The 37 open access Zootaxa articles on spiders contain 254 treatments, of which 212 are species treatments, based on 4,779 specimens (species treatments contain citations of 4,773 specimens, the difference in numbers is due to 26 specimens identified to genus only) (Fig. 3​). Articles contain a mean of 6.9 treatments (5.7 species-rank treatments) per publication and 22.5 specimens per species-rank treatment. 91 (42.9%) treatments are descriptions of new species. There are also 22 new combinations and three new genera. About a third of treatments (72 of all species treatments, 34 of new species descriptions) were based on a single specimen; the treatment with the most specimens had a total count of 851 (Pardosa logunovi in Kronestedt and Marusik 2011). The largest number of species treatments were from the family Agelenidae (28.3%, 60) but the largest number of specimens were Lycosidae (44.3%, 2,116). Adult males made up just over half (51.5%, 2,460) of the specimens, with the remainder divided between adult females (40.7%, 1943) and non-adults (7.8%, 370). The largest number of specimens was collected in Russia (34.8%, 1,660), which when combined with specimens from China (14.9%, 713) accounts for nearly half of the specimens cited. China was the country with specimens appearing in the most treatments (63); only six treatments included specimens from Russia. More than 17% of the specimens were not explicitly associated with an institutional collection (see below); the California Academy of Sciences was cited as the institution archiving the largest number of specimens (14.1%, 674), but Hunan Normal University is the repository for the largest number of primary type (holotype, syntype, lectotype, or neotype) specimens (23.4%, 49) and also had specimens featured in the largest number of treatments (48). Yuri Marusik was by far the most productive specimen collector, alone collecting 869 (19.5%) specimens, not counting the specimens he collected collaboratively with others. Tang Guo collecting alone contributed specimens to the most treatments (17), but Charles Griswold, working alone or with several collaborators, contributed specimens to more treatments (see Discussion: Tracking Individuals). The 1,007 specimens that had the elevation specified were collected across a wide range, with a few (six) collected between 4,001-4,500 m. Overall, most specimens of both sexes but especially males were collected in July, followed by June. The bulk of the specimens cited in this group of publications, which spanned 2002-2014, were collected in the 1990s and 2000s. 2010 was a year of peak activity both in terms of articles and treatments. Peter Jäger was the lead author on the most articles (13.5%, five), but Xin-Ping Wang was lead author on the most treatments (28.3%, 60) and Torbjörn Kronestedt was lead author documenting the most specimens (44.3%, 2,116).

The five spider articles published in Biodiversity Data Journal before August 2014 contain 742 treatments, of which 672 are species treatments, based on 3,432 specimens (species treatments contain citations of 3,399 specimens) (Fig. 4​). Four treatments in this set of articles concern insects (Miller et al. 2013​), the rest concern spiders; if desired, it would be simple to add a filter to show only spider treatments (&FILTER_tax.orderEpithet=Araneae; see Suppl. material 17​). Articles contain a mean of 148.4 treatments (134.4 species-rank treatments) per publication and 5.1 specimens per species-rank treatment. Most of these treatments are faunistic occurence records without descriptive information; one (0.1%) species-rank treatment is a description of a new species. There are no new combinations or descriptions of any new higher rank taxa. About 10% of all treatments (but only 1.3% of species treatments) are missing some higher taxonomic information; this is typically supplied not by the authors but by an automated database query that sometimes fails to interpret the author's intentions (see Döring 2015). 496 (73.8%) of the species treatments explicitly cited specimens. Of these, about two fifths (42.1%, 209) were based on a single specimen; the one new species was based on 69 specimens; the species with the most specimens (10.4%, 352) was Tenuiphantes tenuis, which appeared in three treatments (Čandek et al. 2013, Crespo et al. 2014, Deltshev et al. 2013). The dominant family by both treatments (22.9%, 154) and specimens (33.4%, 1,134) was the Linyphiidae. Adult females made up the majority (60.5%, 2,056) of specimens, with the remainder divided between adult males (38.9%, 1,323) and non-adults (0.6%, 20). The largest number of specimens was collected in Portuguese territory (Madeira island; 43.7%, 1,487). Slovenia was the country with specimens appearing in the most treatments (225); only 48 treatments included specimens from Portugal. Very few specimens (2.4%, 80) were explicitly associated with an institutional collection; these were divided among three collections (two institutions and one personal collection). The single holotype specimen was deposited at Universiti Malaysia Sabah's Institute for Tropical Biology and Conservation, Borneensis (UMS). The team of Kuntner, Gregoric, and Candek collected the largest number of specimen records (43.1%, 823) and their material appeared in the largest number of treatments (143), and members of the team contributed to a substantial number of additional records. The 1,487 specimens that had the elevation specified were collected below 2,000 m, mostly either between 500-1,000 or 1,500-2,000. Most specimens of both sexes but especially females were collected in July (931, 48.8%), more than twice as many as the second month, June (385, 20.2%). All records from these 2013 and 2014 publications were collected during the present decade. Jeremy A. Miller was the only author with more than one article (40%, two), but Klemen Candek was lead author on the most treatments (48.1%, 323) and documenting the most specimens (47.2%, 1,604).

