The data files and scripts necessary to reproduce the results of this study are available as Supplementary material that can be accessed either through Zenodo or via the Github repository.

For this proof-of-concept study, we selected the dung beetle genus Grebennikovius (Coleoptera, Scarabaeinae), recently revised by Montanaro et al. (2024) (Fig. 2a) and endemic to the Eastern Arc Mountains (Tanzania). The genus comprises four species, for which we have generated and analysed semantic descriptions.

To observe and illustrate morphological characters in great detail, we obtained micro-CT images of a specimen of Grebennikovius basilewskyi (Balthasar, 1960). Imaging was conducted at the Finnish Museum of Natural History LUOMUS (University of Helsinki) using a Nikon XT H 225 and the following settings. Multi-metal target with molybdenum setting, 70–100 kV beam energy, 70–100 uA beam current, 1420 ms exposure time and 4476 projection images with four frames of averaging per projection. Detector binning was set to 1x1, gain to 24 dB and white target to 60k. The complete scan time was approximately seven hours and the resulting voxel size of the dataset 2.998 µm. The volumetric dataset was reconstructed from the projection images using Nikon CT pro-3D Version XT 6.9.1 and the dataset was exported to VGSTUDIO MAX 2023.2 (Volume Graphics GmbH, Heidelberg, Germany) in 16-bit format. Excess material was excluded from the dataset. The dataset was visualised using volume renderer (Phong) and aligned correctly. Images from the sample were rendered from anterodorsal, dorsal, lateral, posterior and ventral views. The aedeagus of G. basilewskyi in Fig. 3b was modified from Montanaro et al. (2024).

To describe species semantically, we employed the Phenoscript language powered by its dedicated plugin for VS Code (Fig. 2b). The primary purpose of Phenoscript and its plugin is to streamline the process of creating semantic descriptions. Although other tools can be used for this purpose, such as Protégé, a comprehensive, GUI-based ontology software or Turtle syntax for knowledge graph construction, these methods are much slower than Phenoscript. With the Phenoscript plugin, users can benefit from syntax highlighting and snippets. This makes it easier to select terms from predefined biological ontologies and write semantic statements.

For this study, we used the ontologies listed in Table 1. The process of creating semantic decriptions often requires the addition of new terms to existing ontologies, as was the case in our study, which is discussed in a separate section below.

Writing in Phenoscript closely resembles composing natural language (NL) descriptions, albeit with its own distinct syntax, which is still quite akin to NL. The language documentation and tutorials are available on the Phenoscript repository. The initial step typically involves setting up a YAML configuration file to specify author names, project title and the ontologies to be used. As a next step, Phenospy can generate snippets for the necessary ontology terms. Snippets, which are ontology terms or small blocks of Phenoscript code, can be selected from a drop-down menu in the Phenoscript description, appearing upon typing the first letters. Once the snippets are ready, the user can begin coding semantic descriptions in VS Code using the Phenoscript plugin. For convenience, we present below an overview of the major character patterns used in describing species of Grebennikovius, both in NL and in Phenoscript (see the section "Phenoscript: main patterns of phenotype statements").

Once the Phenoscript description is complete, it can be processed and analysed as outlined in the pipeline shown in Fig. 1, with technical details provided in subsequent sections.

The pipeline (Fig. 1) consists of six steps outlined below, which are facilitated by a makefile tool (Supplementary Material). In a nutshell, a makefile automates the process of sequentially executing various programmes and commands. In our context, it automates the execution of different pipeline steps.

Step 1. Once Phenoscript description is written as a Phenoscript file, it must be converted into OWL format using the Phenospy package, which provides the necessary command-line tools for this conversion. This creates the ABox component of the ontology for further processing.

Step 2. This stage involves validating the OWL file with SHACL (Shapes Constraint Language) to ensure that semantic data satisfy the requirements of the data models employed by the user. SHACL is a conventional tool for validating RDF graph patterns against a set of predefined criteria. As an example, in our context, these criteria require that all phenotypes are linked to species names and include the necessary metadata. We used the SHACL command-line interface provided by the Apache Jena framework. Proceed to the next step if validation succeeds. If it does not, return to the Phenoscript description and correct it.

