Typology
and Discourse Analysis
By John Myhill
0 Introduction
The relationship
between the typology and the analysis of discourse has been characterized by,
on the one hand, the compatibility of a common ideology and, on the other hand,
the practical difficulties inherent in combining the interests of both
sub-disciplines. Compatibility common ideology is the result of sub-disciplines
various views that the study of language should be based on empirical data
analysis rather than a thought experiment. In syntax, semantics, and even
pragmatics, intuitions have played a central role in gathering data, and
interaction with discourse analysis has tended to be controversial from the
outset because of the emphasis in discourse analysis upon the empirical
analysis of linguistic data. A discourse analyst interested in the subdiscipline’s
relationship with, for example, syntax must immediately discuss performance data
and phenomena that mainstream syntacticians simply reject as irrelevant; thus
any syntax combined with discourse analysis can only be nonmainstream syntax.
No such ideological problem arises in the case of typology: intuitions play essentially
no role in the data analyzed by typologists, and typologists are only too happy,
in principle, to consider the possible relevance of discourse phenomena to the problems
they investigate. There has, therefore, been relatively little research which
can be said to have been the product of the interaction between these subdisciplines.
Before proceeding, it will be necessary to describe what I am taking to be
“typology.”
The prototypical
typologies studies have data from a variety of genetically unrelated languages,
analyzed in a general descriptive paradigm that allows for direct, systematic,
and (relatively) just comparing data from various languages and propose
hypotheses about human language in general. This type of study was pioneered by
Joseph Greenberg (1966a, 1966b), who categorized a large number of languages
according to, e.g., most common order of subject, verb, and direct object,
order of adposition and noun, etc., and, on the basis of this categorization, determined
correlations which could be hypothesized as characteristic of human language in
general (e.g. verb-object languages are very likely to have prepositions rather
than postpositions). It is the use of a systematic common descriptive paradigm,
allowing for direct comparison between a wide variety of languages, which
distinguishes the methodology of typology from that of other approaches. Within
the field of discourse analysis, there have been many studies which have compared
different languages but which would not, on this understanding, be considered to
be specifically typological, because they are not focused upon developing a
system for direct, systematic, and universal comparison of a wide variety of
languages as Greenberg’s studies were (e.g. Tannen 1981; Brown and Levinson
1987; Blum-Kulka 1991). Such works are discursive in nature, typically
comparing English with one (or very rarely two) other language(s) and selecting
examples which show how the languages differ in certain respects, or,
alternatively, how they can fulfill similar discourse functions using
constructions which may superficially appear to be different.
Thus, although
such studies are comparative, they are not really directed toward establishing
a systematic universal framework for categorizing discourse phenomena in the
way that Greenberg’s studies established systematic universal framework for
categorizing syntactic phenomena. Because I am discussing the relationship of
typology and discourse analysis, then, I will in the present chapter discuss
those approaches which have been more similar to Greenberg’s in this respect. Section
1 of this chapter will describe general problems associated with methodology combining
typology and discourse analysis. Sections 2 and 3 then discuss two approaches
to these problems, the use of universal conceptual systems of classification and
the use of translation data.
1 Problems of
Typological Discourse Analysis
The study of the
phenomenon of discourse within the framework of the typology presents some
inherent difficulties which can not be found in other areas typology. Traditional
typological studies (Greenberg 1966a; Bybee 1985; Croft 1990) use as their main
source of data reference grammars from a wide variety of languages, and the
linguistic phenomena they consider are those which are likely to be found in a
reference grammar, e.g typical word order (of subject/object/verb,
adposition/noun, etc.), structural characteristics of voice alternations,
phonological inventory, etc. Unfortunately, this is not possible with the sort
of phenomena typically of interest to discourse analysts. For example, if I
were to attempt to do a typological study of the functions of contrastive
connectives similar to English but, I could probably gather a list of
words in a wide variety of languages with some type of generally similar function,
but it would be impossible on the basis of the descriptions of these particles
in reference
grammars to understand and then compare the functions of these different words.
