Martin Haspelmath’s (2013) recent post in this forum discussed the criticism by Morey et al. (2013) of the ISO 639-3 three-letter codes for language identification. In agreement with Martin, I would strongly urge linguists not to swim against the tide but to go with the flow and accept ISO 639-3 as a useful initiative for specific use-cases. The ISO 639-3 codes are not the holy grail that will solve all our problems concerning language-identification, but they have their merits. Most importantly, it is still one of the few resources that at least tries to provide a comprehensive catalogue of the world’s linguistic diversity. If one criticises SIL and their Ethnologue (which is the basis of ISO 639-3), then at least one should also acknowledge that they have been working on this catalogue for decades, and in all this time no other institutionalised linguist has tried to improve, or at least parallel, their effort.
Still, there are many decisions taken in the ISO 639-3 that many linguists are uncomfortable with. While discussing ways to improve upon the current situation, Jeff Good and the undersigned (Michael Cysouw) decided (in 2006, while we were both at the Max Planck Institute for Evolutionary Anthropology in Leipzig) that the underlying problem is not so much the specific decisions of the Ethnologue itself, but the idea that there should be an ideal universally-accepted catalogue of language diversity. In our view, this ideal consensual catalogue is still far off as there are too many unknowns, and it might even be a complete chimera to ever want to achieve such a universally-accepted catalogue. What we need is a way to codify our decisions, and to provide ways to discuss any (dis)agreements on the way. So, it is not so much yet another consensual catalogue that we need: if you need that, just go with the flow of ISO 639-3. What we need is an explicit statement of our current knowledge of the world’s linguistic diversity (with all it’s unknowns) and stringent methods to specify disagreements.
As a solution Jeff and I proposed a triumvirate of concepts that we think can function as a formal basis for any such approach: languoid, doculect, and glossonym. We have been spreading these concepts in informal discourse for years, because it took us way too long to write a scholarly paper on this idea. When we finally wrote the paper, the linguistic review process turned out to be particularly cumbersome, so it took even longer for the paper to appear. But now finally our paper has appeared in Language Documentation and Conservation, Vol. 7 (pp. 331-359). The paper is full of all kind of details that have been amassed throughout the years of working on this paper, so let me try to summarize the main gist here for the busy ones among the current readers (liberally quoting our own paper here).
We begin with the desideratum that languages can only be identified by reference to existing documentation. The motivation behind this proposal is that, while what constitutes a specific language in some abstract sense will perhaps always be controversial, there is no controversy in simply saying that there exists a given book, sound file, manuscript, or article that contains data documenting some language variety, even if there is disagreement about how that variety should be classified. While, at first glance, this desideratum may seem quite obvious, we can merely note here that it is not the basis of the Ethnologue or ISO 639-3. The Ethnologue does include sources as part of the information associated with a given code, but these are treated as an additional source of information about the language. Crucially, we propose a reversal of the code–source relationship as usually understood: The resources are the primary data, with the codes taken to be a kind of metadata applying to the set of resources defining a ‘language.’
The basic entities that we claim should form the foundation for a rigorous definition of languages are thus actual documentary records attesting to the existence of some language variety. We propose to refer to the linguistic varieties documented by these records as doculects (‘documented lect’). All further language-like objects, like dialects, sociolects, languages, genealogical groups, or areal groups, can then be defined by referring to a collection of doculects on which they are based. The definition of all these language-like objects is thus essentially the same: They are all defined via sets of doculects. Although there are very many different kinds of such language-like objects, we propose to use the name languoid to refer to the superset of all those language-like entities. A crucial aspect of this conceptualisation is that the grouping traditionally called ‘language’ does not have an inherently preferred or basic status in this model for classifying the world’s linguistic variation. This allows for rigorous debate about what constitutes a ‘language’ without interfering with the system’s core classificatory apparatus.
Finally, we must also pay attention to the ways that names are used to refer to languoids and doculects, since, ultimately, it is via names that linguists usually talk about different languoids. To preclude any conceptual confusion between the entities and their names, we propose the term glossonym here for the names that are used to refer to languoids.
There are, of course, numerous kinds of information about languages that are of interest to linguists and beyond (e.g., autonymic reference, geographic location, sociolinguistic context, number of speakers). However, our goal here is not to devise an all-encompassing reference work for the world’s languages or anything along those lines. Rather, our goal is construct a conceptual and terminological foundation for talking about languages that can be rigorously applied at a global scale. Our basic claim is that the triad of concepts proposed here (i.e., languoid, doculect, and glossonym) is a minimal system that nevertheless allows for a rigorous definition of language-like entities. Other kinds of information can easily be added on top of this foundation, but are not needed for their definition.
