TCS 2 Implementation Experience Report

On this page

TCS 2 Implementation Experience Report

Title
TCS 2 Implementation Experience Report
Date version created
2025-08-09
Part of TDWG standard
http://www.tdwg.org/standards/117
This version
http://rs.tdwg.org/tcs/doc/implementation-experience-report/2025-08-09
Latest version
http://rs.tdwg.org/tcs/doc/implementation-experience-report
Abstract
The TCS 2 Implementation Experience Report describes how TCS terms are already being implemented and provides a worked out example highlighting Taxon Concept Mappings, which are a feature of TCS that is not used that often.
Contributors
Niels Klazenga (Royal Botanic Gardens Victoria, Australia/Atlas of Living Australia, Australia)
Creator
TDWG Taxon Concept Schema (TCS) 2 Task Group
Bibliographic Citation
Taxon Concept Standard Maintenance Group (2025). TCS 2 Implementation Experience Report. http://rs.tdwg.org/tcs/doc/implementation-experience-report/2025-08-09.

1. Introduction

One of the main objectives of the current effort is to make TCS useful again and the current release provides standard terms for concepts that are already widely used in applications that are out there. No new features have been added in this revision of TCS. Terms from TCS 1 for which we do not have clear idea yet of how they can be implemented have not yet been included in this release.

1.1. Status of this document and its content

This entire document is non-normative.

2. Relation with CoLDP

A recent important application, the Catalogue of Life Data Package (CoLDP), has a data model that is very similar to TCS and includes all TCS terms, the only conflict being that ColDP has a Synonym entity that requires an ID, which TCS does not. This might be a legacy from TCS 1 where has synonym is a relationship between Taxon Concepts. In TCS 2, we have split the has synonym relation into the synonym property for relationships between Taxon Concept and Taxon Name—like taxonName—and the intersects property for relationships between Taxon Concepts. When TCS is expressed in tabular form, a separate table is still needed for synonyms, because of the many-to-many relation between Taxon Concepts and Taxon Names, but this is just a pivot table and there should be no identifiers for synonyms. This is the same way the Name Relation table in the CoLDP schema has been implemented.

CoLDP also offers a Name Usage entity “for simpler sharing”, which is more similar to the Darwin Core Taxon.

3. Relation with Darwin Core

TCS can play the same role for the Darwin Core Taxon that the Darwin Core IRI (dwciri) namespace has for the other Darwin Core entities. Moreover the TCS Nomenclatural Type can be used as the object for the dwciri:typeStatus property.

In order to convert existing Darwin Core Taxon Core data sets to TCS data sets, a taxonConceptID can be obtained by taking the taxonID if the name is accepted or the acceptedNameUsageID if the name is a synonym. If there is no acceptedNameUsageID, but acceptedNameUsage is used instead, the taxonConceptID is the taxonID of the record of which the scientificName is the same as the acceptedNameUsage of the record for which the taxonConceptID is created. If the data set, or record, lacks a nameAccordingTo, the nameAccordingTo is the data set itself.

While for sharing purely taxonomic data TCS is the preferred format, at the interface of taxonomic data and occurrence data, e.g., identifications, Darwin Core will often be preferred. The dwc:taxonConceptID can be used to reference a tcs:TaxonConcept from a dwc:Taxon. For the purpose of distribution maps, etc., dwc:Taxon and tcs:TaxonConcept can be considered equivalent.

4. An example

The following example will (hopefully) serve to demonstrate some of the workings of TCS and how it can be used. This example deals with five different treatments (Tan & Koponen, 1983 [tan_dicranoloma_1983]; Eddy, 1988 [eddy_sphagnales_1988]; Tan, 1989 [tan_dicranoloma_1983]; Norris & Koponen, 1990 [norris_bryophytes_1990]; Klazenga, 1999 [klazenga_revision_1999]) of the moss genus Dicranoloma in Malesia (Malaysia, Indonesia, Brunei, The Philippines and Papua New Guinea). It contains a total of 71 Taxon Concepts (67 at the species level or below) and 133 Taxon Names. All the data used below comes from the JSON-LD file of one of the examples.