Another problem
for typological discourse analysis as compared with more traditional discourse
analysis is the degree of familiarity of the researcher with the languages to
be analyzed; in a typological study the linguist is not going to know all the
languages under investigation very well, while in a traditional discourse stud the
investigator is likely to be a native speaker of or very proficient in the
language(s) under investigation. While there are recorded cases of individuals
knowing a large number of languages, these are typically closely related or at
least related languages; in typological studies, on the other hand, it is
typical to have data from languages from 15–20 different language families. Even
in this case there has to be much more dependence upon observations of textual
patterns (so that longer texts must be used) and much less upon introspective
judgments than would be the case for linguists working in their native
languages. In such a situation, we cannot expect the relatively quick and
impressive types of language-universal generalizations which individual
typological studies of, e.g., word order patterns have been able to produce; in
fact, it is unlikely that any single researcher will be able to conduct studies
of a genetically diverse enough group of languages to allow for the degree of
confidence in universality which typologists are accustomed to. Rather, in
order to achieve an extensive genetic spread, it is necessary for a variety of
discourse analysts, each working in a number of languages, to develop a uniform
means of systematically comparing their results from these different languages.
1.1 Cross-linguistic comparison of discourse function and categories
A side of the
question of where the data to analyze, it is also necessary to typology
discourse analysts consider the nature of the discourse categories that will be
used. this very common for experts describe the category of discourse in
different languages to using the same words to describe something in a
language they are investigating, for example, "Topics,"
"focus," "On the contrary," etc., but this does not mean
that they are referring to the phenomenon of the same discourse. For example,
although the term “topic” has been used to refer to a supposedly
discourse-based category in a wide variety of languages, there is no
cross-linguistic agreement about what a “topic” is. In each language, “topic” actually
refers to whatever discourse properties result in a certain language-specific structure
being used, so that the definition is a result of the language-specific
pattern, and these structures in different languages actually serve clearly
distinct functions.
In order to deal
with this problem of cross-linguistic comparison of function, linguists working
in typological discourse analysis have focused upon developing a set of
criteria which make it possible to give an objective, cross-linguistic
definition of the discourse function of a particular form or construction.
Using these criteria, a linguist can go through a text in a given language,
note all the occurrences of a given form or construction in that language,
determine numerical scores for that form or construction according to various
parameters (e.g. for an NP, how recently its referent has been mentioned,
whether it refers to a human being, etc.), and then compare these scores with
those of other constructions in other languages. The question, of course, is exactly
which scores should be used in which cases, and this is a matter of ongoing research.
A second approach to the problem of comparison is to use translation data; we
can get some idea of the functional similarity of and difference between
constructions in different languages by seeing how often and in what
circumstances they translate as each other. In section 2, I will discuss
parameters used in classification of discourse function; in section 3, I will
discuss the use of translation data.
2 Universal
Systems of Classification of Discourse Function
I will describe
here various text-count methods which have been developed to give an objective,
cross-linguistically applicable description of the discourse function of a given
construction. The use of such text counts does not suggest that speakers
themselves go through any calculations similar to those of the linguist, nor
does it imply that a given text-count score will predict with 100 percent
accuracy which construction will be used on each occasion. Rather, such counts
are purely descriptive tools to allow for cross-linguistic comparison.
