Advancements in technologies such as DNA analysis have given rise to new ways in organizing organisms in biodiversity classification systems. In this paper, we examine the feasibility of aligning two classification systems for flowering plants using a logic-based, Region Connection Calculus (RCC-5) approach. The older “Cronquist system” (1981) classifies plants using their morphological features, while the more recent Angiosperm Phylogeny Group IV (APG IV) (2016) system classifies based on many new methods including genome-level analysis. In our approach, we align pairwise concepts X and Y from two taxonomies using five basic set relations: congruence (X=Y), inclusion (X>Y), inverse inclusion (X<Y), overlap (X><Y), and disjointness (X!Y). With some of the RCC-5 relationships among the Fabaceae family (beans family) and the Sapindaceae family (maple family) uncertain, we anticipate that the merging of the two classification systems will lead to numerous merged solutions, socalled possible worlds. Our research demonstrates how logic-based alignment with ambiguities can lead to multiple merged solutions, which would not have been feasible when aligning taxonomies, classifications, or other knowledge organization systems (KOS) manually. We believe that this work can introduce a novel approach for aligning KOS, where merged possible worlds can serve as a minimum viable product for engaging domain experts in the loop.
With the advent of large-scale technologies and datasets, it has become increasingly
difficult to organize information using a stable unitary classification scheme over time.
An ideal work classification system, as noted in [
In the biodiversity communities, taxonomies, a type of KOS under the classification
schemes [
In this paper, we propose the use of a logic-based approach to compare and
reconcile two major classification systems in the field of plant systematics, with the aim of
demonstrating the feasibility of integrating two classifications that result in numerous
possible solutions. Further, we demonstrate the computational power that can aid us in
aligning KOS which could not have been possible when working with alignments
manually. Specifically, we consider a use case of the flowering plants classification systems
by (1) Arthur Cronquist (1981) [
In line with the traditions of Bentham and Hooker, Takhtajan, and Bessy, the Cronquist
system [
However, rapid breakthroughs in DNA studies and technologies have given rise to
a more recent camp of approaches in classifying plants based on phylogenetics. Early
1 Taxonomic Concept Labels (TCLs): name sec. source
2 In biodiversity classification systems, the sequence of the concepts is not taken into account.
cladistic analysis or phylogenetic systematics, established by Willi Hennig, has put
systematics to the task of finding shared, derived character states among any three groups
of organisms to find their common ancestors, or clades [
Taxonomies are one type of KOS with hierarchical structures, similar to classification
schemes [
PW4 is_a (inferred)PW62 overlaPpsW(i7nferred) 1
is_a (input) 2 is_a (inferred) 2 overlaps (inferred) 1 is_a (input) 2 Fig. 1. A simple taxonomy alignment problem with T1, T2 and 7 possible worlds = T1.a
T2.x T1.b
T1.c
T2.y
T2.z
Input Taxonomies T1.a T2.x T1.a T2.x
T2.z T1.b T2.y T1.b
PW3
T1.c T2.y T1.c T2.z
T1.a
T2.x T1.a T2.x
Nodes congruent T2.z 1
T2 2 T1 T1.c 2
Edges overlaps (inferPreWd)5 1 is_a N(inofdereresd) 2 icso_nag(riunepnutt)T2.y 12 T1.a T2 2 T2.x T1
Edges T1.c 2
T1.b T2.y T1.c
T2.z PW1
T1.b T2.y T1.b T2.z
T1.b
T2.z TN1.oades coTn2g.xruent 3 T1.c
Edges T2.y is_a (inputP)W2 2
Nodes congruent 1
T2 T2.y 2 T1 2
EdgeTs1.c T1o.averlaps (inferred) 1 T2.x is_a (NinofedrerTes2d.)z 2 isc_oang(rinupeuntt) 12
T2 T1.b 2 T1 2 Edges
Nodes congruent 3
Edges is_a (input) 2
Nodes congruent
T2 T1 Edges
Given two taxonomies T1 and T2, and a set of articulations (relationships between
concept X in T1 and concept Y in T2, defined in RCC-5 relations), the Taxonomy
Alignment Problem (TAP) is to derive a merged taxonomy T3. A TAP may have zero, one,
or many solutions – inconsistent, unique, or ambiguous solutions respectively. In
previous work [
It is well-known that the basic taxonomic ranks in biological classification include
kingdom, phylum, class, order, family, genus, and species. In this research, we are mainly
focusing on the alignment of the family-level flowering plants. Among the 295,383
flowering plants species [
Both the Cronquist systems and the APG system have these 40 families or some modifications to the names of these families in their classifications. Each family serves as a concept in our alignment. In our first alignment study, if the family in both systems shares the exact same name, we assume (possible incorrectly) that they are congruent to each other. If there are similar but different names, we will leave the concepts unmapped at first. To be more specific, if concept X in the Cronquist system is exactly the same as concept Y in the APG system, we will mark them as [C.X {=} APG.Y]. See Figure 2 for our initial input taxonomies. The following six articulations are the ones that are uncertain to us because they have different names: [C.Caesalpiniaceae ? APG.Caesalpinioideae] [C.Mimosaceae ? APG.Mimosoideae] [C.Fabaceae ? APG.Faboideae] [C.Aceraceae ? APG.Sapindaceae] [C.Sapindaceae ? APG.Sapindaceae] [C.Hippocastanaceae ? APG.Sapindaceae] ts o c i d u E l a s a B .
