A Gentle Introduction to Metadata
There is little conceptually new about metadata. The shortest definition for the term is ``data about data,'' and, while many of us may not be used to thinking about metadata very much, we create it and make use of it all the time. A reference found at the end of an article, like the one below, is a form of metadata:
Bloomfield, Leonard. 1933. Language. New York: Holt, Rinehart & Winston.
The reference above is information about a book--that is, data about data.
Something important to note about a reference like the one above is that we understand the information it is trying to convey by convention. The first element is the author's name, then the year, then the title, then the city and publisher. Also, our knowledge of the basic structure of references allows us to be fairly sure this reference is a book.
Another (more explicit) way of representing the metadata in the above reference would be as follows:
|Last name of author||Bloomfield|
|First name of auther||Leonard|
|Year of publication||1933|
|Publisher||Holt, Rinehart & Winston|
The standard reference and the above table are both different representations of the same metadata.
A book reference is a fairly well-known kind of metadata, but the general idea of ``data about data'' can be far more inclusive. An annotated bibliography, for example, also constitutes metadata which is very much like a list of references except that it also includes an extra level of description in addition to the basic metadata for the document.
Though they do not match the traditional notion of ``data type'', it's also easy to find metadata about linguistic software, which can often be quite valuable. Below is metadata about the SIL Shoebox field linguist's tool.
|Name||The Linguist's Shoebox|
|Platform(s)||Windows (3.1, 95/98, NT), Macintosh (OS 6, OS 7, 68K, PPC)|
|Categories||data management, parser, text analysis|
|Domains||morphology, syntax, discourse, lexicon/dictionary|
|Description||Shoebox is a computer program that helps field linguists and anthropologists integrate various kinds of text data: lexical, cultural, grammatical, etc. It has flexible options for sorting, selecting, and displaying data. It is especially useful for helping researchers build a dictionary as they use it to analyze and interlinearize text.|
|More information||Shoebox Home|
|Licensing||Shoebox is licensed for $45.00|
One can make metadata for any linguistic resource--there could even be metadata about metadata. For example, a description of a collection of bibliographies would be data about data about data (a.k.a. meta-metadata).
It's also fairly common to include the metadata about a linguistic resource within the resource itself. The title page of a book is one example. An index at the end of a book is also an example of metadata included with a resource.
While the basic idea behind metadata may be old, the rise of the internet will allow it to become more useful than it ever has been since metadata will be much more accessible than before--and if metadata is more accessible, then linguistic resources of all kinds will also be more accessible.
One of the most important uses for metadata is to locate a resource. Thus, a book reference is designed to give enough information to allow someone to find that book.
The other primary use of metadata is resource discovery--that is, finding resources relevant to one's research but which one is unaware of. The subject index of a card catalog is a metadata collection which is good for such a purpose. With the advent of new technologies, there are many new possibilities for the discovery of resources. Online indexes of abstracts, like LexisNexis, are well-known tools which makes use of metadata and new technology to greatly enhance the researcher's ability to find relevant resources.
One of the primary goals of the Open Language Archives Community (OLAC) is to help linguists create metadata and distribute it in order to assist in the discovery and location specifically of linguistic resources.
It's important to realize that metadata is simply data about data and not the data in and of itself. Thus, making metadata public is fairly safe--metadata alone does not give people access to the data. A book reference is not a book, and metadata about online data is also not the online data itself.
The most useful metadata makes it very clear how a resource, once discovered, can be located and accessed. The owner of the resource gets to determine how easy or difficult access should be. Obviously, it's not very useful to let people access metadata for a document that no one will ever be allowed to access. However, it's important to realize that making your metadata publicly available in no way implies that the resource the metadata describes is also publicly available.
Most researchers are already fairly skilled at making metadata. This simply requires understanding the important parameters needed to describe a resource and then recording those parameters in a structured way. Creating a list of references at the end of a book or article requires a basic knowledge of metadata. Also, most well-managed archives have collected their metadata and stored it in a structured way, most typically in a database.
