Content uploaded by Christian Lehmann
Author content
All content in this area was uploaded by Christian Lehmann on Jul 29, 2021
Content may be subject to copyright.
CLIPP
Christiani Lehmanni inedita, publicanda, publicata
titulus
Data in linguistics
huius textus situs retis mundialis
http://www.christianlehmann.eu/publ/
lehmann_data_in_linguistics.pdf
dies manuscripti postremum modificati
28.06.2004
occasio orationis habitae
Colloquio internazionale ‘Di cosa parliamo quando parliamo
di linguistica?’, Roma 1-2 luglio 2002
volumen publicationem continens
The Linguistic Review 21(3/4)
annus publicationis
2004
paginae
275-310
Christian Lehmann
2
Data in linguistics
Christian Lehmann
University of Erfurt
Περ̀ι τω̃ν αφανέων περὶ τω̃ν θνητω̃ν
σαφήνειαν µὲν θεοὶ έχοντι,
ως δὲ ανθρώποις τεκµαίρεσθαι.
1
Alcmaeon of Croton (Diogenes Laertius
VIII
, 83)
Die Theorien vergehen,
aber das Material bleibt bestehen.
2
Einar Löfstedt 1942: IX
Le donné linguistique est un résultat;
et il faut chercher de quoi il résulte.
3
Émile Benveniste 1954(1966):117
Abstract
This article aims to be a contribution to the methodological foundations of linguistics. To an-
swer the question of “what are scientific data?”, a semiotic conception of data is proposed
according to which they are representations of properties of the object area of a science that
serve certain purposes for their users. Kinds of data are distinguished by their ontological
status, degree of abstractness, the type of sign representing them and their originality. The
methodological status of data in the history of linguistic science is briefly reviewed, and their
functions in scientific argument are specified. Various methods of data provision by genera-
tion of data or by use of available data are discussed. Since data are representations, they are
per se a linguistic issue which, however, is even more complicated for linguistic data proper,
because here diverse linguistic levels and diverse levels of abstractness have to be controlled.
Apart from the principal necessity to have clarity on the methodological bases of a sci-
ence, the issue of the nature and function of data in linguistics acquires increased urgency in a
world where the task of documentation of endangered languages is, first and foremost, one of
adequate data provision.
1
On invisible and on earthly things, gods have clarity, while to men it is given to infer.
2
Theories pass away, but the material lasts.
3
The linguistic datum is a result; and we have to search what it results from.
Data in linguistics
3
1. Introduction
Since the etymological meaning of the word data is “(things that are) given,” it is probably an
instance of “nomen est omen” that the notion and role of data in science are generally taken
for granted. A representative sample of contemporary publications on methodology and phi-
losophy of science – introductions, manuals and lexica – reveals that the term nowhere consti-
tutes a lemma and the concept is nowhere introduced explicitly, let alone defined. In linguis-
tics itself, endeavors to clarify the role of data are very recent and as yet few.
4
We will first try to elucidate the notion of data and define it. The conception proposed is
essentially a semiotic one: data – not just linguistic data, but any data – will emerge as a cer-
tain kind of representation of an object. We will then look at essential properties of data in
linguistics and their role in the scientific process, systematize the most important ways of ob-
taining data and finally come to the genuinely linguistic issue of representations of data. The
article is thus a contribution to the methodology of linguistics.
2. The notion of data
2.1. The term
In order to be able to speak about data, we first have to emend the English language a bit.
Since the middle of the 20
th
century, it has become customary in English to use the word data
as a mass noun. As a consequence, it does not pluralize nor combine with an indefinite article,
but instead combines with mensuratives like a piece (of data). For the sake of the following
discussion, we shall undo this linguistic change and reestablish traditional usage: For the con-
cept in question, there is an individual noun datum which forms a morphologically irregular,
but semantically regular plural data. The latter thus does not designate a mass, but a set of
individuals.
Secondly, we have to do a linguistic analysis of the word datum. This is a relational and,
therefore, functional concept in the sense that something is not in and by itself a datum, just as
something is in and by itself a sentence. Instead, something functions as a datum for some-
body. This is explicable both in terms of scientific methodology – and to this we will come in
section 4 – and in terms of the etymology of the word, to which we now turn. The etymologi-
cal meaning of datum is “given,” in roughly the same sense which is still productive in mod-
ern English, for instance in the complex conjunction given that. This conjunction introduces
something whose existence and nature is independent from the deictic center and which the
deictic center cannot but accept as it is. The extensional meaning of the word datum is inde-
4
The largest relevant enterprise is probably the DFG Sonderforschungsbereich 441 on ‘Lin-
guistic data structures’ at the University of Tübingen. Two independent contributions may be
mentioned: Iannàccaro (2000) concentrates on the interference of theory in fieldwork, and
Simone (2001) deals critically with several methodological aspects of linguistics, including,
in particular (section 2.5.2), the role of data in linguistics. I thank Gabriele Iannàccaro for
helpful discussion.
Christian Lehmann
4
terminate insofar as it designates anything that is “given” (in the relevant sense), independ-
ently of its particular nature. Moreover, if datum means “given,” this, of course, evokes the
argument frame of the verb “give”: the giver, in this case the producer or source of the datum;
the recipient, in this case the discoverer or user of the datum; and the transferred object, in this
case the entity which constitutes a datum. At first sight, it might appear that the first two enti-
ties of this argument frame are irrelevant for the scientific concept of data. This is, however,
not so. Maybe in linguistics more than in many other sciences, it is an absolutely crucial issue
who produces the data and who receives them. We will come back to this in section 4.2.
2.2. The concept
In order to clarify the methodological status of data, it is useful to start with the ontology of
naïve realism (cf. Lyons 1977, ch. 11.3): First-order entities comprise physical objects; sec-
ond-order entities comprise states, processes, events and the like. These two kinds of entities
are located in space and time and are observable. Third-order entities comprise abstract enti-
ties such as propositions, which are not bound to space and time and are unobservable.
A fact is a third-order entity – a proposition – which corresponds to a certain second-order
entity. By virtue of this correspondence, the proposition receives the predicate “true.” Particu-
lar propositions in a science may have the status of a fact. For instance, that Caesar once
wrote veni, vidi, vici is a fact, or is taken as a fact, in some particular sciences as well as in our
civilization in general.
The object area of a discipline is what that discipline wants to learn about. The object area
of an empirical discipline has, so we assume whether we are constructivists or not, what will
here be called an ultimate substrate, i.e. a basis in the world surrounding the thinking subject
which he can perceive. However, the “real world” is complex and multifaceted, and no human
cognition is interested in a true copy of it. Every science construes its object area according to
its epistemic interest, by delimiting and idealizing what we can perceive and by distinguishing
between relevant and irrelevant aspects of these percepts. The object area even of an empirical
science (see section 4.1 for the other kinds of science) is, thus, not part of the physical world,
but is a mental representation of part of the world. It does not consist of first and second-order
entities, but of third-order entities. The object area of linguistics is not a set of states of affairs
taking place in the outside world, but a set of situations of linguistic communication con-
strued, delimited, purified and focused upon in ways that vary in the history of science, but
which never equal or exhaust some physical reality.
A particular research is devoted to an epistemic object (also called “phenomenon”
5
),
which is a construct that is part of the object area. Data concern particular aspects of an em-
pirical epistemic object; and by extension the data of an empirical science represent aspects of
its object area.
Consequently, data are neither first nor second-order entities. For instance, a particular
pot-shard is not, in itself, an archaeological datum, and a tape which has a speech recorded on
it is not, qua physical object, a linguistic datum. Instead, a datum corresponds to a fact, i.e. to
a third-order representation of a state of affairs which is considered true. A datum therefore
has an inner side which is a mental representation of some state of affairs. For scientist B to
5
‘Phenomenon’ may also be opposed to ‘datum’; cf. fn. 19.
Data in linguistics
5
accept something which scientist A adduces as a datum means for A and B to share some
mental representation which both consider part of the object area of their discipline.
Naturally, mental representations are not the form in which the data are transmitted and
analyzed in scientific research. Instead, scientific data are processed in the form of semiotic
representations, including linguistic representations, of facts.
6
For illustration, let us briefly
look at data of a few different disciplines.
The object area of demography is the structure and dynamism of a population. Particular
aspects of it, for instance the sex of a particular person that enters some statistics, are publicly
observable. The ultimate substrate of the data of the discipline consists in a set of such par-
ticular states of affairs involving members of the population. These are converted into mental
representations and into symbolic representations of the latter, for instance in a table, each
entry of which refers to an individual and each cell of which represents some property of this
individual, for instance his sex. Such a table represents a set of data in this discipline.
The data on which dendrochronology builds its theories are series of numbers, each of
which represents the width of an annual ring of some tree and is associated with one in a se-
ries of years. The cross-sections of the tree may be stored somewhere for measurements, be-
cause they constitute the ultimate basis of reference for certain relevant observations. The
data, however, are those series of numbers insofar as they represent facts about these objects.
Finally, history has an ultimate basis in reality which comprises such second-order enti-
ties as the event in which prime-minister Begin shook hands with president Sadat on Camp
David on Sept 17, 1978. Apart from a couple of exceptions, historians have not witnessed
these second-order entities. They have, however, recordings of them available, either iconic
records like photos or audio tapes or symbolic representations which historians call sources.
They are, then, the form that data take in history.
No science can be run without such representations of facts. It will, therefore, not be nec-
essary, in what follows, to always distinguish between a fact and its representation. Since
symbolic representations are a genuinely linguistic problem, we will come back to them be-
low (section 6.2).
