning student comprehensive exams)
A: well # [pause] may I ask # what goes into that paper now # because
I have to advise # [pause] a couple of people who are doing the [mm]
B: well what you do # is to [long pause] this is sort of between the two
of us # what you do # is to make sure that your own [pause] candidate

[mm] # is [pause] that your [pause] there's something that your own
candidate can handle # [long pause]
A: you mean that the the the papers are more or less set ad hominem
# are they # [pause]
B: [mm] [long pause] they shouldn't be # [long pause] but [mm]
[pause] I mean one # sets [long pause] one question # now I mean this
fellow's doing the language of advertising # [pause] so very well #
A: yeah#
B: give him one on
A: is this a spare paper (change of topic)
B: yeah....
Text sample 4 (LLrll.l; spontaneous speech — specifically a court
examination of a witness)
A: Mr Potter # did you # [long pause] arrive # about two o'clock # on
the [pause] Sunday # [pause] the date the will was [pause] signed #
[pause]
B: yes # [long pause]
A: and [pause] did you [pause] go # and see your mother straight away
^B: yes I did #

A: what was she then doing # [pause]
B: she was having her lunch # [long pause]
A: what about the brandy bottle # where was that # [long pause]
B: I don't know # I didn't [pause] I didn't see #
A: you didn't see it #
B: well # [pause] no I didn't #
A: I would think I ought to tell your lordship # [pause] so that

nothing # is [pause] is to be hidden # from our side # [pause] that the
plaintiffs supplied # certain statements # to us # [long pause] several
months ago [pause] one of which # was a short statement # from the
doctor # [pause] ....
Text samples 1-4 illustrate the shared characteristics of clusters 1 and 2,
as well as the major difference between these two types. All four samples
are interactional, but text sample 1, which represents cluster 1, is nearly
exclusive in its interpersonal focus, while samples 2-4, which represent
cluster 2, all have a specific informational purpose in addition to their
interpersonal purpose. In sample 1, there is no particular topic that is the
focus of discussion; rather, the participants change topic freely and place
primary emphasis on the interaction itself instead of on the exchange of

26 D. Biber
information. In contrast, samples 2-4 are all informational to some

extent. Sample 2 is from a panel discussion, where a group of discussants
interact with one another debating a series of specific issues; this speech
event is thus explicitly informational and interactional at the same time.
Sample 3 is from a face-to-face conversation in a professional context for
professional purposes, and it thus has a markedly informational as well as
interactional focus. The text that contains sample 4 is from the genre
labeled 'spontaneous speeches' in the London-Lund Corpus, and it is
therefore surprising that this text is grouped into the 'interactional' cluster
2. As sample 4 shows, however, this text is in fact an interaction — a
courtroom examination of a witness which is both informational and
interactional. The grouping of this text into cluster 2 highlights the fact
that the types are defined on the basis of linguistic characteristics rather
than any external criteria.
As noted above, clusters 1 and 2 are very similar with respect to
dimensions 2-5, and text samples 1-4 illustrate these shared character-
istics. All four texts are marked by nonelaborated reference in that they
have very few WH relative clauses, phrasal coordinators, and nominaliza-
tions (the upper features on dimension 3). In fact, all of these texts contain
inexplicit referents that can be understood only by reference to the
immediate situation or shared background of the participants; for
example, there, they, some of them, that, those in sample 1; the present
situation, this country, today in sample 2; that paper, this fellow in sample
3; the brandy bottle, several months ago in sample 4. In addition, all four
texts are nonabstract in that they contain very few conjuncts, passives, or
other past-participial constructions (the features on dimension 5). Like
their respective clusters, text samples 1-4 are relatively unmarked with
respect to dimension 2, 'Narrative concerns' and dimension 4, 'Expression
of persuasion'. Thus, these texts show relatively few past-tense forms,
perfect-aspect verbs, third-person pronouns, etc. (dimension 2) and few
infinitives, prediction modals, suasive verbs, etc. (dimension 4).
The major linguistic difference between clusters 1 and 2 relates to their
characterization on dimension 1: cluster 1 is extremely involved, while
cluster 2 is more moderately involved. The situational characteristics of
the texts grouped into clusters 1 and 2, together with this linguistic
characterization, lead to the interpretation of cluster 1 as highly interper-
sonal, noninformational interaction, and cluster 2 as informational
interaction. The dimension 1 characteristics of text samples 1-4 support
this interpretation. Sample 1 has by far the highest dimension 1 score of
these four, reflected by frequent private verbs (such as know, think), that
deletions (/ know (that) that's all right; I think (that) I'd rather
substitute), contractions (such as they're, that's, can't), present-tense verbs

(are, see), first- and second-person pronouns, demonstrative pronouns

(such as that's about right, those are my scripts), occurrences of it as
pronoun, be as main verb, etc. At the same time, text sample 1 shows a
marked absence of nouns, prepositions, attributive adjectives, and long
words, and it has markedly low lexical variety. This sample is thus
strikingly 'involved', shown by the very high frequencies of the features
with positive weights on dimension 1, and strikingly noninformational,
shown by the very low frequencies of the features with negative weights.
Text samples 2-4 are also 'involved' and interactional, but much less so
than sample 1. Some dimension 1 features occur relatively frequently in
these texts; for example, first and second person pronouns, contractions,
and private verbs. At the same time, though, these texts show relatively
many nouns, prepositions, attributive adjectives, and long words in
comparison to sample 1, reflecting the less involved and more informa-
tional focus of these texts.
In summary, text type 1 (cluster 1) is a type of interaction that is
situated in reference, nonabstract in style, and extremely interpersonal
and involved in focus. Text type 2 (cluster 2) is also a type of interaction
that is situated and nonabstract, but the texts of this type have specific
informational as well as interpersonal purposes. The interpretation of
these two clusters illustrates the way in which text-type categories can cut
directly across genre classifications. For example, some texts classified as
'spontaneous speeches' are highly interactive and belong to text type 2;
texts classified as 'face-to-face conversation' can be either highly interper-
sonal interaction (text type 1) or relatively informational interaction (text
type 2). Text types 3-8 provide many other examples of linguistic
characterizations that cut across external genre classifications.
4.2.2. Clusters3 and4: '"Scientific"exposition'and 'Learnedexposition'.