The 41 articles from both journals contain a total of 996 treatments (884 species treatments) based on 8,231 specimens (species treatments contain citations of 8,172 specimens). About 3/4 of the treatments are published in Biodiversity Data Journal, but more than 58% of the specimens are cited among the much larger total number of Zootaxa articles (Fig. 5​). These charts (along with the charts on author and species, below) demonstrate the capability to combine data elements regardless of whether the semantic encoding is applied prospectively as part of an XML-based publication process or retrospectively through markup of legacy literature. ​

The California Academy of Sciences collection contributed specimens to 34 treatments in six articles, all published in Zootaxa; most of them (29 treatments) in three articles published in 2010 (Fig. 6​). Most treatments concerned the family Agelenidae (79.4%, 27). Eighteen (52.9%) of the treatments were new species, but only three holotype specimens cited were archived in the CAS collection. Nearly all of the 688 specimens cited came from three countries: China (50.7%, 342), Russia (36.6%, 247), and South Africa (12.2%, 82). Specimens collected in China appeared in more treatments by far (27) than CAS specimens from any other country. The collecting team of D. V. Obydov and Y. M. Marusik led among collectors of cited CAS specimens (19.4%, 130). The 315 specimens that had the elevation specified were collected up to 4,000 m with the majority collected above 3,000 m. June was the peak month for cited CAS specimens, and all were collected in the 1990s or 2000s.

The 1,660 specimens collected in Russia were cited in six species treatments (five Lycosidae and one Linyphiidae), included three new species, and published in two articles in 2009 and 2011, both in Zootaxa (Fig. 7​). Adult males made up 73% (1,212) of the specimens, with the remainder divided between adult females (26.8%, 445) and non-adults (0.2%, three). Not counting the 32.2% (535) of specimens not explicitly associated in the source documents with an institutional collection, the largest repository for these specimens is the Institute for Systematics and Ecology of Animals, Novosibirsk, Russia (ISEA; 24.5%, 406). The ISEA had their specimens represented in the largest number of treatments (four). However, the Zoological Museum of Turku University, Finland (ZMHU) and the Zoological Museum of Moscow State University, Russia (ZMMU) tie for the largest number of primary type specimens from Russia cited (37.5%, three each). Yuri Marusik was the most productive specimen collector, alone collecting 581 (35.1%) specimens, not counting the specimens he collected collaboratively with others, and contributing material to the largest number of treatments (four). Most of the 224 specimens that had the collecting elevation specified were collected between 1,001-1,500 m. Most specimens of both sexes but especially males were collected in July, followed by June. The bulk of the specimens cited in this group of publications were collected in the 1990s.