Step 3. Make a TBox file by downloading and merging all the source ontologies used to create semantic descriptions. This step is automated using Phenospy and ROBOT (Jackson et al. 2019), a command-line tool for manipulating and working with biomedical ontologies.

Step 4. Perform ontology reasoning using the ABox (step 1) and TBox (step 3) files. This step is mediated by the materializer tool which uses the whelk reasoner. Ontology reasoning refers to the process of deriving logical conclusions from a set of asserted facts or axioms within an ontology and knowledge graph. Reasoning is used to logically validate the ontology and infer the class membership of the individuals in the ABox.

Logical validation ensures that the ontology contains no contradictions in its structure, definitions or relationships between its entities. If this is the case, the ontology is referred to as "consistent". If the ontology is found to be inconsistent at this stage, it is most likely because there are logical errors within the semantic descriptions that need to be corrected. Additionally, Class inference generates new data from the initial assertions, which can be used for downstream semantic queries. If the validations at steps 2 and 4 are successful, the user can proceed to the next stages.

Step 5. Using Phenospy, automatically generate the annotated NL description from the OWL file. See the section below for more information.

Step 6. Perform semantic queries to extract trait data from the descriptions. See the section below for more information.

NL descriptions were created using Phenospy's algorithm which traverses the knowledge graph encoded in an OWL file and translates the graph patterns into NL. The algorithm searches for character patterns, such as, for example, the presence or absence of anatomical entities, their measurements and then translates them into human-readable NL text.

Generated NL descriptions consist of hierarchical trait statements that usually resemble entity-quality syntax (Washington et al. 2009, Balhoff et al. 2010). Typically, each statement begins with a sequence of locator terms that specify the trait's position (= entity) on the organismal body. Qualities or other phenotypic properties are specified following the locator terms. They can be listed on the same line if only one property is associated with the given locator or on several subsequent lines for multiple properties. In addition, multiple statements associated with the same body part may be nested within one another.

In order to demonstrate the ease with which phenotypic information can be retrieved from our descriptions, we employed two sets of semantic queries. The first set aimed to determine the number of individuals per species associated with an ontology class representing one of the following phenotypic characteristics: colour, shape, size and texture (Table 2, rows 1–4). To achieve this, we executed a SPARQL query using the ontology generated during the step 4 of our pipeline. For instance, applying this query to the statement "head is red in G. armiger" would yield a single individual for G. armiger classified under the "colour" class. We utilised the SPARQL Notebook extension for VS Code to execute SPARQL queries, enabling the organisation of multiple queries within a single file and the annotation of queries using Markdown syntax.

The second set of queries focused on determining the number of individuals associated with the major body parts: head, thorax, abdomen and leg (Table 2, rows 5–8). Within our context, this association signifies that individuals must belong to one of these four classes or be "part of" (BFO:0000050) one of them. For instance, a statement indicating "antennae are long in G. armiger" would result in one individual being associated with the class "head" as antennae are considered part of the "head". To facilitate these types of queries, we incorporated four custom classes into the merged ontology file generated during step 3 (i.e. the Tbox file). These classes were logically defined using Manchester OWL syntax as follows: "X or (part_of some X)", where "X" represents each of the four body parts. By conducting ontology reasoning in step 6 of the pipeline, the individuals from the ABox were automatically classified into these custom classes. Subsequently, these custom classes were used in SPARQL in a manner analogous to the first set of queries.

We expanded the Insect Anatomy Ontology (AISM; Girón et al. (2024), Girón et al. (2023b)) and the Coleoptera Anatomy Ontology (COLAO; Girón et al. (2023a)) to include morphological terms necessary for the descriptions of Grebennikovius species. We created a total of 152 new terms in AISM and COLAO following the editing instructions available on the AISM GitHub repository.

New AISM terms (110 terms, v.2023-04-14 to v.2024-05-11) describe the more general insect body plan and most of them are applicable to a range of insect orders. These terms cover appendage subdivisions (specific tarsomeres, flagellomeres, palpomeres etc.), specific cuticular protrusions (protibial and clypeal teeth, carinae, etc.) and different types of cuticular punctures.

