Language documentation. A program

Abstract

A language is a human skill and its products, comparable to some handcraft like weaving and its products. To document it means to preserve show pieces of it and to provide them with information which enable one to appreciate the skill and to learn (and thus, if necessary, revive) it to some extent.
Documentation of a language is an activity (and, derivatively, its result) that gathers, processes and exhibits a sample of data of the language that is representative of its linguistic structure and gives a fair impression of how and for what purposes the language is used. Its aim is to represent the language for those who do not have access to the language itself. Description of a language is an activity (and, derivatively, its result) that formulates, in the most general way possible, the patterns
underlying the linguistic data. Its aim is to make the user of the description understand the way the language works.
While documentation in this sense has been done for other human skills, for instance in relevant museums, nothing of the sort has ever been suggested for languages. Therefore, many important questions have yet to be answered before one can even think of beginning. A methodologically interesting question is the following: Given
this distinction between documentation and description, what kind and amount of information does the documentation have to add to the raw data so that a linguist should be enabled to come up with a description of the language on the basis of its documentation? From the answer to this and related questions, some proposals for a program of language documentation will be derived.

Full text

CLIPP
Christiani Lehmanni inedita, publicanda, publicata
titulus
Language documentation. A program
huius textus situs retis mundialis
www.christian lehmann.eu/
lehmann_language_documentation.pdf
dies manuscripti postremum modificati
20.02.2001
occasio orationis habitae
DGfS Summer school ‘Linguistic typology’, 30.08. –
11.09.1998, Mainz
volumen publicationem continens
Bisang, Walter (ed.), Aspects of typology and universals.
Berlin: Akademie Verlag (Studia Typologica, 1)
annus publicationis
2001
paginae
83-97

1Versions of this contribution were presented at the Workshop on the ‘Best Record’ of the Max-Planck-
Institute of Psycholinguistics at Nijmegen, 9 – 10 October 1995, and at the DGfS Summer School on
Language Typology at the University of Mainz, 30 August – 11 September 1998. I thank the participants
and Denny Moore for helpful discussion.
Language documentation
A program
Christian Lehmann
Universität Erfurt 20.02.01
Abstract
A language is a human skill and its products, comparable to some handcraft like
weaving and its products. To document it means to preserve show pieces of it and to
provide them with information which enable one to appreciate the skill and to learn
(and thus, if necessary, revive) it to some extent.
Documentation of a language is an activity (and, derivatively, its result) that
gathers, processes and exhibits a sample of data of the language that is representative
of its linguistic structure and gives a fair impression of how and for what purposes
the language is used. Its aim is to represent the language for those who do not have
access to the language itself. Description of a language is an activity (and,
derivatively, its result) that formulates, in the most general way possible, the patterns
underlying the linguistic data. Its aim is to make the user of the description
understand the way the language works.
While documentation in this sense has been done for other human skills, for instance
in relevant museums, nothing of the sort has ever been suggested for languages.
Therefore, many important questions have yet to be answered before one can even
think of beginning. A methodologically interesting question is the following: Given
this distinction between documentation and description, what kind and amount of
information does the documentation have to add to the raw data so that a linguist
should be enabled to come up with a description of the language on the basis of its
documentation? From the answer to this and related questions, some proposals for
a program of language documentation will be derived.1

2
Christian Lehmann
2Cf. Lehmann 1996 for some discussion.
1. Data collection in linguistics
1.1 Data of living vs. corpus languages
As in any empirical science, a basic step in any linguistic work devoted to a language is the
collection of data. To the extent that there has been a methodological tradition in linguistics, there
is no unified role that data collection would play in it.2 First of all, a basic distinction must be
made between corpus languages and living languages. A philology or linguistic description
devoted to a corpus language has to base all of its achievements on a corpus of data which is
either finite or at any rate not extendable at will. Inductive generalizations are limited by the
corpus. Even deductive generalizations must find evidence in the available data, or else they will
be speculations. Consequently, the datum acquires a high value. Much of the scientific work is
devoted to digging out, ordering, purging, preparing, interpreting, archiving, preserving,
representing and transmitting linguistic data. This activity has a high prestige in the philologies.
In Classical Philology, for instance, the best scholars have invested considerable portions of their
lifetime in the edition of the texts. In such disciplines, linguists are either themselves engaged in
this kind of activity or depend entirely on its results.
Things are completely different for a living language. There may be a corpus of data, too; but it
is extendable at will. Consequently, descriptive work often has more data at its disposal than the
investigator wants to look at. If a datum of the desired kind is not at hand, the investigator has
ways of producing it. As a consequence, the value of language data in such a discipline decreases
to a minimum. Data collection seems a completely ancillary and trivial activity in the linguistic
description of a living language.
1.2 Data collection in epistemology
While the situation of the data themselves may be considered as something that a science is
confronted with, but cannot change, there are also reasons in epistemology, or perhaps one should
rather say, in the ideology of science, which contribute to the low prestige of data collection even
in empirical disciplines. The highest goal of any empirical science is a theory. Worthwhile
insights into its subject matter are of necessity general. A particular utterance is of no interest in
itself, but only insofar as it instantiates a syntactic pattern of the language. The syntactic
description of a particular language is of limited value in itself; it acquires higher value insofar
as it represents a linguistic type. A linguistic type in itself is only part of a universal theory of
language, which is what linguistics ultimately aims at. In such an ideology of scientific work,
even descriptive linguistic work has a limited prestige because it is subordinate to explanatory
theoretical work. Collection of data, however, has no prestige at all. It is the lowest step in a
hierarchy, a step that one seeks to leave behind as soon as possible and that one skips more often
than not.
I have called this epistemology of linguistics an ideology, thus implying that there are
metaphysical convictions behind it. One such conviction is that language is an object of nature