Clusters 3 and 4 form a second pair of related text types. Figure 3 shows
that these clusters have very similar characterizations on dimensions 1-4,
differing primarily on dimension 5. Both clusters are markedly nonnarra-
tive (dimension 2) and nonpersuasive (dimension 4), and both are
extremely informational in production (dimension 1) and explicit in
reference (dimension 3). On dimension 5, both clusters are characterized
by an abstract style; they differ in that cluster 3 is extremely abstract in
style, while cluster 4 is only moderately so.
The texts grouped into clusters 3 and 4 at the end of section 4.1 show
that both of these clusters represent types of informational exposition.
Cluster 3 comprises primarily academic prose texts and a few official
documents; the academic prose texts are primarily from natural science,
engineering/technology, and medicine. Cluster 4, on the other hand,

28 D. Biber
comprises a relatively broad range of texts, including academic prose,

press reportage, official documents, press reviews, popular lore, biogra-
phies, hobbies, and religion. Academic prose texts make up approxi-
mately a third of the texts in this cluster; but unlike those in cluster 3, the
academic prose texts in cluster 4 are primarily from the humanities, social
sciences, education, and law. Both of these clusters are expository with an
extreme informational focus. The difference between them relates to the
extreme technical content and style found in cluster 3 versus the more
'learned' presentation of information in cluster 4. Text samples 5 and 6
illustrate the shared characteristics of clusters 3 and 4 as well as the
primary difference between them.
Cluster 3:
Text sample 5 (LOB:J.8; physics journal article)
Thus the first few atomic layers deposited during the gettering period
are highly oxidized, and when the chamber has been 'cleaned up' the
deposit is more metallic. After the evaporation ceases, the deposited
film remains open to oxidation. Thus the deposited film is inhomogene-
ous and approximates to a sandwich layer of oxide/metal/oxide, in
which the outer layers are more highly oxidized than the inner layer.
The exact state of oxidation of the deposited film is unknown and a
further effect of oxidation can be observed upon baking in air. ...
Cluster 4:
Text sample 6 (LOB:J.27; sociology text)
Government in Spain continues to rest on the three institutions of an
hereditary monarchy (rejected by two short-lived republics), the parlia-
ment of the old Castilian Cortes, and an extensive Civil Service, with a
permanent staff except for its highest officials. Spain is at the moment a
kingdom without a king. The Franco regime has committed itself to the
maintenance of the monarchy as an institution by the 1947 Law of
Succession and the Referendum of the following year. Meanwhile the
regime, in its own words, is a representative, organic democracy in
which the individual participates in government through the natural
representative organs of the family, the city council and the syndicate.
Both of these samples show the characteristics of extreme informa-
tional production and explicit reference. Both have a very high concentra-
tion of nouns, prepositions, attributive adjectives, long words, and a quite
varied vocabulary — the bottom group of features on dimension 1; both
have essentially none of the top group of features on dimension 1, such as

private verbs, first- and second-person pronouns, and contractions.

Although the verbs in these texts are consistently in the present tense,
there is a striking absence of verbs in general, coupled with a preference
for noun and prepositional phrases. Taken together, these features result
in an extreme 'informational' characterization on dimension 1. On
dimension 3, neither text makes direct reference to items in the external
situation, and in both cases the full texts make frequent use of WH
relative clauses to elaborate intended references (although these samples
illustrate only one relative construction each). Both samples are written
entirely in the present tense, resulting in their marked nonnarrative focus
on dimension 2. Neither sample makes use of the persuasive features
associated with dimension 4, resulting in their low score on that dimen-
sion.
The striking difference between these two samples relates to their
dimension 5 characterization. Sample 5 is extremely passive in form, with
the agent of predicates being deleted in every case. These forms include
both main verbs (for example, are highly oxidized, has been 'cleaned up\
can be observed) and postnominal modifying clauses (such as layers
deposited during the gettering). This sample also illustrates the frequent
use of conjuncts to mark the logical relations in a text (in this case a
repeated use of thus). Sample 6, on the other hand, is consistently in the
active voice; the only passive form is in the postnominal clause monarchy
rejected by two short-lived republics, and even this case is unlike sample 5
in that the agent is specified. This sample also illustrates a lesser use of
conjuncts, counting on the reader to infer the logical relations among
propositions.
In summary, both text type 3 and text type 4 are expository, extremely
informational, and explicit in reference. The difference between them
relates to their use of an abstract, technical style. This seems to be both a
content and a stylistic distinction. On the one hand, the texts in type 3
focus on highly abstract and technical information; they are therefore
much more concerned with the entities being acted on (the patients) than
with any active agents. They further depend on a frequent use of
conjuncts to specify the logical relations among propositions. Type 4 texts
tend to be less technical in content. In addition, the differences between
these types seem to reflect attitudinal preferences. Thus, the texts in type
3, coming for the most part from engineering and natural sciences, show a
stylistic preference for a presentation of information apart from active
agents, possibly to give the appearance of scientific rigor. In contrast, the
texts in type 4, coming from a broad range of 'literate' prose, show a
preference for a more active style, perhaps reflecting the influence of
prescriptive notions of 'good' style. As was the case with text types 1 and