2.1 Referential distance and topic persistence
The most widely
used text-counts, associated particularly with Talmy Givón and students of his,
are called Referential Distance (RD) and Topic Persistence (TP). We can say
that an NP is generally more topical if its RD is low and its TP is high, but
of course we are really measuring two types of topicality here, anaphoric (RD)
and cataphoric (TP). RD and TP counts make it possible to give a functional
profile of a given construction or NP type. For example, suppose that we are
trying to give a general characterization of the function of the active–passive
alternation in English, e.g. Bill wrote that book vs. That
book was written by Bill. We go through a text, collecting all active
transitive and passive constructions, and then count the average RDs for the Agents
of actives (Bill in Bill wrote that book), the Agents of passives
(Bill in That book was written by Bill), the Patients of
actives (that book in Bill wrote that book), and the Patients of
passives (That book in That book was written by Bill). We then
calculate the mean and median RD and TP scores for active Agents, passive
Agents, active
Patients, and
passive Patients, or list the populations in a table. By doing similar studies
in a variety of languages, we can systematically compare the discourse
functions of active and passive constructions in different languages. This
approach has been useful in providing a typological perspective on functional
alternations, clarifying the discourse motivations underlying these
alternations, and also sharpening the descriptive tools for typological
descriptions; it does not suggest that speakers make such calculations in
deciding which construction to use (although RD can be interpreted as being
generally correlated with cognitive accessibility). In the remainder of this
section, I will discuss a number of studies which have been done using these measures.
2.1.1 RD and TP in analysis of voice systems
Voice alternations
in different languages have been characterized in various descriptive grammars
in a variety of ways, in particular Active vs. Passive, Direct
vs. Inverse, Ergative vs. Antipassive, and (for
Philippine languages) Agent Focus vs. Goal Focus. However, the
basis for such characterizations has often been unclear. Consider, for example,
the following constructions in Tagalog :
(1) Bumasa ang
lalaki ng diyaryo.
read man
newspaper
“The man read a
newspaper.”
Case functions
in Tagalog are marked by prepositions, here ang and ng. It is
clear that ang marks intransitive subjects (e.g. matalino ang lalaki “intelligent
ang man” = “The man is intelligent”). The question here is what general
function to ascribe to ang and ng. One possibility is to say that
ang marks subjects (both intransitive and transitive) and ng marks
direct objects and oblique NPs. Then (1) would be an active construction, with lalaki
as the subject and diyaryo as the direct object, while (2) would be
a passive construction, with diyaryo as the subject and lalaki as
the oblique Agent. Alternatively, we might say that ang is an absolutive
case marker (marking intransitive subjects and direct objects), while ng is
an ergative (transitive subject) and oblique case marker. Then (1) would be an
antipassive construction (grammatically intransitive), with lalaki as
the intransitive subject marked with the absolutive preposition ang, and
diyaryo (which is in this case an oblique rather than direct object) marked
with the oblique preposition ng, and (2) would be an ergative
construction,
with lalaki as
the transitive subject, marked with the ergative preposition ng, an diyaryo
as the direct object, marked with the absolutive preposition ang. In
fact, earlier studies of Philippine languages (e.g. Schachter and Otanes 1972)
used yet anothe type of terminology, referring to ang as marking
“focused” constituents (which causes confusion of another type in terms of
cross-linguistic comparison, since the term “focus” is usually used with some
sort of entirely different meaning) and ng as marking certain nonfocused
constituents, so that (1) is an “Actor Focus” construction while (2) is a “Goal
Focus” construction. Similar labeling problems arise in many other languages
(see Givón 1994).
2.1.2 RD and word order
Linguists have
also applied RD to investigating word order variation. Studies from a variety
of languages have found that preverbal arguments have on average a higher RD
than postverbal arguments (there does not seem to be any corresponding clear pattern
relating TP and word order). However, it is possible to suggest a resolution to
this apparent contradiction (although this is speculative and should be checked
against more data). Supporting this idea is the fact that in the data from the
only SVO language here, Spanish, the researcher specifically excluded
existential-presentative constructions from the counts (see Bentivoglio 1983); if
these constructions are included, the picture changes, as the postverbal
subjects
have a higher
RD (11.99, N = 141) than the preverbal ones (8.22, N = 180).
2.2 Temporal sequencing
Another
criterion for categorizing discourse function in different languages is temporal
sequencing or foregrounding.
Introduced in Labov (1972) (as the concept “narrative clause”), this was first
extended to data in a variety of languages in Hopper (1979).
For example,
Schiffrin (1981) shows that the English historical present is associated with temporally
sequenced clauses, while Hopper (1979) shows that temporal sequencing is
associated with the use of the verbal forms with a di-prefix in Malay.