Punctures are particularly important characters in dung beetle taxonomy, especially at the interspecific level (d'Orbigny 1913, Davis et al. 2008). We recognised two main types of punctures: setigerous and non-setigerous. Each type is further subdivided into simple, ocellate and granulate subtypes. However, we currently avoid using the term “raspish” punctures, which are described as having a small, sharp granule-like protrusion (see “râpeuse” (e.g. d'Orbigny (1913)), “raspish” (e.g. Ziani et al. (2019)), “raspose” (e.g. Barbero et al. (2003)), “raspelartig” (e.g. Balthasar (1963))). This decision is based on the difficulty in distinctly separating them from granulate punctures.

Conversely, the new terms introduced in COLAO (42 terms, v.2023-03-30 to v.2024-02-14) are more specific to beetles, particularly to Scarabaeoidea and Scarabaeinae. These terms include descriptions of elytral striation, pronotal protrusions and depressions and genital structures, such as parameres and endophallites (Génier 2019).

Notably, we use the term “mesometaventral sulcus” instead of the more traditional “meso-metasternal suture”, since the use of meso- and metaventrite should be preferred over meso- and metasternum (Lawrence and Ślipiński 2013). Additionally, “pygidium” is substituted by the more general term “abdominal tergite VIII” (see also Cristóvão and Vaz-de-Mello (2021)).

Traditional positional terminology used in Scarabaeinae taxonomy sometimes does not reflect the true positional relationships between the insect body and its parts. For example, in most scarab beetles (Scarabaeoidea), legs are flattened anteroposteriorly and rotated forwards (fore legs) or backwards (middle and hind legs), making more intuitive (but incorrect) to use dorsal and ventral instead of anterior and posterior for referring to their broad, flattened sides and the reverse for the narrow dorsal and ventral sides (see Fig. 3a). One of the most striking consequence of this reversed terminology is that the protibial margin that bears teeth in scarabs is often called “outer”, “lateral” or even “external”, but, in fact, corresponds to the dorsal side of the insect leg.

Another example can be found in the parameres of the male genitalia (Fig. 3b). Traditionally, left and right parameres are defined, based on an axis in which the articulation between parameres and phallobase is positioned "basally" and the distalmost part of the paramere "apically". Left and right sides are then inferred interpreting this "basoapical" axis as the equivalent of an anteroposterior one. However, if we consider the position of the aedeagus relative to the entire insect body, such axes – and, consequently, left and right sides – are reversed.

The revised interpretation of position has been implemented in the natural language (NL) descriptions by Montanaro et al. (2024) and we have also adopted it here, both for developing new ontology terms and for crafting semantic descriptions. While initially this perspective may seem counterintuitive to traditional taxonomists, it represents the most anatomically consistent way for describing these structures.

In the examples below, we provide traditional NL statements followed by their equivalent Phenoscript statements (in italics), for the main types of semantic traits used in our descriptions. The "note" section briefly explains the rationale for the use of certain semantic constructs.

In a nutshell, a Phenoscript statement consists of a sequence of nodes and edges (i.e. relationships) in a knowledge graph. Edges are defined by a preceding dot "." symbol. Each statement should begin and end with a node. The nodes are followed by edges, which specify the relationships between the nodes. Node–edge sequences may be as long as necessary. The semantic statement is closed with a semicolon ";". Every ontology term is prefixed with the abbreviation of its originating ontology, functioning as a namespace. For instance, the term "aism-cuticular_carina" signifies that "cuticular carina" comes from the AISM ontology. This prefixing helps to clearly identify the source ontology of each term.

Most edges refer to object properties, such as "has_part", "has_characteristic" and their inverses. To improve readability, Phenoscript uses aliases instead of long labels. The following aliases are used:

• ">" has_part;
• "<" part_of;
• ">>" has_characteristic;
• "<<" characteristic_of;
• "|>|" increased_in_magnitude_relative_to;
• "|<|" decreased_in_magnitude_relative_to.

We encourage the reader to consult the Phenoscript language guide for syntax details. For more information about semantic characters, we suggest the Phenoscape Guide to Character Annotation and other relevant materials (Balhoff et al. 2010, Dahdul et al. 2010, Dahdul et al. 2018); note that they were designed for a class-based approach, but can be adapted to an instance-based approach with minor modifications. The Phenoscript version of the statements below is available in Supplementary Material.

