Choosing a single 7-rank classification path

[thompsonmj]

As an example, we can consider the classification paths returned for 'Phleum alpinum'. When the results are filtered to include only the 7 required ranks, we get the following with matching GNR confidence scores of 0.988:

...
  "classification_path": [
    "Viridiplantae|Streptophyta|Magnoliopsida|Poales|Poaceae|Phleum|Phleum alpinum",
    "Archaeplastida|Chloroplastida|Streptophyta|Tracheophyta|Magnoliopsida|Liliopsida|Poales|Poaceae|Phleum|Phleum alpinum",
    "Plantae|Tracheophyta|Liliopsida|Poales|Poaceae|Phleum|Phleum alpinum",
    "Plantae|Tracheophyta|Liliopsida|Poales|Poaceae|Phleum|Phleum alpinum",
    "Plantae|Tracheophyta|Liliopsida|Poales|Poaceae|Phleum|Phleum alpinum",
    "Plantae|Tracheophyta|Liliopsida|Poales|Poaceae|Phleum|Phleum alpinum",
    "Plantae|Magnoliophyta|Liliopsida|Poales|Poaceae|Phleum|Phleum alpinum",
    "Plantae|Tracheophyta|Liliopsida|Poales|Poaceae|Phleum|Phleum alpinum",
    "Plantae|Tracheophyta|Magnoliopsida|Poales|Poaceae|Phleum|Phleum alpinum"
  ],
  "name_string": [
    "Phleum alpinum",
    "Phleum alpinum",
    "Phleum alpinum",
    "Phleum alpinum L.",
    "Phleum alpinum prol. commutatum (Gaudin) Rouy",
    "Phleum alpinum prol. parviceps (Briq.) Rouy",
    "Phleum alpinum Linnaeus",
    "Phleum alpinum L.",
    "Phleum alpinum L."
  ],
  "classification_path_ranks": [
    "kingdom|phylum|class|order|family|genus|species",
    "kingdom|kingdom|phylum|phylum|class|class|order|family|genus|species",
    "kingdom|phylum|class|order|family|genus|species",
    "kingdom|phylum|class|order|family|genus|species",
    "kingdom|phylum|class|order|family|genus|species",
    "kingdom|phylum|class|order|family|genus|species",
    "kingdom|phylum|class|order|family|genus|species",
    "kingdom|phylum|class|order|family|genus|species",
    "kingdom|phylum|class|order|family|genus|species"
  ],
  "sources": [
    "National Center for Biotechnology Information",
    "Open Tree of Life Reference Taxonomy",
    "iNaturalist Taxonomy",
    "Catalogue of Life Checklist",
    "Catalogue of Life Checklist",
    "Catalogue of Life Checklist",
    "The Interim Register of Marine and Nonmarine Genera",
    "GBIF Backbone Taxonomy",
    "EUNIS"
  ],
...

The problems illustrated in this case include:

1. A source returns multiple terms for a given rank, e.g. OTOL's Kingdom, Phylum, and Class ranks. It's not clear how the 'correct' term should be chosen.
2. A simple vote for the most frequently occurring classification_path result is problematic because:
   • A single source can appear more than once, e.g. here the Catalogue of Life Checklist appears 3 times, reporting several synonyms as shown in the name_string field. Whether a term here is a synonym or not isn't available in the output of a GNR query.
   • The sources are interdependent.
3. Results at a given rank (even when not considering OTOL) are discrepant, e.g. "kingdom": ["Plantae", "Viridiplantae"], "phylum": ["Tracheophyta", "Magnoliophyta", "Streptophyta"], "class": ["Magnoliopsida", "Liliopsida", "Magnoliopsida"].

The full response for this example can be explored by visiting: http://resolver.globalnames.org/name_resolvers.json?names=Phleum+alpinum&with_vernaculars=true&wigh_context=false&with_canonical_ranks=true

[egrace479]

There is also the point that small changes impact the rankings. Most notably, if you were to search for

"Lumbricus terrestris Linnaeus, 1758"

in the GNR, you get many "exact match, Score: 0.999" for sources that list "Lumbricus terrestris Linnaeus, 1758":

Those exact matches don't agree at some levels, though I'd hope they are consistent in their choices of class and order across a particular source.

The sensitivity in score, meanwhile, comes---not from content---but from punctuation; sources without the comma ("Lumbricus terrestris Linnaeus 1758") are an "exact canonical match, Score: 0.997". This is also true for those that don't have the year or abbreviate "Linnaeus" in some way ("L.", "(Linn.)", etc). However, in this case, these don't list a full taxonomic tree, just a source link.

Sources that don't note Linneaus at all, are rated "exact canonical match, Score: 0.988", which includes three copies of the same higher-orders from Open Tree of Life:

My understanding is that the name and year provide extra context that would be beneficial--if not essential--to determining to which kingdom a particular genus-species hemihomonym belongs. Is that correct?

EX: We ran into trouble with some hemihomonyms (more info here, this focuses on the genera). A concrete example is "Laelia rosea", which has results in Animalia and Plantae, so presumably we would need the kingdom or person + year (resolves to animal with Schaus & Clements, 1893).

The "exact match" label is entirely based on what is put in, so it's really the score, then looking for conflicting kingdoms in the results based on standardized kingdoms (eg., Animalia = Metazoa, etc), I presume? We'd then need to know what the kingdom is or set up a flag to check manually.

TLDR: If the max Score is below 0.999 or 0.997, we should do an extra check for conflicting kingdoms or have an optional known kingdom parameter. Does this seem reasonable?