Compositions (part-of relations) of concepts occur in modeling primarily if some object can be subdivided along the physical dimensions of space, mass, or time. The composition of a biologi-cal organism may contain parts like “head, body, legs”, but also “cell wall” may be composed of
“proteins, chitin, β-1,3-1,6-glucan”. Space and mass-based abstract composition concepts are independent. In the body composition they fall together, in the chemical composition they are largely independent (the composition of a fluid is exclusively mass-based), and for a composition of distance segments the concept of mass is irrelevant. Space-based concepts include a notion of order and adjacency, mass-based only a summation of parts. Similarly, time-based compositions may or may not have an ordering and duration aspects. A process (e.g., an observation method in science) often includes a clear notion of order and duration of component processes (e.g., a fun-gus is cultivated for a given period, the cell wall destroyed, DNA extracted, DNA purified, PCR performed). However, the sequence of component processes may also be so variable and repetiti-ve as to become irrelevant (e.g., “compare object with color chart”). Similar to spatial
composi-tions, temporal component processes may be nested (e.g., using a microscope while performing some operations).
Part-of relations inform that concepts belong together. A useful property of Part-of relations is that they are transitive (a part-of b, b part-of c → a part-of c). However, it is difficult to do reaso-ning about the properties of the objects having part-of relations. For compositions involving mass it is known that the total mass is the sum of the mass of the parts. For most properties, however, such deductions depend on secondary information. If a is part of b, neither color(a)=red → color(b)=red, nor color(b)=red → color(a)=red is generally true. If the surface color is black for the insect body and red for the head, it is intuitive that the entire object is a mixture of red and black. However, if it is known that the fat body is yellow, the additional knowledge is necessary that this is an internal structure not visible in undissected insects.
The following sections will only discuss the physical arrangement and part-of structures.
These are of primary importance in biological descriptions and are used as the basis for current character decomposition models (p.116). However, observation or experimental methods also have a composition (part-of) structure.
Morphological object composition
Many objects are a composition of other objects. Fig.43 shows an object that consists of one rounded square, two right-angled triangles, six circles, and one equilateral triangle. Aspects of object composition are:
■ Containment: The entire object contains all objects listed. The rounded square further contains the two right-angled triangles.
■ Multiplicity: The right-angled triangles occur twice, the circles six times.
■ Adjacency: All objects are adjacent to the rounded square. The circles are adjacent to the right-angled triangles. No other objects are adja-cent.
■ Ordered sequences: The darker circles are between the lighter circles.
■ Relative position: The circles are at opposite sides of the rounded square.
Fig.44 gives an example for a morphological composition hierarchy for selected parts of the human body. Note that some composition details may have to be resolved by consensus, e.g., whether the foot starts below the ankle or whether the ankle is part of the foot; this is further dis-cussed in section “Competing classifications of object parts” further down.
Entire body Rump
Neck Head
Leg Arm
Ankle Calf Knee Thigh
Foot
Heel
Toe Toenail
Figure 44. An excerpt from a morphological composition hierarchy of human body parts as a UML class diagram.
(Note: The terms “morphology” and “anatomy” are often used interchangeably. In this treatment, however, the term “morphology” refers to the outside organization or composition of an entire
Figure 43. Abstract example of an ob-ject composed of other objects (com-ponents).
object and the term “anatomy” to the inner organization that may require dissecting or disassem-bling the object.)
129. Morphological object composition includes aspects (multiplicity, adjacency, order) that are not immediately included in a part-of hierarchy. Support for these aspects is desirable.
Anatomical object composition
Morphological objects (composite or atomic) usually have a separate inner, anatomical composition. The outer and inner compositions often cannot be organized into a single hierar-chy, i.e., the inner composition cannot be presented nested inside the atomic outer-composition objects. For the object from Fig.43, a hypothetical “vessel” anatomy with a main vessel and branching side vessel is shown (Fig.45). Similar situations occur for nerves or blood vessels in animals or the vascular system in plants. Fig.46 gives another example for an anatomical composition hierarchy for selected parts of the human body.
Entire body
Left lung
Right lung
Bronchial tree Air sacs (alveoli) Left superior lobe
Cardiac notch Left inferior lobe
Right inferior lobe Right middle lobe Right superior lobe Lungs
Larynx Trachea
Figure 46. An excerpt from an anatomical composition hierarchy of humans showing a detailed hierarchy for lungs. Parts of the bronchial tree (and the attached alveoli) occur in all lobes of the lung (shown as shared aggregation, open diamond).