Table 1 visualizes the concepts and examples introduced so far and adds a column for lin-
guistics, which we will take up below. There is a further concept, viz. material, which has to
be distinguished from data and which may be introduced with respect to the first row of Table
1. In its literal sense, the word material designates a collection of physical objects. In some
scientific disciplines, the word may be used to designate (parts of) the ultimate substrate of
their object area, i.e. a set of first-order entities which the data are based on. For instance, for
dendrochronology and archaeology, the physical entities mentioned in the first row of Table 1
constitute the material which research observes and from which it starts. The ultimate sub-
strate of such disciplines is available for repeated direct observation by scientists.
6
Unfortunately, the English word representation is ambiguous: it can mean either a purely
mental entity (German Vorstellung) which may or may not correspond to something outside
the mind, and it can mean a semiotic entity (German Darstellung) in which a perceivable
repraesentans can be distinguished from a – mental or semiotic – repraesentatum. The latter is
what is meant in saying that data are representations. The former meaning will be expressed,
if necessary, by mental representation.
Christian Lehmann
6
For other disciplines including history and linguistics, the real-world entities underlying
their object area are not first-order, but second-order entities such as historical events and
speech events. These are volatile and therefore cannot constitute any material in the literal
sense of the word. In order to be subject to an objective treatment, they first have to be re-
corded. The recordings are on durable material, but this is material in a different sense, be-
cause it is of interest not as a physical object, but only as a representation of the proper object
of research. The word material, therefore, has a different sense in the natural sciences and in
the humanities. In the former, the material is the ultimate substrate of the data; in the latter,
the material is a first recording of the data; i.e. it is a particular kind of data. Consequently, the
last two rows of Table 1 contain data at different levels of representation.
Table 1. Data and representations in some disciplines
discipline
object
dendrochronology
demography archaeology history linguistics
ultimate
substrate
(cross-sections
of) tree trunks
population ruined wall handshake
between two
politicians
speech event
epistemic
object
chronological
position of the
tree
gender dis-
tribution in
the popula-
tion
the original wall relationship
of the two
politicians
utterance in
the speech
event
original
recording
– – contemporary
mention/drawing
of the wall
source men-
tioning the
handshake
video tape of
the speech
event
derived
represen-
tation
series of numbers
representing
measured widths
of annual rings
tables and
charts repre-
senting gen-
der distribu-
tion
design of the
ruined wall
time-line
with major
events,
including
handshake
phonetic
transcription
of the utter-
ance
Up to now I have proceeded as if data were a special kind of thing. However, what consti-
tutes a datum is not its nature but its function. In the context of empirical scientific research, a
datum serves either as the basis for the inductive construction of a hypothesis or as the test for
a theorem arrived at deductively. In order to be able to fulfill this function, a datum must have
a basis outside of and independent from the researcher.
The issue of the externality and independence of the datum from the researcher will re-
peatedly occupy us below (cf. especially fn. 14). It essentially means that there must be meth-
ods of relating the datum to the ultimate substrate. I will come back to this in section 3.4 and
recall here the relationality of the concept “datum,” which I introduced by way of the etymol-
ogy of the term (section 2.1). Nothing is, in and of itself, a datum; instead, it is a datum for
somebody (or for a scientific community) in some perspective. Linguist A has a tape which
records a story in Yucatec Maya. The recording is A’s data. He produces an orthographic rep-
resentation of the story and publishes it as the result of his research. Linguist B uses A’s or-
thographic representation as data for his grammar of Yucatec Maya, which he publishes as the
output of his research. Linguist C is a typologist whose sources of information are grammars.
Data in linguistics
7
He uses B’s descriptive statements on Yucatec Maya clause structure as data, puts them into a
database and arrives at a couple of cross-linguistic generalizations which he publishes as a
typology of clause structure.
The example shows that one person’s analysis may be another person’s data. Something
is not a datum by virtue of corresponding to some elementary observation, like the “protocol
sentences” of the logical positivists. On the contrary, it may be highly abstract.
7
It may never-
theless function as a datum in some research that assigns it the role of unquestionable evi-
dence in the argumentation. Thus, something is not in and of itself a datum, but it is a datum
relative to some particular empirical research.
We must, however, caution against a misunderstanding. The 1970s saw a wave of prag-
maticism where scientific concepts were made relative to the scientist who uses them.
8
Now
of course, linguists just like other scientists sometimes disagree on whether something counts
as a datum or not. The point here, however, is not that the application of a concept to a refer-
ent depends on the user. This is a basic semiotic fact which need not be repeated in the defini-
tion of each concept. We are not here talking about some pragmatic relativity, but about the
intrinsic relationality of the concept. Just as no proposition is, in and of itself, an argument,
but may only be used as an argument under certain conditions, so a representation of some-
thing is not a datum if taken absolutely, but may function as a datum in a certain research.
Now we can define:
A datum is a representation
i
of an aspect of the epistemic object of some em-
pirical research which
i
is taken for granted.
An aspect of x which is taken for granted is a fact about x in the sense defined above. Since a
datum is a representation of something, it is a sign. What I propose, thus, is a semiotic con-
ception of the datum; to be sure, not restricted to the linguistic datum, but intended for the
datum of empirical research in general. Since a datum is a sign, it may be an icon, an index or
a symbol. The examples given in Table 1 include icons, e.g. the videotape of the speech event,
and symbols, e.g. the table representing gender distribution. In some sciences there are also
indices that count as data, as, for instance, a footprint of a saurian in paleontology.
3. Kinds and properties of linguistic data
Let us now pursue the consequences of this conception in order to see by which particular
properties linguistic data differ from data in other disciplines.
3.1. Raw data vs. symbolic representations
Let us first come back to the tape which records a story in Yucatec Maya. The tape is a first-
order entity, and the process in which a certain Maya once recorded the story is a second-
7
In Seiffert (1969f), which is a treatise of scientific methodology in general, the term data is
applied to summary representations of a couple of theories of different scientific disciplines. It
is therefore not the case that linguistic data are statements of particular observables, as Chom-
sky (1964: 28ff.) thinks.
8
Cf., e.g. the definition of the language universal in Lieb (1974: 494): "A property F is uni-
versal in language relative to a person during a time if that person during that time requires
that F should be attributed to all languages by any theory of language."
Christian Lehmann
8
order entity. Now let us set up a research whose epistemic object is this story. Then the sec-
ond-order entity just mentioned is the ultimate substrate of the epistemic object, whereas the
embodiment of the story on the tape is a piece of data which consists in a recording of the
ultimate substrate.
As just said before, the tape-recording is an iconic representation of the data in question.
In general, photos, audio- and videotapes of speech events, no matter whether recorded in
analog or digital technology, are non-symbolic representations of linguistic data, whereas an
orthographic or IPA representation of the same event is a symbolic representation.
9
Higher
level linguistic analysis of any data commonly presupposes their symbolic representation. In
this sense, a recording in the form of a non-symbolic representation constitutes raw data for
linguistic research. For most purposes, the first step in the research will consist in their tran-
scription.
The distinction between raw data and symbolic representations must not be confused with
the distinction between original recording and derived representation (see the last two rows of
Table 1). Some recording is original in the sense that it is not based on another representation.
The original recording of a speech event may be a symbolic or non-symbolic representation,
and either may be produced by a linguist or a layman. A derived representation of the original
recording may conserve its original nature, as when a pure audio record is distilled from a
videotape or when a phonetic transcription is converted into an orthographic one; or it may
shift it into the other type of representation, as when an audiotape is transcribed or a text is
tape-recorded. The production of derived symbolic representations is a typical activity in lin-
guistic data processing, and we will have more to say about it in section 6.2.
3.2. The uniqueness of the linguistic datum
Now consider a particular linguistic datum on that tape, as the occurrence of a certain word on
the tape or the fact that it is immediately followed by a certain other word. These data clearly
go beyond sheer physical traces on the tape in several respects. The first respect was already
mentioned in section 2.2: The fact that word x is immediately followed by word y on that tape
is not a first or second-order entity – a physical pattern on the tape –, but instead a mental
representation of an aspect of the epistemic object as recorded on the tape.
The other properties of this fact are peculiar to linguistics. First, our identification of a
word on the tape presupposes the recognition of a linguistic expression. This is, again, not a
set of physical traces on the tape, but instead an abstraction over certain phonetic objects.
10
Second, a semiotic entity has two sides, only one of which is perceivable. The datum in ques-
tion, however, concerns tokens of certain semiotic entities and consequently the coupling of
the significans with its significatum.
11
The identification of even the most elementary linguis-
tic datum therefore presupposes an abstraction and a semiotic operation.
9
Cf. the related notion of discrete vs. analog communication in Watzlawick et al. 1967: 61-
68.
10
This is what Bühler (1933:30-32) calls the ‘abstractive relevance of linguistic units.’
11
This is what Bühler (1933:24) calls the ‘semiotic nature of language.’
Data in linguistics
9
3.3. Primary and secondary data
Next, let us come back to the example of the typologist who uses the descriptive statements of
a Yucatec grammar as the data for his typology. The example was used to show that the no-
tion of data is relative to a purpose. But it also shows, of course, that data may be elementary
or abstract to different degrees. For something to be used as a datum in a discipline, it suffices
that it be a fact that is empirically relevant to it. It is not necessary that it be an elementary
fact. That a certain word on that Yucatec tape is immediately followed by a certain other word
may, in a certain perspective and for a certain purpose, be considered an elementary fact
which is immediately verifiable by inspection. However, we are used to working with much
more complex and abstract kinds of data. In the past fifty years, linguists have gotten used to
cascades of example sentences which exhibit a regular structural difference and every other
one of which is preceded by an asterisk, accompanied by a commentary in which the linguist
adducing the series refers to them as data. Such a linguistic datum is a semiotic object of a
higher order. Namely, it is an expression of the object language coupled with a statement of
the metalanguage – the latter being highly abbreviated in the form of an asterisk and even its
absence – which predicates a certain property over that object. This statement is taken for
granted and therefore regarded as a linguistic datum.