3
Language documentation
3“Die einzelnen Sprachen sind nicht als Gattungen, sondern als Individuen verschieden; ihr Charakter
ist kein Gattungscharakter, sondern ein individueller.” (Humboldt 1827-29:289)
4Recall the three-step hierarchy of observational, descriptive and explanatory adequacy in Chomsky
1965, ch. 1.4.
such as the objects of biology or chemistry. This means that the object has no individuality, that
it is of scientific interest not as an individual, but as a member of a species. Considering
linguistics as an empirical science very often implies this conception of the linguistic object.
The conception is, however, not justifiable on scientific grounds. Every language is an activity
grounded in the historical situation of its speech community and also a factor in its history.
Insofar, it is a historical object, and linguistics is a historical discipline (cf. Coseriu 1979,
1980[H]). A historical object has an individuality and therefore commands an interest for its own
sake.3 To the extent that each language is unlike any other language, its description is not
transcendable by generalizations or universal theories. Nor is this a defect on account of which
it would loose in value. This would appear so only from the vantage point of a certain ideology
of science. In the perspective of an educated and intellectually active human being, the historical
object has so much more value just because of its uniqueness.
The epistemological conception of language as a natural or as a historical object does not only
shape the kind of interest that we take in it, but even the presupposed data situation. The
utterances that constitute the basic data of a living language are infinite and replaceable only if
we abstract from their individuality. Utterances, however, are historical objects. They represent
a certain diachronic, diatopic, diastratic and diaphasic variety of a language (cf. Coseriu 1980[P]).
Insofar, scientific data which embody utterances even of a living language are not replaceable at
will. The possibility to replicate a given linguistic datum depends on the viability of the language
variety that it is taken from and on the extent to which scientists have access to it. Data from a
language which is the native idiom of many linguists can be replicated much more easily than
data from a language spoken at the brim of civilization. Data from an extinct language cannot be
replaced at all.
The outcome of these considerations is: Linguistic data which are not easily replicable demand
the same care as the text corpus transmitted by generations of philologists. The corpus linguistics
typical of the languages of antiquity, on the one hand, and the field linguistics typically devoted
to languages of the third world which are far removed from traditional European culture, are two
activities within the broad field of linguistics which are often considered as entirely unrelated.
Surprisingly, they now find themselves sitting in the same boat, united by a common concern for
irreplaceable data which are precious because they represent a historically unique facet of human
life.
In the scientific ideology criticized here, linguistic data collection serves the sole purpose of
rendering linguistic description possible. Linguistic descriptions – i.e., mostly, grammars and
lexicons – derive their scientific status not from the quality of the data they present, but
exclusively from their "descriptive and explanatory adequacy".4 Therefore, they do not present
primary data for their own sake, but only to the extent that this is inevitable in order to
plausibilize the analysis or render it more easily intelligible for human readers. Data are very often

4
Christian Lehmann
5General linguistics here is a bit behind the development in the philologies, in anthropology and literary
studies. See Heissig & Schott (eds.) 1998 for a recent survey.
V-mentum `thing for V-ing, thing that serves to V'
ornamentum `thing for adorning'
monumentum `thing to admonish'
documentum `thing for teaching'
T1. Latin derivations in -mentum
handled with utter negligence in linguistic publications. Similarly, in those circles which regard
linguistics as a natural science, it is not a valid criticism in a book review to find fault with the
data, since what is of scientific interest is not the data, but the analysis.
To do away with this ideology implies the rejection of the view of data collection as an inferior
and ancillary prescientific activity. It does not, of course, mean to deny that data collection is a
methodological step presupposed by linguistic description. Quite on the contrary, it is a necessary
precondition for the latter. On the other hand, if we claim that linguistic data have a dignity of
their own, do we mean that data collection is a self-sufficient goal? This is certainly not the case.
There are, in linguistics just as in other sciences, data cemeteries, large amounts of data
assembled by people who thought they served a purpose in themselves and not used by other
people, who thereby prove the former people wrong. The immediate lesson from this is that
linguistic documentation does not reduce to linguistic data collection.
2. Documentation in linguistics
2.1 The concept of documentation
The idea and the term of language documentation are novel in linguistics.5 Linguists have not
thought of their activity as a documentary one. While there are a variety of things that one can do
to and with a language, such as describing or explaining it, documenting it is not an approach to
language that would be familiar in linguistics. It seems therefore appropriate to start with some
conceptual clarification.
In the past decades, we have come to associate with the concept of documentation the whole field
of information storage and retrieval, of library resources, archiving methods and computer
databases. Many a linguist has a more or less extensive linguistic documentation project which
is essentially an annotated bibliographic database. Documentation in this sense means something
like the systematic collection and representation of records of key information on some class of
objects by which one can either get more information on an object or access the object itself.
For present purposes, this sense of the term, although certainly relevant, is not of central interest.
Instead, a bit of etymology will be helpful here. The Latin word documentum is formed on the
derivational pattern in T1.

