Taxonomic databases are perpetuating approaches to citing literature that may have been appropriate before the Internet, often being little more than digitised 5 × 3 index cards. Typically the original taxonomic literature is either not cited, or is represented in the form of a (typically abbreviated) text string. Hence much of the “deep data” of taxonomy, such as the original descriptions, revisions, and nomenclatural actions are largely hidden from all but the most resourceful users. At the same time there are burgeoning efforts to digitise the scientific literature, and much of this newly available content has been assigned globally unique identifiers such as Digital Object Identifiers (DOIs), which are also the identifier of choice for most modern publications. This represents an opportunity for taxonomic databases to engage with digitisation efforts. Mapping the taxonomic literature on to globally unique identifiers can be time consuming, but need be done only once. Furthermore, if we reuse existing identifiers, rather than mint our own, we can start to build the links between the diverse data that are needed to support the kinds of inference which biodiversity informatics aspires to support. Until this practice becomes widespread, the taxonomic literature will remain balkanized, and much of the knowledge that it contains will linger in obscurity.

Biodiversity informatics, identifiers, DOI, literature, taxonomy, dark taxa, data cleaning, data integration

Bergman (2001) coined the term “deep web” to refer to the part of the web that is largely hidden from search engine crawlers, such as dynamically generated web pages. A major challenge facing web crawlers is how to “surface” that deep web so that it becomes accessible to search engines. By analogy, while much of the scientific literature is readily discoverable, taxonomic literature remains largely obscure.

As an example of the consequences of this obscurity, consider the fate of the name Leviathan as used for a recently discovered fossil whale described in Nature (Lambert et al. 2010a). Unbeknownst to the authors, the name Leviathan was previously used for an extinct mammoth (Koch 1841). Once the homonymy was uncovered, an erratum was published in the same journal (Lambert et al. 2010b). The erratum cites the original publication of Leviathan (Koch 1841), but if the reader visits the corresponding page on the journal Nature’s website there is no link to the digitised version of this publication, despite it being available in the Biodiversity Heritage Library (http://biodiversitylibrary.org/). The lack of a link is symptomatic of the poor state of digital access to taxonomic literature. Reading the list of literature cited in a modern taxonomic paper online, it is striking that while citations of papers on phylogenetics, ecology, or methodology typically include links directly to that article (for example, using Digital Object Identifiers, DOIs), the citations to taxonomic literature are mostly devoid of such links. In an age when most readers will expect any publications worth reading to be online, the absence of links to the primary taxonomic literature consigns it to a ghetto where only the most determined and well-resourced readers will dare to venture. For many readers the lack of a link means a dead-end in their search for information.

Unless we want the taxonomic literature to linger in obscurity we need to make it easily findable and accessible. An obvious starting point would be if taxonomic databases linked to the digitised taxonomic literature. However, most taxonomic databases are little more than online collections of 5 × 3 index cards, a technology Linnaeus himself pioneered (Müller-Wille and Charmantier 2011). Often databases simply present the user with lists of names, divorced from the associated taxonomic literature (such as the original publication of the name). When literature is cited, it is typically as a text string, lacking either an identifier or a link that the researcher can use to locate the publication. This is not to deny the value of the heroic efforts of indexers such as Charles Davies Sherborn (Cleevely 2009), but it is striking how persistent conventions from the print world remain, despite the Internet removing many of the physical constraints of that medium. For example, the recent publication of the Rotifer List of Available Names (LAN) (Segers et al. 2012) comprises a list of references in abbreviated form (“microcitation”) instead of the full publication details, and the list lacks any bibliographic identifiers. Part of the goal of publishing such a List of Available Names is to enable others to scrutinise it, yet the authors of the list provide virtually no assistance to the reader in locating the corresponding literature.

If we accept that the key documents of taxonomy are the publications that contain the names, descriptions, nomenclatural changes, and taxonomic revisions, then a major challenge is to “surface” these documents so that readers can discover them. This means changing practices that have served the community well in the pre-digital era, but which are now hindering its progress. One of the key changes will be the adoption of globally unique identifiers for the taxonomic literature.