Claims made in an empirical science, including both hypotheses arrived at by inductive
generalizations over empirical data and theorems derived deductively on the basis of a theory,
must be testable on data belonging to its object area. Suppose our typologist launches the
(false) hypothesis that if a language has numeral classifiers, it lacks nominal number. We can
check this hypothesis for Yucatec Maya on the data contained in the typologist’s database,
where this language is marked for possession of both numeral classifiers and nominal number.
However, we want to go further and verify whether Yucatec does have numeral classifiers.
Here we see that although a descriptive statement may be used as a datum at some level, it
bears no direct correspondence to anything observable. We therefore have to distinguish be-
tween primary and secondary linguistic data. Primary linguistic data are (original or derived)
representations of specific speech events with their spatio-temporal coordinates, i.e. of objects
with a historical identity. Secondary data are more abstract in some respect. At a first level of
abstraction, we get what Lyons (1977:29-31) calls “system-sentences.” These are sentences in
written representation that lack spatio-temporal coordinates and, therefore, a historical iden-
tity. They are being used as types rather than as tokens, but come along with a claim of being
usable in some actual speech situation, thus, a claim of being potential primary data. Yet more
abstract are facts concerning (primary or secondary) data, including metalinguistic statements
on properties of speech events or system sentences and higher order generalizations over such
properties. This includes, in particular, the starred example sentences and their non-starred
counterparts mentioned before and, more generally, so-called “negative data” (or what Ian-
nàccaro (2000:68 et pass.) calls “antiesempio”), i.e. claims on the non-existence of certain
phenomena.
A primary linguistic datum necessarily represents a certain spatio-temporal variety of a
language. Certain branches of linguistics, including much of descriptive linguistics and a for-
tiori typology and universals research, are usually not interested in intralinguistic variation
and want to make statements concerning a language as a whole. Such statements may be
based on primary data, but abstract from their spatio-temporal setting. More often than not,
they are based on system-sentences and other kinds of secondary data.
Christian Lehmann
10
I have now introduced two related distinctions: one between primary and secondary data,
and one between original and derived representations. The latter distinction obtains inside the
category of primary data. In section 6.2, we shall see that the production of derived represen-
tations generally involves abstraction. Therefore, the transition from primary to secondary
data in a sense continues an increase in abstraction that also holds for the progression from
original to derived representations.
3.4. Data and operational procedures
Table 2 summarizes the kinds of entities, in particular of data, introduced so far. It is to be
read as a decision tree from top to bottom. The criteria of classification and their values are
arranged in colored boxes.
Table 2. Types of data
At the bottom level, the two distinctions according to type of sign and according to originality
are independent of each other. Both of them obtain within the class of primary data. Secon-
dary data are, of necessity, symbolic and derived.
In general, for a datum to be accepted as such in the discipline, there must be operational
procedures of relating secondary to primary data, and primary data to the ultimate substrate.
Such procedures are part of the methodology of that discipline, viz. of the methods that allow
scientists to control the relationship between the theory and the data. The methods establish
the relationship in both directions. On the one hand, they are standardized procedures
12
that
may be applied routinely to a set of raw data and allow the scientist to convert these into sym-
12
the “survey protocols” and “analytical techniques” of Simone (2001, section 2.3f)
Data in linguistics
11
bolic representations and analyze the latter, i.e. to develop hypotheses on them. On the other
hand, such methods constitute the operationalization of theoretical constructs, i.e. they specify
the conditions under which a concept may be applied to a phenomenon and under which a
theorem is considered falsified by a datum. If there are no such operational procedures, then
firstly there is no basis on which the datum can be taken for granted, which means it is not a
datum in the sense of our definition; and secondly, there is no way of relating a theory to a
perceptible epistemic object, which means it is not an empirical theory. This is a field in
which linguistics has not excelled during its existence. In section 6.2, some elementary stan-
dards of working with derived data are proposed.
4. Role of data
4.1. Methodological status of data in science
Only in an empirical science can data have any import at all. If there is sometimes talk of data
in a logical discipline such as philosophy and mathematics (cf. fn. 7), it makes sense only to
the extent that some specific research occasionally does make use of empirical methods; and
the same can be said about hermeneutic disciplines such as literary studies. On the other hand,
in an empirical discipline such as biology and chemistry, data are the ultimate basis, the point
of reference and touchstone for any scientific statement.
The case of linguistics is, again, more complicated, since it shares properties with all three
kinds of sciences mentioned.
13
To the extent that the object of linguistics is a construct of our
mind, linguistics is a logical science. To the extent that it is observed in the world around us,
it is an empirical science. And to the extent that the object requires understanding, linguistics
is a hermeneutic discipline. These facets of our field have influenced diverse conceptions of
data in it. If linguistics is a logical science, then it needs no data at all. Witness of this attitude
are a couple of linguistic theories, e.g. the one of Coseriu (1958), which were launched with-
out consideration of a single linguistic datum or in which linguistic data only play the role of
illustrations which render the theoretical statements more readily intelligible for the con-
sumer. If linguistics is an empirical science, then it depends on data, both in inductive work
that analyzes and generalizes over the data and in deductive work which tests hypotheses on
the data. A good example of this approach is the kind of sociolinguistic work represented in
Labov (1982). Finally, if linguistics is a hermeneutic discipline, then its object is not data in
the sense of facts ultimately reducible to observable entities, but instead mental representa-
tions conveyed and modified between understanding subjects, including the participating sci-
entist. Stolze (1992) is an example of this approach.
In the history of the discipline, various brands of linguistics have gravitated towards one
or another self-image. Accordingly, the role played by empirical data has varied enormously.
With some simplification, we may distinguish three phases. The first is the logical strand of
linguistic research, which stems from antiquity. For Plato, Aristotle and the Stoics, grammar
was a precondition for rational argumentation. This conception prevailed in medieval modism
13
Simone (2001) is rather negative on the status of linguistics as an empirical science. The
term he applies to linguistics, ‘meso-science’ (intended to mean ‘half-science’), however, also
has a positive interpretation, viz. ‘center-science.’
Christian Lehmann
12
and in rationalist general grammar. Since the advent of modern linguistics at the beginning of
the 19
th
century, the logical approach to language has been strong in typology, e.g. in the work
of Friedrich Schlegel and Vladimir Skalička, and of course in those approaches that aim at
elaborating linguistic theories, including the work of Louis Hjelmslev, Eugenio Coseriu and
Noam Chomsky. In this tradition, linguistic data play no role in research, since the linguist
only externalizes what is already in his mind. Representations of speech events are neither
needed as a basis of inductive generalizations nor as touchstones of empirical verification or
falsification of hypotheses deduced from the theory, since it is not an empirical theory.
14
If
tokens of linguistic units like words or sentences appear in scientific treatises, it is as illustra-
tive examples in order to facilitate understanding. Consequently, the logical trend in linguis-
tics has not produced a culture of linguistic data. On the contrary, to the extent that it pre-
vailed in modern linguistics over a certain period, it suppressed any methodology that would
care for a responsible treatment of data, with the consequence that knowledge and skills in
this area are relatively underdeveloped in today’s descriptive linguistics, if compared with
neighboring disciplines such as anthropology or sociology.
Only slightly younger is the hermeneutic phase of linguistics, starting with the school of
Alexandria, with Dionysios Thrax (ca. 160 – 95 BC) and Apollonios Dyskolos (first half of
2
nd
cent. AD), and continuing in the philologies to this day, but in linguistics itself essentially
discontinued since its inception proper, at the beginning of the 19
th
century. In this tradition,
linguistic data mostly take the form of manuscripts. They are preserved essentially on account
of their content, not because of the linguistic data they contain. The latter, again, are not
viewed as tokens of a type, as instantiating patterns of the linguistic system, but instead as
expressing unique messages, sent by a member of a certain society and culture to the scientist
as the recipient, but a member of a different society and culture. Although this tradition has
developed an admirable skill and diligence in editing, transmitting, archiving and interpreting
texts, no notion of data as representing the object area of linguistics has developed.
The view of linguistics as an empirical science is a latecomer in the history of the disci-
pline. There were timid beginnings in historical-comparative linguistics since Franz Bopp, a
century later ousted to a large extent by European structuralism as launched by Ferdinand de
Saussure. Awareness of the linguistic datum, its nature, role and dignity, evolved first in those
branches of linguistics that actually executed fieldwork. These were essentially European dia-
lectology at the end of the 19
th
century and, in the first half of the 20
th
century, American
structuralism in its contact with anthropology. The achievements of the latter were essentially
annihilated, in the way just alluded to, by generative grammar. They were, however, taken up
and refined in the seventies by the modern disciplines of sociolinguistics and psycholinguis-
tics. These imported the methodology of such thoroughly empirical sciences as sociology and
experimental psychology into linguistics, including their methods of obtainment and manipu-
lation of data. Here we meet, for the first time, an elaborate conception of the linguistic da-
tum. It is, however, outside the institutional core of the linguistic discipline, whose object is
14
This assertion itself, although not its reason, is freely conceded by the advocates of the in-
vestigation of the ‘I-language’ instead of the ‘E-language’ (cf. Simone 2001:58 against
Chomsky 1986:22). Note that the construct of I-language (internal language) is meant to
avoid the requirement stated in section 2.2 that a datum must have a basis outside of and in-
dependent from the researcher.
Data in linguistics
13
linguistic systems and which produces descriptions, comparisons and theories of this object. It
has only been for a relatively short time that descriptive linguistics, taking up insights of
European dialectology and American structuralism, has been struggling to raise its methodo-
logical standards concerning the treatment of data to the level established in socio- and psy-
cholinguistics. In the past few decades, more and more linguists have dedicated themselves to
fieldwork and to the documentation of endangered languages, involving the recording, repre-
sentation, elaboration and archiving of primary data for their own sake. There are so few rele-
vant traditions and established methodological standards in our field that the urgent necessity
of documenting endangered languages has given a decisive thrust to the elaboration of tech-
niques concerning the processing of linguistic data.
