Franz. 2014. Explaining taxonomy's legacy to computers – how and why?

Explaining taxonomy's legacy
to computers – how and why?
Nico M. Franz 1,2
Arizona State University
http://taxonbytes.org/
1 Concepts and tools developed jointly with members of the Ludäscher Lab (UC Davis & UIUC):
Mingmin Chen, Parisa Kianmajd, Shizhuo Yu, Shawn Bowers & Bertram Ludäscher
2 The Meaning of Names: Naming Diversity in the 21st Century
September 30, 2014; Museum of Natural History, University of Colorado
On-line @ http://www.slideshare.net/taxonbytes/franz-2014-explaining-taxonomys-legacy-to-computers-how-and-why

Alternative title:
Concept taxonomy –
now with logic reasoning.

Definitional preliminaries, 1
Taxonomic concept: 1
The circumscription of a perceived
(or, more accurately, hypothesized)
taxonomic group, as advocated by
a particular author and source.
1Not the same as species concepts, which are theories about what species are, and/or how they are recognized.

Definitional preliminaries, 2
Provenance: 1
Information describing the origin, derivation,
history, custody, or context of an entity (etc.).
Provenance establishes the authenticity, integrity
and trustworthiness of information about entities.
1 See, e.g.: http://www.w3.org/2005/Incubator/prov/wiki/What_Is_Provenance

Concept taxonomy in three introductory phrases
• An emerging solution to the challenge of tracking stability
and change across multiple taxonomic name usages.

• Fully compatible with Linnaean nomenclature (Codes).

• Fully compatible with Linnaean nomenclature (Codes).
• The focus is on building sound provenance chains amenable
to computational representation and reasoning; irrespective
of whether the nomenclatural/taxonomic history of a perceived
lineage of organisms was perfectly stable since the times of
Linnaeus, or continues to undergo major alterations.

Overview of today's presentation
• The challenge (1.0): Limitations of the name  taxon reference model.
• The challenge (2.0): How to track taxonomic concept provenance?
• Introducing Euler/X – overview of workflow and user/reasoner interaction.
~ 8 mins.

Overview of today's presentation
• The challenge (1.0): Limitations of the name  taxon reference model.
• The challenge (2.0): How to track taxonomic concept provenance?
• Introducing Euler/X – overview of workflow and user/reasoner interaction.
• How does it work?
• Use case 1: Dwarf lemur classifications sec. 1993 & 2005.
• From simple to complex merge taxonomies.
• How can we represent taxonomic concept overlap?
• Scalability & information gain: How many articulations?
• Why? Insights into the performance of names as concept identifiers.
• Use case 2: Andropogon glomeratus sec. auctorum.
• In conclusion – feasibility, accessibility, and what it means.
~ 8 mins.
~ 15 mins.

The challenge (1.0):
Often, we make statements like this:

"Andropogon glomeratus
is a species of grass (Poaceae)
that occurs in the Southern U.S." Photo by Max Licher (ASU Herbarium); Cottonwood, Arizona.
http://swbiodiversity.org/seinet/imagelib/imgdetails.php?imgid=431755

Thereby we stipulate a direct
name  taxon reference relationship.

Proposition 1: names refer (directly) to taxa
Taxonomic name
Taxon (species)
that occurs in the Southern U.S."
Biological data
Reference relation:
name refers to entity

Proposition 1: names refer (directly) to taxa
Taxonomic name
Taxon (species)
Biological data
Reference relation:
name refers to entity
Data transmission:
facilitated by name

Yet, the legacy of taxonomy is more complicated:
the name  taxon relationship can change.1
This poses some representation challenges…
1 See Franz et al. 2008. On the use of taxonomic concepts in support of biodiversity research and taxonomy; pp. 63–86.
In: The New Taxonomy, Systematics Association Special Volume 74. Taylor & Francis, Boca Raton.

Challenge 1: by necessity, a name refers only to a type (specimen)
Taxonomic name
Identity of the name/reference
relation is regulated by Codes
(e.g., Typification)

Challenge 2: the discovery of 'true' taxon boundaries is contingent
Taxonomic name
Taxon (species)
The boundaries of taxon identity
have the property of contingent,
scientific hypotheses = concepts

Challenge 3: name/taxon (concept) changes are semi-independent
Taxonomic name
Taxon (species)
Precise,
reliable
mapping?