The taxonomic community’s experience with globally unique identifiers has been mixed. Several factors have contributed to this. The first is the saga of Life Science Identifiers (LSIDs) (Martin et al. 2005) which seemed a promising technology for identifying data in biology, but in the end the biodiversity community were the only major adopters. This was compounded by the lack of reusing existing identifiers. Every project employing LSIDs created their own identifiers for their data, and rarely, if ever, used LSIDs from other projects. For example, both the Index of Organism Names (ION, http://www.organismnames.com/) and ZooBank (http://zoobank.org/) have records for the genus name Tyrannobdella, each with their own LSID (urn:lsid:organismnames.com:name:4439403 and urn:lsid:zoobank.org:act:43D55B49-C888-4D6B-AF6F-61238EC1339B, respectively). Neither database acknowledges the existence of the other by using the other’s identifier. Furthermore, neither ION nor ZooBank use the most obvious identifier for the PLoS One paper that published Tyrannobdella (Phillips et al. 2010), namely the DOI: 10.1371/journal.pone.0010057.ION represents the reference as a text string:

“Tyrannobdella rex n. gen. n. sp. and the evolutionary origins of mucosal leech infestations. PLoS ONE, 5(4) 2010: e1057, 1–8.”

ZooBank mints its own identifier for the PLoS One paper: urn:lsid:zoobank.org:pub:8D431ED1-B837-4781-A591-D3886285283A (since this was written ZooBank has added the DOI for this article). Ironically, the only thing that links these two records together is the taxonomic name “Tyrannobdella”.

A consequence of the failure to reuse existing identifiers is that the biodiversity informatics community has created a large amount of data identified by a technology few people understand (LSIDs, which by default wouldn’t work in a web browser) and with very few cross-links. This lack of links means each database is effectively another silo, and hence many of the expected benefits of serving biodiversity data in RDF (Page 2006) have not materialised.

This experience may encourage a healthy scepticism about the utility of identifiers, but I would argue that this is because we’ve overlooked the importance of their reuse. If different databases insist on minting their own identifiers and not using (or linking to) existing identifiers, then our data will remain in silos. Reusing identifiers will help establish links between databases, and it is these links that will be the basis of many of the hoped-for inferences we can make in biodiversity informatics (Page 2008).

The first step towards improving the current generation of taxonomic databases would be to associate the taxonomic literature with existing digital identifiers, such as DOIs. Admittedly, this will not always be straightforward. Although DOIs are the bibliographic identifier of choice, and CrossRef provides tools for locating an existing DOI for a reference, it is not always straightforward to find a DOI for a publication. Part of the difficulty in citing the older literature is that many of the conventions we take for granted in modern scientific articles are lacking. Modern articles have titles, and are published in journals that usually have an unambiguous name, volume number, and pagination. This triplet is usually unique, and makes it relatively easy to locate an article in a bibliographic database (Page 2009). However, these conventions need not apply to older publications. For example, (Bennett and Jarvis 2004) cite the following paper:

Ogilby W (1838) On a collection of Mammalia procured by Captain Alexander during his journey into the country of the Damaras. Proceedings of the Zoological Society of London 1838:5–15.

This journal has been digitised by both Wiley and BHL. Wiley makes pages 5-15 available as an article with the doi: 10.1111/j.1096-3642.1838.tb01402.x and attributes the authorship to Richard Owen, not W. Ogilby. On inspection we see that pages 5–15 comprise two articles, one by Ogilby and one by Owen. The first paragraph of page 5 contains the text:

“A selection of the Mammalia procured by Captain Alexander during his recent journey into the country of the Damaras, on the South West Coast of Africa, was exhibited, and Mr. Ogilby directed the attention of the Society to the new and rare species which it contained.”