30 D. Biber
2, the classification of texts into types 3 and 4 cuts across genre categories.
For example, several social science and humanities academic texts are
grouped into type 3 because they are relatively technical in content and
adopt the abstract and technical style ofthat type; conversely, a few natural
science and engineering academic texts are grouped into type 4, adopting
an active, nonabstract style in contrast to the norms for their subgenres.
4.2.3. Cluster 5: 'Imaginative narrative'. Figure 4 shows that cluster 5

is situated in reference (dimension 3), nonabstract in style (dimension 5),
and slightly involved (dimension 1), but the primary distinguishing
characteristic of this cluster is its extreme narrative emphasis, shown by its
high score on. dimension 2. Not surprisingly, the texts grouped into this
cluster are mostly fiction, or 'imaginative narrative'. Text sample 7
illustrates the involved type of narrative common in fiction, while sample
8 illustrates a nonfictional type of involved narrative from a judge's final
statement in a court case.
Cluster 5:
Text sample 7 (LOB:L.12; Mystery fiction)
I'd finished making the bed by then. As I pushed it back against the
wall I heard something drop on the floor.
That was when the percolator in the living-room started making
bubbling noises. There was nothing on the floor that I could see. I told
myself it must've fallen down between the bed and the wall.
... Wasn't urgent anyway. Maybe my cigarette-case ... or Sonia's
powder compact ... I'd look for it later.
So I got up from my hands and knees, went into the living room and
fixed myself a cup of coffee.
Text sample 8 (LL:12.4b; prepared speech — court case)
A:
I have to decide in this case # [pause] what # [pause] if any maintenance
# [pause] should be paid # [pause] by the husband as I shall call him #
[pause] to the wife # [long pause] he's in fact § no longer the husband #
[long pause] he was originally petitioner # [pause] because there's been a
decree # [pause] absolute # [long pause] and he has remarried # [pause]
the decree # [long pause] was pronounced in favour # of the respondent
wife ft [pause] on the grounds of the husband's admitted adultery #
[pause] his charge # of adultery # [pause] against her # with the main
correspondent # [long pause] failed # after a [pause] somewhat lengthy
[pause] hearing # [pause] her charges of cruelty # against him # [pause]
likewise failed # [long pause]

Sample 7 is typical of the majority of texts grouped into cluster 5. Most

of this text is simple narration in the past, which uses frequent past-tense
forms, third-person personal pronouns, and perfect-aspect verbs, result-
ing in a high score on dimension 2. The sample is also relatively 'involved'
even though there is no direct interaction of participants; for example, this
sample shows a very frequent use of first-person pronouns and contrac-
tions. A sample that included direct dialogue between participants would
have even more dimension 1 'involved' features than sample 7. Most
cluster 5 texts are similar to sample 7, being fictional with an extreme
narrative focus and a moderately involved characterization.
Text sample 8 illustrates how nonfictional texts can have a similar
mixing of narrative focus and involved presentation. This sample is from
a final summation and judgment in a court case. As background to the
final judgment, the judge summarizes the events that are relevant to the
case. As such, this speech event is largely narrative, and sample 8 contains
many of the features characteristic of a high dimension 2 score, such as
frequent past-tense verbs, perfect-aspect verbs, and third-person pro-
nouns. In addition, this text is moderately 'involved', with the speaker
using first-person pronouns, contractions, private verbs, etc. There are
only a few nonfiction texts grouped in cluster 5, but sample 8 illustrates
how these other texts can also have a primary narrative focus combined
with an involved presentation.
4.2.4. Cluster 6: 'General narrative exposition'. Cluster 6 is the largest

cluster, with 117 core texts and another 33 peripheral texts. Figure 4
shows that this cluster combines expository/informational and narrative
features, making it similar to clusters 3 and 4 in some respects and to
cluster 5 in other respects. With respect to dimension 1, cluster 6 is similar
to clusters 3 and 4 in being markedly informational and noninvolved;
with respect to dimension 2, cluster 6 is similar to cluster 5 in that it has a
moderately high narrative concern. On dimensions 3 and 5, however,
cluster 6 is not similar to any of these other three clusters; it is unmarked
on these dimensions, rather than 'elaborated' and 'abstract' like clusters 3
and 4, or 'situated' and 'nonabstract' like cluster 5. The distinctive
characteristics of cluster 6 are thus a marked informational focus
(dimension 1) and a moderate narrative concern (dimension 2).
The texts grouped into this cluster likewise combine these two concerns.
They are primarily informational and expository but often use narration
to convey information. Unlike cluster 5, the narrative portions in these
texts are not imaginary or for entertainment; they are rather an integral
part of the expository information being conveyed.
Cluster 6 is the most general of the text types identified by the present

32 D. Biber
study. As noted above, there is a total of 150 texts grouped into this
cluster; these texts represent 19 different genres, including press reportage,
press editorials, general fiction, biographies, humor, press reviews, aca-
demic prose, and religion. This is thus a very general type of exposition; it
is not markedly learned or technical, not markedly elaborated in reference
or abstract in style, and it often uses narration as part of its exposition.
Text samples 9, 10, and 11 illustrate the distinctive characteristics of
this cluster. Sample 9 is from an editorial and illustrates the use of
narrative forms to convey expository information. Sample 10 is from
press reportage, in which the information being conveyed comprises a
narration of past events. Finally, sample 11 is from a humor text and is
representative, of the fictional and biographical types of writing that use
the features of this cluster for entertainment purposes.
Cluster 6:
Text sample 9 (LOB:B.20; editorial letter)
Communism had little or nothing to do with the riots in South Africa
or the more recent disorders in Rhodesia. In fact, former leaders of the
Communist Party in the Union have left the country. Some are now in
the Rhodesian copper belt and at least one of them is in London.
In contrast, Moscow has embarked upon a special operation in
Ruanda-Urundi, which borders on the Belgian Congo. This state of
some 21,000 square miles and a population of 4,630,000 has been a
United Nations trust territory under the administration of Belgium, but
a few days ago she announced that she was giving up the trusteeship.
Text sample 10 (LOB:A.24; press reportage)
Four hundred angry Soccer fans chanted 'Sack the manager' outside
Newcastle United Football Club's ground yesterday.
United had just been thrashed 4-0 by Everton, and now look certain
to be relegated to the Football League's Division Two. Newcastle's
manager is ex-winger Charlie Mitten.
At half-time, with United two goals down, one disgusted fan climbed
the club's flagpole and hauled the Union Jack to half mast.
It was a riotous day for soccer....
Text sample 11 (LOB:R.2; humor)
He had long sensed injustice in the distinctions drawn between ordinary
wage-earners and those self-employed. By the time his monthly salary
arrived, the Inland Revenue had already taken their share, and there
were precious few reductions in tax except for wives, children, life-
insurances or any of the other normal encumbrances which Cecil had