Presence Phenotypes

• Dorsal margin of metafemur with carina (Fig. 5b:16).
  • uberon-male_organism > aism-metafemur > bspo-dorsal_margin > aism-cuticular_carina;
  • Note. Simple character statement indicating presence of a structure, i.e. a cuticular carina on the dorsal margin of the metafemur of our organism.
• Interstria 5 tuberculated distally (Fig. 4a:7 and Fig. 5c:7).
  • uberon-male_organism > colao-elytral_interstria_5 > bspo-distal_region > aism-cuticular_tubercle;
  • Note. Simple character statement analogous to the previous one.
• Axial and subaxial endophallites fused.
  • uberon-male_organism > colao-fused_axial_and_subaxial_endophallites;
  • Note. The ad hoc COLAO term fused_axial_and_subaxial_endophallites allows to write a complex phenotypic character as a simple presence statement, i.e. the endophallite originated by the fusion of the axial_endophallite with the subaxial_endophallite.

Absence Phenotypes

• Posterior longitudinal hypomeral carina absent (Fig. 4b:9).
  • uberon-male_organism !> colao-posterior_longitudinal_hypomeral_carina;
  • Note. Simple absence statement in which the negation (!) expresses the lack of the entity colao-posterior_longitudinal_hypomeral_carina in the organism's body.
• Fourth protibial cuticular tooth absent (Fig. 5a:14).
  • uberon-male_organism > aism-protibia !> aism-dorsal_protibial_cuticular_tooth_4;
  • Note. This character goes along with the character "Protibia with 3 teeth". In fact, the statement that the protibia has three teeth does not logically imply that the fourth one is absent. To assert it unequivocally, the presence of the tooth must be negated.

Count Phenotypes

• Maxillary palpus with 4 palpomeres (Fig. 4b:8).
  • uberon-male_organism > aism-maxillary_palpus_with_4_palpomeres;
  • Note. Simple character statement indicating the presence of a structure. Interestingly, this apparently simple line automatically encapsulates, through the AISM term maxillary_palpus_with_4_palpomeres, the presence of each individual palpomere (i.e. AISM maxillary_palpomere_I to maxillary_palpomere_IV).
• Legs with 5 tarsomeres (Fig. 4:6).
  • uberon-male_organism > (aism-protarsus_with_5_protarsomeres, aism-mesotarsus_with_5_mesotarsomeres, aism-metatarsus_with_5_metatarsomeres);
  • Note. Expressing the presence of pro-, meso- and metalegs would intuitively require to write three separate statements, one for each leg type. However, PhenoScript allows to write node lists, i.e. encapsulating into one line, in the form of a list between parentheses, multiple characters that are simultaneously ascribed to the organism.
• Protibia with 3 teeth (Fig. 5a:14).
  • uberon-male_organism > aism-protibia > (aism-dorsal_protibial_cuticular_tooth_1, aism-dorsal_protibial_cuticular_tooth_2, aism-dorsal_protibial_cuticular_tooth_3);
  • Note. Similarly to the previous one, this statement expresses that the protibia bears protibial teeth 1 (distalmost) to 3 (proximalmost). The absence of a possible protibial_tooth_4 is expressed by the pattern "Fourth protibial cuticular tooth absent" (see "Absence Phenotypes").