Subsequent authors have transformed this sentence into the article title “On a collection of Mammalia procured by Captain Alexander during his journey into the country of the Damaras”. Note also that in this case, there is a mismatch between the granularity at which taxonomists cite the literature and the granularity at which Wiley has assigned the identifier (the DOI corresponds to two articles). Perhaps the most obvious example of this mismatch is exemplified by the BHL, which typically recognises units at the scale of journal volume, or individual pages, but not at article level (Page 2011a).

Discovering existing identifiers for the taxonomic literature will sometimes be difficult, for a multitude of reasons. For example, taxonomic databases often store an abbreviated (or even corrupted) version of the citation, the citation may be translated from its original language, or the journal may have been renamed and the new name applied retrospectively to older issues (Page 2011c). All of this makes creating the mapping tedious, but this mapping need only be done once.

As a final motivation to surface deep taxonomic data, consider the rise of “dark taxa” in genomics databases (Page 2011b). A growing percentage of “taxa” in GenBank lack a formal scientific name; in 2010 dark taxa comprised over 80% of invertebrate taxa added that year (Parr et al. 2011). Many of the most recent dark taxa are a product of DNA barcoding projects, and at the time of writing these sequences have been “suppressed” by GenBank, that is, they are still in the database but do not feature in search results. But there is still a background trend towards increasing numbers of unidentified sequences in GenBank. A significant challenge will be determining whether these dark taxa represent newly discovered taxa, or come from known taxa but have not been identified as such (Hibbett and Glotzer 2011; Nagy et al. 2011).

It is clear that some dark taxa do, in fact, have names. For example, consider the frog “Gephyromantis aff. blanci MV-2005” (NCBI tax_id 321743), which has a single sequence AY848308 associated with it. This sequence was published as part of a DNA barcoding study (Vences et al. 2005). If we enter the accession number AY848308 into Google we find two documents, one the supplementary table for (Vences et al. 2005), the other the a subsequent paper by (Vences and Riva 2007) that describes the frog with this sequence as a new species, Gephyromantis runewsweeki. This is a relatively straightforward example, and the taxonomic description is freely available online. But it still required significant time to track down the species description for this one example.

A key question facing attempts to find names for dark taxa is whether the methods available can be scaled to handle the magnitude of the problem. One could argue that newer technologies such as DNA barcoding make classical taxonomy less relevant, and perhaps the effort in digitising older literature and exposing the taxonomic names it contains is misplaced. A counter argument would be that the taxonomic literature potentially contains a wealth of information on ecology, morphology and behaviour, often for taxa in areas that have been subsequently altered by human activity. Furthermore, as technologies such as barcoding uncover previously overlooked variation, older taxonomic names previously sunk in synonymy may yet become relevant. For example, several taxa have been synonymised with the silvery mole-rat Heliophobius argenteocinereus Peters, 1846 (Peters 1846) but DNA sequence data has revealed several clades within that species (Faulkes et al. 2011). Consequently, rather than coin new names for these clades we can rescue older names from synonymy. Hence DNA barcoding may give a new lease of life to old names.

Names may have a special place in the hearts of taxonomists (Patterson et al. 2010) but the pace of biodiversity discovery is outstripping our ability to put names on taxa, as evidenced by the rise of dark taxa in GenBank. There are increasing calls to adopt less formal taxonomic naming schemes (Schindel and Miller 2010), or to focus on describing biodiversity without necessarily naming it (Deans et al. 2012; Maddison et al. 2011). Underpinning much of this call to “ramp up” the rate of biodiversity description will be identifiers, assigned to the entities that taxonomy deals with, including specimens, genotypes, phenotypes, publications, and, yes, taxonomic names. As I have argued previously (Page 2008), in many ways taxonomists have been doing this already but without using web-friendly identifiers. Examples include lists of collection acronyms (Leviton et al. 1985) and author names. The issue now is how do we scale these activities to accommodate the deluge of data we are accumulating as we digitise life and our efforts to document it?

I thank Ellinor Michel for the invitation to speak at the Sherborn meeting, and for her patience as I eventually got around to writing the promised manuscript. I thank the reviewers, Donat Agosti, Ken Johnson, and Rich Pyle for their constructive comments on the manuscript.