[open-linguistics] How to represent LLOD diagram categories at datahub ?

[Bettina Klimek]

Dear all,

at the last telco I agreed to categorize all the data sets in the  
datahub. In the telco we set up the following three categories:

Lexicon ? based on data containing words
Corpus ? based on data containing text
Linguistic Database ? based on data containing languages

After I looked up every single data set in the datahub several  
problems occurred:

1) some data sets seem to be wrong/spam
2) the source URLs in some data sets do not work or are missing
3) there seem to be less data sets at datahub than there are bubbles  
in the cloud diagram
4) some data sets are not linked yet

Concerning these issues I would like to know if there is someone  
managing and maintaining the correctness and completeness of the data  
set entries at datahub? I think this is a matter of quality that  
should not be ignored.

5) a categorization in terms of the three categories defined above  
does not work for all the data sets

While going through all the data sets manifold questions concerning  
the LLOD cloud as such occurred to me. The most important one is: what  
is linguistic data? It is not defined anywhere on the website and the  
data sets reveal all kinds of data which differ in their information  
provided, purpose of representation, data types they supply to the  
user as well as in the linguistic content. To me it seemed as if the  
categorization of the data sets is taking the second step without the  
first. The first step should be to  set up a definition of what  
linguistic data is and as a result which kind of data sets are to be  
expected as belonging into the cloud.

When thinking of the category labels I tried to find some that would  
be broadly accepted by the users. However there seems to be no ideal  
user. A user could be a linguist interested in human languages, a  
computer linguist, a computer scientist or someone interested in NLP  
(even more different users are thinkable). The data sets mirror the  
diversity of users within the data they provide. I think it is  
impossible to find category labels which can cover all the special  
research areas linguistic data could be used for. That is why I tried  
to establish categories everybody understands intuitively. Thus I have  
assigned the data sets to the following 5-6 categories, which are  
defined as follows:

LLOD data set categories:
1) Data representing words = Lexicon
2) Data representing texts = Corpus
3) Data representing information about languages = Language Database
4) Data derived derived from but conserving original data = Ontology
5) Data representing further knowledge about the data (such as  
bibliographical references) = Meta Data
6) maybe: category for further relevant data sets not directly  
concerned with linguistic data

As you can see I kept the first two categories because they are  
straightforward and users intuitively know what a lexicon and a corpus  
are. I established category #3 for those data sets dealing with  
certain information on a large number of languages which cannot easily  
be understood as lexicon or corpus. This holds for PHOIBLE and ASJP.  
The former lists phonological segments that are below the word and  
text level and which are provided for many languages. The latter is a  
language classification based on lexical similarities (which are not  
given though). Another case is glottolog being a language  
classification as well but additionally providing relevant  
bibliographical sources. Since these references are set up in a  
distinct application (Langdoc) glottolog is one special case covering  
two categories: Language Database and Meta Data. Another case, even  
more diverse, is the Multext-East data set which seems to cover the  
categories Lexicon, Corpus as well as Ontology. We can debate about  
the assignment of a data set to two categories. Maybe it is better to  
force everybody uploading data sets to select for only one default  
category.

Furthermore the assignment of the data sets to the categories is based  
on what is to be provided primarily! The ASIt for instance evokes the  
impression that some kind of Italian corpora are the content of the  
data set. However it is rather interested in presenting the analysis  
of various corpora (which are not provided as such!) in form of  
rdf-schema. Data sets like these are the reason why I set up category  
#4. These are problematic because they are based on human language  
data but converted into a machine readable language and by doing so  
totally changing the user interested in the data. The linguist would  
be interested in a user friendly searchable language corpus and the  
computer linguist/scientist is only concerned with the rdf data for  
instance. To solve this problem I assigned data sets primarily  
providing human language data (even if there is also an rdf verision)  
to the categories #1 to #3 and data sets being concerned with machine  
language in any way (only using human language material in order to do  
so) to category #4.

I think it is essential to put these definitions into the cloud (maybe  
as tooltips), because the Meta Data category for instance will be  
associated with different information depending on whether a user is  
concerned with human language or with computer languages.

My classification of the data sets can be seen at:
https://docs.google.com/spreadsheet/ccc?key=0AkVaxylrRsewdGwtckloTlYyZ25iY2duTkQwdlZBaHc&usp=sharing

There you can also see some ?unsure data sets? such as DBpedia. I  
think these are corresponding to the ?resources used to assist and  
augment language processing applications, even if the nature of the  
resource is not deeply entrenched in Linguistics, but only as long as  
the usefulness is well motivated? as explained on:  
http://www.semantic-web-journal.net/blog/call-multilingual-linked-open-data-mlod-2012-data-post-proceedings. I couldn?t come up with an appropriate category label for them yet (maybe ?indirectly relevant resources??!). And I am not sure if they should be in the cloud at all because the data sets are already so heterogeneous that adding not primarily linguistic data sets might cause even more confusion to the user and maybe also frustration when he won?t find only linguistic data. Keeping the LLOD cloud as a pure linguistic data cloud and providing a possibility to link to the LOD cloud (which already contains the ?unsure? data sets) could be a practical option  
here.

This is just another proposal for a possible classification. I think  
it is efficient enough to also cover data sets which are to come in  
the future. I know it is very broad and not very specific, but since I  
am of the opinion that further sub-classifications need to be  
developed with regard to the different linguistic user types, we can  
work on that more carefully later. As a first impression of the cloud  
the 5-6 categories I proposed here should be sufficient to get an  
overview of the main cloud content.

Looking forward to your comments,
Bettina




----------------------------------------------------------------
This message was sent using IMP, the Internet Messaging Program.