Qualitative Phenotypes

• Organism oval-shaped and flattened (Fig. 4 and Fig. 5).
  • uberon-male_organism >> (pato-ovate, pato-flattened);
  • Note. Simple quality statement in the form of a node list, expressing on the same line that the organism is both ovate and flattened.
• Clypeus with two sharp, upturned triangular teeth (Fig. 5a:13).
  • uberon-male_organism > aism-lateral_clypeal_tooth_1 >> (pato-upturned, pato-sharp);
  • Note. Composed statement which first expresses the presence of lateral_clypeal_tooth_1 (which is bilaterally paired by definition), then it qualifies it as both upturned and sharp.
• Posteromedial margin of head with small smooth area (Fig. 5a:11).
  • uberon-male_organism > aism-vertex > bspo-posterior_region > bspo-medial_region >> pato-smooth;
  • Note. Composed statement expressing the presence of a posteromedial region of vertex, defined by postcomposition as the posterior_region of medial_region of vertex. This region is then qualified as smooth.
• Posterior margin of pronotum with row of large, oval, ocellate punctures (Fig. 4a:2).
  • uberon-male_organism > aism-pronotum > bspo-posterior_region > aism-row_of_punctures > aism-ocellate_cuticular_puncture >> (pato-increased_size, pato-ovate);
  • Note. Complex statement which first describes the presence of a row of punctures on the posterior margin of the pronotum composed by (an undefined number of) ocellate_cuticular_puncture and then qualifies such punctures as increased_size (= large) and ovate. This feature is particularly evident in G. armiger and G. pafelo, but a row of large punctures can also be observed in G. basilewskyi.
• Scutellar shield absent (Fig. 4a:3).
  • uberon-male_organism > colao-scutellar_shield >> pato-concealed;
  • Note. We have chosen the term concealed instead of simply expressing the absence of scutellum because this structure is not visible from above, but present beneath the elytra.
• Mesometaventral sulcus rounded medially (Fig. 4b:10).
  • uberon-male_organism > colao-mesometaventral_sulcus > bspo-medial_region >> pato-curved;
  • Note. We adopt the term "mesometaventral sulcus" to refer to the transverse sulcus between the mesoventrite and the metaventrite. The statement defines its medial region and then qualifies it as curved, a PATO term with a meaning equivalent to the NL "rounded".
• Mesotibia expanded distally (Fig. 4a:5).
  • uberon-male_organism > aism-metatibia > bspo-distal_region >> pato-dilated;
  • Note. Composed statement expressing the presence of the metatibia, then defining its distal_region and lastly qualifying it as expanded through the semantically equivalent PATO term dilated.
• Protibial apex bearing a row of short, thick setae (Fig. 5a:15).
  • uberon-male_organism > aism-protibia > aism-antero-distal_margin > aism-cuticular_seta >> (pato-multiple, pato-increased_thickness);
  • Note. Composed statement in which the presence of many cuticular_seta is described by qualifying them as multiple. The relatively high thickness of setae is then given by increased_thickness.
• Parameres symmetrical, elongated.
  • uberon-male_organism > aism-parameres >> (pato-symmetrical, pato-elongated);
  • Note. Statement qualifying parameres both as symmetrical and elongated. Here we used aism-parameres instead of aism-paramere to mean that the region of cuticle formed by the two parameres is symmetrical and, therefore, the left and right parameres have shapes that mirror each other. Using aism-paramere would have been incorrect, since an individual paramere is not symmetrical per se.

Absolute and Relative Measurement Phenotypes

• Body length: 4.5 mm.
  • uberon-male_organism >> pato-length .iao-is_quality_measured_as iao-measurement_datum:md-c4c164 .aism-has_unit unit-millimeter; iao-measurement_datum:md-c4c164 .iao-has_measurement_value 4.5;
  • Note. Absolute measurements require to define a measurement_datum, to which the measure of the focal quality (in this case, length) will be assigned. To measurement_datum are then ascribed: 1) an absolute measurement unit (in this case, millimetre) through the property .aism-has_unit; and 2) a measurement value (in this case, 4.5) through the property .iao-has_measurement_value. Due to its syntax, this statement must necessarily be broken into two lines, separated by semicolons. For this reason, an alphanumeric tag must be added to iao-measurement_datum to make sure it is recognised as the same entity between the two lines.
• Pronotal surface covered with ocellate setigerous punctures separated by 1—2 diameters (Fig. 4a:1).
  • uberon-male_organism > aism-pronotum:id-1549f8 >> aism-interpunctural_distance .iao-is_quality_measured_as iao-measurement_datum:md-1b36aa .aism-has_unit pato-diameter << aism-ocellate_setigerous_cuticular_puncture < aism-pronotum:id-1549f8; iao-measurement_datum:md-1b36aa .iao-has_measurement_value 1.5;
  • Note. Relative measurements are substantially similar to absolute ones, but they use qualities of other entities as measurement units. In this example, the interpunctural_distance on pronotum is measured using the quality diameter inhering to ocellate_setigerous_cuticular_puncture. For simplicity, we prefer to give the average measurement value (1.5) instead of specifying the interval (1–2). As explained in the previous pattern, alphanumeric tags identify aism-pronotum and iao-measurement_datum occurring in separate lines as the same entities.