4.2. Functions of data
The functions of data in empirical research are various. They derive from the relationality of
the concept of the datum as introduced in section 2.1: The datum is related to its producer or
source and to its user or recipient. We will start with the former.
4.2.1. The relation of data to the producer
In the past fifty years, scientific standards regarding sources of data and their identification
have raised considerably. Up to the middle of the twentieth century, it was customary in de-
scriptive linguistic work – although not in dialectology and the philologies – to use examples
without any indication of their source. If the language being described was the author’s native
language, one could bet that he had produced the examples himself, because it was not cus-
tomary to base descriptive work on corpora or on fieldwork. In the past decades, the use of
corpora has become both easier and more wide-spread, and it is now standard to identify the
source of the data. Diligence in this respect still varies a lot, but there is a trend towards more
accuracy concerning the spatio-temporal coordinates of the production of the data so that the
consumer gets a chance of controlling the particular variety of the language represented by
them.
Here also belongs the issue of the representativity of the data (on which more appears in
section 5.3). Most linguistic descriptions purport to deal with a language. Then, of course, the
question arises how the data must be sampled in order to justify such a claim. A good exam-
ple of sound methodology in this respect may be found in frequency dictionaries (e.g.
Gougenheim et al. 1967). Methods of delimiting the linguistic variety to be documented and
to define a balanced sample that represents this variety are firmly established in the tradition
of their production.
The producer of the data is normally a person in his capacity as a speaker of the language
in question. Respect for his role has increased, too. While his renaming from “informant” to
“consultant” may safely be neglected as an outflow of political correctness, the career that he
has made from the background to the foreground of linguistics is much more noteworthy.
While he was not even mentioned in the earliest linguistic publications based on fieldwork, it
Christian Lehmann
14
is now standard to render him due attention, and not seldom has he advanced to the position
of co-author of the linguistic description (e.g. Hofling and Tesucún 2000).
15
What is perhaps even more significant is the fact that speech communities are no longer
content to serve as mere sources of data. On the one hand, they want to have a say in the re-
search project, determining what is to be recorded and published and what not; and this de-
velopment has, alas, not always served to support scientific work. On the other hand, they
have developed a genuine interest in the data that are produced and processed in linguistic
projects, as that may be the form in which the language survives after it is no longer used by
its speech community. The data here fulfill an important function in documenting the way of
life of a society for posterity, so that future generations of the community may at least learn
how their forefathers lived, and maybe even revive the traditional language (cf. Lehmann
2001). As a consequence of this new function of linguistic data, an awareness of the necessity
to develop standards of quality has arisen, and we currently witness an unprecedented up-
growth of research projects that develop standards and technological facilities designed to
represent, process and archive linguistic data.
16
The progress in all of these respects is, of
course, a direct consequence of the fact that linguistics has, during the same period, increas-
ingly become an empirical science.
The quality of data concerns their user, and to this we will come in the next section, but it
also concerns their producer. Ceteris paribus, data whose content and form satisfies esthetic
and spiritual standards are more highly valued than junk data. What appears to be a truism for
philologists and for members of a community whose traditional language is threatened by
extinction is an unwonted thought for most linguists. People who are professionally more in-
terested in structure than in content tend to neglect such qualitative criteria. However, where
resources are limited – and they certainly are in the documentation of an endangered language
– quality of the data is the essential selection criterion besides representativity (see Lehmann
2001, section 5), and it is the professional task of linguists to respond to such demands.
4.2.2. The relation of data to the user
4.2.2.1. The relation of the data to the researcher
As to the user of the data, we may distinguish between the researcher who takes them for
granted and the addressee of the research. Starting with the researcher, we can distinguish
between the research itself and the report on the research delivered to the consumer. In the
research itself, the data are used either as the basis for induction or as the test of theorems that
were deduced. In the report, the data play the argumentative role of evidence for the theory
(cf. Simone 2001, section 2.6).
Let us first come back to the literal meaning of the term datum, “given.” In real life, the
researcher is, of course, not merely a passive recipient of the data. It was said in section 2.2
that data are not a specific kind of thing waiting to be discovered by the scientist. It would
15
Iannàccaro (2000, section 6f) insists that informants just like linguists have their linguistic
theories which may shape the data they furnish.
16
The D
O
B
E
S (Dokumentation bedrohter Sprachen) program of Volkswagen-Stiftung, which
has been running since 2000, is a representative example.
Data in linguistics
15
therefore be naïve to assume as a normal course of things that the ever-attentive scientist hits
upon a set of data and then feels impelled to develop a theory that accounts for them. This
may happen from time to time, but even then he has the choice of ignoring the data. In general
it is the scientist’s epistemic interest that triggers the research, including the supply of data.
The essential difference between an empirical and a logical science is that the object area
of the former has an ultimate substrate outside the scientist, and insofar its properties do not
depend on him. Of course, the scientist defines his object area, delimits it, and as we shall see
in section 5.2, he can bring the existence of the data about. But even if he does so and in his
research controls a number of variables, he does not determine the dependent variable, i.e.
that property of the data which he is investigating. This is, at the same time, an essential dif-
ference between empirical and hermeneutic methodology. Cases in which the empirical re-
searcher nevertheless influences his data fall into a variety of categories. The category of
fraud is methodologically least interesting. The other problems with objectivity of data are
essentially bound up with the way they are obtained and will therefore be discussed in section
5.
The greatest practical problem for the researcher is usually that the kind of data required
for the particular research project is not available, so that part of the project is precisely data
provision. While this is usually not a problem for languages spoken in the research team, it
may cost considerable time and money in other cases. The expenses tend to get disproportion-
ate in typological projects, both because these need data from many diverse and often remote
languages and because a typological project should exploit finished descriptions instead of
having to procure and analyze primary data in the first place. This gives rise to the question of
why it is not possible, in descriptive and, a fortiori, in typological linguistics, to make use of
available data instead of having to provide them in the course of the research in question.
One must say at the outset that linguistics is not the only science that has this problem.
Probably 95% of all the projects in the empirical sciences, from chemistry over neurobiology
and sociology to psychology, produce their own data on which they base their research. Very
seldom indeed does a project take up the data of another project to either examine them criti-
cally or to use them with a different epistemic interest. On this background, the linguist could
just reject the demand of using other people’s data.
17
The motivation for requiring the avail-
ability of independent data in linguistics probably stems from the fact that the processing of
linguistic data to the point that they can serve as the basis of higher-level analyses is ex-
tremely laborious. While the natural sciences are accustomed to automatizing their data proc-
essing to a high extent so that it costs more material and machine-power than manpower,
automatization of data processing in linguistics is rather underdeveloped. First of all, trips to
remote places and months of fieldwork may be necessary to obtain a sizable corpus of raw
data. Secondly, processing linguistic raw data to the point that they appear as texts with sev-
eral tiers of analyses plus translation, as described in section 6.2, requires a linguist who
knows the language. And finally, even such a person takes years to fully analyze a reasonable
corpus of texts. It thus becomes intelligible that researchers, especially those with more theo-
retical interests, do not always want to start from scratch, and funding agencies tend to feel
the same.
17
And, true enough, some researchers do not make their data available to others.
Christian Lehmann
16
Such considerations lead to the elaboration of representative corpora of languages, such
as Svartvik and Quirk (1980) for English, the Mannheimer Corpus for German, A
DMYTE
for
Spanish, Archivio dell’Italiano Parlato and so on, and in the past two decades they even gave
rise to a new branch of linguistics, viz. corpus linguistics. As a result, both very large corpora
for some European languages and standards of elaborating such corpora have been developed.
And there can be no doubt that such corpora are increasingly being made use of in empirical
research. The problem is, however, that progress in science leads to new kinds of questions,
questions which the producers of the corpus could not foresee and therefore did not think
about tagging their texts for. Currently we have arrived at the point where users require the
corpus, with all its sophisticated derived representations, to come along with the raw data that
it is based on, i.e. they want to check the data on the original videotapes (cf. section 6.1). At
this moment, it is hard to prognosticate whether satisfying this demand will solve the problem
of the insufficiency of “free” corpora which the investigator has not tailored to his specific
problem. It must be admitted that apart from some corpora that are frequently used by differ-
ent researchers, there are also large data cemeteries, collections of data that were gathered and
processed with enormous expenditures for some specific research project and which are com-
pletely useless after the end of that project. And there is no doubt either that the use of infor-
mation technology, i.e. digital representation of the data, has tended to aggravate rather than
alleviate this problem.
The deficiency of available data on most languages of the world is possibly even direr
when it comes to secondary data. This is both a problem of quality and of quantity. Linguis-
tics is still a relatively young science; in the beginnings scientific standards were not particu-
larly high, and there has not been sufficient manpower to describe the thousands of existing
languages in breadth and depth. As a result, many questions of contemporary typology and
universals research do not find an answer in available grammatical descriptions.
Many of the problems in this area are methodological problems, in particular problems of
standardization. The form in which linguistic data should be presented and in which corpora
should be arranged, and the structure that a grammatical description should observe and the
kinds of questions that it should answer are routine issues that are amenable to standardization
to a much higher extent than many linguists appear to believe. At the moment, we are still at
the stage where 30% of the linguists think they can get away with a phonetic representation
which ignores IPA and where everybody thinks up his own conventions of interlinear mor-
phological glossing. It would contribute to the emancipation of our science if basic methodo-
logical operations that are really routine could automatically obey an established standard.
This would free linguistic work for those tasks that really require mental energy.