One of the primary goals of OLAC is to create a standard way to document the data of linguistic resources to assist in resource location and discovery. Importantly, OLAC does not seek to dictate how linguists must design their metadata. Rather, it seeks to build a community-driven consensus about linguistic metadata. (An article giving an overview of OLAC is available on the OLAC web site at http://www.language-archives.org as is an article describing the proposed OLAC metadata set.)
Also, even an OLAC-compliant archive can store its metadata in a non-OLAC format. Such a resource would simply ``translate'' its metadata into OLAC metadata in order to make it more widely available.
The technical term for translating metadata from one format to another is mapping. Many archives have either created non-OLAC metadata or will have good reasons for wanting to do so. However, for the most part mapping their metadata to OLAC metadata will be fairly straightforward since OLAC metadata uses very general descriptors like ``creator'', ``subject'', and ``format''. Since OLAC metadata is specifically designed for linguistic resources, it also includes a few descriptors like ``subject language'' which are of particular value to language researchers.
An example will help to illustrate how mapping from a particular metadata format to the OLAC standard could work. The record below is taken from the internal metadata database for the Comparative Bantu Online Dictionary.
|Contributor||Van Der Veen, Lolke|
|Author||Van Der Veen, Lolke|
|Citation||Van Der Veen 1994|
|Formats||Word 5, Text|
|Remarks||The Word version has a nice introduction which I made into an info file. --JG|
This record contains a range of information, not all of which is of equal importance for resource location and discovery. The language, Gevove, is of obvious interest, and the name of the creator of the data is also fairly important. Other parameters like ``Recon Status'' (meaning how much lexical reconstruction had been done on the resource) are less important for the general user (though they are of importance to the archive).
The information in the above metadata record was mapped to OLAC metadata by the author of this article in the following way:
|Contributor||Van Der Veen, Lolke|
|Formats||Word 5, Text|
The first thing to note about the mapping from CBOLD internal metadata to OLAC metadata is that information was lost--for example, there is no mention of ``Recon status'' in the OLAC version. OLAC standard metadata does not contain a ``Recon status'' field since this is too specific for general language resources.
It would have been possible to include information like ``Recon status'' in the OLAC metadata because the OLAC standard allows for some flexibility. For example, though it does not define a ``Recon status'' field, it does define a Description field, where any information not belonging to another field can be put.
If it is possible to put information like ``Recon status'' into OLAC metadata via the Description field, why wasn't it done? As the creator of the above metadata records, I can say that this was because I chose not to include such information when I mapped CBOLD internal metadata to OLAC metadata in order to make my task simpler. I made sure the most important aspects of the metadata (like subject language) appeared in OLAC metadata, of course, but also chose to leave some information out.
This ``incomplete'' mapping of CBOLD-internal metadata to OLAC metadata illustrates an important point: Ideal metadata would contain all information possible about a resource. However, OLAC does not enforce this. Rather, it leaves it up to the user what information to put in the metadata. OLAC standard metadata must have a particular structure and make use of OLAC-defined descriptive parameters, but the creator of the metadata otherwise has a lot of freedom.
Many archives have already designed ways to distribute their metadata, most typically through some sort of online access to their database. The Linguistic Data Consortium, for example, has on online catalog to allow access to its metadata.
Online archive access managed by the archive is certainly very valuable, but it has the problem that the user must know about the archive in the first place in order to locate resources in that archive. Ideally, metadata from all linguistic archives could be accessed and searched in a centralized way so that a linguist need not know about a relevant archive in order to locate useful resources--this requires a generalized system of metadata distribution.
The OLAC metadata standard is specifically designed to make general distribution of linguistic metadata possible. This is the first step to the creation of search engines designed specifically for the needs of language researchers. There are various ways to distribute OLAC metadata, some of which are simpler than others. These ways are described in the OLAC implementers document. The simplest way is to use the OLAC Repository Editor.
Search engines making use of OLAC metadata already exist, the most important being the one hosted by the LINGUIST List at http://www.linguistlist.org/olac/. The value of these search engines is largely contingent on the metadata made available to them. The more researchers who create OLAC standard metadata, the easier it will be for all researchers to find the data they need.