so far avoided. He read the film star's sorry story and frowned at the
provisions of Schedule D taxation which not only allowed her to claim
relief on the most unlikely purchases, but also postponed demanding
the tax until her financial year was ended, audited and agreed by the
Inspector.
All three of these text samples illustrate the informational features
associated with dimension 1: frequent occurrences of the bottom features
(such as nouns, prepositional phrases, attributive adjectives) plus mark-
edly infrequent use of the upper features (such as private verbs, contrac-
tions). This is true of the editorial (sample 9), which is primarily
informative and expository in purpose, as well as the humor text (sample
11), which is primarily entertaining and narrative in purpose. Further,
despite the different purposes of these texts, they all use narrative forms
associated with dimension 2 (such as past-tense forms, perfect-aspect
verbs, third-person pronouns). This tendency is most pronounced in the
humor text sample, but it is found in all three samples. On the other three
dimensions, these samples illustrate the unmarked characterization of
cluster 6: not markedly 'elaborated' or 'situated' in reference, and not
marked with respect to persuasion or abstract style.
The text type represented by this cluster has a special place in the
present typology: it is the most general and nondistinct of the eight types.
Although the texts in this type share a general linguistic characterization,
having a carefully crafted, informational presentation and making rela-
tively frequent use of narrative forms, the linguistic characterization of
this type tends to be relatively unmarked on all five dimensions. As text
samples 9-11 show, the texts in this type can have different purposes,
although the underlying logical development used to achieve those
purposes seems relatively similar. That is, all of these samples use a
narrative line and careful informational elaboration to achieve their end.
In the case of editorials, that end is analysis of some political or social
situation; in the case of press reportage, that end is informing through a
factual report of events; in the case of humor (as well as fiction and
biography), that end is entertainment through a report of events. These
texts belong to the same type in their surface characterizations and, to a
lesser extent, in their underlying organizations; they show considerable
variation, however, with respect to their specific purposes. I will return to
the discussion of text type 6 relative to the other types in section 5.
4.2.5. Cluster 7: 'Situated reportage'. Cluster 7 is the smallest and most

distinct text type identified in the study. Figure 4 shows that it is not
marked with respect to dimension 1, but it is markedly nonnarrative,

34 D. Biber
nonpersuasive, and nonabstract (with respect to dimensions 2, 4, and 5).

The most distinctive characteristic of this cluster is on dimension 3, where
it is characterized as extremely situated in reference. The core texts
grouped into this cluster are all sports broadcasts. This text type thus
characterizes the on-line reportage of events which are in progress and
occur in a fairly rapid succession. Text sample 12, taken from a broadcast
of a soccer game, illustrates the distinctive characteristics of this text type.
Cluster 7:
Text sample 12 (LL:10.2; sports broadcast — soccer)
A:
Dunn ft down the line ft a bad one ft it's Badger that gets it ft he's got
time to control it ft [pause] he feeds in fact ft Tom Curry ft one of the
midfield players ahead of him ft [pause] Curry has got the ball ft on that
far side ft chips the ball down the centre ft [pause] again ft a harmless one
ft [pause] no danger ft out comes Stepney ft [pause] and now left-footed ft
his clearance ft [pause] is again ft a long [pause] high ft probing ball ft
down centrefield ft onto the head of [long pause] Flynn ft Flynn to
Badger ft Badger on the far side ft
Sample 12 illustrates the distinctive characteristics of cluster 7: neither
involved nor informational (that is, relatively few occurences of either
upper or bottom features from dimension 1), markedly nonnarrative
(no past-tense verbs, perfect-aspect verbs, or other features from dimen-
sion 2), nonpersuasive (none of the features associated with dimension 4),
and markedly nonabstract in style (none of the passive constructions
associated with dimension 5). Although these characterizations all repre-
sent features that are markedly infrequent, they reflect the very specialized
purpose and production situation of these texts: a speech event that
describes events actually in progress to a large audience that is not
present. For example, the distant relationship between broadcaster and
audience results in the lack of involvement features; the rapid on-line
production of text results in relatively few informational features; and the
reportage of events in progress results in the nonnarrative characteriza-
tion of these texts. The most distinctive positive characterization of type 7
texts is the extremely high use of expressions referring directly to the
physical and temporal situation of communication. Thus, sample 12
contains numerous expressions such as down the line, ahead of him, on that
far side, down the centre, which require direct reference to the playing field
for understanding. The very frequent use of these expressions results in
the extremely situated characterization of type 7 on dimension 3.
We might wonder why this text type is much more 'situated' than type