G P A s d li o n g a M .
G P A s e l a i m a L .
G P A e d a l C d ir e t s A .
G P
A il.tsneaaenPGG il.scspaeaPADG ll.soaneaSPAG l.trsseeaAPAG A a t y h p o li o n g a M .
C 1 3 8 7 ts o c o n o M .
G P A s e l a li o n g a M .
G P A s e l a l u c n u n a R .
G P A s s le le iilhagp itrcabu a u .M .C G G P P A A s il.rscsaeaB ll.vsaeaGM l.sabeaFG G P P P A A
A tems (green=Cronquist; yellow=APG), with the Fabaceae and the Sapinadaceae relationship unknown on both sides (red dotted lines)
Due to the above six uncertain relationships, we expect that the articulations
between the concepts of the two classifications in our initial input taxonomies are
underspecified and will result in numerous possible worlds. Furthermore, the bold
assumption to mark names that are spelled the same in both taxonomies as congruent is often
not enough, as noted in [
The first round of alignment between the two classification systems resulted in 555 possible worlds—meaning that we have 555 different ways for reconciling these two classifications (see Figure 3 for a few examples). The alignment resulted in so many solutions because the articulations for the Fabaceae family (the beans family) and the Sapindaceae family (the maple family) were left ambiguous and underspecified.
In the Cronquist system, the three families placed under the order Fabales are: Caesalpiniaceae sec. Cronquist Mimosaceae sec. Cronquist Fabaceae sec. Cronquist Caesalpinioideae sec. APG IV Mimosoideae sec. APG IV
Faboideae sec. APG IV While in the APG IV system, the order Fabales only consists of one family Fabaceae, and under the Fabaceae there are three sub-families:
First, the family/subfamily names between the two systems are not exactly the same,
the spelling is different in the suffix ( -ceae vs. -deae); furthermore, the ‘umbrella’
family Fabaceae in the APG system is spelled the same as one of the three bean families in
the Cronquist system (also Fabaceae). Similarly, we raise doubts on whether the
Sapindaceae in the Cronquist system is entirely equivalent to the Sapindaceae sec. APG
IV; though they share the exact same name, the Aceraceae sec. Cronquist, and
Hippocastanaceae sec. Cronquist, also in the order of Sapindales, were totally void in the APG
system. Therefore, in our initial attempt we could not firmly state what the articulations
among these families were, echoing the claim that “names are not enough” in [
After consulting the domaiCn.Caeryoxphpylelalrest on our secondCr.Poortulacaceoaef alignm=ent, we AwPGe.Proertulaacbacelaee und to refine our alignments betweeC.nPoltyghonealetswo taxonomies and make the rel=ationshiApPGs.Paolmygoonancegae the beaCn.Csaryfoaphmyllidialeies explicit. DespC.iStaelicatlehs e slight differenCc.ePoslygionnactehaee suffix, the three families under Cronquist indeed were congruent to the three sCu.Sbali-cafcaeameilies wit=hin the AAPPG.GSalicaceae sys
tem (Figure 4), and that the Sapindaceae sec. CronquisC.tV,iolAacecaeeraceae se=c. Cronquist, and
Hippocastanaceae sec. CronquCi.sCtapaparraelesall combined as one as Sapindaceae sec.AAPGP.ViolaceIaVe
G
now. C.Malvales C.Cucurbitaceae = APG.Cucurbitaceae
New developments, whether in technologies or in paradigms, always challenge the
views of past major knowledge organization systems. In the biodiversity domain,
molecular data when examined under careful sampling, can provide valuable pointers to
the classification of plants. However, to quote [
This paper serves as an exploratory research on the comparison and alignment of two KOS, specifically classification schemes. Our approach suggests that classifications can coexist with each other while disambiguating the names among concepts in a merged possible world. Our approach also demonstrates the capability of computationally solving complex logic-based alignments for cases where, e.g., due to underspecified relations in the input KOS alignments, manual efforts would likely fail to yield all 555 different ways to merge and reconcile the two KOS. = = = = = = = = = = =
C.Portulacaceae APG.Portulacaceae C.Scrophulariales APG.Lamiaceae C.Lamiaceae C.Lamiales