5
Language documentation
6Thus, an important specific purpose of language documentation is to serve as a record of the past and
as an element of ethnic identity for future members of the community that has lost its identity as a speech
community but which still recalls that their ancestors had a language of their own.
7In putting it this way, I do not, of course, exclude the possibility of combining the efforts for
documenting a language with those for preserving it. It would be quite conceivable to use the
documentation of a language as a chrestomathy (see §?). On the other hand, elaborating teaching materials
of a language is obviously a different task than documenting it for posterity.
8Cf. Lehmann 1992 on the conception of a language museum.
Accordingly, the literal sense of documentum is `thing that serves to teach'. It also means
`example, proof'. By its etymology, document is thus a functional concept which evokes the
question: who teaches what to whom? We will come back to such questions below. At the
moment, it suffices to note that documenting a language involves producing and storing such
records as may serve to teach one about the language.
The observant reader will have noted that I might have said "... such records as may serve to teach
the language." Although this may be an important mediate goal of language documentation, it is
not its direct purpose. To teach a language involves a presentation of linguistic material which
proceeds from simple and everyday phrases to complex and peripheral constructions and which
helps the learner in assimilating the material and in stepwise building a competence. In order to
achieve this goal, the user has to get involved in speaking the language himself. This idea of
learning by doing is irrelevant to documentation. The primary purpose of language
documentation is to represent the language for those who do not have direct access to the
language itself. What is essential here is not the motivation of the user. He may be a linguist or
a layman, he may wish to find out about the way the language works, may be curious about its
peculiarities6 or may want to apply universal theories to it, or he may wish to learn it. A language
documentation may be put to all these uses, but they are all tributary to the primary use of just
representing the language itself. The essential issue here is, consequently, representativeness and
accessibility.7
2.2 Documentation of a skill
Language is among those activities in which human beings are creative both as individuals and
as members of a historical society. Insofar, it is comparable to such skills as house-building,
music, pottery (cf. Silverman & Parezo (eds.) 1995). By pursuing the analogy, we may note that
teaching somebody the traditional pottery of a society normally involves making him sit down
at the potter's wheel and work the clay. Docum e nting the pottery, on the other hand,
obviously starts from the orderly presentation of selected show pieces of different kinds. In
addition, there may be various kinds of information. There may be photos or a video showing a
potter in the various production phases. There may be explanations about the steps to take, from
the quality of the clay and where it is found up to the temperature and duration of the burning.
There may be information on the uses that the ready pieces of pottery are put to or on potters and
their life. All these kinds of information would appear, e.g., in a good pottery exhibit of an
ethnological museum.8 It is easy to see that while such a documentation is not a course in pottery,

6
Christian Lehmann
it might play a fundamental role in designing one. Imagine a situation where the traditional
pottery of a society has become extinct. If there is a comprehensive documentation of the kind
indicated, then a skilled craftsman should be able to learn and thus revive that pottery from
working through the documentation. Also, a ceramologist should be able, by relying solely on the
documentation, to derive a scientific description of this variety of pottery.
We may now abstract from the example. The documentation of a human skill is goal-oriented in
the sense that it pays primary attention to the things produced by that skill. If the products are
volatile, as they basically are in the case of music and language, then they first have to be
recorded on an enduring medium. That is, documentation of a language does not involve life
representations of native speakers, just as documentation of a music style does not necessitate the
presence of musicians. What we demand is that the show pieces be of high quality and
representative of the variation.
In some cases, a good presentation of ready products may be sufficient for an experienced
craftsman or even a layman to recognize how these things are made. For instance, by studying a
traditional wooden Maya house, it should be possible – given sufficient dexterity – to build one
oneself. In other cases, this would clearly not suffice. For instance, even a trained musicologist
would be unable to find out how a certain variety of Papuan ethnomusic is made if he could just
listen to a high-fidelity record of some piece, but had never seen any of the instruments. Similarly,
the comprehensive documentation of a Hottentot language would include the presentation of a
sagittal cut through the mouth which demonstrates the production of a click.
In some areas, the user of a documentation will still be unable to understand what is going on if
he were just presented with products and movies of the production process. Imagine the
documentation of a religious ceremony. We may see the priest signing the cross over a person.
If we are not familiar with the culture, we will just observe him making a hand movement without
knowing what he was really doing. What we need is an interpretation of the act. While this is
obvious for all symbolic actions, it is even true for many other actions whose purpose is not
evident from directly observable effects. As an example, think of a documentation of traditional
healing methods. From such examples, we may generalize that the documentation of a skill does
not reduce to the presentation of primary data. The user of the documentation must be put in a
position of appreciating what is being done. This requires the addition of interpretive information
which is not inherent in the primary data, but generated by the author of the documentation.
One may regret this necessity in the interest of objectivity. However, the mere collection and
reproduction of raw data is not a scientific activity at all. Moreover, objectivity in documentation
is an illusion, anyway. Since the documentation does not repeat the documented reality itself, but
only represents a sample of it, there is necessarily a process of selection, which in itself is not
objective and which, in fact, can be highly tendentious. A well-known example is the presentation
of hieroglyphic inscriptions of the ancient Mayas by early epigraphers. The calendar was the only
thing in these inscriptions that they had deciphered. Therefore inscriptions with extensive dating
abounded in their documentations, and these were the only parts that were provided by a
translation. As a result, americanists around the middle of our century believed that the whole
spiritual life of the Mayas revolved around measuring the time and observing the calendar.