A iypology of English texts 35
1, 'Intimate interpersonal interaction'. Both text types are noninforma-

tional, and in both types the participants share the same temporal
context. In fact, in some respects we might expect type 1 to be more
situated in reference than type 7 texts: in type 1 texts, participants actually
share the same physical situation, and addressees can request clarification
in cases of misunderstanding; in typical type 7 texts, the speaker (a radio
broadcaster) does not actually share a physical situation with the
listeners, and there is no possibility of clarification. There seem to be two
reasons for the observed characterization of type 7 as extremely situated
relative to type 1. First, the interactive texts in cluster 1 do not involve the
same informational demands and are not produced under the same time
constraints as those in cluster 7. That is, on-line reportage of sports events
involves a rapid production of speech describing all relevant events as
they occur. In such a situation, there is great demand for situated
reference, because there are very many different referents to keep track of
but very little opportunity for elaborated referring expressions. Interper-
sonal interaction (type 1), on the other hand, involves considerably fewer
referents and provides considerably more opportunity for elaborated
reference. Second, the expected style of sports broadcasts is one that gives
the impression of an extremely rapid and exciting succession of events,
even if this is not actually the case. Thus, even in the reportage of a
baseball game, where events occur much more slowly than in a soccer
match, many of the same features seen in sample 12 are frequently used
(see Ferguson 1983). Due to the distinctive linguistic characteristics
associated with these unusual production demands, sports broadcasts are
isolated as a separate text type of 'situated reportage' by the present
analysis.
»
4.2.6. Cluster 8: 'Involved persuasion'. Cluster 8 also represents a
relatively specialized text type. In their linguistic characterization, the
texts in this cluster are moderately involved, nonnarrative, elaborated in
reference, and nonabstract in style (Figure 4). The most distinctive
characteristic of cluster 8 occurs on dimension 4, where these texts are
markedly persuasive in form. The texts grouped into this cluster (see the
breakdown at the end of section 4.1) are also characteristically argumen-
tative or persuasive in their primary purpose. The majority of these texts
are spoken: some are interactional and informational, such as the
interviews and the telephone conversation between disparates; others are
informational monologues, such as the spontaneous and prepared
speeches. The remaining texts are written, informational texts, such as
popular lore, professional letters, religion, humor, and editorials. Overall,
the linguistic characterization of these texts is primarily persuasive and

36 D. Biber
secondarily involved, while the texts themselves are primarily argumenta-

tive or persuasive in purpose, leading to the interpretive label 'Involved
persuasion'.
Text samples 13-15 illustrate several of the different ways that texts
can be persuasive as well as involved. Text sample 13 is from a session of
parliament, in which a number of MPs interact with the Secretary of
State. This text contains a number of short interactive monologues,
with each speaker attempting to persuade the others. Text sample 14
is from a sermon. This is strictly a monologue, with the preacher
attempting to 'exhort' and persuade the audience. Sample 15 is from a
professional letter. This text is moderately interactive in that it is a
response to a previous letter from a specific individual. It is also
persuasive, in that it responds to specific questions and proposes a
specific course of action.
Cluster 8:
Text sample 13 (LL:11.4; spontaneous speech — MPs in Parliament —
interacting with each other and the Secretary of State)
Q: would he not agree that it is essential at the moment ft that more
[pause] should be free for exports and less absorbed within our public
sector ft [long pause]
A: well ft I think I would accept on the latter point ft that more of our
resources must go ft into [mm] into the balance of payments ft ....
Q: would he agree that [mm] ft [pause] an absence of such a statement
ft [pause] continues to generate uncertainty in the industry ft and
perhaps he might like to take this opportunity to [mm] ft re-emphasize
his support ft for the second force airline ft [long pause]
A: well I would certainly ft [pause] regret it if ft [pause] parts or ft
or indeed the whole of the [mm] review ft [pause] was to dribble out ft
that's not my intention at all ft [pause] we shall of course ft [pause]
indeed we are ft [pause] studying it [pause] very carefully ft [pause] ....
Text sample 14 (LL:12.1c; prepared speech — sermon)
A:
we must ft [long pause] have our corporate life together ft as a church ft
[long pause] .... we can fight ft [pause] and we must fight ft [pause]
against the world ft the flesh ft and the Devil ft [pause] as individuals ft
[pause] but we must also fight ft [pause] as the whole church of God ft
[long pause].... we must have God's guidance ft and grace ft [pause]....
we must go out realizing ft [pause] that without God's grace ft [pause]
we are utterly powerless ft [long pause]

Text sample 15 (Z.I; professional letter)

Furthermore it really would be inappropriate for me to put words in
your mouth. In short, you should really take the format of the
resolution and put in your own thoughts .... the association is already
sampling opinion on a number of other matters and it may be possible
to add this one. If it is not possible to add your concern this year, it
would certainly be possible to add it next year.
In all three of these samples, there is a high frequency of the 'persua-
sive' features associated with dimension 4. In sample 13, there are
prediction modals (would, shall), necessity modals (should, must), possibil-
ity modals (might), suasive verbs (agree), and conditional subordination.
Sample 14 shows an extreme use of necessity modals (must) as well as the
use of possibility modals (can). Sample 15 also shows frequent use of
modals (should, may, would) as well as conditional subordination. In
addition, all three samples are relatively involved, as shown by features
such as first- and second-person pronouns, emphatics and hedges, private
verbs, etc. These samples thus illustrate a relatively specialized text type
having a primary persuasive or argumentative purpose. It is interesting
that the speakers and writers of these texts are fairly uniform in adopting
an involved presentation rather than a strictly informational presentation.
That is, these texts use both overt persuasive markers (dimension 4) and
identification with the listener/reader and an informal, colloquial style
(dimension 1) to make their point.
In contrast, some of the peripheral texts grouped in this cluster are
strictly informational while being overtly argumentative and persuasive.
Text sample 16, from a philosophy journal, illustrates this type of text.
Text sample 16 (LOB:J.54; academic prose — philosophy)
... .the impression must be describable without reference to any event or
object distinct from it. It must be possible to characterize that internal
impression without invoking any reference to the so-called object of
desire.... The supposition, then, that desiring or wanting is a Humean
cause, some sort of internal tension or uneasiness, involves the follow-
ing contradiction: As Humean cause or internal impression, it must be
describable without reference to anything else ...; but as desire this is
impossible. Any description of the desire involves a logically necessary
connection with the thing desired. No internal impression could
possibly have this logical property. Hence, a desire cannot possibly be
an internal impression.
This text is directly persuasive. It overtly considers arguments and
counterarguments and forcefully argues in favor of a point of view. In the