Consequence: the name  taxon reference model is often too simple
Taxonomic name
Taxon (species)
Biological data
Precise,
reliable
mapping?
Reference
limitations!
Name-based data transmission:
reliability is also contingent

If we accept a contingent, changing
name  concept  taxon reference model,
then perhaps we should always say this:

Proposition 2: concept labels refer (directly) to taxonomic concepts
..is the (Latin) name (string), nomenclaturally anchored with a
type specimen, that can participate in the (more precisely in-dividuated)
concept label "Andropogon glomeratus sec. Barkworth
et al. 2014" (reference: Manual of Grasses for North America),
which in turn refers to..

..a feature-based circumscription ("Plants cespitose, upper portion dense, …
Pedicellate spikelets vestigial or absent, sterile. 2n = 20.") – the taxonomic concept
as advocated by this reference – which may or may not align
accurately with a (presumably existing and) relatively stable
evolutionary lineage of organisms in nature for which..

..a feature-based circumscription ("Plants cespitose, upper portion dense, …
Pedicellate spikelets vestigial or absent, sterile. 2n = 20.") – the taxonomic concept
as advocated by this reference – which may or may not align
accurately with a (presumably existing and) relatively stable
evolutionary lineage of organisms in nature for which..
..biological occurrence data are on hand.

Hence:
The challenge (2.0):
If we individuate taxonomic concepts
and their labels consistently, ..

1889
1933
1948
1950
1968
1979
1983
2006
2014
Chain of A. glomeratus concepts, 1889-2014.

..then how can we track concept provenance?

1889
1933
1948
1950
1968
1979
1983
2006
2014
?
Provenance representation challenge:
How is each concept articulated to another?

Proposed solution:
We articulate them with (RCC-5)
concept-to-concept relationships..

1889
1933
1948
1950
1968
1979
1983
2006
2014
Congruence [==]
Congruence [==]
Proper inclusion [>]
Inverse proper inclusion [<]
Overlap [><]
Congruence [==]
Exclusion [|]
Future Floras: Congruence? [==]
RCC-5 = Region Connection Calculus
with five basic relations.

…and utilize logic reasoning to
infer consistent merge taxonomies.

Merge – A. glomeratus sec. Blomquist (1948) / sec. Campbell (1983)
Congruence [==]
Merge View Legend

We now have a tool for this: Euler/X
https://bitbucket.org/eulerx

Euler/X toolkit in a single screenshot (desktop version, IX-2014)

Euler/X applies logic reasoning
to support the following workflow:

User/reasoner interaction: achieving well-specified alignments
T1 = Taxonomy 1
T2 = Taxonomy 2
A = Input articulations
[==, >, <, ><, |]
C = Taxonomic constraints

T1 = Taxonomy 1
T2 = Taxonomy 2
A = Input articulations
[==, >, <, ><, |]
C = Taxonomic constraints
 Articulations are asserted
by taxonomic experts.

Data format for an Euler/X alignment input file
T2 Year Author

T2 Year Author
Parent
concept Child
concepts

T2 Year Author
Parent
concept Child
concepts
T1

T2 Year Author
Parent
concept Child
concepts
T1
T2 to T1
Articulations
(as provided
by the user)

Input visualization of the 2005/1993 concept trees & articulations
Input articulations
2005 concepts
1993 concepts

No!

No Possible World merge [empty canvas, nothing to report]

No!
Yes

Nine Possible World merges for an under-specified use case input

Yes
Yes

MIR =
Maximally Informative Relations
[==, >, <, ><, |]
for each concept pair
Yes
Yes

Use case 1: dwarf lemurs sec. 1993 & 2005 1
Chirogaleus furcifer sec. Mühel (1890) – Brehms Tierleben.
Public Domain: http://books.google.com/books?id=sDgQAQAAMAAJ
1 Franz et al. 2014. Two influential primate classifications logically aligned. (unpublished)

The 2nd & 3rd Editions of the Mammal Species of the World
1993 2005
Primates sec. Groves (1993)
 317 taxonomic concepts,
233 at the species level.
Primates sec. Groves (2005)
 483 taxonomic concepts,
376 at the species level.
Δ = 143
species-level
concepts

Primate 1993/2005 concept alignments:
From simple to complex merge taxonomies.