7
Language documentation
description of language meta-level
documentation of language object-level
language
T2. Documentation and description of a language
Human skills are embedded in the life of the society. Some aspects of traditional pottery will be
understood only if we know about the position of the potter in the society. Similarly, certain
passages in the documented texts and certain words will be understood only if they are provided
with information about the reality being denoted. Although this is a trivial truth, its appreciation
in linguistics has varied. In German linguistics around the second world war, there was a
movement called `Wörter und Sachen' (words and things) which seeked to elucidate word
meanings by factual knowledge about the designated objects of traditional culture and vice versa.
American structural linguistics in its traditional combination with ethnology has cultivated this
line of research, too. The two strands mentioned make us see that the endeavor to fully document
a language transcends the boundaries of linguistics. A language documentation would not be
feasible if ethnographic information were systematically excluded from it.
3. The relation between documentation and description
Documentation of a language is an activity (and, derivatively, its result) that gathers, processes
and exhibits a sample of data of the language that is representative of its linguistic structure and
gives a fair impression of how and for what purposes the language is used. Its aim is to represent
the language for those who do not have access to the language itself. Description of a language
is an activity (and, derivatively, its result) that formulates, in the most general way possible, the
patterns underlying the linguistic data. Its aim is to enable the user of the description to compare
this language to other languages.
In the clearest cases, the difference between documentation and description is a difference in the
logical level. The documentation consists of specimina or representations of its object, while the
description is a discourse on the object. If the object is a language, then the description bears a
metalinguistic relation to the object, while the documentation remains on the object-language
level. This relationship is shown in T2. However, as we shall see shortly, this is really a
prototypical distinction. In practice, the purpose of a documentation is often not served
sufficiently if no explanations are given (cf. §2.2). Explanations, however, belong strictly to the
description. Moreover, in the cases of objects like languages, the documentation involves
representations of the object; and a representation of an object is not easily assignable to the
object-level or meta-level.
Other differences between documentation and description are derived from this basic one. There
is a difference in generality in the sense that the documentation cares for the specific object, while
the description seeks to generalize over it. And there is a corresponding difference in abstractness:
the documentation is concrete, its components are easily perceived by the layman, while the
description is abstract and made for the specialist.

