=Paper=
{{Paper
|id=None
|storemode=property
|title=Glottolog/Langdoc: Defining Dialects, Languages, and Language Families as Collections of Resources
|pdfUrl=https://ceur-ws.org/Vol-783/paper7.pdf
|volume=Vol-783
|dblpUrl=https://dblp.org/rec/conf/semweb/NordhoffH11
}}
==Glottolog/Langdoc: Defining Dialects, Languages, and Language Families as Collections of Resources==
https://ceur-ws.org/Vol-783/paper7.pdf
Glottolog/Langdoc: Defining dialects, languages,
and language families as collections of resources
Sebastian Nordhoff & Harald Hammarström
Max Planck Institute for Evolutionary Anthropology
Abstract. This paper describes the Glottolog/Langdoc project, an at-
tempt to provide near-total bibliographical coverage of descriptive re-
sources to the world’s languages. Every reference is treated as a resource,
as is every “languoid”[1]. References are linked to the languoids which
they describe, and languoids are linked to the references described by
them. Family relations between languoids are modeled in SKOS, as are
relations across different classifications of the same languages. This setup
allows the representation of languoids as collections of references, render-
ing the question of the definition of entities like ‘Scots’, ‘West-Germanic’
or ‘Indo-European’ more empirical.
1 Dialects, languages, and language families
The question of what is a dialect and what is a language is a very old one, and
up to now, there are no agreed upon criteria how to resolve it. While it is a
hotly debated topic among the general public, there is general consensus among
linguists that this question is of relatively minor interest. The classical quota-
tion summarizing the problems of defining a language was popularized by Max
Weinreich: “A language is a dialect with an army and a navy”. This highlights
the socio-political dimensions of declaring something a ‘dialect’ or a ‘language’.
To give an illustration: Before the break-up of the former Yugoslavia, Serbo-
Croatian was considered a single language, whereas now Bosnian, Croatian, and
Serbian are considered three distinct languages despite their grammars not hav-
ing undergone any change. The reason for this change in status is clearly political
and not linguistic.
On the level of language family, the disputes are less important, but disagree-
ment still exists. Some linguists argue for instance that Quechua is a language
family comprising 2, 6, or 46 languages, while others argue that Quechua is one
language with a certain number of dialects. Political considerations also play
a role here: a pan-Quechuan identity advocated by the Academia Mayor de la
Lengua Quechua is easier to vindicate if they share a common language rather
than if they share a common language family.
The difficulties of defining a language have a direct bearing on Semantic Web
applications. ISO 639-31 is a standard for identifying languages in electronic
resources. This works fine as long as this standard governs user preferences,
1
http://www.sil.org/iso639-3/
2
locales and the like. But when it comes to automatic reasoning, ISO 639-3 is
actually insufficient due to the lack of a clear definition of the languages it assigns
codes to. For instance, German as spoken in Northern Germany does not have
a distinction between /e:/ and /æ:/, while in the South, this distinction can
be found. When pulling together resources about the language with the code
‘deu’ (German), counts of the number of vowels will be conflicting. This is not
due to different analyses, but rather to different beasts being analyzed. URIs
relying on ‘deu’, ‘eng’, ‘fra’ etc thus do not identify exactly one resource, but
rather point to a random member of a set of resources. Western languages are
often standardized, the obvious example being French, but such standardization
is only found for a small percentage of the 7000+ languages with an ISO 639-3
code. For languages without a written standard, the problems alluded to above
in the German example become much more serious. Standardization furthermore
does normally not refer to phonological issues like the distinction between /e:/
and /æ:/.
ISO 639-3 is a good choice when dealing with the major languages of the
world, but it is not granular enough for the Semantic Web when world wide
linguistic diversity is the research topic, as for instance in the field of linguistic
typology. We address this problem by using a resource-based definition of lin-
guistic varieties (‘languoids’), based on collections of resources treating them.
Unique identifiers are provided for languoids at every level, allowing a step-less
differentiation where required. Differences in categorization are also taken care
of by tying URIs to the namespace of their creators, so that German as defined
by Smith might exclude Swiss German, but German as defined by Miller might
include it.
2 Definitions
We use the following technical terms for our purpose
– a lectodoc is a document containing information about a linguistic variety
(a lect). This can be a grammar, a dictionary, a word list, or sociodemo-
graphic information about the linguistic group (“The ABC live on the XYZ
river as hunter-gatherers and speak DEF”).