In practice, both the Ethnologue and the Glottolog are perfectly compatible with our proposals. Both organise sources, which can mostly be equated with doculects (though not always, see Section 6.3 of Cysouw & Good 2013). Any language or family or other grouping proposed in those resources are then simply defined as the collection of those doculects. Unfortunately, Ethnologue has made the relation between codes and sources a bit less transparent in the last update of their website (e.g. the code blo has sources as listed at a different URL, unfortunately without direct link). And these sources are of course only the SIL-based sources. Glottolog is almost perfect, as the Langdoc-part consistently identifies sources interlinked with languoid. The only aspect missing is to be able to specify sources as consisting of multiple doculects (cf. the Langdoc-treatment of Nies 1976 with Section 6.3 and 7.3 of Cysouw & Good 2013).
For many further details and for precise definitions of our terminological triad, please refer to Cysouw & Good 2013. It is freely accessible online, so there is nothing to hold you back to read on there, when you already have made it to the end of this post!
Cysouw, Michael & Jeff Good. 2013. Languoid, Doculect, Glossonym: Formalizing the notion “language”. Language Documentation and Conservation 7. 331-359. http://hdl.handle.net/10125/4606
Haspelmath, Martin. 2013. Can language identity be standardized? On Morey et al.’s critique of ISO 639-3. Diversity Linguistics Comments. http://dlc.hypotheses.org/610
Morey, Stephen & Post, Mark W. & Friedman, Victor A. 2013. The language codes of ISO 639: a premature, ultimately unobtainable, and possibly damaging standardization. Handout of a talk given at the PARDISEC RRR Conference, December 2013.
Nordhoff, Sebastian & Hammarström, Harald & Forkel, Robert & Haspelmath, Martin (eds.) 2013. Glottolog 2.2. Leipzig: Max Planck Institute for Evolutionary Anthropology.
(Available online at http://glottolog.org, Accessed on 2014-01-04.)
Thanks a lot for sharing this with all folks you actually
recognize what you’re speaking about! Bookmarked. Kindly additionally talk over
with my site =). We could have a hyperlink change agreement
Amuesha = Yanesha’ [ame] is not an error, it is trivial in the iso6393 namespace. Like I said.
I am puzzled by the statement “Glottolog is almost perfect, as the Langdoc-part consistently identifies sources interlinked with languoid. The only aspect missing is to be able to specify sources as consisting of multiple doculects (cf. the Langdoc-treatment of Nies 1976 with Section 6.3 and 7.3 of Cysouw & Good 2013).”
Glottolog can (and does in thousands of cases) specify sources as consisting of multiple doculects in the spirit of 6.3 and 7.3 of Cysouw & Good (2013). Look yourself at the Link to Nies and look at the BibTex and the lgcode field to see that it IS specified as such (and consequently all the doculects in Nies 1976 do appear under their respective languoids).
The ‘almost’ refers to the fact that in Glottolog there is no way to know which parts of Nies 1976 refer to which language (except by using scholarly interpretation). Nies uses the following names for the various doculects in this source:
A) campa nomatsiguenga
G) huitoto meneca, mɨnɨca
H) bora muinane
Looking in the Glottolog, I find the following languoids:
5) Murui Huitoto
Now, as a linguist with a little imagination I can deduce the following links, but this linkage is far from obvious, and surely not possibly to apply automatically:
(as an aside, note that the attribution in Glottolog of “Murui Huitoto” is probably an error, it should be “Minica Huitoto”. A different error is found on the website of SIL peru, which lists “Nüpode Huitoto” instead of the correct “Bora Muinane”).
The central point of the formal definition of a doculect (which is not in the blog-post above, but which is in the complete paper) is the following: a doculect is a pairing of a resource with a glossonym, where the glossonym is the non-normalised version of the name used in the source. Nies 1976 should be ideally included as 8 different doculects, with the names in the exact spelling as used in the source. Then the trace-back to the original data is completely transparent, and there is no discussion about the identification of the original doculect. Using the definition, it is impossible to make errors (except for typos).
Still: I really like the glottolog in very many ways. It is a big step forward!
If you look at the BibTex (like I said), you do have all non-trivial mappings between the names used in Nies and the languoid assigned, i.e.:
Bora Muinane = Muinane [bmr]
Campa Nomatsiguenga = Nomatsiguenga [not]
Piro = Yine [pib]
But it is only the non-trivial mappings that have an explicit X corresponds to Y. The rest of the names used in Nies are actually uniquely identifiable (in the sense of the namespace used in e.g. iso6393) and could be resolved with a computer.
(This is except for the Murui case which indeed seems to an error, i.e., a plain error, not an error of design). So, like I said, glottolog does follow the spirit of Good & Cysouw 6.3/7.3 (but not in the letter of G&C 2013 since non-normalized original names are not used if the mapping is trivial).
As I said, I consider this to be “almost perfect”. Note that my emphasis is on “perfect” (I really like Glottolog!), and your defence seems to be a reply to the “almost” :-).
The (minor) criticism remains: Glottolog does not offer IDs for doculects in our strict sense, although they can be derived from the “lgcode” field (i.e. each combination of source+lgcode can be interpreted as a doculect).
(Note that there is a second smallish error: it should be “Amuesha = Yanesha’ [ame], as Nies uses the name “Amuesha”.)
A perfect Glottolog IMHO should assign IDs for each combination of source+lgcode = doculect.