War is hell, right? Lately we have been dealing with a particular
combination of two circles of the metadata Inferno: the first (limbo)
and sixth (heresy):
The limbo I'll define as a poorly designed metadata spec: the MARC
holdings standard. The poor design in question is the ambiguity of
enumeration/chronology subfield assignment, specifically this rule:
When only chronology is used on an item (that is, the item
carries no enumeration), the chronology is contained in the
relevant enumeration subfield ($a-$h) instead of the chronology
subfields ($i-$m).
http://www.loc.gov/marc/holdings/hd863865.html
This means that as a programmer trying to parse enumeration and
chronology data from our holdings data *that uses a standard* I cannot
reliably know that a subfield which has been defined as containing
"First level of enumeration" will in fact contain enumeration rather
than chronology. What's a programmer to do? Limbo, limbo.
Others in this thread have already described the common heresy involved
in MARC cataloging: embedding data in a record intended for a single
institution, or worse, a specific OPAC.
Due to the ambiguity in the spec and the desire to just make it look the
way I want it to look in my OPAC, the temptation is simply too great. In
the end, we have data that couldn't possibly meet the standard as it is
described and means that we spend more time than we expected parsing it
in the next system.
In our case we work through these issues with an army of code tests. Our
catalogers and reference staff find broken examples of MARC holdings
data parsing in our newest discovery system, we gather the real-world
MARC records as a test data set and then we write a bunch of Rspec tests
so we don't undo previous bug fixes as we deal with the current ones.
The challenge is coming up with a fast and responsive mechanism/process
for adding a record to the test set once identified.
-Steve
Bess Sadler wrote, On 1/27/12 8:26 PM:
> I remember the "required field" operation of... aught six? aught seven? It all runs together at my age. Turns out, for years people had been making shell catalog records for items in the collection that needed to be checked out but hadn't yet been barcoded. Some percentage of these people opted not to record any information about the item other than the barcode it left the building under, presumably because they were "in a hurry". If there was such a thing as a metadata crime, that'd be it.
>
> We were young and naive, we thought "why not just index all our catalog records into solr?" Little did we know what unholy abominations we would uncover. Out of nowhere, we were surrounded by zombie marc records, horrible half-created things, never meant to roam the earth or even to exist in a sane mind. They could tell us nothing about who they were, what book they had once tried to describe, they could only stare blankly and repeat in mangled agony "required field!" "required field!" "required field!" over and over…
>
> It took us weeks to put them all out of their misery.
>
> This is the first time I've ever spoken of this publicly. The support group is helping with the nightmares, but sometimes still, I wake in a cold sweat, wondering… did we really find them all?????
>
>
> On Jan 27, 2012, at 4:28 PM, Ethan Gruber wrote:
>
>> EDIT ME!!!!
>>
>> http://ead.lib.virginia.edu/vivaxtf/view?docId=uva-sc/viu00888.xml;query=;brand=default#adminlink
>>
>> On Fri, Jan 27, 2012 at 6:26 PM, Roy Tennant<[log in to unmask]> wrote:
>>
>>> Oh, I should have also mentioned that some of the worst problems occur
>>> when people treat their metadata like it will never leave their
>>> institution. When that happens you get all kinds of crazy cruft in a
>>> record. For example, just off the top of my head:
>>>
>>> * Embedded HTML markup (one of my favorites is an<img> tag)
>>> * URLs to remote resources that are hard-coded to go through a
>>> particular institution's proxy
>>> * Notes that only have meaning for that institution
>>> * Text that is meant to display to the end-user but may only do so in
>>> certain systems; e.g., "Click here" in a particular subfield.
>>>
>>> Sigh...
>>> Roy
>>>
>>> On Fri, Jan 27, 2012 at 4:17 PM, Roy Tennant<[log in to unmask]> wrote:
>>>> Thanks a lot for the kind shout-out Leslie. I have been pondering what
>>>> I might propose to discuss at this event, since there is certainly
>>>> plenty of fodder. Recently we (OCLC Research) did an investigation of
>>>> 856 fields in WorldCat (some 40 million of them) and that might prove
>>>> interesting. By the time ALA rolls around there may something else
>>>> entirely I could talk about.
>>>>
>>>> That's one of the wonderful things about having 250 million MARC
>>>> records sitting out on a 32-node cluster. There are any number of
>>>> potentially interesting investigations one could do.
>>>> Roy
>>>>
>>>> On Thu, Jan 26, 2012 at 2:10 PM, Johnston, Leslie<[log in to unmask]>
>>> wrote:
>>>>> Roy's fabulous "Bitter Harvest" paper:
>>> http://roytennant.com/bitter_harvest.html
>>>>>
>>>>> -----Original Message-----
>>>>> From: Code for Libraries [mailto:[log in to unmask]] On Behalf
>>> Of Walter Lewis
>>>>> Sent: Wednesday, January 25, 2012 1:38 PM
>>>>> To: [log in to unmask]
>>>>> Subject: Re: [CODE4LIB] Metadata war stories...
>>>>>
>>>>> On 2012-01-25, at 10:06 AM, Becky Yoose wrote:
>>>>>
>>>>>> - Dirty data issues when switching discovery layers or using
>>>>>> legacy/vendor metadata (ex. HathiTrust)
>>>>>
>>>>> I have a sharp recollection of a slide in a presentation Roy Tennant
>>> offered up at Access (at Halifax, maybe), where he offered up a range of
>>> dates extracted from an array of OAI harvested records. The good, the bad,
>>> the incomprehensible, the useless-without-context (01/02/03 anyone?) and on
>>> and on. In my years of migrating data, I've seen most of those variants.
>>> (except ones *intended* to be BCE).
>>>>>
>>>>> Then there are the fielded data sets without authority control. My
>>> favourite example comes from staff who nominally worked for me, so I'm not
>>> telling tales out of school. The classic Dynix product had a Newspaper
>>> index module that we used before migrating it (PICK migrations; such a
>>> joy). One title had twenty variations on "Georgetown Independent" (I wish
>>> I was kidding) and the dates ranged from the early ninth century until
>>> nearly the 3rd millenium. (apparently there hasn't been much change in
>>> local council over the centuries).
>>>>>
>>>>> I've come to the point where I hand-walk the spatial metadata to links
>>> with to geonames.org for the linked open data. Never had to do it for a
>>> set with more than 40,000 entries though. The good news is that it isn't
>>> hard to establish a valid additional entry when one is required.
>>>>>
>>>>> Walter
>>>
|