C.Oleaceae APG.Oleaceae C.Scrophulariaceae APG.Scrophulariaceae
Nodes
C congruent APG
Edges is_a (inferred) is_a (input) overlaps (inferred) 12 15 42 9 12 69 C.Hamamelidae C.Rosidae APG.RosidClade APG.BasalEudicots APG.Ranunculales APG.Magnoliaceae APG.Magnoliales APG.Magnoli ds C.Magnoliaceae C.Magnoliales APG.Asparagales
C.Liliales APG.Arecaceae APG.Arecales C.Arecaceae C.Arecales APG.Alismatales APG.Araceae C.Araceae C.Arales
APG.CoreEudicots APG.Hamamelidaceae APG.Saxifragales C.Hamamelidaceae C.Hamamelidales
C.Fagales APG.Fagales APG.Fabaceae APG.Fabales C.Fabales APG.Rosales APG.Sapindaceae APG.Sapindales C.Sapindales APG.Myrtales APG.Onagraceae
C.Myrtales C.Onagraceae APG.Malpighiales APG.Malvaceae APG.Malvales C.Malvaceae C.Malvales C.Violales APG.Apiaceae APG.Apiales C.Apiaceae
C.Apiales APG.Brassicaceae APG.Brassicales C.Brassicaceae C.Capparales APG.Ericaceae APG.Ericales C.Ericaceae C.Ericales
C.Asteridae APG.Ranunculaceae C.Ranunculaceae C.Ranunculales APG.Papaveraceae C.Papaveraceae C.Papaverales APG.Orchidaceae C.Orchidaceae C.Orchidales C.Iridaceae APG.Iridaceae APG.Liliaceae APG.Liliales C.Liliaceae
C.Fagaceae APG.Fagaceae C.Betulaceae APG.Betulaceae C.Fabaceae APG.Faboideae C.Caesalpiniaceae APG.Caesalpinioideae
C.Mimosaceae APG.Mimosoideae APG.Moraceae C.Moraceae C.Urticales APG.Rosaceae C.Rosaceae
C.Rosales line=inferred overlapping relationship from Euler/X)
Somewhat ironically, the limitation in this study also lies in the strong inferential power—in which when a parent node only has one child, the RCC reasoner in Euler/X will collapse the concepts and merge them as the same node. For example, Euler/X derived that the family Ericaceae and the order Ericales are exactly the same and merged Cronquist.Ericaceae, APG.Ericaceae, Cronquist.Ericales, and APG.Ericales as congruent. However, we argue that this limitation is due to the fact that we have only chosen some 40 major flower families instead of all 416 families. We could add missing children or artificial children here, or include the full 416 angiosperms families, but this is beyond the scope of our study. For the purposes of demonstrating the logic-based taxonomy alignment approach, our smaller use case is sufficient and more tractable .
It is also worth noting that domain expert opinions are still needed to differentiate and lead us to the single solution we are looking for. We foresee our logic-based approach for aligning KOS as an essential preliminary processing steps and a minimum viable product before bringing the KOS interoperability alignment problems to the domain experts. KOS alignment problems are usually complicated with a slow learning curve; domain experts, in our case, plant systematists, may not fully comprehend at first the reason we need to align different KOS. If we approach them directly with two classifications and ask them one by one what relationships between each concept are, they will probably feel befuddled by the situation. Once we have the Euler/X-generated possible worlds in the first round of alignments, whether a few, tens, hundreds, or even thousands of PWs, the alignment problem will become concrete to the domain experts and consulting them for validation of the articulations would be much easier.
This study, we also believe, has further implications on making our classification systems more “full of beans” (here we take on the positive connotation, meaning full of energy), meaning that it may open doors to enable semantic interoperability, and enrich diversity in classification systems when we work with KOS alignments using the logicbased RCC-5 approach. We believe that in the future we can implement this approach for semantic interoperability issues among classifications in the information science community, or even other higher-level KOS such as ontologies. 7
The first author wishes to thank Dr. Stephen R. Downie for the wonderful introduction to plant systematics. This research is the outcome of the course project of IB335 and the Independent Study taught by Dr. Downie. The authors also thank Dr. Nico M. Franz and Ms. Ly Dinh for support and kind feedback on this research.