8
Christian Lehmann
9It is just the highest and finest which is not recognizable on the basis of those separated elements and
can only be perceived or sensed in connected discourse. It is only this which we have to regard as the true
and first instance in all investigations which are supposed to penetrate into the living essence of the
language. The dismemberment in words and rules is only a dead artifice of scientific analysis.
It is beyond dispute that linguistic descriptions cannot be understood by laymen. Let us assume
that this is a virtue, because otherwise it would be less clear what the specifically scientific about
them is. There are, however, many linguistic descriptions – grammars and dictionaries – which
are commonly regarded as sufficient pieces of their kind within the confines of the discipline. One
might then ask: if we have such a description of a language, then what do we need a
documentation for? In particular, if we have a generative grammar, then it can generate data for
us, so there is no use in storing data at all.
In theory, documentation and description of a language are mutually independent. One should,
in fact, document a language in such a way that future linguists can derive a description from it
(cf. Himmelmann 1993); and one should describe a language in such a way that future linguists
can produce data on the basis of it. In practice, however, these demands on the quality of
documentations and descriptions exceed the capacity of human linguists. Already Wilhelm von
Humboldt found occasion to write the afterwards oft-quoted passage on grammars available to
him:
Gerade das Höchste und Feinste läßt sich an jenen getrennten Elementen nicht erkennen
und kann nur ... in der verbundenen Rede wahrgenommen oder geahndet werden. Nur sie
muß man sich überhaupt in allen Untersuchungen, welche in die lebendige Wesenheit der
Sprache eindringen sollen, immer als das Wahre und Erste denken. Das Zerschlagen in
Wörter und Regeln ist nur ein totes Machwerk wissenschaftlicher Zergliederung.
(Humboldt 1836:418f)9
Things have not changed much since Humboldt's times. Most of the available linguistic
descriptions, among them some which are held in high esteem in specialist circles, do not allow
one to produce a natural text based on them. This is not to say that such descriptions are without
value. It merely means that they serve a different purpose. Therefore, while we may hope for ideal
descriptions to come forth in the future, for the time being it is safer to produce a language
documentation if one wants a lively representation of how the language really works.
Moreover, the dividing line between documentation and description is not sharp. As I said before,
the documentation does not reduce to a body of raw data as produced by the speakers, but
includes representations of the data, representations produced by the linguist, e.g. a phonetic
transcription, an interlinear morphemic gloss, a translation. If this is so, then the documentation
contains an analysis. It presupposes a description, and vice versa. For these reasons, it is neither
possible nor advisable to separate the documentation from the description.
The description of a language does not reduce to an account of the language system, i.e. a
phonology, a grammar and a lexicon. It also contains an account of the ethnographic, social,
genetic and historical situation of the language (cf. Lehmann 1989). If a full account of the
language system is already something that has been achieved for very few languages indeed, then
a description in this comprehensive sense is truly boundless. I will come back to the problem of

9
Language documentation
limiting the task in §6. At this point, the parallelism between the documentation and the
description should be noted. It was said at the end of §2 that the documentation is not confined
to the formal linguistic aspects of the phenomena, but extends to ethnographic aspects. The same
is, of course, true for the description. Although documentation and description of a language
belong to different levels, they may be analogous as to their internal systematic structure.
4. Representation and interpretation of linguistic data
For a living language, linguistic raw data are video films or, in default of these, audio recordings
of communicative events. They have to be represented at various linguistic levels. The
professional linguist distinguishes a large number of levels, among them the phonetic,
phonological, morphological and syntactic levels. In addition, there are two kinds of translation,
an interlinear morphemic gloss and a free translation. However, the documentation of a language
does not require representations to be this diversified. On the one hand, this would presuppose
phonological, grammatical and semantic analyses, which would require to postpone the
documentation until after the completion of the description. This is undesirable in view of the
many languages which are in urgent need of documentation. On the other hand, the
documentation is not made exclusively for the professional. If the layman can draw profit from
a documentation without a syntactic analysis, then the linguist should be able to do so, too.
With the acoustic rendering of the original utterance, the user has immediate access to the
significans of the language sign. The next thing he wants is an understanding of the document,
i.e., he needs access to the significatum. This can be represented in form of a free translation.
With this, the user can grasp the sense of the text. So far, the representations only account for the
particular utterance. They remain at the level of la parole. La langue becomes relevant only as
a vehicle, just as in everyday life. In linguistic documentation, the language system is paid
attention to for its own sake. In addition to the sheer sound and the sheer sense of the text, we
want to see its linguistic structure represented. At the same time, we want to keep the
documentation as free as possible of descriptive theory, in the spirit of division of labor between
documentation and description and because documentation can, in practice, not always rely on
an available description and because the documentation should be usable by laymen. The
scientifically responsible way of representing the structure of an utterance with the lowest
possible degree of sophistication is to identify the morphs and match each with its meaning. Thus,
the elementary representation of the structure of the significans is an allomorphic representation,
and the elementary represention of the structure of the significatum is an interlinear morphemic
gloss. Accordingly, T3 includes three levels of linguistic representation above the level of the raw
data.

10
Christian Lehmann
10 The latter part of the task is, alas, not trivial, as becomes particularly clear in the case of polysynthetic
languages.
linguistic
representation
free translation parole
interlinear morphemic gloss langue
allomorphic representation
raw data video/audio recording parole
T3. Representation of primary linguistic data
Since people without linguistic training tend to find morphemes and interlinear glosses too
abstract, it would seem desirable to produce an alternate representation for them in which the
allomorphic representation reduces to an orthographic one and the interlinear morphemic gloss
reduces to a word-for-word translation.10 Moreover, a good documentation has to solve the
technical problem of presenting the three linguistic representations simultaneously with the
acoustic rendering.
Let us briefly come back to the notion that one of the purposes of a language documentation is
to serve as the basis for future descriptions of the language. To my knowledge, there are no
investigations of the question of what quantity and quality of data of a language a trained linguist
needs in order to come up with a description of it. There is good evidence from professional
experience that bare raw data with no additional information whatsoever are insufficient. For
instance, all cases of successful script decipherment involved some kind of historical or
archaeological information or even a bilingue in addition to the texts themselves. Wherever such
information is not available, as in the case of the Indus Valley script, decipherment fails.
Similarly, I once participated in an attempt to do a linguistic analysis of a tape recording of an
unknown language. It failed essentially, apart from the first sentence, which we figured out on the
basis of the external information that it was a fairy tale and the subsequent hypothesis that the
sentence probably meant "Once upon a time, there was an X." With video recordings, conditions
for linguistic analysis might be more favorable. However, the purpose of a language
documentation is not to serve as a riddle for smart linguists, but to allow insight into the way the
language works. The easiest way of achieving this is obviously to identify the meaningful
elements and indicate their meaning.
5. Criteria for documentation
5.1 Quality
The criteria for the selection of texts that go into the documentation are essentially two: quality
and representativeness. The quality of a text concerns both its content and its form. As for the
former, the documentation of a language is part of the documentation of its culture. Therefore,
texts whose content is important in the culture, for instance prayers or myths or instructions to
the youth, are more highly valued than, say, routine everyday conversations.