The 869 specimens collected solely by Y. M. Marusik were cited in five treatments (four Lycosidae and one Linyphiidae) including two new species, and were published in three articles, all appearing in Zootaxa (Fig. 8​). Most specimens (66.9%, 581) were collected in Russia. The specimens that Marusik collected in Russia contributed to four treatments. Most of the specimens were males (66.9%, 581) collected in June or July during the 1990s. Only one collection of 20 specimens had the elevation specified in the source publication. Not counting the 65.8% (572) of specimens not explicitly associated in the source documents with an institutional collection, the largest repository for these specimens is the Institute for Biological Problems of the North, Magadan, Russia (IBPN; 15.7%, 136). The three primary type specimens collected by Marusik are deposited in the Zoological Museum of Moscow State University, Moscow, Russia (ZMMU).

This article is the seventh in a series on Holarctic members of the wolf spider genus Pardosa (Lycosidae). The publication contains seven species treatments and one higher (species group) treatment based on 1,957 specimens in the family Lycosidae; three of the treatments are new species (Fig. 9a​). Adult males made up 70% (1,370) of the specimens, with the remainder divided between adult females (29.8%, 583) and non-adults (0.2%, 4). Specimens cited in this article are mostly from Russia (82.5%, 1,615), with Mongolia (9%, 176), Canada (7.3%, 142), China (0.9%, 17), and the United States of America (0.4%, 7) also represented; five of the treatments include specimens from Russia. Not counting the 35.5% (695) of specimens not explicitly associated with an institutional collection, the largest repository for these specimens is the Institute for Biological Problems of the North, Magadan, Russia (IBPN; 23.7%, 464). Four of the 12 primary type specimens cited were archived in the Zoological Museum of Moscow State University, Russia (ZMMU). Specimens cited in five treatments are deposited at both NHRS (Swedish Museum of Natural History, Stockholm, Sweden) and IZAS (Institute of Zoology, Chinese Academy of Sciences, Beijing, China). Nearly all of the 204 specimens for which the collecting elevation was specified were taken from 1,001-1,500 m. Y. M. Marusik was by far the most productive specimen collector, alone collecting 809 (41.7%) specimens, not counting the specimens he collected collaboratively with others. Most specimens of both sexes, but especially males, were collected in July, followed by June. The bulk of the specimens cited in this article were collected in the 1990s. Three treatments are based on hundreds of specimens; four treatments are based on no more than a few dozen specimens, but none is based on a singleton.

Pardosa zyuzini is one of seven species treatments and one of three new species in Kronestedt and Marusik (2011). 25.8% (504) of the specimens in the article are attributed to this species (Fig. 9b​). Adult males made up 71.4% (360) of the specimens, with the remainder being adult females (28.6%, 144). Specimens were collected in Russia (66.5%, 335) and Mongolia (33.5%, 169). The largest repository for these specimens is the Institute for Biological Problems of the North, Magadan, Russia (IBPN; 39.1%, 197). The holotype is deposited in the Zoological Museum of Moscow State University, Russia (ZMMU). Yuri Marusik was by far the most productive specimen collector, alone collecting 216 (42.9%) specimens, not counting the specimens he collected collaboratively with others. Nearly all of the 204 specimens for which the collecting elevation was specified were taken from 1,001-1,500 m. Most specimens were collected in July, but the largest number of females were found in June. Most of the specimens were collected during the 1990s.

Records of this linyphiid spider based on a total of 352 specimens appeared in three articles, all published in Biodiversity Data Journal (Fig. 10​​). Nearly all of these were from the Portuguese island of Madeira with the remaining records from Macedonia and Slovenia. Most of the specimens were female (63.4%, 223). None of these records explicitly cited a collection in their structured text, although some cite a depository in the body of the article (see Discussion). Collector name, collecting date, and type status were also omitted from most of the structured text.