Morphological and anatomical compositions are separate hierarchies, but not independent. It may be desirable to be able to express which anatomical parts are contained inside a morphologi-cal object, even if an anatomimorphologi-cal part stretches through multiple morphologimorphologi-cal parts. In the ex-ample of the human body (Figs. 44 and 46) the two hierarchies are so far related only by using
“Entire body” as their base class, which does not allow one to determine which anatomical ob-jects may be found in the neck. In Fig.47, additional relations between morphological and anato-mical objects are drawn in the form of dependency relationships. Both larynx and lungs may be associated uniquely with a morphological object, whereas the trachea (windpipe) runs through both the neck and the thorax. Relations like that between “Neck” and “Trachea” could perhaps be modeled by creating a new relationship type, a “partial-part-of” relation.
Figure 45. A “dissection” of the object from Fig.43, showing an inner, anatomical composition hierarchy that is aligned with the outer composition hierar-chy.
Entire body Head
Neck Thorax
Arm Leg
Lungs Trachea
Larynx
Left lung Right lung
Figure 47. Interference between morphology and anatomy at the example of the human respira-tory system shown as dependencies (dashed arrows).
130. Anatomical (inward) composition hierarchies and are not necessarily nested inside a mor-phological (outward) composition hierarchy. They can therefore not be displayed in a single tree and support for multiple composition hierarchies is a requirement.
131. Mechanisms to express dependency relations between multiple composition hierarchies may be desirable.
Object decomposition
Decomposing objects, i.e., recognizing that an object is a composition and identifying the parts of this composition, is not always trivial. In many cases the decision whether to treat something as a separate object part (participating in an object composition), or whether to treat it as a pro-perty of the main object, depends on tradition rather than explicit rules. For example, most bota-nists will consider hairs on stem, leaves, etc. an attached structure (that would be modeled as a part in a composition). However, a dentate leaf margin will usually be considered a shape proper-ty rather than a larger number of separate teeth objects added to the margin (Fig.48). Similarly, structures responsible for a color effect may either be considered a composition or implicit in a property (Fig.49).
Pullan & al. (2005) analyze “striated area on petal apex: present”, concluding that it is a com-plex combination of “structure (petal apex), property (presence) and state information (striated)”.
Other authors may express the same information as: “area on petal apex: striated”. Again, the fundamental problem is whether a part (which may then have a name) is recognized, or whether
A)
B)
C)
Figure 48. Teeth of a dentate margin (top) are usually not considered components (but implied in the proper-ty), whereas hairs on a margin or surface are often con-sidered component objects. However, such teeth may have a substructure, e.g., a differently colored mucro-nate tip (circular insert).
Figure 49. Object color may be due to surface structures creating physi-cal colors effects (A), pigments (B), or soluble molecules (C). All may be viewed as a composition or not.
simply the anonymous region at the apex of the leaf is described through its properties. Rules how to systematically design a terminology are desirable for such cases.
Unfortunately, whereas in the previous example the second solution (using region instead of part) seems to be preferable, in many other cases no general “region” description is available as an alternative. In the case of complex butterfly or moth wing patterns it is often not possible to describe a component of the wing pattern without reference to its properties. The color of the third wavy band cannot be described as “structure X: wavy and red”.
Part of the rules governing object decomposition may be to consider something a part in a composition rather than a property if it has further object parts and properties. However this rule does not reflect the current customs in biological terminology. For example, teeth not only have properties that are directly shape-related (length, width, or angle), but may also have independent properties: different hairs may be found on the teeth, leaf veins may protrude at the tip, or the tip may be differently colored and mucronate (abruptly ending in a sharp point, insert in Fig.48).
When designing a descriptive system that intends to make consistent use of object composi-tions, the fact that logical object decomposition rules and biological practice often disagree has to be taken into account. It is important to provide a mechanism to map property expressions (leaf hairy, margin dentate, fruiting body unilocular or multilocular) to object compositions (and their multiplicity) to allow comparisons and to provide readable and intuitive descriptions.
In some cases, independent properties may cause interactions that influence decisions about object decomposition. In Fig.50 the left object is clearly recognized as a simple hexagonal shape.
However, if the striping pattern has additional darker lines added to it at appropriate positions, partitioning the object into two triangular and one rectangular object is tempting. Such problems of human perception are often embedded in biological descriptive terminology. For example, it will be difficult to define an unambiguous decomposition of contours contained in butterfly wing patterns that does not depend on a small taxonomic group.
?
Figure 50. Ambiguous object decomposition.
Whereas in the previous example the preferred de-cision would probably be not to decompose into parts, in the next example decomposition is expected. Fig.51 shows three “pseudocreatures” with increasing reduc-tion of head and neck shapes. Ultimately, the recogni-tion of a head is based only on the presence of eyes.
