Publications of the past four millennia on linguistic topics were written in diverse languages: Akkadian, Sanskrit, Chinese, Japanese, Greek, Latin, Arabic, French, German, Spanish, English, to name but a few. Like anything in human history, this depends on time and place coordinates.

In linguistics, the problem of the choice of language is complicated by the variable of the object language: If the object language differs from the metalanguage, again the question arises whether the intended readership is familiar with the object. Some other disciplines in the humanities, including philology, literature and history, have an analogous problem.

The metalanguage

The choice of a metalanguage in science is made at two levels:

The language of publication

Anyone who publishes a text is forced to choose a language for it. This has always limited the possible readership of the text and will continue to do so for a long time. Therefore, the principal criterion in choosing the language of publication – in linguistics, the metalanguage – is the composition and language competence of the intended readership.

A world is possible in which every scientist, including every linguist, can read one language which is the working language of science. Medieval science in Europe was close to this situation, as essentially all scientists could read in Latin. Since the end of the 20th century, the language of science has been becoming English; and this may continue to be so for a couple of decades to come.

This development is also bound up with globalization. In the nineteenth century, a Japanese linguist did not expect to be read by a German reader, and vice versa. They were simply not included in the intended readership. This has changed with globalization. An author publishing on a general topic will generally not wish to exclude entire populations of readers.

However, this does not entail that all scientific publications should, for the time being, be in English. A publication may be meant as a contribution to a linguistic discipline devoted to a particular object language. All members of such a discipline may then know this language or a language which has traditionally been used to describe this object language. If the author knows this alternate metalanguage better than English, he may prefer to use it. This is widely the case in descriptive linguistics and philology devoted to certain languages and cultures, like Sinology, Japanology, Arabistics, Romanistics, Germanistics, Slavistics and certainly some others. If the author directs a publication devoted to one such language specifically to colleagues occupied with the same language, he will use that language – or the vehicle traditionally used in the discipline in question – as his metalanguage. By specifically addressing this readership, he may consciously be excluding readers outside the circle. As a simple example, consider a book on the history of the German language. If this is intended as a reference work for researchers working in Germanistics, it may be written in German. If it is intended as a coursebook for students of Germanistics worldwide, it may be written in English or even in the language of some region where Germanistics exists.

The same holds for a description of a minority language. If it is intended to be usable by the speech community, then it could, in principle, be composed in that language. However, this will often miss the target, because that language may lack linguistic terminology altogether, while possible users of the description – for instance, regional teachers – may have some formal education in the regional dominant language. For instance, a description of a Tupi language that is meant to be of use to its speech community may be written in Portuguese.

The language of reading

Even in a situation where there is synchronically one language of science, this does not mean that everything of relevance in the discipline was published in this language. The language of the discipline has changed over the past millennia. In the natural sciences, the assumption prevails that everything thought and published before today is either outdated or sufficiently incorporated into currently true theories, so there is no need to read those older scientific publications. This is not the situation in the humanities (or in philosophy, for that matter). There has been little progress in the sense of accumulation of truth. Instead, views and approaches have come and gone; and some which were widely held in the fifteenth century are currently en vogue again. No progress will ever be made if that earlier thinking is not taken knowledge of. (More on this elsewhere.)

The same is true of linguistic descriptions composed in past times. The language under description may have died out meanwhile, or it has just changed since. Such languages and language states can be incorporated into modern – comparative or historical – work only if researchers can read those descriptions. Nor should one hope for a translation of such older work. Even if perfect translation software were available, one would still be confronted with a scientific tradition and research methods which were bound up with the language and culture of the author. It suffices to think of transcription techniques used before the advent of IPA.

All of this presupposes that members of the discipline can read the languages of publication of earlier times. A scientist who only reads the current language of science is a pathetic historyless figure.1 Consequently, the answer to the question of this subsection is: A scientist should be able to read publications composed in the principal languages of (the history of) his discipline. For a linguist, a handful of languages does not appear to be asking too much. A linguist who only reads and refers to publications written in his own native language demonstrates limited language skills and lack of engagement in internationalization with equal rights.

Translating and glossing the object language

An empirical linguistic work presents data and examples taken from certain languages. The question is whether such object-language expressions are translated and provided by interlinear morphological glosses.

It may be helpful to throw a glance at two extreme positions on this issue. On the one hand, it was customary in 19th and still in early 20th century publications on Indo-European linguistics to presuppose knowledge of Indo-European languages in the readership and, consequently, not to translate, let alone analyze examples of those languages. These publications have a limited readership today, and the authors' attitude appears conceited. On the other hand, since the turn of the 21st century, some publishers have been asking authors to translate and provide with an interlinear gloss all non-English examples, thus, including languages of international communication like Spanish and French. This promotes monolingualism among linguists and demonstrates English language imperialism. It seems clear that a rational solution to the problem has to be more fine-tuned.

Expressions of object languages which have not been widely used as languages of science – thus, the far majority of living and dead languages – are translated into the metalanguage of the publication. This is valid for any type of linguistic publication, from the introductory course book to the specialized article. It remains to decide whether they have to be provided by an interlinear gloss, too. This has not been customary in text books, for instance, a course in Navajo. This may, again, be a question of the intended readership: If the publication is intended for readers who are willing to invest continual effort into the study of the language in question – not necessarily professional linguists –, the interlinear gloss may be superfluous or not helpful.

If the object language is or was a language of science, then the question of whether expressions in it will be translated and provided with interlinear glosses again depends on the intended readership. The content of the work may presuppose knowledge of the language, so data and examples in that language will not be translated. This is so if the publication addresses researchers of a discipline devoted to that language, as noted above. In this case, the work may also be of little relevance to scientists outside the specific discipline. A work dedicated to the language of a particular writer may be an example.

If none of these conditions hold, then data and examples of the object language will be translated and provided by an interlinear gloss. The above requirement that a scientist should be able to read work composed in a language that was or is a working language of his discipline may extend to the need to translate and gloss examples taken from these languages. Here, however, one meets the limits of the polyglossy of members of the linguistic discipline. The above enumeration of languages that were or are working languages of linguistics comprises a dozen – thus more than the average linguist knows. Different linguists know different languages. A Japanese linguist may be able to read in Chinese, a Polish linguist may be able to read in German. This is no basis for deciding that examples in Chinese or German do not need to be translated.

The question of whether an interlinear morphological gloss is necessary is a variant of the more general question of what kind of linguistic analysis is necessary in a particular linguistic treatise. It obviously depends on the level of the language system being treated. If it is grammar, then the gloss is needed. If it is discourse or phonetics, it may not be needed.

* This page has profited from a discussion on the Lingtyp list of 25 - 26 june, 2020.

1 A particularly irresponsible behavior has been demonstrated since the last turn of the millennium by authors launching big theories on linguistic change while blissfully ignoring two centuries of historical linguistics, including authors from Brugmann via Paul and Meillet to Benveniste, who did not publish in English, but solved most of the now fashionable problems with more circumspection and on a more solid empirical basis.