Microcebus rufus sec. 2005 – same name, congruent concepts [==]
1. Input concepts & articulations
Merge View Legend

Microcebus rufus sec. 2005 – same name, congruent concepts [==]
2. Merge visualization
Grey rectangle, round corners
 Taxonomic congruence
Merge View Legend

Mirza coquereli sec. 2005 – name change, congruent concepts [==]
Merge View Legend

Microcebus murinus (et al.) sec. 2005 – "lumping / splitting" [> , <]
Merge View Legend

Microcebus murinus (et al.) sec. 2005 – "lumping / splitting" [> , <]
Yellow octagon
 Unique to T1 (1993)
Green rectangle
 Unique to T2 (2005)
Merge View Legend

Microcebus (part) & Mirza sec. 2005 – monotypic parent concepts
Mirza & M. coquereli sec. Groves (2005)
are two co-extensional concepts in T2

Microcebus (part) & Mirza sec. 2005 – monotypic parent concepts
Mirza & M. coquereli sec. Groves (2005)
are two co-extensional concepts in T2
Three concepts
are congruent!

How can we represent concept overlap?

Microcebus (all) & Mirza sec. 2005 – concept overlap [><]
Merge visualization: containment, with overlap [-e mnpw --rcgo]
Dashed blue line
 Overlap [><]

Unique to 1993.Microcebus
(2005  Mirza/coquereli)
Dashed blue line
 Overlap [><]

(1993  undescribed)
Dashed blue line
 Overlap [><]

(1993  undescribed)
Dashed blue line
 Overlap [><]
Shared, congruent
child concepts

We can resolve the merge overlap products.

Merge visualization: "merge concept" representation [-e mncb]
Red lines
 Newly inferred articulations
(to and from merge concepts)

Red lines
2005.Microcebus*1993.Microcebu
s
 Shared merge concept

1993.Microcebus2005.Microcebus
 Merge concept unique to 1993
2005.Microcebus1993.Microcebus
 Merge concept unique to 2005
2005.Microcebus*1993.Microcebu
s
 Shared merge concept
Red lines

Scalability & information gain:
How many input articulations are sufficient?

Cheirogaleoidae sec. 2005 – how many articulations are sufficient?
T2: 27 concepts; T1: 14 concepts; 22 input articulations

17 'non-new' 2005 species-level concepts
 Articulated to 1993 species-level concepts

4 'new' 2005 species-level concepts
 Exclusion (|) from 1993 family-level
concept

1 additional highest-level articulation
 2005.Cheirogaleoidae > 1993.Cheirogaleidae
 Eliminates 15 additional Possible Worlds

No genus-/subfamily level
articulations are needed

Well-specified merge: 378 Maximally Informative Relations
 ~ 17x information gain through reasoning

Well-specified merge: 378 Maximally Informative Relations
 ~ 17x information gain through reasoning
Primates: 483x317 = 800 concepts
402 articulations
153,111 MIR
 ~ 380x information gain!

Why?
Performance of names as concept identifiers.

MSW 2nd/3rd Edition name/concept identity relations
 56.4% of the paired name lineages are taxonomically reliable.
 Computers need concept resolution to track taxonomic provenance.

Use case 2
And Andropogon glomeratus sec. auctorum? 1
Photo by Max Licher (ASU Herbarium); Cottonwood, Arizona.
http://swbiodiversity.org/seinet/imagelib/imgdetails.php?imgid=431755
1 See Franz et al. 2014. Names are not good enough: reasoning over taxonomic change in the Andropogon complex.
Semantic Web – Interoperability, Usability, Applicability – Special Issue on Semantics for Biodiversity. (in press)

In brief: Things are very messy.

Question 1: Which concept labels have included the name string
"Andropogon glomeratus" in past eight classifications?
Tabular alignment of eight Andropogon classifications: 1889 to 2006
 6 / 8 classifications are taxonomically unique for the concept of
A. glomeratus sec. auctorum.
 No two concepts including the "A. glomeratus" name string are
taxonomically congruent.