Some examples of Taxon Concepts with their accordingTo, accepted name (taxonName) and synonyms (synonym), when entered into a graph database, are shown in figure 1.

Figure 1.

Figure 1. Taxon concepts (green dots) with their accordingTo (blue dots) and all associated names (orange dots). Taxon Concepts are connected to names via the taxonName and synonym properties and Taxon Names are connected to each other through the basionym and replacedName properties. A. Dicranoloma assimile sec. Klazenga 1999 (figure-1a.svg), B. Dicranum assimile sec. Norris & Koponen 1990 (figure-1b.svg), C. Dicranoloma assimile sec. Tan 1989 (figure-1c.svg), D. Dicranoloma assimile s.s. sec. Eddy 1988 (figure-1d.svg), E. Dicranoloma assimile s.l. sec. Eddy 1988 (figure-1e.svg), F. Dicranoloma assimile sec. Tan & Koponen 1983 (figure-1f.svg).

The Taxon Concepts in figure 1 all have the same Taxon Name, Dicranoloma assimile (Hampe) Renauld, so will show at least some overlap. At the same time, they have different synonyms, so they are not exactly the same. If we put all Taxon Concepts and associated references and Taxon Names in the graph, we get a rather nebulous picture of the relationships between the items in the graph (figure 2).

Figure 2.

Figure 2. Graph with all Taxon Concepts (green dots), Taxon Names (orange dots) and references (blue dots).

Removing the references makes for a much clearer picture (figure 3) but there is still a big agglomerate of connected taxon concepts that contains almost half of the Taxon Concepts. While it is easy for humans to see where this cluster can be broken up, it will not be for machines and, more importantly, even humans cannot do anything about it without changing the Taxon Concepts while only working with Taxon Concept–Taxon Name relationships (i.e. accepted names and synonyms) alone.

Figure 3.

Figure 3. Graph with all Taxon Concepts (green dots) and Taxon Names (orange dots).

To address the problem, let’s leave the administration for what it is and see what else there is to see in the data. For this, we make use of the fact that all scientific names have type specimens, so we can use the accepted names and synonyms as a ‘specimen circumscription’ (cf. TCS 1).

To do this properly, we first have to group names with the same type. In TCS, we’ve got the basionym and replacedName properties, representing terms that are used in the domain, to connect names with the same type. However, the basionym and replacedName properties will never connect the same names, and they are always different names from the subject (a name cannot be its own basionym or replaced name), so these properties cannot always be used to connect all homotypic names to the same name.

In the current example we’ve got names that have no other combinations (figure 4A) and combinations that group around a basionym (figure 4B), for which linking up the names through the basionym will work fine, but we also have some more complicated situations, like a replaced name and a replacement name, both of which are the basionym of another name (figure 4C), or a replaced name with a replacement name where the replacement name is the replaced name of another name (figure 4D). In order to link up these names, we have to do something different.

Figure 4.

Figure 4. Taxon Names and their relationships. A. Name without other combinations (figure-4a.svg), B. basionym with two combinations (figure-4b.svg), C. replacement name and replaced name, both of which are the basionym of another combination (figure-4c.svg), D. replacement name of which the replaced name is the replacement name of another name (figure-4d.svg).

There was no appetite in the Task Group for an originalName property, analogous to the Darwin Core originalNameUsage, in TCS, probably because of the problematic semantics, so we have to infer it from the basionym and replacedName relations. This is probably for the best, as for our purposes it is immaterial which name in a group becomes the “original name”, so we do not have to be concerned with the semantics and can do what works best in the situation. In this case, the “original name” is the earliest name (like in dwc:originalNameUsage) but (in contrast to dwc:originalNameUsage) this name does not have to be a legitimate name. So, it will be a different name depending on whether illegitimate and/or invalid names are included in the data set or not. Figure 5 shows the examples from figure 4 with the originalName relations included.

Figure 5.

Figure 5. Same Taxon Name graphs as in the previous figure, but with originalName property connecting combinations to the same “original name” (figure-5a.svg, figure-5b.svg, figure-5c.svg, figure-5d.svg).