11
Language documentation
11 See Lenk 1996 for some methodological considerations.
Quality of the form means linguistic correctness and aesthetic beauty of a text, as it results from
the linguistic skill of its authors. In the philologies, most attention has always been devoted to
literary texts. While these may be problematic from the point of view of representativeness, there
is no doubt that they range high in quality. From this perspective, the presence of literary texts
in general purpose language documentations and their prominent role in published linguistic
descriptions is certainly justified, although not, of course, to the extent of excluding any other text
genre.
For spoken texts, the quality of linguistic performance includes such aspects as textual cohesion,
stylistic appropriateness, richness in system resources, correctness of constructions and, last not
least, accuracy of articulation. The latter is, so to speak, the acoustic counterpart to calligraphy,
which would have to be sought for if literary texts of a community with a long writing tradition
were to be documented. Needless to mention, the linguist, too, bears part of the responsibility,
namely for the technical quality of the recording.
The problem of the quality of the text is bound up with the issue of purism and the freedom of
the linguist to edit the recorded text. The first thing to be said here is that the more the linguist
cares for optimal production conditions, the less need will there arise for editing. Secondly,
editing can only mean the production of further representations of the recorded text; under no
circumstances must the original recording be changed. Finally, whenever possible, responsible
editing should engage the authors of the text themselves. The decision on the form in which their
text is to be transmitted to posteriority is first and foremost theirs.
5.2 Representativeness
The task to compose a representative text corpus of a language is, again, a novel one in
linguistics.11 Its traditional counterpart in the philologies is known by the name of chrestomathy.
This is a collection of texts designed to facilitate the access to a language for someone who learns
it as a second language. Traditionally, it is essentially an anthology of literary texts. I have already
commented on the task of supporting the learner of a language in §2, and on the role of literary
texts in §5.1.
Representativeness of a language documentation means that it adequately represents the purposes
and ways the language is made use of in the speech community and the structural properties and
possibilities of the linguistic system. As for the latter, there is no direct way of guaranteeing
representativeness. For one thing, the presupposition is that documentation of a language should
be possible without prior analysis of the linguistic system. For another, since the virtues of the
linguistic system are language-specific, there would by definition seem to be no universal method
of bringing them out. It therefore seems best to directly approach the first requirement, viz. the
representativeness of the sample as regards uses of the language. If this is defined on a genuinely
linguistic basis, then it should also, derivatively, provide for structural representativeness.
This purely linguistic basis for the composition of a representative text corpus must be oriented
by parameters which are universal in all languages. These are the parameters which constitute the
speech situation. The essential components of the speech situation are the speech act participants,

12
Christian Lehmann
12 I assume, along the lines of Kabatek 1999, that the parameter of distance vs. proximity, which is
postulated as the central universal parameter of linguistic variation in Koch & Oesterreicher 1990:14, can
be reduced to other dimensions such as those mentioned.
the context, the communicative task, the topic talked about, the code, the channel and the message
(cf. Jakobson 1960). Each of them constitutes a parameter of variation. The parameters together
define a multidimensional space, in which text genres can be placed. T4 gives an overview of the
variation generated by these parameters and of the ways they may be used for the characterization
of text genres.
Some comments are in order here. First, not all of the components of the speech situation are
made use of in T4. The message does not appear because this is just the dependent variable whose
full range of variation we hope to obtain by varying the other parameters. The code does not
appear because it is here equated with the language system, and this we want to keep constant.
Again, it is assumed that code variation below the level of the language system, such as choice
of different sociolects or registers, will follow as a dependent variable if we vary the other
parameters, for instance the social group of the speech act participants or the formality of the
speech situation.12 The same goes for code switching. This is, of course, not excluded by the
premise of focussing on one language system. It, too, should come out as a consequence of
varying some of the other parameters. Likewise, diachronic variation (to the extent that it appears
at the synchronic level) should be obtainable by suitable combinations of the parameters of T4.
For instance, narrative monologs typically conserve older strata of a language, while creative
dialogs are typically progressive in the use of linguistic resources.