Jeremy A. Miller was the lead author on two publications in Biodiversity Data Journal and one open access publication in Zootaxa containing treatments on spiders (​Fig. 11​). Collectively, these articles contain 20 treatments (17 species rank) including 6 new species and 2 new combinations. Four of these treatments concern Hymenoptera (Ichneumonidae, 6 specimens), the remainder concern the spider families Penestomidae, Symphytognathidae, Dictynidae, Tetragnathidae, and Theridiidae, based on a total of 208 specimens. Six treatments, including one new species, were based on singletons. Most specimens were collected in South Africa (57.5%, 123), followed by Malaysia (32.2%, 69). The specimen sex ratio exhibited a pronounced female bias, accounting for 105 of 129 sexually mature specimens; an additional 85 specimens recorded were not sexually mature. The largest number of specimens (32.2%, 69) were deposited in Universiti Malaysia Sabah's Institute for Tropical Biology and Conservation, Borneensis (UMS), but the largest number of primary type specimens cited (54.5%, six) were deposited in South Africa's National Collection of Arachnida in Pretoria (NCA). The largest number of specimens was collected during March, followed by February. Most of the specimens were collected since 2,000, but a modest number were collected during the 1910s and a few as far back as the 1890s. Only 23 specimens had associated elevation data, but of those, most were collected between 1,500-2,000 m.

Specimen data associated with spiders (among other taxa) are often structured to reflect multiple individual specimens in a single vial. A lot of specimens may include adult males, adult females, and sexually immature specimens for which sex is undetermined; all of these specimens may be associated with a single specimen code. We find that abundance and sex capture useful information about specimen data. For example, the phenology of males and females may be different, so the ability to determine what time of year a particular sex has been collected can be valuable to an investigator planning field work. Darwin Core expects a lot of specimens to all be of the same type (e.g., sex) and does not offer a simple way to represent heterogeneous objects in a single record (see https://github.com/tdwg/dwc/issues/35, https://github.com/tdwg/dwc/issues/36). To work around this, Plazi's SRS can parse the number of male and female specimens entered in the "sex" field (e.g., 1 male, 4 females), and either sum these values to determine the total abundance or use the individualCount field for this purpose (the individualCount can be higher than the sum of males and females when non-adult specimens are present; the difference between individualCount and the sum of males and females is automatically added to the category "other"). We recommend that authors of taxonomic treatments in BDJ complete both the individualCount field and sex field with their materials citations.

Information on the institution where particular specimens were deposited was frequently absent from the structured materials citations data in BDJ articles (Fig. 4​). Even when the institution is specified in the body of the article, that information becomes disassociated from the digital records themselves as they become disseminated to cybertaxonomic resources, such as Plazi and GBIF. We therefore recommend that authors include this information with their structured records. Similarly, collector name, collecting date, and type status should be included with records whenever possible.

Biodiversity Data Journal was launched with the motto: "making small data big." Realizing the potential of this vision requires that authors contribute structured specimen data with a sufficient level of granularity and detail. Otherwise, the power of data aggregation is curtailed.

The extent to which primary specimen data can be extracted from legacy literature depends on how the data are structured. Ideally, the data that refer to 'materials citations' link particular specimens to particular institutions and collecting events. Occasionally, specimen counts, institutions, and collecting events are disassociated. For example, in the treatment on Pardosa zyuzini (Kronestedt and Marusik 2011; Fig. 12), 154 paratype specimens are cited from the type locality in Mongolia and are distributed among five institutions. Exactly how many specimens are in which institution is not reported. From such ambiguously structured text, we can count the number of specimens reported for this species, and deduce that specimens from particular localities are present in each of the specified institutions, but we cannot count the number of specimens in each institution. Such records appear in the “Specimens by collection code” chart as “missing” (note that the term collectionCode as used by Plazi is equivilent to institutionCode in Darwin Core; this unfortunately confusing termanology should be reconciled in the future). In the XML file, the locality and collecting event data are associated with each of the five institutions (collectionCode), and the specimen count data are associated with the locality and collecting event, but no collectionCode.​

XML structured documents in Plazi are available for aggregation by GBIF. The treatments from Zootaxa included in this study supply records for 120 species that otherwise have no records in GBIF. However, the records for 98 of these species are not yet visible on GBIF. In most cases, the reason appears to be that the species is not in the GBIF backbone taxonomy (Döring 2015​). This means that more than half of the treatments marked up for this study contributed structured data for species that otherwise had no such data available. The fact that most of these data are not currently available for aggregation and reanalysis on GBIF is unfortunate and in need of attention. This project concerns recent legacy literature and while these names have been incorporated into the World Spider Catalog (http://www.wsc.nmbe.ch/), the catalog has only recently been restructured in a way that facilitates synchronization with online resources such as GBIF; this has not yet been implemented. It may be that this problem will rectify itself in due course. Alternatively, with the development of appropriate mechanisms, Plazi could contribute taxonomic names directly to GBIF or to the Catalogue of Life, which provides taxonomic data to GBIF and other online biodiversity databases.