One issue that was encountered is that the basionym and replacedName properties cannot be used to link autonyms to a group of names with the same type. This issue is resolved by the basedOn property, for which the only semantics is that it links two names with the same type and that the subject is a more recent name than the object. Thus, basedOn can be used to link names to a homotypic group of names that cannot be linked using basionym and replacedName, e.g. invalid names and autonyms.

There is one other bit of data preparation that needs to be done, which is adding the accepted name and synonyms of infraspecific taxa to the specimen circumscription of their parents.

Now we’ve got everything in place to start mapping Taxon Concepts to each other. To do so, we first get all the pairs of Taxon Concepts that intersect, i.e. have names in common. Then, for each pair, we see which names are in the intersection (A ∩ B) and which names are in one concept and not in the other (A – B or B – A). This helps us find the more precise relationships between concepts. For the names that are in the relative supplements, there is another check to do, as there is only a conflict if the name is used elsewhere in the other treatment. With this proviso, if there are only names in the intersection and there are no names that are in one concept and not in the other, the concepts are congruent (A isCongruentWith B); if the subject (A) has names that are not in the object (B) the subject includes the object (A includes B); if there are names in the object that are not in the subject, the subject is included in the object (A isIncludedIn B); and if there are names that are in the subject and not the object as well as names that are in the object but not the subject, the concepts partially overlap (A partiallyOverlaps B). Examples of each of the intersection types found in this example are given in figure 6.

Figure 6.

Figure 6. Examples of Taxon Concept Mappings using type specimens of names that the concepts apply to. ==: isCongruentWith; >: includes; <: isIncludedIn; ><: partiallyOverlaps. A ∩ B: specimens included in both subject and object Taxon Concepts; A – B: specimens included in the subject Taxon Concept but not in the object Taxon Concept; B – A: specimens included in the object Taxon Concept but not in the subject Taxon Concept.

The Taxon Concepts and their mappings based on the inclusion of type specimens of names are shown in figure 7. The groups of connected concepts, or single concepts, are, of course, the same as in figure 3 where we linked Taxon Concepts via the Taxon Names, because they are based on the same data. What we have gained, however, is that (1) we now have the more precise mapping relations and (2) a more expressive language to align taxon concepts (and reconcile taxonomies) than just ‘accepted name’ and ‘synonym’. Most importantly (3), the taxon concept mappings are things we are allowed to muck around with, as they are not part of the taxon concepts.

Figure 7.

Figure 7. Taxon Concepts with mappings based on the type specimens of names that are included in the concepts (ostensive mappings).

Taxon concept information has two components (cf. Senderov et al., 2018 [senderov_openbiodiv-o_2018]), an ostensive component, which is a list of (some of the) organisms that belong to the taxon concept, and an intensional component, mostly expressed as characters, which is a definition of the allowable variation within (all) the organisms that make up the taxon. In TCS 1 these components are represented by the ‘SpecimenCircumscription’ and ‘CharacterCircumscription’ elements (for TCS 2, we are going to take a bit more time to think about circumscription, so these are not included in this release yet). For ease of use of terms, we call the taxon concept mappings based on the ostensive and intensional components of taxon concept information ‘ostensive mappings’ and ‘intensional mappings’ respectively. For many purposes, and certainly in this example, ostensive mappings can be inferred from the information that is already in the data set, so that is what we have just done, while intensional mappings need to be explicitly added to the data set, which is what we are going to do next.

In this example, in most cases the intensional mapping is the same as the ostensive mapping. Reasons why intensional mappings can be different from the ostensive mappings, and thus the data is misleading, include the following:

  1. Different interpretations of specimens. There is no reason why type specimens cannot be misidentified by later workers. This is the case here for the type specimen of Dicranoloma defoliatum Froehl., which was included in Dicranoloma brevicapsulare sec. Eddy 1988 by Eddy (1988), in Dicranoloma blumei sec. Tan 1989 by Tan (1989) and in Dicranoloma brevisetum sec. Klazenga 1999 by Klazenga (1999). The partiallyOverlaps mappings between these three concepts that were caused by the type of Dicranoloma defoliatum were changed to isDisjointFrom mappings in the intensional mappings.