4.2.2.2. The relation of the data to the addressee
Finally, we have to look at the function of the data for the addressee of some scientific report.
If the addressee is more interested in the theoretical results of the research than in the data it is
based on, then he may wish to join the author of the report in taking the data for granted. At
the level of interpersonal communication, his consideration of the data will then depend on
the confidence that he has for this author and his sources of data. In anticipation of this atti-
tude of the consumer, many linguistic publications do not include their data or at most rele-
gate them to an appendix.
On the other hand, it is possible that the addressee of the research report wants to check
the data, be it that he mistrusts the author, be it that he develops an interest in the data that
Data in linguistics
17
goes beyond the function that the author had destined them for. Such a reader requires full
explicitness in the presentation of data, including both verifiable information on their prove-
nience and such a representation of their linguistic structure that is sufficient for their con-
trolled understanding. Here we again hit upon the issue of the “datum for its own sake.” Data
are representations of such aspects of the research object that correspond to the epistemic in-
terest of the researcher. He cannot foresee the diverse epistemic interests of his addressees,
and he cannot exhaustively represent in his data every aspect of the epistemic object. This is a
practical problem that can only be solved by steering a middle course. If the researcher makes
use of a published corpus, then he has the right to simplify the representation of the data for
the sake of his epistemic interest, provided that he refers the addressee to the original. If the
research is based on original data, then it is the duty of the researcher to make them available
in such a form that the addressee of his report can fully control them (Himmelmann 1993).
This involves standardized linguistic representations of the data, which we will come back to
in section 6.2.
For the addressee of a research report, data may also fulfill a function of visualization and
illustration. However, this is only a secondary function of linguistic data. Their primary func-
tion as stated in section 2.2 is to serve as the basis of induction and as the test for deduction.
The expository or illustrative function is actually the primary function of examples, not of
data. In other words, the mere fact of being used as an example in the exposition does not
confer the methodological status of a datum to some linguistic expression. Some examples
come along with a claim of being data, others not.
5. Obtainment of data
In the last section (4.2.2), the methodological relation between the data and the researcher was
introduced. We now have to deal more in detail with the ways of obtaining linguistic data (cf.
Iannàccaro 2001). We will concentrate on primary data in the sense of section 3.3 and on the
ways in which the method of data provision affects their quality.
5.1. Introspection
The linguist who is a speaker of the language he describes can himself produce and interpret
data of this language. The method of using one’s own language competence in both these lin-
guistic operations and in the associated metalinguistic operations like grammaticality judge-
ments, paraphrases and the like, is called introspection. It has a venerable tradition in linguis-
tics, as the forefathers of logical linguistics, Plato and Aristotle, already relied on it (see sec-
tion 4.1).
Taking up the distinction between production and understanding, we can observe that the
role of introspection differs. While some linguists have requested, in more or less unsystem-
atic ways, grammaticality judgements of fellow speakers for example sentences the linguist
had thought up, the interpretation of primary data originating in his speech community has
mostly been regarded as the professional task of the linguist that he must respond to and for
whose completion he does not depend on others. Speech recognition experiments might ap-
pear to be an exception to this generalization. These, however, do not aim at learning about
the meaning that members of the speech community assign to some utterance produced by
Christian Lehmann
18
another member. Instead, the meaning of the test sentences is presupposed among the con-
trolled variables, and the research interest is in the mechanisms that subjects apply in inter-
preting them.
Even if it is granted that a linguist (who uses the hermeneutic method) is a professional
interpreter, it is worth noting that this capacity of the linguist has its limits even for his native
language. I am referring to linguistic varieties not covered by his competence. Most conspicu-
ous here is the problem of interpreting child language data, where the hermeneutic intuition of
the researcher may fail and possibilities of metalinguistic interaction with the native speaker
are limited.
18
As for the production of example sentences upon introspection, this has been an estab-
lished custom in descriptive linguistics to our day. To the extent that linguistics is a logical
discipline, this is unobjectionable (cf. section 4.1). However, introspection has been treated as
a safe empirical method, under the pretense that a linguist’s speech behavior and grammati-
cality judgements should be (at least) as good as those of any other native speaker, and more-
over the data he produced could be counterchecked by the other linguists whom he addresses
with his research report, so that objectivity was guaranteed. It has become clear for some time
now that these suppositions are false and that this use of introspection is a misuse of the con-
cept and associated ethos of empirical science. First of all, linguists are members of a closed
circle in their speech communities who share a sociolect that is narrowly delimited, subject to
traditional normative standards and nothing less than representative of their language. Second,
the procedure in which a linguist produces data on which he constructs a theory which he then
tests by these data is, of course, circular. The data do not actually fulfill any control function,
and the procedure has nothing to do with scientific method. And last but not least, few lin-
guists have escaped the temptation to dress the data they produce according to the theory they
cherish.
The net result of all of this is that introspection is necessary and useful as a heuristic tool
in linguistic work, but it is not part of empirical methodology, and the data thus produced
have no status in empirical research besides illustrating what the linguist theorizes.
5.2. Production vs. discovery of data
Coming to the serious ways of obtaining linguistic data, there are essentially two of them:
data may be found or may be produced.
19
Of course, data that are found have been produced
at some point. However, what is important here is that the researcher may participate causally
in the production of the data or may just come across pre-existent data. Disciplines differ cru-
cially in this respect. In traditional archaeology, data accepted by the methodology must be
pre-existent and discovered.
20
Data produced under the influence of the scientist have the
18
While interpretation of child language data is certainly possible to various degrees, their
production upon the researcher’s introspection is outright impossible.
19
A related distinction is made in Iannàccaro (2000, section 3) and Iannàccaro (2001:25):
‘phenomena,’ which are found (without being searched for), are distinguished from ‘data,’
which are searched (or produced).
20
Modern archaeology employs methods of the natural sciences, and here things are again
different.
Data in linguistics
19
status of fakes. In neurolinguistics and experimental psychology, on the other hand, all the
data that are of relevance are produced under the control of the scientist. And in a natural sci-
ence such as neurobiology, a certain rat brain is not an interesting datum in itself, but only if a
certain area of it exhibits a certain change of color produced by the experimenter. The imme-
diate conclusion from this is that both kinds of data are well-established in scientific method-
ology. It may therefore cause no astonishment if both of them are used in linguistics, too.
In the purely empirical sciences, which do not have the hermeneutic component which
linguistics has, reliability of research methods is a must, and it entails as a corollary that re-
sults must be reproducible. Consequently large amounts of similar data are produced so that
one can apply statistical methods to them. This kind of approach does exist in linguistics, too,
chiefly in neurolinguistics, experimental psycholinguistics and in statistical linguistics. There
are, however, limitations and drawbacks to such methods which will be mentioned in section
5.4.2. Much linguistic research is devoted to data that are not reproducible, be it for contin-
gent reasons, because the factual preconditions for their production can no longer be met, be it
for theoretical reasons, because they do not have the status of tokens, but of a type (cf. section
3.3). Although these two situations are methodologically totally different, they do share the
uniqueness of their data. This is typical for a science that has a share of hermeneutics: The
epistemic object is a manifestation of the mind of an individual human being, and, insofar, it
is not measured and no induction is applied to it, but instead it is understood.
In other disciplines, the distinction between produced and discovered data is handled with
utmost diligence. In neurobiology, no ambiguity ever arises over the issue of whether the
color of the brain area was there before or after the experiment. In this respect, there is fault
with much linguistic work. Especially in work dealing with grammar, there is a tradition go-
ing back to antiquity for the researcher to use, side by side, example sentences that he found
in a corpus and examples that he has produced himself. And if examples are taken from spon-
taneous recordings, they are normally edited, because speech errors and the like are of no in-
terest to the grammarian. Normativism lurks behind every corner, and objective data become
indistinguishable from illustrations of what the researcher thinks should be the case.
Primary linguistic data are tokens of linguistic signs. Human beings have two converse
relations to them: either as a speaker or as a hearer. The linguist takes the same two perspec-
tives, as shown in Table 3. In the perspective of the hearer, he is confronted with utterances
produced by somebody else. He analyzes their form and structure, interprets this and thus
arrives at the meanings and functions carried by the data. This is the semasiological perspec-
tive, which is typical of structural linguistics. In the perspective of the speaker, the linguist
starts from some cognitive or communicative function which is to be fulfilled by linguistic
signs. The utterances produced in this way are functional variants of each other, and so the
linguist sees which structural means the language uses to fulfill such a function. This is the
onomasiological perspective, typically taken in functional linguistics.
Table 3 Two perspectives on linguistic data
viewpoint basis semiotic operation perspective
hearer forms and structures interpretation semasiological
(“structural”)
Christian Lehmann
20
speaker cognitive and communicative
functions
production onomasiological
(“functional”)
A comprehensive description of a language system is arranged either by structural or by
functional criteria, and partial descriptions are devoted either to some structural device or to
some functional domain. The descriptive linguist is therefore in need of data that share either
their structure or their function. Natural language users, however, are only attentive to struc-
tures and functions in their respective contexts. They do not by themselves aim at producing
discourse that only fulfills a given function, or at only understanding discourse that exhibits a
certain structure. Such one-sided interest is exclusive of a professional approach. The linguist
does have ways of selecting data of the kind that correspond to his approach. However, just as
natural language users are both speakers and hearers, the linguist must be aware that the se-
masiological and onomasiological approaches complement each other; to take only one of
them gives a biased picture of the language. With this in mind, we will examine, in the fol-
lowing two sections, the use of linguistic data in the two approaches.