One limitation of the way specimen data are currently structured is that the "collectorName" field does not parse out individual collectors when more than one person is involved. The problem that this creates is evident in the case of the most prolific collector. Marusik collected specimens alone and in collaboration with others, including D. V. Obydov. Some records listed Marusik as the first collector, others listed Obydov first. These are counted in the “Specimens by collector name” chart as three different collectors (1: Y. M. Marusik, 2: Y. M. Marusik & D. V. Obydov, 3: D. V. Obydov & Y. M. Marusik). In such cases, we may wish to attribute half of the specimens to each collector (or more generally, divide the number of specimens collected by the number of collectors and attribute that fraction to each member of the team). Alternatively, we might want to count the number of specimens that each individual contributed to collecting, which would result in a total count of specimens equal to or greater than the actual number collected. Similarly, we see in the “Treatments by collector name” chart for all data that Charles Griswold appears, alone or with collaborators, in four of the top ten rankings. A search of the underlying data reveals that Griswold, alone or with collaborators, contributed specimens to 22 treatments. As currently structured, it is not easy to track this.

In addition to specimen collector, individuals have two other main roles in taxonomic research: as publication (co-)authors and as taxonomic authorities (in which case the name of one or more individuals is associated with a taxonomic name). Both of these can be conducted as solo or collaborative activities. Summaries of individual contributions to taxonomic research should allow us to derive data on all three functions so that scientists get due credit for their efforts.

Data exchange between taxonomic research and institutional collections is less sophisticated currently than it could be. Collections-based institutions have an interest in monitoring how their collections are used in taxonomic research. XML-structured documents make it possible to import data from marked up literature into collections databases, making such databases more current and complete. Institutions can be compared by the number of specimens cited, or the number of treatments contributed to. Such metrics could produce healthy competition among institutions, stimulate specimen circulation in support of taxonomic research, and provide quantitative metrics in support of further funding. Similar metrics could form a basis for nuanced comparison of individual researchers. We might be interested not just in the number of treatments published, but in the number of specimens examined per treatment. We might be interested in the number of specimens and taxa collected by an individual that appear in published taxonomic research. Instead of looking just at individuals, these data could be aggregated by categories such as institution or country.

As more structured data from legacy literature become available and are shared with collections databases, it will become necessary to recognize instances of the same object cited in a database and one or more publications. Taxonomic research typically proceeds through the re-examination of specimens previously studied and cited by other taxonomists, including but not limited to primary type specimens. Without a mechanism to identify these as the same object, the total number of specimens known could become difficult to discern. Likewise, the same specimen recorded in a collections database and cited by a taxonomic paper should be recognized as one object, not two. This highlights the need for globally unique identifiers for specimens, whose application remains problematic in biodiversity informatics (Guralnick et al. 2014, Guralnick et al. 2015). Only 40% of the records investigated for this study (1,144 of 2,894, representing 26% of the specimens, 2,156 of 8,231) had some form of specimen code reported in the literature. Instiututionally unique identifiers can suggest that two matching records refer to the same object, but a robust global system for tracking specimens and their associated data and usage history remains an elusive challenge.

Some recent publications have used specimen data from taxonomic literature to investigate aspects of biodiversity knowledge. We adapt our spider dataset to two of these questions: 1) the time lag between species discovery and description, and 2) the prevalence of rare species in species descriptions.