  2. Ambiguous types. Eddy (1988) treated Dicranoloma laevifolium sec. Eddy 1988 as a species on its own. Tan (1989) and Norris & Koponen 1990 included the type of Dicranum laevifolium Broth. & Geh. in Dicranoloma billarderii sec. Tan 1989 and Dicranum billarderii sec. Norris & Koponen 1990 respectively. None of these workers probably saw the type material in Brotherus’s herbarium, which Klazenga (1999) considered to belong to Dicranoloma braunii sec. Klazenga 1999, but based their concepts on the description and on other specimens identified as Dicranoloma laevifolium. It cannot be discounted that we have to do with a mixed type gathering here and that the material in Brotherus’s herbarium and the material in Geheeb’s herbarium (which is inaccessible) belongs to different species, which is a situation that is not unheard of in bryophytes. In any case, intensionally Dicranoloma braunii sec. Klazenga 1999 does not overlap (is disjoint from) Dicranoloma laevifolium sec. Eddy 1988, Dicranoloma billarderii sec. Tan 1989 and Dicranum billarderii sec. Norris & Koponen 1990. The mappings between the latter three concepts remain unchanged from the ostensive mappings.

  3. Missing data. This is the case when a name is used in some works and not in others. In this example Tan & Koponen (1983) recognise Dicranoloma brachypelma sec. Tan & Koponen 1983, while Eddy (1988) and Klazenga (1999) include the type of Leucoloma brachypelma Broth. in Dicranoloma braunii sec. Eddy 1988 and Dicranoloma braunii sec. Klazenga 1999 respectively. Tan (1989) and Norris & Koponen (1999) do not mention the type of leucoloma brachypelma at all. Because of this, we cannot say with certainty how Dicranoloma braunii sec. Tan 1989 and Dicranum braunii sec. Norris & Koponen 1990 relate to the other three concepts of Dicranoloma braunii (Bosch & Sande Lac.) Paris.

  4. Different geographic scopes. The five treatments have a different geographic focus. While the treaments by Eddy (1988) and Klazenga (1999) are for the entire Malay Archipelago, Tan & Koponen (1983) focus on the Philippines, Tan (1989) on Borneo (especially Gn. Kinabalu in Sabah) and Norris & Koponen (1990) on Papua New Guinea (the Huon Peninsula in Morobe Province). This might explain why Tan (1989) and Norris & Koponen (1990) did not list Leucoloma brachypelma, the type of which is from Java, but not why Tan & Koponen (1983) do have it.

    The example also includes species whose main distribution is outside Malesia. Tan & Koponen (1983) recognise Dicranoloma brassii sec. Tan & Koponen 1983 as endemic in New Guinea, while Norris & Koponen (1990) and Klazenga (1999) include the type of Dicranoloma brassii E.B.Bartram in Dicranum dicarpon sec. Norris & Koponen 1990 and Dicranoloma dicarpon sec. Klazenga 1999. Dicranoloma dicarpon is a mainly Australasian taxon that reaches into Papua New Guinea (and has two records, the syntypes of Dicranoloma formosanum Broth., in Taiwan), so, while ostensibly Dicranoloma brassii sec. Tan & Koponen (1983) is congruent with Dicranum dicarpon sec. Norris & Koponen 1990 and Dicranoloma dicarpon sec. Klazenga 1999, intensionally it is included in those concepts. Eddy (1988) and Tan (1989) include the type of Dicranoloma brassii in Dicranoloma assimile sec. Eddy 1988 and Dicranoloma assimile sec. Tan 1989 respectively, so concepts of Dicranoloma assimile (Hampe) Paris, which are in all five treatments, are affected too.

    In another example, Dicranoloma platycaulon sec. Tan 1989 includes the type of Dicranoloma euryloma var. rugifolium E.B.Bartram, which is also included in Dicranum rugifolium sec. Norris & Koponen 1999 and Dicranoloma assimile sec. Klazenga 1999. Dicranoloma platycaulon is restricted to New Zealand and South-eastern Australia, so Dicranoloma platycaulon sec. Tan 1989 is not included in Dicranum rugifolium sec. Norris & Koponen 1990 or Dicranoloma assimile sec. Klazenga 1990, as the data seems to suggest, but partially overlaps these concepts.