13
Language documentation
1. Speech act participants (speaker, hearer, bystander)
Nature (supernatural vs. human being vs. animal vs. none; monolog, dialog, palaver ...).
Social group (sex, age, social status, profession, ethnic affiliation ...).
Roles: symmetric vs. asymmetric (kin, chief vs. citizen ...), intimate vs. stranger.
2. Context of speech act (situation)
Place (church, mill, pub ...).
Time (daytime vs. night ...).
Formality: distance vs. proximity (friendly encounter, work, ritual ...).
Real-life embedding (game, drama ...).
3. Task
Illocution:
narrative (myth, proverb, joke, riddle ...);
instructive (working routine, game instruction ...);
discursive (political/forensic speech, sermon, blessing, curse ...);
interrogative (examination, interrogation ...);
poetic (poem, song ...).
Conventionality:
ritualized (baptism, courtship ...);
conventional (greeting and leave, route directions, official address ...);
creative (event report, dispute, poem ...).
4. Topic
(Traditional/modern; work/leisure, past/future, family/village, mythical figures/persons
/animals/plants ...).
5. Channel
Medium: oral vs. written.
Directness: face-to-face vs. technical transmission.
T4. Components of speech situation and text genres
Given these assumptions, a representative text corpus of a language may be defined as one which
exhibits sufficient variation on each of the parameters of T4. This notion may be refined by
spelling out the way in which the parameters cross-classify and by characterizing some
parameters or values as more central than others. We will come back to this in the last section.
The two essential criteria of quality and representativeness are logically independent and therefore
not seldom in conflict, as already observed in §5.1. Everyday utterances which are typical of
informal encounters are certainly representative of the ways of communication in a society, but
they may be of low quality with respect to cultural importance of their content or accuracy of
articulation. If the language approaches extinction, most of the members of the dissolving speech
community will be semi-speakers, who produce low quality in terms of richness in linguistic
structure. The proper choice under such conditions obviously depends on the purpose which the
documentation is to serve. A comparison with other fields of culture which preserve creations of

14
Christian Lehmann
13 As I am writing this (october 1999), the Volkswagen Foundation has launched a program of
documentation of endangered languages based on just these principles.
the past, such as literature or archaeology, would lead one to conjecture that high quality of the
texts will prove most important for future generations.
This implies that the criteria proposed here for the documentation of a language also bear on the
selection of languages to be documented in the first place. At first sight, it seems clear that
priority should be given to endangered languages. There can, in fact, be no doubt that the
documentation (and the description) of those few hundred languages which are not endangered,
but which nevertheless currently absorbs 99% of all linguistic resources, is actually not urgent at
all and can safely be postponed until all the languages now endangered have become extinct (cf.
Lehmann 1998). On the other hand, it would be hasty to conclude that the priority of documenting
a language is the higher the more the language is endangered. This would be so only if we had
infinite means for language documentation at our disposal. As long as manpower and funds are
limited, large quantities of languages will, despite all our efforts, die out before they have been
sufficiently documented. If, under these circumstances, we decided to dedicate all available means
to whatever language is, at the moment, most endangered, we would always be documenting
moribund languages which only have semi-speakers left. On the one hand, the ratio of gains and
losses is particularly bad for these languages because data gathering is exceedingly difficult and
time-consuming. On the other hand, such a decision would lead us into conflict with our two
documentation criteria. For a language which only has semi-speakers left, neither quality nor
representativeness in documentation can be achieved, because those few speakers do not
remember the language well enough, since they have not used it for most purposes for a long
time.
The conclusion from this is that, as far as assignment of available manpower and funds is
concerned, priority should be given, ceteris paribus, to those languages which are yet lively
enough to allow at least for a minimum standard in quality and representativeness of
documentation. If we have to accept that many languages will die out without documentation,
then let them be those languages which we cannot decently document, anyhow. I know that this
proposal will hurt several colleagues’ feelings; but I know of no viable alternative.13
6. A program for language documentation
There are many languages which are in urgent need of documentation because they are becoming
extinct. Professional experience tells us that it may take a linguist's lifetime or even more to fully
describe a language. If a documentation can be the basis for future descriptions, and if a
documentation can be completed considerably faster than a description, then documentation
projects would be the first choice for many languages in the world today. However, since the
concept of language documentation is novel in linguistics, we have no basis in experience for
making a sensible estimate of the average duration of a language documentation project. Whoever
has worked on the edition of a field-recorded text knows that this is very time-consuming. It
would be easy to spoil the whole enterprise of language documentation from start by setting the