Fontaine et al. (2012) investigated the pace of biodiversity description based on a sample of species described in 2007. Applying the same approach to this spider dataset, we find that the mean delay between the first collection and description of the 92 new species in the data set is 17.2 years (standard error: 2.37) with a median of seven years, ranging between 102 (the lycosid Pardosa zyuzini Kronestedt & Marusik 2011) and zero years (i.e., published the same year it was discovered; the symphytognathid Crassignatha danaugirangensis Miller et al. 2014). Fontaine et al. did not break out data on spiders separately, but our values fit closely with their “other invertebrates” category. This means that the average shelf life between discovery and description for spiders is about four years less than for insects, and about 15 or more years less than for vertebrates and plants.

Lim et al. (2011) surveyed taxonomic literature to investigate the prevalence of rare species. Rare species are a conspicuous part of community structure, particularly prevalent among tropical arthropods (Coddington et al. 2009, Novotný and Basset 2000). Lim et al. (2011) found that rare species are prevalent in the taxonomic literature. Rare species include singletons (taxa known from a single specimen) and uniques (taxa known from a single collection event). Based on 25 articles on an assortment of arthropod taxa (including spiders) published in the monograph series of the American Museum of Natural History (2000-2010), 17.7% of new species are described from a single specimen. The Zootaxa spider dataset describes new species from singletons at more than twice this rate (34/91 = 37.4%). Another way of looking at rarity is by looking at the number of localities a new species has been found in (the incidence-based as opposed to the abundance-based approach). The American Museum of Natural History invertebrate data indicated that 27.5% of new species are described from a single locality; the Zootaxa spider data were again richer in rarity with 38.5% (35/91).

A key difference between these previous efforts and our literature markup project is that the primary data upon which our analyses are based are readily accessible to all from Plazi (http://plazi.cs.umb.edu/GgServer/srsStats) and can be built upon as treatments are added. Meier and Dikow (2004) and Dikow et al. (2009) presented several ideas for how specimen data from taxonomic revisions could be applied to biodiversity research. These include comparing areas for conservation prioritization, data-driven assessment of species conservation status, and predicting the number of species in a group yet to be described. While their ideas call for denser coverage of geographic areas than permitted by our dataset, their work continues to serve as a guidepost to what the future of XML marked taxonomic literature could achieve.

The rate of publication in taxonomy continues to rise, as does the potential for accelerated transition from discovery to publication (Miller et al. 2014). Therefore, while addressing the challenge of legacy literature XML markup, we should not continue adding to the backlog. There have been great recent advances in taxonomic publishing. Pensoft (http://www.pensoft.net/about.php​) journals have led the way with an XML-based approach that facilitates the reuse, aggregation, and dissemination of content to an increasing variety of cybertaxonomic databases and resources. Whether integral to the publication process or added later, structured taxonomic treatments promote transparency and repeatability in biodiversity science while facilitating the aggregation and reanalysis of data.

The modest pilot demonstration presented here is focused on spider literature, but the approach can be applied to any taxonomic literature. Our accumulated biodiversity knowledge includes an estimated 2-3 billion specimens in natural history collections and 500 million pages of printed text (Thessen and Patterson 2011, Ariño 2010​). These are the data we need to answer questions that are relevant to our world today, like setting conservation priorities and anticipating the effects of climate change on biodiversity and ecosystem functions that affect the lives of people. Computer models of the biosphere are becoming increasingly sophisticated and powerful, and this field appears to have tremendous potential for growth in the near future (Purves et al. 2013). To fully realize the benefits of these nascent technological advances, we are going to have to wrestle with the gap between the knowledge we have accumulated in libraries over 250 years of research and the data that are available in structured digital form so they can be used by computers. In short, we have half a billion pages worth of biodiversity knowledge and no way to query it. XML tagging, whether applied to prospective publications as in Biodiversity Data Journal or to legacy publications using tools like GoldenGATE, makes the primary data available in digital structured form. This enhances the scientific quality of the work because it is easier for other scientists to re-evaluate and test conclusions based on the data. But the real power comes when data from many articles are combined, queried, and reused for new purposes. Potential applications for these data span the scientific, policy, and public spheres (Arzberger et al. 2004, Hardisty et al. 2013​). When we all have better access to the information that already exists in the global biodiversity library, this helps us do a better job of exploring what we don’t know and wisely applying what we do.