  5. New taxa. Two new taxon names, Dicranum cutlackii Norris & T.J.Kop. and Dicranoloma steenisii Klazenga were published in the period from 1983 to 1999 in which the five treatments were produced. I have concluded that Dicranum cutlackii sec. Norris & Koponen 1990 and Dicranoloma steenisii sec. Klazenga 1999 are taxa that were overlooked before, rather than split off from other concepts, so they do not intersect with concepts from earlier treatments, but it is easy to think of situations where it is the other way. For new concepts that do not have any names in common with existing concepts, it is impossible to say based on the name data alone whether they are truly new concepts, which leave existing concepts unchanged, or whether they have been split off from already recognised taxa, so they affect the concept that keeps the existing name. Extra data needs to be included in the data set to tell these situations apart.

The intensional mappings have been added to the JSON-LD and are shown in figure 8.

Figure 8.

Figure 8. Taxon Concepts with mappings that were added to the data set by the author of the latest revision (intensional mappings).

We can see that the large agglomerate of connected concepts has been nicely broken up now that the intersections that are caused by misidentifications and ambiguous types have been removed. The mappings that do intersect now accurately reflect the horizontal relationships between the taxon concepts (I can see that).

To see what this looks like with a taxonomic backbone, let’s throw in some higher taxa and hierarchical relationships (figure 9).

Figure 9.

Figure 9. Taxon Concepts with intensional mappings and with higher-taxon concepts (yellow dots) and hierarchical (parent) relationships from the latest treatment (Klazenga, 1999).

Now we can see that in the top-left corner there is a group in which the latest revision (where the concepts of higher taxa come from) recognises three taxa and another one in the bottom-right corner in which two taxa are recognised. There is even a group at the bottom just left of the middle in which the latest revision recognises two species in different genera. The rest of the groups contain a single concept from the latest revision. If in these groups all concepts are connected to each other, all concepts in the group are congruent with each other, but if there are things sticking out, there is also overlap or inclusion. For example, in the group in the top-right corner, all other treatments recognise two or three taxa, but the latest treatment lumps those all together into one.

One more thing we can do is grouping the taxon concepts that are congruent to each other. This is what Avibase does with the Avibase IDs. The taxon concepts in these groups (or with the same Avibase ID if we were talking about birds) have the same (or essentially the same) extension, which is the entire set of organisms the concept applies to. We could also say that they are the same taxon, as a taxon is the extension of a taxon concept.

The Avibase publication (Lepage et al., 2014 [lepage_avibase_2014]) speaks of “deep concepts”. These “deep concepts” are already in the taxon concepts and are not a new type of objects that the AviBase ID aplies to. What Avibase does, and what we are doing here, is deduplication, not creating a new type of object.

The “aggregate Taxon Concepts” thus created are shown in figure 10.

Figure 10.

Figure 10. As previous figure, but with “aggregate Taxon Concepts”, representing Taxon Concepts that are congruent with it and each other. Individual Taxon Concepts are related to the aggregate Taxon Concepts through isCongruentWith relationships (not shown). Aggregate Taxon Concepts can be related to each other via includes, isIncludedIn and partiallyOverlaps relationships.

While the deduplication may not have made a big difference in this example, as there are relatively many conflicting concepts, it is easy to see how much of a difference it can make at a larger scale or in less messy groups.

For this example, as will mostly be the case at the level of individual taxonomic revisions, it was easy to create the intensional mappings for all pairs of taxon concepts. At a much larger scale this will of course not work anymore. However, as indicated, in many cases the ostensive mappings we can get from the name matching are correct and they are mostly the best we’ve got. Therefore, it is important to persist these mappings as results of the taxonomy reconciliation, so that experts can go in and say ‘this mapping is incorrect’ or ‘this mapping was missed’, thereby correcting the worst mismatches and missed matches and incrementally improving the data. Letting in mappings from other sources, the further upstream the better, would be great too. For this all to work taxon concepts need to be identifiable in taxonomies (and identifications) and need to have strong identifiers.