Myhill (1992) argues that, in languages with a relatively high frequency of VS
order, sequencing is particularly associated with VS word order, while SV order
is associated with unsequenced clauses. On the other hand, in languages with a
lower frequency of VS order, this correlation is not found. This is shown by
the data in table 8.3 (see also data from Old English in Hopper 1979). The
Biblical Hebrew data here are particularly striking, in that they show that when
the language changed to a lower frequency of VS order, the association between
temporal sequencing and VS order disappeared. The concept of temporal sequencing
therefore makes it possible to make a typological generalization regarding word
order type.
2.3 Other types of text-counts
Linguists have
proposed other types of text-counts which can be useful in giving a profile of
the discourse function of a construction. Myhill and Xing (1996) propose a definition
of the term “contrast” which can be objectively applied to naturally occurring usages
so as to categorize individual clauses as contrastive or not (and also to distinguish
between different subtypes of contrast), so that one or another contrastive function
can be shown to be statistically associated with the use of a certain word order,
intonation pattern, or particle (e.g. Japanese wa, Korean -(n)In).
In Forrest’s (1994) study of voice alternations in Bella Coola, in addition to
counts associated with NP information status such as RD and TP, she also uses a
text-count distinguishing between NPs which refer to major characters in a
story and those which do not, and shows that variation on this parameter
correlates with the use of one or other voice construction.
3 Translation
Data
Translation
provides another means of comparing discourse functions in different languages.
It is useful in that it gives some idea of the functional similarity or
difference between constructions in different languages. For example, in
Dryer’s (1994) study of voice in Kutenai, he asked a bilingual Kutenai–English
speaker to translate a Kutenai text into English. He found that, out of 70
clauses using a certain Kutenai construction clearly associated with highly
topical Patients, only nine were translated into English as passives, the rest
being translated as actives, suggesting that this Kutenai construction is
functionally like an inverse rather than a passive. Sometimes, translation data
show that text-counts such as RD and TP do not give a true picture of the
functional similarity or difference between different constructions in
different languages. For example, Sun and Givón (1985) use data such as RD and TP
to argue that object-fronting constructions in Chinese and Biblical Hebrew
serve basically the same function. However, Myhill and Xing (1993) show that,
if we look at translation data, we see that the object-fronting constructions
in these languages are frequently not translated as each other; for example, of
82 OV constructions in a Biblical Hebrew database, 48 (59%) do not use
an OV construction in the Chinese translation, while of 193 OV constructions in
the Chinese translation, 159 (82%) do not use an OV construction in the
Hebrew original. In other words, in the majority of cases, an OV construction
in one language would not be used where an OV construction would be used in the
other language. This shows that the Hebrew and Chinese OV constructions clearly
differ significantly in discourse function, in spite of their RD and TP scores.
References
Bentivoglio, Paola. 1983. Topic
continuity
and discontinuity in discourse: a
study of spoken Latin-American
Spanish. In Givón 1983a, pp.
255–312.
Blum-Kulka, Shoshana. 1991.
Interlanguage pragmatics: the
case of
requests. In R. Phillipson, ed., Foreign/
Second Language
Pedagogy Research,
pp. 255–72. Clevedon:
Multilingual
Matters.
Brown, Penelope and Stephen C.
Levinson.
1987. Politeness: Some
Universals in
Language Usage. Cambridge:
Cambridge University Press.
Bybee, Joan L. 1985. Morphology.
Amsterdam: John Benjamins.
Chafe, Wallace L. 1976.
Givenness,
contrastiveness, definiteness,
subjects,
topics, and point of view. In Li
1976,
pp. 25–56.
Chafe, Wallace L., ed. 1980a. The
Pear Stories.
Norwood, NJ: Ablex.
Chafe, Wallace L. 1980b. The
deployment
of consciousness in the
production
Typology and
Discourse Analysis 173
of a narrative. In Chafe 1990a,
pp. 9–50.