5.3. Spontaneous data and the semasiological approach
The semasiological approach to linguistic description presupposes a large corpus. Two en-
tirely different methodological situations must be distinguished here. The first is defined by a
corpus language. For languages such as Hittite and Accadian, we dispose of large, but finite
corpora. We cannot change this empirical situation, so if we want to describe such a language,
we have to take the semasiological approach. The other situation can be characterized as the
exploration of a corpus whose production was triggered by the researcher in the first place. In
this, he may pursue diverse purposes. The linguist may want to document a language in dan-
ger of extinction. Then the task is to produce a corpus of the language that should survive it
and be available to future scientists and laymen alike. Or he may need data that represent a
particular variety of a language, as for instance when child language data are recorded in or-
der to investigate a particular problem of language acquisition.
The two situations have a problem in common, which is the representativity of the data in
the corpus. In the case of a corpus language, obviously only the written language is repre-
sented. The lack of oral communication means that not only the basic mode of communica-
tion, including all of the phonetics and most of the phonology, remains unknown but also –
especially for a language from an ancient society, where only a small percentage of the popu-
lation was literate – what is quantitatively the bulk of communicative events remains unrepre-
sented. Written communication is often restricted to very specific genres, not only in antiq-
uity. For instance, for many earlier linguistic stages or extinct languages of the Americas, we
only have catechisms and similar religious literature. On the basis of such data, one can hope
to reconstruct only a very approximate image of the language.
If it is the linguist who triggers the spontaneous production of a corpus, the problem of
representativity should, in principle, be solvable. If research is limited to a well-defined lin-
guistic variety, as in the example mentioned before, the main problem is usually a practical
quantitative problem, in the sense that the sample of different idiolects must be large enough
to warrant generalizations concerning the linguistic variety as such. In the case of the docu-
mentation of a language, the problem of representativity has not been solved to this day. Lin-
Data in linguistics
21
guists who described languages in fieldwork have concentrated, again and again, on a very
few genres of texts, viz. myths, tales, autobiographic stories, in short: narratives. To be sure,
this happened for good reasons: These speech events are easy to record, and they tend to be
well structured and to contain a relatively high portion of complete and grammatical sen-
tences. However, they are not at all representative of communication in the community, be-
cause they are essentially monological, while most communication, especially in scriptless
communities, is dialogical or polylogical. The problem of what constitutes a representative
corpus of a language is a new one in linguistics and is a challenge both for linguistic theory
and methodology and for practical linguistic work. (See Lehmann 2001.)
There are essentially two ways of linguistically analyzing a corpus. If one is interested
only in a particular structural feature, for instance the genitive and its functions, then one
scans the corpus for all genitive forms, produces a concordance of them, classifies the exam-
ples found and comes up with a semasiological analysis of this structural device. If, on the
contrary, one aims at a comprehensive description of the language system, the choice method
is to analyze each successive sentence of the corpus as regards its internal structure at all lev-
els and its linguistic and extralinguistic context, to assemble the structural categories, relations
and processes of the language in this way, to classify them by structural criteria and then to
assign each structural device its functions. The approach is well-established in the philology
and linguistics of corpus languages. Linguistic descriptions of languages such as Accadian
and Hittite have been elaborated essentially in this way.
The approach was formalized in structural linguistics during the first half of the twentieth
century. The two principal operations of segmentation and classification are applied at the
lower levels of linguistic structure, and the result is a structural description that covers at least
the phonology and morphology. At the higher levels of linguistic structure, i.e. the syntactic
and textual levels, the recognition of patterns and the correct assignment of a given token to a
pattern involve procedures of interpretation that are not easily formalized. Here again the
hermeneutic approach comes in, which offers both the advantage of perceiving contextual
relations, disambiguating polysemous or homonymous structures and figuring out the deeper
sense of an utterance, and the corresponding drawback of subjectivity.
Given that in language, structure serves function, no comprehensive analysis of linguistic
structure without reference to its function is possible. Some schools of American structural
linguistics made it a point to apply “cryptanalysis,” i.e. to come up with a structural analysis
of primary data without knowing their meaning. These attempts must be considered failed.
The story of script decipherment is rich in illustrative test cases. All cases of successful script
decipherment involved some kind of historical or archaeological information or even a bilin-
gue in addition to the texts themselves. Wherever such information is not available, as in the
case of the Indus Valley script, any linguistic analysis fails.
It is quite a different issue whether one needs to be a native speaker of a language in order
to analyze it. The answer given to this question by the success story of linguistics is a clear
“no.” Apart from some peripheral items such as nursery rhymes, whose knowledge may serve
as the ultimate touchstone of the native speaker, many persons (including linguists) have ac-
quired a full, native-like command of a second language. And this is not even necessary in
order to do a linguistic description of a language, since understanding and controlled interac-
tion with native speakers may serve the same purpose. Witness are dozens of excellent gram-
mars of languages not mastered by their authors.
Christian Lehmann
22
Human beings, including linguists, are not genetically equipped to focus on linguistic data
that have a certain structure, and are fallible in this task. Even researchers of good will are not
immune to uncontrolled distortion of their data. It has been shown (Cutler 1981), for instance,
that collections of speech errors that were observed under non-controlled circumstances are
often not reliable because of the subconscious hermeneutic interaction of the person who
notes them. That is, people, including linguists, subconsciously change their perceptual input.
It is also well-known both to anthropologists and to linguists that the kind of participating
observation that is typical of much fieldwork inevitably distorts the data; and even if the
fieldworker is aware of it, he cannot eliminate the bias altogether.
The procedure of scanning the corpus for data that have a given structure is, in principle,
automatizable. Tools that do this service are available at diverse levels of sophistication. The
basic level affords just a search for certain allomorphs or word-forms and makes a concor-
dance of them, while the most advanced level involves algorithms of speech recognition and
grammatical analysis. Needless to say, tools of the latter kind are available only for a handful
of languages. For all the other languages of the world, the corpus will first have to be tagged
for the kind of structural information that one may want to retrieve. This, however, presup-
poses just the kind of structural analysis that we are talking about.
The advantage of working with a corpus is, of course, the enhanced objectivity of the data
and of all the research that is based on it. In comparison with the other approaches, the possi-
bilities for the researcher to manipulate the data are minimized. Another great advantage is
that a corpus the researcher has not produced himself may be varied, heterogeneous, full of
surprises and a constant source of inspiration. Exposing oneself to spontaneous data is, in fact,
the safest way of discovering those categories of a language that are peculiar to it and that the
researcher did not expect. The heterogeneity of spontaneous data has, it is true, two sides.
Multiplicity and richness is the positive side. The negative side is wild variation. It is the task
of the linguist to systematize and interpret variation. But a good deal of the variation present
in a corpus is not due to corresponding differences in function or in the context, but is just
dysfunctional: idiolectal idiosyncrasies, dialect differences that one would factor out if one
could control them, false starts and other kinds of speech errors with their repairs etc. And the
undeniable drawback of a corpus is its incompleteness. Certain lexical items, morphological
forms and syntactic constructions will be lacking even from a very large corpus. However,
this just confirms what was said above about the complementariness of the semasiological
and onomasiological approaches.
5.4. Generation of data and the onomasiological approach
In the onomasiological perspective, the researcher wants to know how a certain cognitive or
communicative function is fulfilled in the language, and the task is to obtain primary linguis-
tic data of utterances that fulfill it. This presupposes a theory of the cognitive and communica-
tive basis of language which is subdivided according to functional domains such as Concept
Formation, Reference, Determination, Possession, Spatial Orientation, Temporal Orientation,
Participation, Interpropositional Relations etc. (cf. Lehmann in press, section 2.2). Each do-
main is spelled out down to the level of typological grammatical categories. Depending on the
specific research interest, for instance collecting new data in the domain or classifying avail-
able data by some functional parameter, the concepts are then operationalized in the form of
questionnaires, example sentences, test frames and the like. Since our interest here is the ob-
Data in linguistics
23
tainment of data, the methods that are more appropriate for the classification of data, espe-
cially test frames, will be foregone, but a few of the others will be singled out.
5.4.1. Elicitation and translation
In working with informants, an established elementary method of obtaining data in a prede-
fined functional domain is to elicit them with the help of metalinguistic procedures. If a mor-
phological paradigm is wanted, then the grammatical parameter in question is assigned each
of its values in turn, and the informant is asked to provide the corresponding forms. An analo-
gous procedure can be applied at the level of syntax, by transforming a sentence into a mini-
mally different one according to some relevant functional parameter.
The translation method consists in preparing example sentences of the background lan-
guage (i.e., the regional lingua franca that the linguist and the informant use for communica-
tion) and to ask the informant to translate them into his language. The example sentences have
systematic paradigmatic relations to each other so that they cover the expected variation in the
functional domain in question.
21
However, in its simple form, the translation method is intrinsically invalid. If the task is to
find out those grammatical categories of the target language which render certain functional
categories, it is methodologically inappropriate to present the latter in the disguise of the
grammatical categories of another language, as this obviously leads to interference from the
latter. On the other hand, the method has a couple of advantages, and it is therefore worth
refining. One way of doing this consists in translation questionnaires (see, for example, Dahl
2000, appendices). Here, characteristic little stories or situations are constructed, in which the
sentence to be translated is embedded. The context is configured in such a way as to force the
association of that sentence with the cognitive category which is at stake and whose expres-
sion in the target language is to be tested. In the original version of the questionnaire, the
category would appear in its English grammatical manifestation, but that is suppressed by
presenting its host word as a mere lexeme, without any grammatical categories and, in par-
ticular, without any hint to the grammatical category being tested. This is, of course, done in
order to minimize interference from the background language used. The following is a typical
example from such a questionnaire:
Perfect questionnaire (Dahl (ed.) 2000:803, #37)
It is cold in the room. The window is closed. A asks B:
You
OPEN
the window [and closed it again]?
