[tdwg-content] Canonical name parsing

Wed Mar 14 16:19:10 CET 2012

Hi Peter.

I certainly agree that aggregators only represent one use case here but, having seen a lot of the mess of real-world data, I don't believe that simply adding a new term will fix this problem for the users you describe.  To get the results you want, we would need a sufficiently large majority of data sets to follow the rules perfectly that we could ignore those that were non-conformant.  This would mean we should mandate that every data set must use the new element (with or without the existing scientificName element) and that they must present scientific names in the expected way (or else have their data considered non-compliant). Until now, the philosophy on publishing Darwin Core data has been to make it as easy as possible for data providers to expose their data, even at the expense of greater complexity for consumers.  I suspect that we would have a lot less data available for use now if we had taken a more stringent approach.

In some ways, this proposal reminds me of the structures in ABCD which seek to offer users verbatim and more normalised ways to represent several types of information.  This actually makes consuming all the possible forms of such data very complex, since a record may contain all variant forms or just any one of them.  If multiple forms are available, which one should be considered the primary version?

I suspect that things may also get complicated as soon as you discuss botanical subspecies, varieties, subvarieties, forms and subforms.  There are recommended ways to abbreviate the rank markers in these cases but some variation can be expected.

Of course aggregators should be providing more robust services for accessing exactly what you want in a consistent, predictable way and I would suggest that the best place to attack the problem is to define exactly what a typical user needs to see and then for GBIF and similar projects to work on delivering predictable data downloads and web services that clean out all of these nomenclatural inconsistencies - and perhaps also add value in other ways such as augmenting the data with associated environmental values (as the Atlas of Living Australia does).  This would allow us all to work together on developing a consistent and predictable algorithm for handling interpretation of name strings, including synonymy, misspellings, virus names and everything else that makes this such a difficult problem.

Best wishes,

Donald

----------------------------------------------------------------------
Donald Hobern - GBIF Director - dhobern at gbif.org 
Global Biodiversity Information Facility http://www.gbif.org/ 
GBIF Secretariat, Universitetsparken 15, DK-2100 Copenhagen Ø, Denmark
Tel: +45 3532 1471  Mob: +45 2875 1471  Fax: +45 2875 1480
----------------------------------------------------------------------

-----Original Message-----
From: peter.desmet.cubc at gmail.com [mailto:peter.desmet.cubc at gmail.com] On Behalf Of Peter Desmet
Sent: Wednesday, March 14, 2012 3:41 PM
To: Tim Robertson [GBIF]
Cc: Donald Hobern (GBIF); dev Developers; TDWG content mailing list; TDWG TAG mailing list; Christian Gendreau
Subject: Re: Canonical name parsing

Hi Tim,

I agree, aggregators like GBIF and Canadensys will have to deal with clean and dirty data in each field anyway: they need code libraries to deal with this and it is good that these are being developed. But, that doesn't help someone who wants to use data from a Darwin Core Archive with his data in Excel or a Roderic Page who wants to get things done for a prototype.
Having to use Java libraries or even the Name Parser [1] (though both
great) is a barrier to data use. Darwin Core (Archives) is not only used for machine to machine interaction, humans use it too, and I think we should allow easy hacking (I mean this in the good sense), especially for something as important as the scientific name.
In addition, as a data publisher (e.g. for our VASCAN checklist) I
*do* have the information to provide a clean and simple to use canonicalScientificName, but I just can't share it via the otherwise excellent biodiversity sharing standard Darwin Core. I think that's a pity.

Peter

[1] http://tools.gbif.org/nameparser/
[2] http://data.canadensys.net/vascan

PS: Yes, Canadensys will use the GBIF interpretation libraries. Since we develop in Java as well, using those libraries is as easy as the proverbial "one line of code". We're looking forward in testing them and providing patches to enhance them. Open source FTW! :-)

On Wed, Mar 14, 2012 at 07:32, Tim Robertson [GBIF] <trobertson at gbif.org> wrote:
> Hi Peter,
>
> I'm replying off the TDWG list, since it is a bit of a tangent to your discussion.  If you feel it is relevant, please CC the list again.
>
> At GBIF as you know, we have to interpret all kinds of quality of content.  I tend to agree with Donald that this would not really help in consumption, as in my experience we will have to deal with both clean and dirty data in each field *anyway* when this is used at network scale.  I would rather see us evolve the interpretation libraries to handle all the corner cases, which we need to develop anyway.  We already do a pretty decent job at extracting canonicals.  This is further enhanced when you couple the extracted canonical with a fuzzy match against the "authoritative names" we can now index thanks to the availability of checklists in DwC-A format.
>
> I know you are a Java shop.  Are you using the GBIF interpretation libraries [1] at the moment?  If not, is there a reason why you don't?
> They are used in all GBIF projects (portal, checklistbank etc), and the more we enhance them, the better it is for everyone.  We have a significant test coverage [2,3] and there have been quite some man months (years?) spent already in their development and with some real regular expression experts (most notably Markus D. and Dave M.).  All our work is Maven-ized, versioned and available in our Maven repository [4].
>
> I hope these are interesting to you.  We would welcome any patches to enhance them, or assistance in identifying the corner cases and capturing those as unit tests.
>
> Hope this helps,
> Tim
>
> [1] 
> http://code.google.com/p/gbif-ecat/source/browse/trunk/ecat-common/src
> /main/java/org/gbif/ecat/parser/NameParser.java
> [2] 
> http://code.google.com/p/gbif-ecat/source/browse/trunk/ecat-common/src
> /test/java/org/gbif/ecat/parser/NameParserTest.java
> [3] 
> http://code.google.com/p/gbif-ecat/source/browse/trunk/ecat-common/src
> /#src%2Ftest%2Fresources [4] 
> http://repository.gbif.org/index.html#nexus-search;quick~ecat-common
>

--
Peter Desmet
Biodiversity Informatics Manager
Canadensys - www.canadensys.net

Université de Montréal Biodiversity Centre
4101 rue Sherbrooke est
Montreal, QC, H1X2B2
Canada

Phone: 514-343-6111 #82354
Fax: 514-343-2288
Email: peter.desmet at umontreal.ca / peter.desmet.cubc at gmail.com
Skype: anderhalv
Public profile: http://www.linkedin.com/in/peterdesmet