Question 2: Which previously named concepts are congruent with
Andropogon glomeratus sec. Weakley (2006)?
Tabular alignment of eight Andropogon classifications: 1889 to 2006
 What Weakley (2006) refers to as "A. glomeratus" was previously referred to as:
1889: A. macrourus var. hirsutior + A. macrourus var. abbreviatus
1933: A. glomeratus (in part, I)
1948: A. glomeratus (?)
1950: A. virginicus var. hisutior + A. glomeratus (in part, II)
1968: A. virginicus (in part)
1979: A. virginicus var. abbreviatus (in part)
1983: A. glomeratus (in part, I)

Case 1: 1948.Blomquist vs. 1950.Hitchcock & Chase (Δ = 2 years)
T2: 7 concepts (1950); T1: 7 concepts (1948) – containment view
Merge: 3 congruent regions, 3 with same name
6 unique regions, 4 with non-unique name

 A. glomeratus sec. 1950 and A. glomeratus sec. 1948 are overlapping, as each concept
includes a non-congruent variety-level concept.
 Interestingly, the shared concept region has no unique name in either taxonomy. It is
'un-named', at least within the context of the 1950/1948 classifications.

T2: 7 concepts (1950); T1: 7 concepts (1948) – merge concept view
 The shared, overlapping region is more informatively resolved and labeled in the merge
concept visualization; the region 1950.A._glomeratus * 1948.A_glomeratus contains no
subelements that carry the name "A. virginicus" in either classification.

Case 2: 1889.Hackel vs. 2006.Weakley (Δ = 117 years)
T2: 12 concepts (2006); T1: 12 concepts (1889)
Merge: 8 congruent regions, 0 with same name (!)

Case 2: 1889.Hackel vs. 2006.Weakley (Δ = 117 years)
T2: 12 concepts (2006); T1: 12 concepts (1889)
Merge: 8 congruent regions, 0 with same name (!)
 Hackel & Weakley agree very substantively on what entities are
'out there in nature'; however, more than a century of Code-compliant
name changes has obscured their agreements.

Case 3: 1983.Campbell vs. 2006.Weakley (Δ = 23 years)

Case 3: 1983.Campbell vs. 2006.Weakley (Δ = 23 years)
 One of the simpler merge taxonomies in this use case, although
8 / 15 merge regions have taxonomically misleading names (i.e.,
congruence/different names; non-congruence/same names).
 This ratio is near-average through nine pairwise alignments.

In conclusion – feasibility, accessibility, and what it means.
• Feasibility of tracking taxonomic concept provenance in computational logic:
• We are making leaps and bounds in feasibility (and in scalability) right now.
• However, many interesting challenges remain (e.g., user/reasoner interaction).

• Accessibility and acceptance of the RCC-5/reasoning approach:
• We need more use cases, and users – the Euler/X approach works!
• It can be applied to any new or legacy systematic publication, biodiversity
database, checklist, classification, phylogeny, or other kinds of taxonomic
syntheses (print or virtual) and versions thereof; complementing the Linnaean
system while providing superior individuation of taxonomic content.
• Having a sound web service is the next critical step in advancing the approach.

• Accessibility and acceptance of the RCC-5/reasoning approach:
• We need more use cases, and users – the Euler/X approach works!
• It can be applied to any new or legacy systematic publication, biodiversity
database, checklist, classification, phylogeny, or other kinds of taxonomic
syntheses (print or virtual) and versions thereof; complementing the Linnaean
system while providing superior individuation of taxonomic content.
• Having a sound web service is the next critical step in advancing the approach.
• What does it all mean?
• The legacy of taxonomic name and concept authoring is amenable to
computational logic and provenance tracking. We can likely derive much data
integration power from further developments in this direction.