38 D. Biber
above sample, this is shown principally by an extremely frequent use of

modals (must, could, can). At the same time, this text has a very
informational characterization with respect to dimension 1 (frequent
nouns, prepositional phrases, etc.). Texts like sample 16 are grouped into
cluster 8 because they are overtly persuasive, but they are peripheral to
this text type because they do not combine the typical use of persuasive
and involved features.
In summary, then, the texts in text type 8 are primarily distinguished by
their persuasive and argumentative emphases. This orientation is typically
combined with an involved, often interactive, style, which aids the
persuasive force of the text by developing a sense of solidarity with the
listener or reader. In other cases, though, these texts can be overtly per-
suasive while having a marked informational focus, as in text sample 16.
5. Discussion and conclusion
The typology developed here is relatively complex, and sometimes the

resulting 'text types' are surprising. For example, there is no single
interactive or dialogue text type. Rather, the analysis identifies two major
interactive types: Intimate interpersonal interaction (type 1), concerned
primarily with the immediate interpersonal interaction, and Informa-
tional interaction (type 2), which has a primary informational emphasis.
Similarly, there is no single expository text type. Instead the analysis
identifies three expository types: Scientific exposition (type 3), which is
extremely informational, elaborated in reference, and technical and
abstract in style and content; Learned exposition (type 4), which is similar
to Scientific exposition except that is is markedly less abstract and less
technical in style; and General narrative exposition (type 6), which is a
very general text type that combines narrative forms with expository,
informational elaboration. In the same way, there is no single narrative
text type. Instead, the analysis identifies General narrative exposition
(type 6) and Imaginative narrative (type 5), which is a relatively involved
text type having a primary narrative focus. The remaining two text types
are relatively distinct. Type 7 is labeled Situated reportage, a text type
reporting events actually in progress. Type 8 is labeled Involved persua-
sion; these are texts with a primary argumentative and persuasive purpose
and style, which are also typically involved in presentation.
This typology replicates the major text distinctions identified in Biber
and Finegan (1986). That study, which was based on an earlier three-
dimensional model of textual variation (Biber 1986), is primarily meth-
odological, exploring the feasibility of the approach adopted here for

identifying underlying text types. Although this earlier study considered

a narrower range of texts and variation along only three dimensions, it
shows a very high degree of convergence with the salient functional
distinctions made in the present analysis.7 In particular, both studies
identify text types having the following functions: (1) intimate or
immediate interaction, (2) informational interaction, (3) imaginative
narrative, (4) and (5) two varieties of formal exposition ('scientific' and
'learned' in the present typology; with and without narrative in the 1986
study), (6) general, informal exposition, and (7) situated, immediate
reportage. Type 8 of the present study ('Involved persuasion') was not
identified in the earlier study, and a type labeled 'Interactional narrative'
in the 1986 study was not replicated here. These latter two types require
further study.
In the introduction, I noted that a typology of texts is needed as a
theoretical basis for discourse and register studies. The typology devel-
oped here is immediately useful in this regard. For instance, numerous
studies have described discourse characteristics of 'narrative' or 'exposi-
tion' — but the present typology shows that there is no single narrative or
expository type. Rather, the typology identifies expository characteristics
in three different text types and narrative characteristics in two text types.
(In addition, Involved persuasion constitutes a fourth text type that has
some expository features.) These text types have different linguistic and
communicative characteristics, and each deserves study on its own terms.
Texts chosen from one or another of these types will not adequately
represent 'narrative' or 'exposition' as wholes; such a study would require
analysis of texts chosen from all of the relevant types.
The typology also shows that the relationship between genres and text
types is not straightforward, but it is not intended to invalidate the genre
distinctions. I regard genre and text type categorizations as having
different theoretical bases, so that they are both valid but distinct text
constructs. As noted in the introduction, genres correspond directly to the
text distinctions recognized by mature adult speakers, reflecting differ-
ences in external format and situations of use. The present study does not
attempt to identify all of the basic genre distinctions in English; it only
shows that the theoretical bases of genres are independent from those for
text types. Genres are defined and distinguished on the basis of systematic
nonlinguistic criteria, and they are valid in those terms. Text types, on the
other hand, are defined on the basis of strictly linguistic criteria (similari-
ties in the use of cooccurring linguistic features). As noted, text types
often cut across genre categorizations. For example, large numbers of
face-to-face conversations are grouped into both text type 1 and text type
2; academic prose texts are split among four text types: 3, 4, 6, and 8.

40 D. Biber
Although the texts in these genres are similar in their nonlinguistic

characteristics, they belong to different 'types' in terms of their linguistic
characterizations. The two perspectives are thus complementary.
By focusing on linguistic cooccurrence patterns, rather than on the
distribution of individual features, the present study identifies some
surprisingly subtle functional distinctions among texts. For instance,
relative differences in the use of the cooccurring features that define
dimension 1 correspond to the subtle functional differences between type
1 and type 2: both types are interactional, but 1 is strictly interpersonal
while 2 is more informational. Similarly, differences in the use of the
defining features of dimension 5 reflect the specialized functional distinc-
tion between type 3 and type 4: both are informational and expository,
but 3 is abstract in style and represents technical specializations while 4 is
nonabstract in style and represents specializations in humanities and the
social sciences. Further, these surface cooccurrence patterns seem to
reflect differences in rhetorical organization as well as communicative
function. For example, the texts in types 3 and 4 all tend to have
expository developments, while the texts in types 5 and 6 tend to have
narrative developments, even though type 6 tends to have an informa-
tional or expository purpose. More research is required on the extent to
which underlying rhetorical structure correlates with these functional
types, but a quick survey indicates a surprisingly close relationship.
There are, however, sometimes diverse purposes among the texts in a
type. This is especially the case for type 6, which is labeled General
narrative exposition. I noted above that this is the most general and
nondistinct of the text types. The texts in this type share the use of
narrative forms and informational presentation for general expository
purposes. There is some variation, however, among the specific purposes
of these texts. Returning to text samples 9-11, we can see an analytical
purpose in sample 9 (editorial), a simple factual reporting purpose in
sample 10 (press reportage), and an entertainment purpose in sample 11
(humor). Type 5 (Imaginative narrative) is similar in this respect. The
texts in this type have a primary narrative purpose and use narrative
forms frequently, and most of these texts have fictional entertainment as
their specific purpose. A few of the texts in type 5, though, are intended to
inform rather than entertain (for example, the court case summary
illustrated in sample 8 as well as two sermons). The other text types seem
relatively consistent even in terms of their specific purposes, but types 5
and 6 show that the functional unity of the types is strongest at the level of
general communicative functions (such as narration, exposition, inter-
action) rather than specialized intent (such as informing versus entertain-
ing).