Relative Comparison Phenotypes (within the same species)

• Head punctures becoming smaller anteriorly (Fig. 5a:12).
  • uberon-male_organism > aism-clypeus > bspo-anterior_region > aism-setigerous_cuticular_puncture >> pato-diameter |<| pato-diameter << aism-setigerous_cuticular_puncture < aism-frons;
  • Note. This pattern compares the quality of an entity (the diameter of some setigerous_cuticular_puncture located on the anterior region of clypeus) with an equivalent quality of another entity (the diameter of some setigerous_cuticular_puncture located on the frons). The property decreased_in_magnitude_relative_to (alias: |<|) between the two entities states that the former quality has a smaller value than the latter.
• Ventral membranes of parameres equally sclerotised.
  • uberon-male_organism > aism-left_ventral_conjunctiva_of_paramere >> pato-thickness .ro-similar_in_magnitude_relative_to pato-thickness << aism-right_ventral_conjunctiva_of_paramere < uberon-male_organism;
  • Note. This statement compares the degree of sclerotisation of some left_ventral_conjunctiva_of_paramere (preferred quality term: pato-thickness) with that of the right_ventral_conjunctiva_of_paramere. In AISM and COLAO, the term "conjunctiva" is preferred over "membrane" as conjunctiva is explicitly defined for cuticular elements in insects. Since the two quality values are similar, the property .ro-similar_in_magnitude_relative_to was chosen.

Relative Comparison Phenotypes (between species)

• Hind wing of Grebennikovius armiger shorter than hind wing of Grebennikovius basilewskyi.
  • uberon-male_organism::grebennikovius_armiger > aism-hind_wing >> pato-length |<| pato-length:id-d38816[exclude = True] << aism-hind_wing:id-6b490e[exclude = True] < uberon-male_organism:grebennikovius_basilewskyi[exclude = True];
  • Note. This statement compares two qualities ascribed to two different OTUs, Grebennikovius armiger and Grebennikovius basilewskyi. To allow the use of entities or qualities belonging to one OTU (G. basilewskyi) within another OTUs' (G. armiger) description, the function “exclude” must be used. This relative comparison is composed by two parts: 1) the length of the hind_wing of the organism uberon-male_organism::grebennikovius_armiger and 2) the length of the hind_wing of uberon-male_organism:grebennikovius_basilewskyi. The comparison is expressed through the property decreased_in_magnitude_relative_to (alias: |<|) written between the two lengths.

Other Phenotypes

• Elytral interstria 8 carinated, the carina located medially to the lateral region of elytron (Fig. 4a:4 and Fig. 5b:4).
  • uberon-male_organism > colao-elytron_with_9_striae:id-5770f5 > colao-elytral_interstria_8 > aism-cuticular_carina .aism-medial_to bspo-lateral_region < colao-elytron_with_9_striae:id-5770f5;
  • Note. This statement describes the position of an anatomical entity (a cuticular_carina located on the elytral_interstria_8) with respect to another entity (the lateral_region of the elytron itself). The positional property aism-medial_to is used to state that the first entity is medial to the second.
• Margin of abdominal tergite VIII entirely grooved (Fig. 5c:17).
  • uberon-male_organism > aism-abdominal_tergite_VIII:id-791f84 > bspo-anatomical_margin .ro-coincident_with aism-cuticular_groove < aism-abdominal_tergite_VIII:id-791f84;
  • Note. In dung beetles, the abdominal tergite VIII (a.k.a. pygidium) often has a more or less strongly impressed groove running parallel to the tergite's margins. This structure is, therefore, defined here as a cuticular_groove coincident_with the anatomical_margin of the tergite. The RO term coincident_with is, in this case, the most appropriate one to describe such positional relationship since it refers to linear structures that are both parallel and adjacent/coincident.

Our paper assessed the utility of Phenoscript, an emerging language for semantic descriptions and its associated tools for producing semantic data using an individual-based approach. Our results demonstrated the effectiveness of Phenoscript in creating semantic descriptions thanks to its syntax that is similar to NL expressions. In addition, the syntax was also improved over the previous version of the language (Mikó et al. 2021). The general concept of Phenoscript makes it a versatile tool, extendable beyond applications in biosystematics to ecological and other domains. Phenoscript allows taxonomists to bypass traditional NL descriptions and, instead, create semantic ones. These descriptions can then be converted into NL text for publication and into an ontology format for further analysis and dissemination.