Acknowledgments
• Robert Guralnick, Susanna Drogsvold & all CU Museum of Natural History "The
Meaning of Names" conference organizers!
• Euler/X team: Mingmin Chen, Parisa Kianmajd, Shizhuo Yu, Shawn Bowers
& Bertram Ludäscher
• Juliana Cardona-Duque (weevils), Naomi Pier (primates) & AlanWeakley (grasses)
• taxonbytes lab members: Andrew Johnston & Guanyang Zhang
• NSF DEB–1155984 & DBI–1342595 (PI Franz); IIS–118088 & DBI–1147273
(PI Ludäscher)
Franz Lab: http://taxonbytes.org/ https://sols.asu.edu/

Select references on concept taxonomy and the Euler/X toolkit
• Franz & Peet. 2009. Towards a language for mapping relationships among
taxonomic concepts. Systematics and Biodiversity 7: 5–20. Link
• Chen et al. 2014. Euler/X: a toolkit for logic-based taxonomy integration. WFLP
2013 – 22nd International Workshop on Functional and (Constraint) Logic
Programming. Link
• Chen et al. 2014. A hybrid diagnosis approach combining Black-Box and White-
Box reasoning. Lecture Notes in Computer Science 8620: 127–141. Link
• Franz et al. 2014. Names are not good enough: reasoning over taxonomic change in
the Andropogon complex. Semantic Web – Interoperability, Usability, Applicability –
Special Issue on Semantics for Biodiversity. (in press) Link
• Franz et al. 2014. Reasoning over taxonomic change: exploring alignments for the
Perelleschus use case. PLoS ONE. (in review)
• Euler/X toolkit: https://bitbucket.org/eulerx/euler-project
• Euler web service (in progress): http://euler.asu.edu/
• Concept taxonomy @ taxonbytes: http://taxonbytes.org/tag/concept-taxonomy/

The good: names refer to type specimens necessarily
Source: Witteveen. 2014. Biology & Philosophy. (in press)

The challenge: names refer to non-type specimens contingently
Names
Non-types
Source: Dubois. 2005. Zoosystema 27: 365-426.

We may categorize kinds of nomenclatural
and taxonomic change, and opportunities,
to track each, as follows:

Nomenclatural/taxonomic change & provenance tracking square
E.g.: - A binomial name is formed incorrectly.
- A homonym is discovered, requiring name change.

E.g.: - A type specimen is lost, a neotype must be designated.
- "One fungus (a-/sexual), one name" – Melbourne Code.

E.g.: - A heterotypic synonymy is established (inferred).
- a Priority-carrying name is newly 'transferred'.

E.g.: - A junior genus-level name is transferred among tribes.
- An informal clade name is redefined across treatments.

Many
changes
Some
changes
Many
changes
MOST
CHANGES
???
Question: Which changes are most common in a particular group?
Answer: Concept-level resolution is needed to assess this.

Question: What is the proper scope of reference for representing our
progress in inferring the tree of life?

Suggested answer: Even though the name  taxon mapping is the
ultimate aim..

..in effect we only need to represent the name  concept mapping.
Congruence over time will suggest that we are 'getting taxa right'.

R32 lattice of RCC-5 articulations (lighter color = less certainty)

Higher-level primate classifications
– 1993 versus 2005:
Many recurrent names,
little taxonomic congruence.

Primates sec. 1993 & 2005
Order to Subfamily-level
 Not much is grey.

Strepsirrhini
sec. 2005
Haplorrhini
sec. 2005
Catarrhini
sec. 2005

Use case 2: Perelleschus sec. 2001 & 2006 1
Perelleschus salpinflexus sec. Franz & Cardona-Duque (2013)
DOI:10.1080/14772000.2013.806371
1 Input articulations: Franz & Cardona-Duque. 2013. Description of two new species and phylogenetic reassessment of
PerelleschusWibmer & O'Brien, 1986 (Coleoptera: Curculionidae), with a complete taxonomic concept history of
Perelleschus sec. Franz & Cardona-Duque, 2013. 2013. Systematics and Biodiversity 11: 209–236.
Merge analyses: Franz et al. 2014. Reasoning over taxonomic change: exploring alignments for the Perelleschus use
case. PLoS ONE. (in press)

Goal: align two phylogenies with differential taxon sampling
T1: Perelleschus sec. 2001
• Phylogenetic revision
• 8 ingroup species concepts
• 2 outgroup concepts
• 18 concepts total

Goal: align two phylogenies with differential taxon sampling
• Phylogenetic revision
• 2 outgroup concepts
• Exemplar analysis
• 1 outgroup concept

Logic representation challenge:
Perelleschus sec. 2001 & 2006 concepts
have incongruent sets of subordinate members,
yet each concept has congruent synapomorphies.