Finally, it must be emphasized that the text types identified here are in
fact 'prototypes'. That is, these types represent the 'typical' text forms and
functions of English rather than absolute distinctions among texts. The
linguistic variation among texts was studied here in terms of a continuous
five-dimensional space, where the types are dense concentrations of texts
within that space. The types are based primarily on the areas of markedly
high density, the 'core' texts, and secondarily on groupings of 'peripheral'
texts. Because the peripheral texts do not occur in dense concentrations,
they are assigned to the closest type; they are sometimes relatively
dissimilar to that type, although they are even less similar to any other
type.
Even if we limit the discussion to the core text types, the analysis here
shows that the differences among types must be considered in relative
terms. I noted in the discussion of Table 2 in section 4.1 that the types are
not equally distinct. Some text types, like types 1, 2, and 7, are quite
distinct from the other types; others, like types 3 and 4, are relatively
similar to each other. In fact, there is a continuous range of variation
among texts. It is theoretically possible for a text to have any score on
each dimension, defining a continuous, multidimensional space of varia-
tion. It turns out, though, that there are regions that have very high
concentrations of texts within that space, and these regions are identified
as the text prototypes in English. In between these prototypes, there are
particular texts that combine functional emphases and linguistic forms in
complex and relatively idiosyncratic ways. These texts are not aberra-
tions; they rather reflect the fact that speakers and writers exploit the
linguistic resources of English in a continuous manner.
There are thus two complementary perspectives on linguistic variation
among texts. One perspective focuses on the continuous nature of text
variation; the other perspective, which forms the basis of the present
study, identifies the relatively few distinct types that are frequently used in
English. In theory, texts could be evenly distributed across possible
linguistic and functional characterizations. This is not the case, however.
Rather, the majority of texts are distributed across a few sets of linguistic
form/function classes, and these marked concentrations of texts are
interpreted as the major text 'types' of English. These types reflect marked
tendencies of speakers and writers to construct texts around a limited set
of functions and cooccurring linguistic forms. The typology thus gives
structure to the multidimensional space of textual variation, even though
it does not negate the continuous nature of that space.
Additional research on the dimensions of variation in English might
help identify other, more specialized text types. The typology developed
here, however, presents eight basic prototypes of texts in English. As such,

42 D. Biber
the typology provides an important step toward modeling the ways in

which texts can differ from one another, providing the theoretical basis
for discourse comparisons in English and a foundation for cross-linguistic
research to identify universal dimensions of variation among texts.
Received 2 December 1987 University of Southern California

Revised version received
29 March 1988
Notes
* I would like to thank Pat Clancy, Ed Finegan, and an anonymous Linguistics reviewer
for their many helpful comments on an earlier draft of this paper. Correspondence
address: Department of Linguistics, University of Southern California, University Park,
Los Angeles, CA 90089-1693, USA.
1. For example, past tense has a mean value of 40.1 and a standard deviation of 30.4 across
all of the texts, and thus an absolute frequency of 113 translates into a standardized
score of 2.4:
(113-40.1) / 30.4 * 2.4
That is, a frequency of 113 is 2.4 standard deviations from the mean of 40.1.
2. The same text corpus was used to determine the dimensions of variation and to develop
the present typology. The written texts are taken from the Lancaster-Olso-Bergen
Corpus of British English (known as the LOB Corpus); the spoken texts are taken from
the London-Lund Corpus of Spoken English. These two corpora are supplemented by
private collections of personal and professional letters.
3. The FASTCLUS procedure from S AS was used for the clustering. Disjoint clusters were
produced since there was no theoretical reason to expect a hierarchical structure. Peaks
in the cubic clustering criterion and the pseudo F statistic, both produced by the
FASTCLUS procedure, were used to determine the number of clusters to extract for
analysis. These statistics provide a measure of the similarities among texts within each
cluster in relation to the differences between the clusters. In the present case, both
measures showed a peak for the eight-cluster solution, indicating that this solution
provided the best fit to the data.
4. Core texts are those that have a distance of ten or less from their cluster centroid;
peripheral texts have distances greater than ten. This distance was chosen because it
excluded the major outliers in each cluster.
5. Clusters can have intermediate mean dimension scores for two reasons: the cluster is
characterized by frequent occurrences of both positive and negative linguistic features
on that dimension, or the cluster is characterized by the marked absence of both positive
and negative features. Either distribution of features results in an unmarked character-
ization with respect to the dimension in question.
6. Text samples are labeled as follows:
CORPUS:GENRE.TEXT-NUMBER
For example, text sample 1 is labeled LL:1.8, because it is from the London-Lund
Corpus, genre type 1 (face-to-face conversation), and text 8 number within that genre.
In the spoken-text samples, # marks intonation unit boundaries.

