Thanks, Osma. I totally agree with you. I absolutely see language coding
for subjects, for text of notes, etc. as making a lot of sense when
there are choices. For the question I posed about titles, obviously one
would display the only title in the record. These are the kinds of
"cases" I was hoping to solicit.

One other thing in terms of language tagging of titles is being able to
automatically process initial articles so humans don't have to do that.
We had such a function in the Univ of California catalog in the 1980's,
based on the language of text, and it was pretty accurate for modern
languages. We used it mainly to find errors in the human coding in MARC
245 fields. It seems to me that a machine-generated initial article
decision could save time, with humans only having to code the ones that
the machine gets wrong or flags as unclear. (Although I do think that we
are getting further from alphabetical displays as time goes on, so even
this may be less necessary.)

The Google translation function is pretty good at guessing the language
of a short text string. Bibliographic records are odd texts and don't
yield the same kinds of clues as "natural language," but I'd be
interested to see a "guess the language" run against a bunch of titles.
Adding a 6th rule Ranganathan's 5, I'd say: "Save the time of the
cataloger." (It worries me when people assume that this is all manual work.)


On 1/12/18 12:41 AM, Osma Suominen wrote:
> Karen Coyle kirjoitti 11.01.2018 klo 18:46:
>> ... when your audience is English speakers, you probably don't want
>> Latin display? Well, with the example of Quo Vadis it wouldn't be very
>> useful to suppress the display of the title because the title isn't in
>> English. If you aren't using the language to determine what to display,
>> then what is the purpose of coding the language? If a user types "quo
>> vadis" into a search box, do they have to say what language it is they
>> are searching?
> In my experience, the language of RDF literals generally becomes
> important once there are several literal values in different languages
> and the UI needs to decide which of them to display. For example, a SKOS
> concept may have skos:prefLabel values in several languages, or a FRBR
> Work entity could have several titles ("1984", "Nineteen
> Eighty-Four"@en, "Vuonna 1984"@fi etc). When displaying such data in a
> UI, I've generally followed this algorithm:
> 1. If there is a label in the UI language, display that one.
> 2. Otherwise, if there is a label without a language tag, display that.
> 3. If the above two methods failed, pick a label in any available
> language more or less at random.
> The other labels that were not selected for display are either not shown
> at all, or shown separately with less emphasis, depending on the context.
> In typical bibliographic records there is usually just one title. I
> don't think suppressing the display of the only available title such as
> "Quo Vadis" makes sense (and the above algorithm would show it in step
> 3). But if there are parallel titles in different languages, then
> choosing the right one based on the language preference of the user
> seems like the right thing to do.
>> I'm pushing back on the assumption that having each string coded for
>> language is a necessity because I want a rational reason for adding this
>> labor to the already complex task of cataloging. This skepticism comes
>> out of my many years of processing bibliographic data that had
>> underlying assumptions about needs that weren't borne out in practice.
>> I'm not against coding for language, but I am against doing so without
>> thinking through what it means for catalog users.
> There's indeed plenty of reason for skepticism in this area!
> One argument for ubiquitous language tagging I can think of is
> supporting intelligent text indexing. If the text index knows about the
> language of literals, it can perform language-specific stop word lists
> and normalizations such as stemming or lemmatization, which help match
> e.g. singular words in search queries to plural forms in the data, or
> vice versa. For example, the jena-text index which can be used with the
> Apache Jena Fuseki triplestore has a facility [1] for this which makes
> use of language-specific analyzers in the Lucene text index.
> Language-specific text indexing generally improves the relevance of
> search results (improves recall with little loss of precision) so is
> beneficial for end users.
> -Osma
> [1]

Karen Coyle
[log in to unmask]
m: +1-510-435-8234
skype: kcoylenet/+1-510-984-3600