– a doculect is a documented lect, i.e. a linguistic variety described in ex-
actly one lectodoc. This means that on this level of analysis, every grammar
of German represents a (slightly) different doculect, by virtue of being a
different document. Very similar doculects will typicall be grouped into a
languoid.
– a languoid [1] is a collection containing doculects or languoids. This recur-
sive definition means that every languoid is reducible to doculects, which are
anchored in lectodocs, identifiable resources.
– Glottolog is a project providing information about and URIs for 60k lan-
guoids
– Langdoc is a project providing bibliographical data and URIs for 200k
lectodocs
3
– Glottolog/Langdoc[2] is the connection of Glottolog and Langdoc URIs.
It can be found at http://glottolog.livingsources.org
These terms and their roles in the Glottolog/Langdoc project will now be
explicated in more detail.
3 Lectodocs and Doculects
To reuse a foundational metaphor of linguistics: lectodocs and doculects are two
sides of the same coin. A lectodoc is a document, e.g. a book, which describes
a doculect, e.g. Catalan. It is impossible to dissociate one and the other. It is
true that speakers of Catalan do not need a document to know that they are
speaking Catalan and to make use of their language, but here we are dealing with
linguistic research on a world-wide scale, which crucially depends on documents
storing information about languages. The doculect is thus the thing described,
the content, while the lectodoc is the container of the description, the document.
As a shorthand test: a doculect is something you can speak while a lectodoc is
something you can print. While this distinction is not crucial for many linguists,
in the context of Linked Data, it is very important in order to not pollute the
Semantic Web with illicit inferences.
In the Glottolog/Langdoc project, lectodocs are annotated for author, year,
title etc using vocabulary from Dublin Core2 and BIBO.3 Lectodocs are further
annotated in the Glottolog/Langdoc namespace for ‘document-type’ (grammar,
dictionary, wordlist, etc). Lectodoc-URIs make use of numerical identifiers since
existing identifiers like ISBN do not cover our whole document space. For in-
stance, M.A. theses from Brazil do normally not have ISBN numbers.
4 Languoids
Languoids replace the traditional concepts of dialects, languages, and language
families in the Glottolog/Langdoc project. Languoids are mathematically sets,
which can contain other languoids, or doculects. Languoids may not be the empty
set.
As an illustration, take the languoid l9267 ‘Kamu’ (a language of Australia).
This is defined as the set of the doculects {d278503}, a singleton set in this case.
The leading ‘d’ of the only member stands for doculect and the digits refer to
the ID of the corresponding lectodoc, in this case Harvey, M. (1990) A sketch
grammar of Gamu.
The sister language Madngele is defined by the doculects contained in the
lectodocs of the set {d282482}, again a singleton. d282482 can be dereferenced
as Zandvoort, F. B. (1999) A grammar of Matngele.
Higher languoids are represented by sets of languoids. As an example, the
mother of the two languoids just mentioned is l9266, Eastern Daly. It is defined as
2
http://dublincore.org/
3
http://bibliontology.com/
4
the union of the two daughter languoids: {l9267,l9268}. In this case, the leading
‘l’ stands for ‘languoid’ and the digits refer to the ID of the languoid.4
Languoids are annotated for their children (sublanguoids) with skos:narrower,5
parents (superlanguoids) with skos:broader and immediately contained lectodocs
with rdfs:isDefinedBy. Languoids are furthermore annotated for common names
and codes. One crucial aspect of Glottolog/Langdoc is that languoids have cre-
ators, i.e. linguists who have proposed a given languoid. This languoid then
resides in their namespace. Tying languoid names to creators allows us to disen-
tangle common confusions with regard to higher groupings. The string ‘Dravid-
ian’ is a common moniker for a language family of South Asia, but the internal
structure of the tree can vary, as the following two partial trees show
(1) Ethnologue 2005:Dravidian = {Central Dravidian, Northern Dravidian,
South-Central Dravidian, Southern Dravidian, Unclassified Dravidian}
(2) Zvelebil 1970:Dravidian = {Central Dravidian, North Dravidian, South
Dravidian, Telugu, Toda-Badaga, Tul.u}
Central Dravidian, North(ern) Dravidian and South(ern) Dravidian occur
in both superlanguoids. But the two superlanguoids have different numbers of
sublanguoids, 5 and 6 respectively, making them not isomorphic on a graph
theoretic level. The labels Unclassified, South-Central, Telugu, Toda-Badaga, and
Tul.u are also not shared, a further indication that the semantics of Dravidian
as conceived by Zvelebil and Dravidian as conceived by the Ethnologue cannot
be equated.