Pre-existing knowledge about possible head and neck reductions is essential to allow such an object compo-sition. This knowledge is often missing when people try to identify organisms whose fundamental arrange-ment is new to them. When developing a key covering both insects (tagmatized into head and thorax) and spiders (single cephalothorax) the treatment of parts in a character decomposition model involving a mandato-ry part-information would be a serious problem. In in-sects the eyes are part of the head, in spiders they be-long to a region of the cephalothorax.
Figure 51. If head and neck shapes are increasingly reduced, the recog-nition of “head” and “body” is ultimately based on other features (and knowl-edge that a reduction exists).
132. Whether physical objects should be considered atomic or a composition depends on per-spective and conventions, and may depend in complex ways on interaction with other compositions and properties. It is desirable to add mechanisms that help in communicating the perspective and conventions between designer and consumer or a descriptive terminol-ogy.
133. An object composition may often be considered a property of the parent object. The infor-mation system needs mechanisms to relate (or “map”) property and object composition expressions.
Competing classifications of object parts
In general, the decompositions (or “partitionings”) of biological organisms into parts are bound to a guiding principle considered optimal for a given purpose. The system may be based on morpho-anatomical aspects (as already discussed), but also on functional, biochemical, ontogenetic, life history stages (including multiple generations), phylogenetic, or similarity aspects. Different schools (and most notably terminology in different languages) frequently use different classifica-tions.
Different decomposition systems may result in partially congruent concepts, i.e., a given part may occur in different systems. The names for such object parts under different classifications may or may not be identical. Where they are, they can no longer be used to infer the classification system. For example, a rhizome (an underground stem) may be classified as part of the root-sys-tem (functional concept) or sroot-sys-tem (morpho-anatomical concept). While a petiole (leaf stalk) is commonly considered part of the leaf, the peduncle (flower stalk) is truly part of the stem system, but its length will commonly be noted together with other characteristics of the flower (or per-haps with the branching of the inflorescence, Table 39). Similarly, spines, thorns, and prickles have exact botanical definitions, but are in practice used interchangeably under a similarity or functional concept rather than a morpho-anatomically one. Note that in this case “spine or thorn”
is a generalization (kind-of) concept, which generalizes multiple object components (part-of con-cepts; see also “Generalization of object parts (compositional concepts)”, p.153).
Table 39. Examples of ambiguous or competing classifications of plant parts.
True morpho-
part of flower Other classification: “inflorescence”
(combining flower and stem characters) Spine
(from entire leaf)
kind of leaf “spine/thorn” e.g., Cactaceae Spine
(from stipules) part of leaf
kind of stipules “spine/thorn” e.g., Robinia or other Fabaceae, many Euphorbiaceae
Spine (from petiole)
part of leaf kind of petiole
“spine/thorn” e.g., in Fouquieria splendens Thorns
(from stem/shoot)
part of stem kind of stem
“spine/thorn” in Bougainvillea the thorns are modified inflorescences!
Prickle derived from stem
part of stem kind of epidermis
“spine/thorn” e.g., Rosa (rose, the “thorns” are prickles), Smilax (catbrier)
A classification may be useful in the majority of cases, but inappropriate in certain organisms.
Fig.52 presents examples where a common leaf classification is difficult to apply in certain taxo-nomic groups where leaf morphology varies continuously.
Figure 52. Examples of species with gradually changing leaf shapes, defying a strict classifica-tion into ground leaves/rosette, stem leaves, and inflorescence leaves (bracts). From left to right:
Alliaria petiolata, Sisymbrium austriacum, S. volgense, and S. orientale (simplified after Rothmaler & al. 1985).
The questions whether something is considered a composition or a property (see previous sec-tion), which classification system to use, and whether a classification system for object parts is appropriate are a major source for instability in descriptive terminology. They generally cannot be resolved by logic, but require consensus and conventions. This consensus often depends on a given taxonomic group, but may also depend on a school of thought.
The major problem with this is that while the context can be described for descriptive termi-nology and descriptions using this, it is usually unavailable during identification (when the class name is not yet known). If identification depends on correct decomposition, or correct application of a composition hierarchy it may fail. Although this problem is in principle shared by both char-acter matrix (p.104) and character decomposition models (p.116), the character decomposition models may suffer more if opinions about object decomposition evolve, and (at least current) de-composition models are unable to deal with multiple dede-composition concepts.
134. Multiple morphological concepts and corresponding object composition hierarchies may exist. It is desirable to support alternative concepts of object parts and composition hierar-chy.
135. The conventions whether something is considered a property or a composition, or which composition hierarchy should be preferred often depends on context, especially taxonomic scope.