The example presupposes a functional concept which may be described as “temporal lo-
calization of an event in the immediate past prior to the speech act such that, not the state logi-
cally resulting from the event itself, but a physical consequence of it persists at speech act
time”; and it is asked which structural category the target language uses to express it. Some
21
In its simplest form, the method goes back at least to dialectology (see, for example, Wei-
jnen et al. 1975-9). In a more modern form, it underlies the series Archivo de Lenguas
Indígenas de México launched by Jorge A. Suárez and now edited by Yolanda Lastra
(1974ff). Here, the documentation of a language consists of the translation of a set of several
hundred standardized sentences into the target language. The sentences are chosen in such a
way as to maximize chances that their translations will exhibit the central grammatical catego-
ries and vocabulary of the language.
Christian Lehmann
24
languages (like German) would use the perfect tense here; others (like Spanish) would use the
simple past; yet others (like Yucatec Maya) would prefer the perfective aspect.
With both the elicitation and the translation methods, the responses of the informant are
recorded, analyzed with his help and counterchecked with other native speakers. Both meth-
ods are frequently applied in fieldwork on underdescribed languages. They are popular be-
cause they are inexpensive in every respect. To a certain extent, they are necessary to system-
atically complete data obtained by other methods. However, as has long been known, they
are, to some extent, both unreliable and invalid. They are unreliable because the linguist, the
informant and their relationship are sources of error which render the data faulty. Elicitation
and translation are more a hermeneutic than an empirical method, as the two persons in their
interaction jointly construe some meaning. The two methods are invalid to the extent that they
are meant to reveal the grammatical categories that the language possesses. In fact, they only
reveal such categories that the analyst expects and therefore codes in his questionnaires, ex-
ample sentences and paradigmatic operations. It is therefore crucial that the onomasiological
method does not rely on the grammatical categories of the analyst’s language, or on any
grammatical categories at all, for that matter. Instead, it must rely on a universal (i.e., lan-
guage-independent) system of cognitive and communicative functions. To the extent that lin-
guistics does not (yet) dispose of such a system, it cannot be guaranteed that these two meth-
ods will discover the grammatical categories of the language. Consequently, what is true of
any scientific method at all is a fortiori true of the methods of elicitation and translation: They
must never be applied in isolation, but must always be complemented by other methods.
5.4.2. Induced speech
Sometimes data that are relevant to the research topic are too rare in the corpus or otherwise
hard to come by. Another set of methods within an overall onomasiological approach involves
induced speech, that is, the elicitation of linguistic behavior by non-linguistic stimuli. The
Max Planck Institute for Psycholinguistics at Nijmegen has been developing, over the years, a
sizable set of tools, kits and experiments to be employed for this purpose in diverse cognitive
and communicative fields. One type of method involves the representation of little scenes
with puppets or by silent movies, which are then to be described or retold by the native sub-
jects. There may also be communicative problems to be solved, such as the task of orienting a
fellow in space or instructing him to mount a device. All of these methods presuppose a cer-
tain functional domain and a set of cognitive or communicative operations in it. The setup of
the experiment is designed and the task is defined in such a way as to maximize chances that
the linguistic solution to the task will make use of the grammatical devices that the object
language possesses in that area.
Similarly, speech errors are valuable data for reconstructing the mechanisms underlying
speech production. There are large corpora of speech errors, but the conditions under which a
datum was entered into the corpus are often opaque, so that no statistical methods can be ap-
plied. Then experiments may be conducted in which subjects are prompted to produce speech
errors of a certain kind, for instance metatheses, which are then sufficient to develop system-
atic hypotheses on their origins (Baars 1980).
The advantage of methods of induced speech against those methods which involve
metalinguistic elicitation and translation is that they exclude interference from other lan-
guages. However, the experimental setting itself is not always entirely natural, leaving aside
that the mere fact of being in an experimental situation is bound to trigger uncommon linguis-
Data in linguistics
25
tic behavior. Moreover, methods of induced speech have the disadvantage that they are rela-
tively costly in terms of time and money.
6. Representations of data
The ultimate substrate of linguistics is speech events. These are directly observable and may
be recorded. However, just like in other sciences, the data of linguistic research are almost
never tokens of the ultimate substrate itself, but only representations of it. This is essentially
due to the volatility of speech events, but also to the fact that linguistics is only interested in
the linguistic aspects of speech events – those aspects that are semiotic in nature. We saw in
section 2.2 that data are representations of the epistemic object, and consequently they are
signs. Linguistics differs from other disciplines in that its epistemic object itself is semiotic in
nature, so that the object and representations of it may become indistinguishable. The two
central modes of representation are the auditory and the visual mode, and to some extent they
are interconvertible with preservation of those features that most linguists are interested in. As
a consequence of this situation of the data, linguists have worried very little about whether a
particular datum was an original or a derived representation. In this section, we will sort out
the relationships of the various representations to each other.
6.1. Raw data
Language is an activity, not a (static) object. The ultimate substrate of linguistics consists of
second order entities, not of first order entities. This is true regardless of whether the speech
event in question is one of speaking or of writing. Therefore, the closest, most faithful render-
ing of the ultimate substrate is a sound movie. A sound movie represents the process in which
the original utterance was produced, with its hesitations and editing operations. It represents
the complete phonetics of a spoken utterance, including pauses and prosody. It shows the
paralinguistic communicative behavior of the speaker, with his mimics and gestures. The
movie represents the whole speech situation, with the addressee and his reactions and the ex-
tralinguistic context which is presupposed and referred to by the deixis and which is some-
times changed by speech, for instance in commands. In short, the movie renders most of what
speakers naturally make use of in producing and interpreting speech.
Needless to say, a movie is only a representation, not the original. At any given point in
time, the spectator only sees the scene in one perspective. Most of the time, this is the per-
spective of the addressee, not of the speaker. Since only the auditory and visual senses are
involved, the spectator does not feel or smell what the speaker and hearer feel or smell. And
there are various other reductions and distortions in a movie. Nevertheless, it currently pre-
sents the most faithful way of rendering a speech event. For many purposes inside and outside
the linguistic discipline, especially for the documentation of endangered languages, but also
for various didactic purposes, the best data are raw data in this sense.
Although the auditory and visual modes are interconvertible to a certain extent, the proc-
ess of converting a sound movie or an audio tape into a symbolic representation and deriving
various other symbolic representations from the latter is unidirectional. That is, as long as the
raw data are available, one can always fall back on them and distill better secondary represen-
tations from them. The converse does not hold: Once the original recording is lost and only
Christian Lehmann
26
symbolic representations are left, certain questions about the original speech event will always
remain unanswered.
6.2. Symbolic representations
The epistemic object of linguistics has many facets and is capable and in need of many levels
of abstraction in order to be fully understood. Processing linguistic data therefore essentially
involves their representation at diverse symbolic levels. Depending on their particular epis-
temic interest, linguists represent an utterance at least
22
at the levels enumerated in Table 4.
Table 4. Levels of representation of linguistic data
n° level of representation code and symbols of representation
1 segmental phonetic IPA
2 prosodic phonetic intonation curves, stress levels, etc.
3 lexical-phonological morphophonemes, morpheme boundaries
4 orthographic standard orthography
5 morphological interlinear gloss with vocables of background language
6 grammatical grammatical categories and relations
7 semantic translations in various languages
All of these are written representations, which means they necessitate a change of mode.
This is the first step towards the reification (or hypostatization) of the epistemic object of lin-
guistics. There is no way of avoiding it in scientific work, but one must be aware of it, or else
one will fall victim to what Harris (1980:6-18) calls “scriptism.” The notion of grammar
(τέχνη γραµµατική) as the “art of writing” is deeply rooted in linguistics.
Practically all of these representations, except the intonation curves of n° 2, are symbolic.
They bear complex relations to each other which need not be analyzed here in full. Some of
them, especially n° 1 – 3, render properties of the significans of the language sign, while oth-
ers, especially n° 7, render properties of its significatum, and yet others, especially n° 5 and 6,
represent aspects of its structure. Some representations, especially n° 1 and 2, render proper-
ties of the raw data as closely as a symbolic representation possibly can, while others, espe-
cially n° 3 and 6, are abstractions from more concrete representations. Most representations
render individual linguistic items, while others, especially n° 6, show classes instead of indi-
viduals.
Correspondingly, the conversion of one representation into another one first and foremost
changes the data. Most of these changes are reductions; a few are refinements. For instance,
the conversion of n° 1 into n° 3 involves loss of phonetic information, while the conversion of
n° 3 into n° 5 is accompanied by an upgrading because it involves resolution of homonymy.
Each representation may be used as linguistic data for some purpose; each renders different
questions and answers possible. It is only necessary to keep in mind that while a datum by
definition represents only an aspect of the epistemic object, derived representations reduce
and distort the original even more. Methods which relate derived representations ultimately to
22
It is not important that the list of Table 4 be complete; it suffices for it to be representative.
Data in linguistics
27
original representations of primary data are an important subset of those procedures called for
in section 3.4, which guarantee linguistics the status of an empirical science.
Thus, the two main operations which produce derived linguistic representations are
abstraction and the semiotic operation of coupling a significans with its significatum. As we
saw in section 3.2, the necessity of applying just these two operations constitutes the
uniqueness of the linguistic datum. It is for this reason that the analogies between linguistics
and other sciences sought in section 2.2 come to their end here, and Table 4 has no
counterpart in any other science.
Abstraction involves reduction of variation. In converting representation n° 1 into n° 3,
phonetic variation is neutralized, and in converting n° 3 into n° 5, allomorphy is eliminated.
Some of this variation is part of the linguistic system; some of it is lectal or just irregular.
Coughs, hesitations, slips of the tongue, false starts, etc. tend to be suppressed in the produc-
tion of more abstract representations, at the latest at levels n° 6 and 7. The production of de-
rived linguistic representations therefore also involves editing the original representation (cf.
Simone 2001:57). This is principally done with a view to the norm. The clearest cases may be
observed when texts that were recorded in the field are prepared for publication.