What can be done is to state that the two representations are similar, given
that both creators had the shared goal of representing those languages spoken in
South Asia which are related to Tamil, Telugu, Kannada etc. The URIs provided
by Glottolog/Langdoc allow to formulate relations between two languoids, e.g. as
skos:closeMatch, which has the definition: “used to link two concepts that are
sufficiently similar that they can be used interchangeably in some information
retrieval applications”6 with the emphasis on some. An example would be
(3) glottolog:l109398 skos:closeMatch glottolog:l2712 .
This kind of predication would not be possible with ISO 639-3 because ISO
639-3 does not provide a code for Dravidian, and ISO 639-3 would never dis-
tinguish between two different representations of the same language family. It is
true that ISO 639-57 provides some labels for language families, but these are
by no means exhaustive, and do not allow us to distinguish between different
creators, which is crucial as shown for the Dravidian example above.
4
It is also possible to have sets with both languoids and doculects, but this case is
rare. It arises for instance when a work treats a language family as a whole. This
will not be developed in detail here.
5
http://www.w3.org/2004/02/skos/
6
http://www.w3.org/TR/2009/CR-skos-reference-20090317/#mapping
7
http://www.loc.gov/standards/iso639-5/
5
It has been stated above that the two trees/graphs of Dravidian are not
isomorphic and can therefore not be arguments of owl:sameAs. However, it is
clear that the two types of Dravidian are closer to each other than either is
to a tree of Austronesian for instance. This can be modeled by reducing the
tree representations to flat lists of lectodocs attached below the nodes to be
compared. If the two sets of lectodocs are identical, owl:EquivalentClass is
warranted since the concepts have the same extension. If one is a proper subset
of the other, more computation is necessary, taking into account other trees. This
will not be discussed here, but will often result in applying skos:closeMatch.
5 Advantages
The Glottolog/Langdoc approach has a number of advantages:
– URIs are given for every languoid, where other systems like ISO 639-3/5
restrict themselves to languages and selected families
– The availability of unique identifiers at every level allows to circumvent the
issue of singling out a more prestigious level of ‘language’, short-circuiting
some perpetual discussions.
– Namespaces for authors assure that different languoids are not lumped to-
gether if the name is the same but the semantics differ.
– Every languoid has to be grounded in a resource; languoids without resources
(which are commonly found in Ethnologue, e.g. Loarki8 ), are not permitted.
This means that intersubjectivity can be attained: two researchers can make
sure that they are talking about the same thing, which was not always the
case in the past.
– The set-theoretic model allows to reduce the complexity of the model through
transitive closures when that much granularity is not required.
– The set-theoretic approach furthermore allows searches like “Give me a dic-
tionary of a Dravidian language” without knowledge of all the languages
contained within Dravidian. In traditional systems, one would have to gather
all relevant ISO 639-3 codes by hand and iterate through them in individual
queries.
6 Summary
Glottolog/Langdoc makes world-wide linguistics fit for the Semantic Web by
providing near-complete bibliographical coverage of lesser known languages. It
furthermore provides unique identifiers for its references as well as for languoids
of every level, allowing third-party projects to harvest, digest, and further an-
notate the resources.
8
http://www.ethnologue.com/showl anguage.asp?code=lrk
Bibliography
[1] Good, J., Hendryx-Parker, C.: Modeling contested categoriza-
tion in linguistic databases. Proceedings of the EMELD 2006
Workshop on Digital Language Documentation: Tools and Stan-
dards: The State of the Art. Lansing, Michigan. June 20-22, 2006
http://www.linguistlist.org/emeld/workshop/2006/papers/GoodHendryxParker-
Modelling.pdf (2006)
[2] Hammarström, H., Nordhoff, S.: Langdoc: Bibliographic infrastructure for
linguistic typology. Oslo Studies in Language to appear, to appear (2011)