Describing object-part composition
How can object or class-specific compositions be recorded as part of a description? How can descriptions of composition be compared with one another? The traditional solution is illustrated in Fig.53. The terminology contains a list of potential part concepts and for each of these a “part-is-present/absent character” is defined. This solution is used both by character/character state models (DELTA, CBIT Lucid, SDD, etc.) and character decomposition models (Nemisys/ Geni-sys, Prometheus description model).
Rosette petiole Rosette lamina Rosette leaf Stem petiole Stem lamina Stem leaf Stem Peduncle Sepal Petal Stamen Style Gynoecium Flower
Figure 53. An abstract, generalized list of plant parts (left) is filtered (or mapped) through a list of present/absent characters to describe an actual plant species (right).
In DELTA or CBIT Lucid information about a composition hierarchy is available to humans (through prior knowledge, character labels or headings), but not to machine-reasoning. In SDD it is expressible through a machine-interpretable terminological concept hierarchy (p.125), and in decomposition models (p.116; Nemisys/Genisys, and Prometheus description model) it is an ex-plicit part of the data storage model. In these models, the flat list of object parts from Fig.53 is replaced by a general object composition hierarchy (Fig.54, left side). By applying the informa-tion from “part-is-present/absent-characters” (provided these character are recognizable through metadata), a realized tree can be created (Fig.54, right side). This process is perhaps more similar to the flat character model than may be expected; Pullan & al. (2005) describe it as “parts of the ontology that correspond to the form of the particular specimen/taxon being described are flagged as being present”. This may be due to the need to record not only presence, but also multiplicity (discussed in the next section, see especially p.145).
Stem Flower
Stamen Petal
Rosette leaf
Rosette lamina Gynoecium
Sepal
Style
Rosette petiole Stem leaf
Stem lamina Stem petiole Peduncle
Stem Flower
Stamen Petal
Rosette leaf Gynoecium
Style
Rosette lamina Generalized Plant Composition Realized Plant Composition
Figure 54. In description models providing an explicit compositional hierarchy (SDD, Nemisys/ Genisys, Prometheus), hierarchy information may be inherited by the actual composition for a specific taxon (compare Fig.53).
DELTA, CBIT Lucid, or SDD do not provide metadata whether a character is a “presence/ ab-sence character” that informs about realized composition. In contrast, in explicit character de-composition models the combination of a part with the “presence/absence” property may be used to deduce this (depending, however, on the “correct” use of the model – an iridescent layer may be present/absent, but not iridescence itself).
Rather than relying on implied deductions from a composition hierarchy, DELTA, the Nemi-sys/Genisys outline, and SDD provide character applicability definitions (compare “Character ap-plicability rules”, p.76), which often are due to presence or absence of object parts. The character dependency mechanism is able to infer from “petal is scored as present”, and “petal presence de-pends on flower present”, that “flower must be present”. Although parts of the desirable inference from composition knowledge are thus covered, it is not possible to deduce the composition with safety. Reasons for character dependency other than optional composition exist and only humans can make appropriate deductions.
A part-hierarchy like the one shown in Fig.54 is intuitive and very useful for flowering plants.
However, the more structurally diverse a taxonomic group is, the less intuitive will such a hierar-chy be. Defining a generalized part-hierarhierar-chy for diverse taxonomic groups like Chlorobionta (ranging from microscopic green algae, over mosses and ferns to flowering plants) or Chordata (including primitive tunicates, cephalochordates, and vertebrates from fishes to mammals) is a challenging task. The results may be too abstract to be usable without additional annotations.
It seems a valid question whether the object composition hierarchy should be in the terminolo-gical domain at all, or perhaps rather in the description domain. The composition hierarchies (“part-of”) could be stored as a special form of descriptive data, similar to the absence/presence/ multiplicity information. The relations shown in the right-hand side of Fig.54 would then be part of descriptions, rather than inherited from the generalized composition hierarchy in the terminol-ogy domain, and may look like “petal is part of flower (5 times), stamen is part of flower (2 times)”. The information in such description-compositions could easily be compared, and, if
It seems a valid question whether the object composition hierarchy should be in the terminolo-gical domain at all, or perhaps rather in the description domain. The composition hierarchies (“part-of”) could be stored as a special form of descriptive data, similar to the absence/presence/ multiplicity information. The relations shown in the right-hand side of Fig.54 would then be part of descriptions, rather than inherited from the generalized composition hierarchy in the terminol-ogy domain, and may look like “petal is part of flower (5 times), stamen is part of flower (2 times)”. The information in such description-compositions could easily be compared, and, if