23
However,
the norm is not something arrived at inductively in an empirical science, but something set by
groups of speakers including linguists (see section 5.2 on normativism). The whole process is
directed towards the distillation of system sentences; but if these are what is wanted, there are
shorter ways of getting them, namely by introspection (see section 5.1). In processing linguis-
tic data, two rules must therefore be observed: First, the editing must be transparent; second,
derived versions must not replace, but accompany the original version.
Raw data are the most theory-free form of data that one can get in linguistics. The produc-
tion of all the representations of Table 4 involves some analysis and consequently presup-
poses some theory.
24
Especially at levels n° 5 and 6, representations are conceivable from
which the reader can abduce the entire grammatical theory of the author. This only serves to
once more underline the point that data must not be confused with primary data. Whether
anybody regards any of the representations of Table 4 as linguistic data depends on his pur-
poses and on his conviction that the representation can be related back to primary data by
standard methodological procedures. There is, alas, no clear-cut distinction between data and
constructs; a representation is, by definition, a construct. The most one can say here is that the
progression from raw data to derived representations and finally to secondary data replaces
the primary data by increasingly abstract constructs.
On the other hand, some of the relations between the various levels of representation are
highly regular, so that one representation can be derived from the other by the application of
rules. This means that, despite the restrictions of section 4.2.2, some of the processing of lin-
guistic data is automatizable. Here I am referring, in the first place, to the achievements of
23
For instance, Manuel J. Andrade recorded many Yucatec Maya texts on discs in the 1930s.
Refugio Vermont-Salas provided a close phonetic transcription of them in 1971, which was
microfilmed, but never published. Hilaria Máas Collí produced an orthographic transcription
in 1984, with heavy editing of the original text, and this one was published (Cuentos mayas
yucatecos. Mérida, Yuc.: Universidad Autónoma de Yucatán, 1990).
24
The data are “theory-laden”; see, for instance, Iannàccaro 2000:53f and Simone 2001:60.
Christian Lehmann
28
speech technology and corpus linguistics which pertain to tagging and markup,
25
morphologi-
cal analysis, interlinear glossing etc.
26
Most of these procedures will probably require interac-
tion or control of a linguist for the rest of the lifetime of linguistics, but their automatization is
nevertheless useful for the reasons mentioned in section 4.2.2: computers perform them with
more consistency and efficiency and free linguists for more exacting work.
7. Conclusion
I noted above the fact that neglect of data has been the rule in linguistics, at least since the
beginning of structural linguistics. However, an indulgent interpretation of this fact is possi-
ble: It appears that the theories and methodological concepts of structural linguistics first had
to be developed and tried out on object languages that linguists controlled by introspection,
for which data provision was no problem because any wanted amount of data could be gener-
ated at any moment, and which have an age-old descriptive tradition so that descriptive tools
did not need to be developed from scratch. At the start of the twenty-first century, linguistics
has become mature and now enters a new phase of its development. Thanks mostly to field-
work on diverse languages, to descriptions that are both functional and structural, and to typo-
logical comparison, the discipline is now in a position to approach in a responsible way the
rest of the world’s languages whose methodological situation is less comfortable.
This moment in the history of the discipline happily coincides with new and urgent de-
mands being made on it from outside, viz. from the speech communities of languages threat-
ened by extinction. As if awakening from sleep in a scientific greenhouse, the discipline has
suddenly become aware of the fact that its capacity is urgently needed for the documentation
and description of most of the languages of the world, both for the sake of their speech com-
munities and their interest in their cultural tradition and for the sake of the very database of
the discipline itself. Language documentation has become a slogan in today’s linguistics. As
is usual in such cases, some members of the scientific community who are more flexible in
publicizing the work they had always been doing than in adapting to urgent demands from
outside have adopted the new term as a more effective label under which to sell traditional
linguistic description. Most of us, however, have understood that the new situation demands a
rethinking of our methodological bases. In endangered languages, data constitute a value for
their own sake because they are irreplaceable. Consequently, we need to develop methodo-
logical standards for their scientific and practical treatment so that future generations can
make the best possible use of them.
References
A
DMYTE
(1993 ff). Archivo digital de manuscritos y textos españoles. Madrid: Micronet.
25
SGML, XML and the Corpus Encoding Standard developed by EAGLES represent impor-
tant progresses in the representation of analyzed linguistic data.
26
From among the producers of relevant software, the Summer Institute of Linguistics
(www.sil.org) must be singled out.
Data in linguistics
29
Baars, Bernard J. (1980). On eliciting predictable speech errors in the laboratory. In: Errors in
linguistic performance. Slips of the tongue, ear, pen, and hand, Fromkin, Victoria A.
(ed.), 307-318. New York & London: Academic Press.
Benveniste, Emile (1954). La classification des langues. Conférences de l'Institut de Linguis-
tique de l'Université de Paris 11:33-50. Reprint in: Benveniste, Emile (1966). Problèmes
de linguistique générale, 99-118. Paris: Éd. Gallimard (Bibliothèque des Sciences Hu-
maines).
Bühler, Karl (1933). Die Axiomatik der Sprachwissenschaften. Kant-Studien 38:19-90.
Chomsky, Noam (1964). Current issues in linguistic theory. London: Mouton (Janua lingua-
rum series minor, 38).
Chomsky, Noam (1986). Knowledge of language. Its nature, origin, and use. New York:
Praeger (Convergence).
Coseriu, Eugenio (1958). Sincronía, diacronía e historia. El problema del cambio lingüístico.
Montevideo: Facultad de Humanidades y Ciencias.
Cutler, Anne (1981). The reliability of speech error data. Linguistics 19:561-582.
Dahl, Östen (ed.) (2000). Tense and aspect in the languages of Europe. Berlin & New York:
Mouton de Gruyter (Empirical Approaches to Language Typology, EUROTYP 20-6).
Gougenheim, G., P. Rivenc, R. Michéa, and A. Sauvageot (1967). L'élaboration du français
fondamental (1er degré). Étude sur l'établissement d'un vocabulaire et d'une grammaire
de base. Paris: Didier (Nouvelle éd. refond. et augm.).
Harris, Roy (1980). The language makers. Ithaca: Cornell University Press.
Himmelmann, Nikolaus P. (1993). Material ambitions. Languages of the World 7(2):66-68.
Hofling, Charles A. and Tesucún, Félix Fernando (2000). Itzaj Maya grammar. Salt Lake
City: University of Utah Press.
Iannàccaro, Gabriele (2000). Per una semantica più puntuale del concetto di “dato linguisti-
co”: un tentativo di sistematizzazione epistemologica. Quaderni di Semantica 21(1):51-
79.
Iannàccaro, Gabriele (2001). Alla ricerca del dato. In: Dati empirici e teorie linguistiche. Atti
del XXXIII Congresso Internazionale di Studi della Società Linguistica Italiana, Federico
Albano Leoni, Rosanna Sornicola, Eleonora Stenta Krosbakken and Carolina Stromboli
(eds.), 23-35. Roma: Bulzoni (SLI, 43).
Labov, William (1982). The social stratification of English in New York City, 3rd printing.
Washington: Center for Applied Linguistics.
Lastra, Yolanda (ed.) (1974ff). Archivo de lenguas indígenas de México. México, DF: Cole-
gio de México.
Lehmann, Christian (2001). Language documentation: a program. In: Aspects of typology and
universals, Walter Bisang (ed.), 83-97. Berlin: Akademie Verlag (Studia Typologica, 1).
Lehmann, Christian (in press). Documentation of grammar. In: Lectures on endangered lan-
guages: 3. From Kyoto Conference 2001, Miyaoka, Osahito (ed.),.xxx-xxx. Osaka: Osaka
Gakuin University (ELPR Publication Series C004).
Lieb, Hans-Heinrich (1974). Universals of language: quandaries and prospects. Foundations
of Language 12:471-511.
Löfstedt, Einar (1942). Syntactica. Studien und Beiträge zur historischen Syntax des Lateins.
Erster Teil: Über einige Grundfragen der lateinischen Nominalsyntax. Lund: Gleerup
(Acta Reg. Societatis Humaniorum Litterarum Lundensis, X:1) (2., erw. Auflage).
Christian Lehmann
30
Lyons, John (1977). Semantics. 2 vols. Cambridge: Cambridge University Press (Rep. 1990-
1991).
Seiffert, Helmut (1969-70). Einführung in die Wissenschaftstheorie. Bd. 1: Sprachanalyse -
Deduktion - Induktion in Natur- und Sozialwissenschaften (1969). Bd. 2: Geisteswissen-
schaftliche Methoden: Phänomenologie - Hermeneutik und historische Methode -
Dialektik (1970). München: Beck (Beck'sche Schwarze Reihe, 60-61).
Simone, Raffaele (2001). Sull’utilità e il danno della storia della linguistica. In: Storia del
pensiero linguistico: linearità, fratture e circolarità. Atti del Convegno della Società Ita-
liana di Glottologia, Verona, 11-13 novembre 1999, Giovanna Massariello Merzagora
(ed.), 45-67. Roma: il Calamo.
Stolze, Radegundis (1992). Hermeneutisches Übersetzen. Linguistische Kategorien des
Verstehens und Formulierens beim Übersetzen. Tübingen: G. Narr (TBL, 368).
Svartvik, Jan and Randolph Quirk (1980). A corpus of English conversation. Lund: CWK
Gleerup.
Watzlawick, Paul, Janet Beavin Bavelas, and Don D. Jackson (1967). Pragmatics of human
communication. A study of interactional patterns, pathologies, and paradoxes. New
York: Norton & Co.
Weijnen, Anton et al. (eds.) (1975-9). Atlas linguarum Europae (ALE). 3 vols. Assen: van
Gorcum.