15
Language documentation
standards too high. Therefore, we have to allow for different degrees of completeness in various
dimensions.
The Prague school conception of center and periphery in language will be helpful here. While
T4, if suitably specified, can serve to define dimensions on which maximum variation is desired,
we can, at the same time, arrange the values of each parameter according to their degree of
centrality in the communicative life of a speech community. For instance:
- On parameter 1, a constellation in which both the speaker and the hearer are adult human
beings of the same social group, for instance a married couple, may be taken to be highly
representative, while other constellations are increasingly marginal.
- On parameter 2, a real-life situation is more central than a play.
- On parameter 3, a conventional task such as giving route directions is more central than a
creative task such as composing a poem.
- On parameter 4, oral face-to-face communication will be taken as a core case, while telephone
conversations or letter writing are more at the periphery.
In this way, we will be able to define a gradience between core and periphery in the language. To
spell this out is among the most urgent tasks of our discipline. A documentation project may then
start at the core and proceed to the periphery. However generous or limited the resources of such
a project are, such a procedure would seem to always guarantee the best documentation possible
under the given circumstances.
References
Chomsky, Noam 1965, Aspects of the theory of syntax. Cambridge, Mass.: MIT Press (Special
Technical Report No. 11).
Coseriu, Eugenio 1979, "Humanwissenschaften und Geschichte. Der Gesichtspunkt eines
Linguisten." Jahrbuch der Norwegischen Akademie der Wissenschaften. Historisk-filosofisk
klass, 10. november 1978; 3-15.
Coseriu, Eugenio 1980, "`Historische Sprache' und `Dialekt'." Göschel, Joachim & Ivi , Pavle &
Kehr, Kurt (eds.) 1980, Dialekt und Dialektologie. Ergebnisse des Internationalen
Symposiums "Zur Theorie des Dialekts", Marburg/Lahn, 5.-10. September 1977. Wiesbaden:
Steiner (Zeitschrift für Dialektologie und Linguistik, Beihefte N.F. Nr. 26 der Zeitschrift für
Mundartforschung [sic!]); 106-122.
Coseriu, Eugenio 1980, "Vom Primat der Geschichte." Sprachwissenschaft 5:125-145.
Heissig, Walther & Rüdiger Schott (eds.) 1998, Die heutige Bedeutung oraler Traditionen. Ihre
Archivierung, Publikation und Index-Erschließung. Opladen: Westdeutscher Verlag
(Abhandlungen der Nordrhein-Westfälischen Akademie der Wissenschaften, 102).
Himmelmann, Nikolaus P. 1993, "Material ambitions." Languages of the World 7/2:66-68.
Humboldt, Wilhelm von 1827-29, "Über die Verschiedenheiten des menschlichen Sprachbaues."
Humboldt 1963: 144-367.
Humboldt, Wilhelm von 1836, Über die Verschiedenheit des menschlichen Sprachbaues und
ihren Einfluß auf die geistige Entwicklung des Menschengeschlechtes. Berlin: Königl. Ak.
Wiss.; in Kommission: Bonn etc.: F. Dümmler. Abgedr.: Humboldt 1963:368-756.

16
Christian Lehmann
Humboldt, Wilhelm von 1963, Schriften zur Sprachphilosophie. [= Werke in fünf Bänden, hrsg.
v. A. Flitner und K. Giel, Bd.III]. Darmstadt: Wissenschaftliche Buchgesellschaft. 4.
Nachdruck 1972.
Jakobson, Roman 1960, "Closing statement: Linguistics and poetics." Sebeok, Thomas A. (ed.)
1960, Style in language. Cambridge, Mass.: MIT Press; New York & London : J. Wiley &
Sons; 350-377.
Kabatek, Johannes 1999, L'oral et l'écrit – quelques aspects théoriques d'un «nouveau»
paradigme dans le canon de la linguistique romane. Tübingen: unpubl. ms.
Koch, Peter & Oesterreicher, Wulf 1990, Gesprochene Sprache in der Romania: Französisch,
Italienisch, Spanisch. Tübingen: M. Niemeyer (Romanistische Arbeitshefte, 31).
Lehmann, Christian 1989, "Language description and general comparative grammar." Graustein,
Gottfried & Leitner, Gerhard (eds.) 1989, Reference grammars and modern linguistic theory.
Tübingen: M. Niemeyer (Linguistische Arbeiten, 226); 133-162.
Lehmann, Christian 1992, "Das Sprachmuseum”. Linguistische Berichte 142:477-494.
Lehmann, Christian 1996, "Dokumentacija jazykov, nakhodjaš ikhsja pod ugrozoj vymiranija.
(Pervoo erednaja zada a lingvistiki)." Voprosy Jazykoznanija 1996/2:180-191.
Lehmann, Christian 1998, "Das große Sprachensterben." Fakultät für Linguistik und Literatur-
wissenschaft der Universität Bielefeld (ed.), 25 Jahre für eine neue Geisteswissenschaft.
Bielefeld: Aisthesis; 131-151.
Lenk, Elena 1996, Erstellung eines Textkorpus für eine Sprachdokumentation mit LDS.
Bielefeld: Universität (AVG Arbeitspapier Nr. 13).
Silverman, Sydel & Parezo, Nancy J. (eds.) 1995, Preserving the anthropological record. New
York: Wenner-Gren Foundation for Anthropological Research. 2. ed.