Definitional preliminaries 1
Ostensive alignment: the congruence among higher-level
concepts is assessed in relation to their entailed members.
 Ostension: giving meaning through an act of pointing out.
1 See Bird & Tobin. 2012. Natural Kinds. URL: http://plato.stanford.edu/archives/win2012/entries/natural-kinds/

Definitional preliminaries 1
Ostensive alignment: the congruence among higher-level
concepts is assessed in relation to their entailed members.
 Ostension: giving meaning through an act of pointing out.
Intensional alignment: the congruence among higher-level
concepts is assessed in relation to their properties.
 Intension: giving meaning through the specification of properties.
1 See Bird & Tobin. 2012. Natural Kinds. URL: http://plato.stanford.edu/archives/win2012/entries/natural-kinds/

Ostensive alignment – members are all that counts
Challenge 1: Differential outgroup sampling
(2 / 1 concepts)
T2: 2006.PHY & 2006.PHYsubcin
T1: 2006.PHY only
Input constraints
Ostensive alignment
2001 & 2006

(2 / 1 concepts)
T1: 2006.PHY only
Solution: Locally relax coverage with "nc"
= "no coverage"
Input constraints
Ostensive alignment
2001 & 2006

(2 / 1 concepts)
T1: 2006.PHY only
Solution: Locally relax coverage with "nc"
= "no coverage"
Result: 2006.PHY == 2001.PHY
 Outgroups are held congruent.
Input constraints
Ostensive alignment
2001 & 2006

Input constraints Challenge 2: Ostensive alignment
Ostensive alignment
2001 & 2006

Challenge 2: Ostensive alignment
Solution: 11 ingroup concept articulations
are coded ostensively – either as
<, ><, or | – to represent non-congruence
in the representation
of child concepts
Input constraints
Ostensive alignment
2001 & 2006

Challenge 2: Ostensive alignment
Solution: 11 ingroup concept articulations
are coded ostensively – either as
<, ><, or | – to represent non-congruence
in the representation
of child concepts
Result: 2006.PER < 2001.PER
2006.PER | 2001.[5 species concepts]
etc.
Input constraints
Ostensive alignment
2001 & 2006
5 x |
2 x ><

Intensional alignment – representation of congruent synapomorphies
Input constraints
Challenge 3: Intensional alignment
Intensional
alignment
2001 & 2006

Input constraints
Solution: An Implied Child (_IC) concept is
Intensional
alignment
2001 & 2006
added to the undersampled (2006)
clade concept; and the (5) "missing"
species-level concepts are included
within this Implied Child

Input constraints
Solution: An Implied Child (_IC) concept is
Intensional
alignment
2001 & 2006
added to the undersampled (2006)
clade concept; and the (5) "missing"
species-level concepts are included
within this Implied Child
11 ingroup concept articulations are
coded intensionally – as == or > –
to reflect congruent synapomorphies
of 2001 & 2006

Input constraints
Result: The genus- and ingroup clade-level
Intensional
alignment
2001 & 2006
concepts are inferred as congruent:
2006. PER == 2001.PER
2006.PcarPeve == 2001.PcarPsul
etc.

Review – representing ostensive versus intensional alignments
Ostensive alignment
2001.PER includes more
species-level concepts
than 2006.PER [>].

Review – representing ostensive versus intensional alignments
Ostensive alignment
2001.PER includes more
species-level concepts
than 2006.PER [>].
Intensional alignment
2006.PER reconfirms the
synapomorphies inferred
in 2001.PER [==].

The other piece in the puzzle: Concept-to-voucher identifications
Source: Baskauf & Webb. 214. Darwin-SW. URL: http://www.semantic-web-journal.net/system/files/swj635.pdf

Franz. 2014. Explaining taxonomy's legacy to computers – how and why?

Recommended

Recommended

More Related Content

What's hot

What's hot (20)

Similar to Franz. 2014. Explaining taxonomy's legacy to computers – how and why?

Similar to Franz. 2014. Explaining taxonomy's legacy to computers – how and why? (20)

More from taxonbytes

More from taxonbytes (19)

Recently uploaded

Recently uploaded (20)

Franz. 2014. Explaining taxonomy's legacy to computers – how and why?