There is considerable overlap in the texts used in these two studies (approximately
60-70%), which biases the results in favor of converging typologies.
References
Besnier, Niko (1986). Register as a sociolinguistic unit: defining formality. In Social and
Cognitive Perspectives on Language, Jeff Connor-Linton, Christopher Hall, and Mary
McGinnis (eds.), 25-63. Los Angeles: University of Southern California.
Biber, Douglas (1986). Spoken and written textual dimensions in English: resolving the
contradictory findings, language 62, 384-414.
—(1987). A textual comparison of British and American writing. American Speech 62,
99-119.
—(1988). Variation across Speech and Writing. Cambridge: Cambridge University Press.
—, and Finegan, Edward (1986). An initial typology of English text types. In Corpus
Linguistics 77, Jan Aarts and Willem Meijs (eds.), 19-46. Amsterdam: Rodopi.
—, and Finegan, Edward (1988). The drift of English genres from the 18th to the 20th
centuries. Paper presented at the Georgetown University Round Table on Languages and
Linguistics, Georgetown. (To appear in conference proceedings, edited by Thomas J.
Walsh.)
Brown, Penelope, and Fräser, Colin (1979). Speech as a marker of situation. In Social
Markers in Speech, Klaus R. Scherer and Howard Giles (eds.), 33-62. Cambridge:
Cambridge University Press.
Chafe, Wallace L. (1982). Integration and involvement in speaking, writing, and oral
literature. In Spoken and Written Language: Exploring Orality and Literacy, Deborah
Tannen (ed.), 35-54. Norwood, N.J.: Ablex.
Connor-Linton, Jeff, Hall, Christopher, and McGinnis, Mary (eds.), (1986). Social and
Cognitive Perspectives on Language. Southern California Occasional Papers in Linguistics
11. Los Angeles: University of Southern California.
Ervin-Tripp, S. M. (1972). On sociolinguistic rules: alternation and co-occurence. In
Directions in Sociolinguistics, John J. Gumperz and D. Hymes (eds.), 213-250. New York:
Holt, Rinehart, and Winston.
Ferguson, Charles A. (1983). Sports announcer talk: syntactic aspects of register variation.
Language in Society 12, 153-172.
Finegan, Edward, and Biber, Douglas (1986). Two dimensions of linguistic complexity in
English. In Social and Cognitive Perspectives on Language, Jeff Connor-Linton, Christo-
pher Hall, and Mary McGinnis (eds.), 1-24. Los Angeles: University of Southern
California.
Grabe, William (1984). Towards defining expository prose within a theory of text
construction. Unpublished Ph.D. dissertation, University of Southern California.
Hymes, Dell (1972). Foundations of Sociolinguislics: An Ethnographic Approach. Philadel-
phia: University of Pennsylvania Press.
Longacre, Robert (1976). An Anatomy of Speech Notions. Lisse: de Ridder.
Redeker, Gisela (1984). On differences between spoken and written language. Discourse
Processes 7, 43-55.
Smith, Edward L. (1985). Text type and discourse framework. Text 5, 229-247.
Tannen, Deborah (1982). Oral and literate strategies in spoken and written narratives.
Language 58, 1-21.


A Typology of English Texts : Douglas Biber

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

A Typology of English Texts : Douglas Biber

Uploaded by

Copyright:

Available Formats

A typology of English texts*

This paper develops a typology of texts in English with respect to a five-

Linguistics 27 (1989), 3-43 0024-3949/89/0027-0003 $2.00

projected research, a perspective is needed on the kinds of language and their

It is easy to illustrate the need for a typology of texts; many discourse

Brought to you by | Takming University of

All of these typologies have been proposed on functional rather than

Brought to you by | Takming University of

English include novels, newspaper articles, editorials, academic articles,

2. Background: five dimensions of variation

The notion of linguistic cooccurrence is central to linguistic analyses of

Brought to you by | Takming University of

identified quantitatively (by a statistical procedure known as factor

Brought to you by | Takming University of

Summary of the cooccurrence patterns underlying the dimensions:

Dimension 1. 'Involved versus Dimension 2. 'Narrative versu

Brought to you by | Takming University of

Dimension 4. Overt expression of Dimension 5. 'Abstract versus

Most of the dimensions consist of two groupings of features, which

Brought to you by | Takming University of

past time, third-person animate referents, and reported speech (public

Brought to you by | Takming University of

3. Multidimensional characterizations of texts

Brought to you by | Takming University of

-15 + Academic prose; Press reportage

Brought to you by | Takming University of

4. A typology of texts in English

4.1. Identification of formally distinct text types

Groupings of texts that are similar in their linguistic form can be

Brought to you by | Takming University of

Table 1. Distribution of texts across 23 genres

Written — genres 1-15 from the LOB corpus

Brought to you by | Takming University of

Brought to you by | Takming University of

Table 2. General summary of the clusters

Cluster number Frequency of Frequency of Nearest cluster Centroid

Brought to you by | Takming University of

Brought to you by | Takming University of

Involved Narrative Elaborated Overt Abstract

Cluster 5 is extremely narrative, moderately involved, situated, and

Brought to you by | Takming University of

Involved Narrative Elaborated Overt Abstract

extremely situated in its reference, in addition to being markedly nonnar-

Brought to you by | Takming University of

As discussed above, the dimensions represent frequency scales for sets

Brought to you by | Takming University of

2 + 1 Personal letters (50%)*

Brought to you by | Takming University of

The labels in this breakdown summarize the interpretations of the

Brought to you by | Takming University of

4.2. Interpretation of the clusters as text types

4.2.1. Clusters 1 and 2: 'Intimate interpersonal interaction' and 'Informa-

Brought to you by | Takming University of

A: some of them are rather large # [pause]

B: want any sugar # [pause]

Brought to you by | Takming University of

^B: yes I did #

A: I would think I ought to tell your lordship # [pause] so that

Brought to you by | Takming University of

information. In contrast, samples 2-4 are all informational to some

Brought to you by | Takming University of

(are, see), first- and second-person pronouns, demonstrative pronouns

4.2.2. Clusters3 and4: '"Scientific"exposition'and 'Learnedexposition'.