The proposed computational pipeline automates the production of semantic descriptions and can be applied to any other taxon using a desktop computer. Due to our focus on dung beetles, the developed approach can be specifically applied to them with ease, enabling further semantic-based research in this group. Thus, the proposed semantic approach opens up possibilities for new types of publications, where taxonomists can semantically re-describe known species in order to unlock their traits for other cross-disciplinary research within biology.

Semantic descriptions, however, are currently slower to write than traditional NL ones for a number of reasons. A shortage of comprehensive educational resources and a relatively small community create a high initial barrier to learning semantic methods in evolutionary biology (Deans et al. 2015).

At present, composing semantic descriptions involves the addition of many new terms to ontologies, due to ontology incompleteness. In this study, we had to add 152 terms to AISM and COLAO to cover all the necessary morphological terminology for complete species descriptions. Eventually, we expect this task will diminish as usage of ontologies increases and they become saturated with terms.

Significant time is spent thinking about how to code particular traits semantically. Despite the establishment of the necessary protocols (Balhoff et al. 2010, Dahdul et al. 2010, Dahdul et al. 2018), we still require new logical models for certain types of traits. For example, the species descriptions in this study were based on single specimens, which worked well since the species have limited variation. This approach will not be optimal for many other species with at least moderate morphological variation. Thus, new data models and solutions are needed to model intraspecific variation efficiently.

In order to demonstrate the queriability of the descriptions, we conducted semantic queries as a proof of concept. Currently, such queries can be used to retrieve phenotypic information semi-manually for analysis and comparison across species. The process involves creating a query targeting specific traits and applying it to a set of semantic phenotypes. However, automatic comparison, such as identifying common or different traits between species, is not feasible with current methods and remains a topic for future research.

The conversion of semantic descriptions to natural language is not trivial. Our study encountered difficulties in accurately translating certain character patterns, underscoring the need for improved methods in this area. It is also essential to develop new methods for post-processing and analysing semantic descriptions. Particularly in taxonomy, innovative approaches for species diagnosis and comparison would be highly beneficial.

A nanopublication is a concept in scientific data management that is particularly relevant in the context of big data and FAIR principles (Kuhn et al. 2013, Kuhn et al. 2018). It represents a minimal unit of publishable information that can be used to describe anything, such as "species X feeds on species Y". Technologically, nanopublication is a small knowledge (RDF) graph (Kuhn et al. 2021) that is similar to the semantic phenotypes produced in the present study.

Although the concept of nanopublications is still emerging, it promises to revolutionise information sharing and analysis (Kuhn and Dumontier 2017). Creating a nanopublication involves generating an RDF graph through a specialised service, such as nanodash, as used in this study. Once created, the nanopublication is immediately accessible to the public, facilitating its use in big data analysis.

Currently, nanopublications created using the nanodash service are not subject to peer review. Thus, authors are encouraged to take full responsibility for the content. However, the nanodash service does distinguish peer-reviewed from non-peer-reviewed nanopublications. Moreover, its integration with BDJ facilitates the direct integration of nanopublications into conventional academic publications, where they do undergo the peer-review process.

As semantic phenotypes and nanopublications have technological similarities and both aim to be computable and FAIR, integrating them into one framework would be beneficial. In our research, we generated both semantic phenotypes and nanopublications. At the moment, these datasets are not integrated and stored separately, not in a unified semantic graph or triple store. As a result of this disintegration, we are unable to simultaneously query species traits and the data from nanopublications. Therefore, there is a need for additional methodological advancements in order to achieve effective integration. Data from this study can be used to explore and develop new methods for such an integration.

Semantic phenotypes offer a significant improvement in the generation, analysis and sharing of taxonomic data, marking a substantial move towards FAIR phenotypes and computable information. To fully integrate semantic phenotypes and descriptions into standard practices in taxonomy and biology, further advancements in computational methods are needed, along with the development of platforms for managing semantic phenotypes and active engagement from the scientific community.