Another aspect that is particularly relevant to the description of biological objects is change over time (develops-from and stage-of relations). This may cause change to the properties of object parts (and consequently the concept of what a part is), but it may also affect the composition of objects. Temporal development in biology has four aspects:
■ Behaviorally related changes (e.g., a turtle retracts head and legs under threat),
■ continuous ontogenetic change during the development of individuals,
■ discontinuous, named life cycle stages, especially if different generations of individuals are involved in a life cycle,
■ phylogenetic change over evolutionary time.
In this discussion “changes over time” like deformation, discoloration, or decoloration as a result of specific preservation methods are excluded. These are treated under observation or measure-ment methodology, see p.171.
It is interesting to note that UML prior to version 2.0 had no mechanisms to model such in-formation. Mechanisms to model the dynamic behavior of programming objects over time (UML state, sequence, and activity diagrams) exist, but not a built-in modeling pattern to represent a static view of knowledge about potential change of objects.
Continuous ontogenetic change
Almost all object parts change over time during the development, maturation, or senescence of biological organisms. In many cases the changes (growth, slight deterioration when aging, etc.) are so regular and generic that they are not separately mentioned in descriptions. In cases, where unexpected or rare changes occur, the knowledge about temporal development processes is ex-pressed in descriptions in two major ways:
■ It may be embedded in the terminology. For example, bud, shoot, cotyledons, and leaves in a plant, larvae, pupae, imago in insects denote certain developmental stages. Some terms like
“annual”, “biennial”, or “perennial” embed knowledge about the life cycle of a plant, which can be gathered with certainty only by long-term observation – but which can in many cases be inferred based on other characters (woodiness, development of root system, etc.).
■ The changes of object parts or properties are explicitly described: “leaf shape rhomboid in young leaves, later elliptical”, “flowers blue when opening, later red”, or the wandering of eyes that occurs when larval flatfishes (flounder, sole, halibut, etc.) change from symmetrical shape to the asymmetric flattened shape they obtain when settling on the sea bottom. These explicit descriptions of temporal development can either be expressed as modifiers of catego-rical terminology, or as free-form comment text.
Named life cycle stages and generations
Many organisms have more or less distinct phases in their temporal development called develop-mental or life cycle stages. These may be an artificial classification of a continuum (e.g., embryo/ new-born/infant/youth/adult, or seed/seedling/sapling/adult plant) or associated with distinct events like molting, pupation, change of habitat (free-swimming to sessile, different host orga-nisms), or the changes in nuclear phase (e.g., haploid vs. diploid). The first of these cases is called “growth stages” by Pujar & al. (2006) and defined as “distinct morphological landmarks in a continuous developmental process”.
In many organisms different life cycle stages form no special problem for descriptions be-cause – even though property values like measures of size, proportions, patterns, coloration, etc.
may change – the fundamental property and object composition model applies to all stages. Ex-amples are infants and young individuals of most mammalian and avian species, larval stages of ametabolous insects, or the shoot stages of most plants. Describing all in a single description may work in some cases (e.g., adult and senescent animals) but not in all cases. For example, cater-pillar instars of a single species or seasonal generations of butterflies like Araschnia levana may display substantially different colors and patterns. To avoid overgeneralization, such generations and life cycle stages have to be treated in separate descriptions (requiring a life cycle annotation of the description scope, see “Secondary classification resulting in description scopes”, p.215).
In other organisms, however, the life cycle stages differ so profoundly that the applicability of property and object composition models depends on the development stage. This is quite similar to dependency of such model on taxonomic groups (and occasionally different sexes). Examples are hemi- and holometabolic insects (e.g., changing from larva over pupa to butterfly), marine hydroids (Cnidaria: Hydroida; changing from swimming planula larvae over sessile polyps for-ming colonies to swimfor-ming, jellyfish-like medusae), anamorphic and teliomorphic stages of many fungi, or ferns (alternating between an inconspicuous prothallus in the haplophase and the diploid fern).
In the context of identification, determining the life cycle stage is usually a post-recognition feature, i.e., it requires identification at least to the level of a higher taxon from which some gen-eralization can be deduced. In many examples, even a specific identification is required. Where different character sets are required to describe life cycle stages, a generalization method may be required for successful identification (compare “Problems with specialized, context-dependent names for object parts”, p.157).
Phylogenetic change
As mentioned above (p.153), the taxonomic hierarchy of non-extinct organisms is a generaliza-tion hierarchy. However, the development of lineages of evolugeneraliza-tionary time (anagenetic evolugeneraliza-tion) is a part-of hierarchy. A paleontological species may therefore be a part of the evolutionary clade leading to a current species. Since this is difficult to ascertain (the paleontological specimen could belong to a side-line now extinct), these part-of hierarchies are irrelevant in practice.
For all three changes over time, the periods on the time axis (e.g., embryo–adult, larva–pupa–
butterfly, Triassic–Jurassic–Cretaceous–Tertiary–Quaternary) form a sequence that can be ar-ranged in a part-of hierarchy. The changes of parts that are correlated with this time axis (e.g., wing sacs to wings) are, however, best viewed as kind-of generalizations.
165. In addition to the common composition (part-of) and generalization (kind-of) relations, relations expressing change over time (ontogenetic, life cycle, evolutionary history) are desirable in descriptive information models designed for biological objects.
Properties
Definitions
The composition and generalization concepts of object parts discussed so far are one half of the character decomposition models (see p.116). The other half are called properties in the Nemisys/ Genisys models (compare Table 9, p.63; basic property under data type aspect) and the Prome-theus description model. In the following an attempt to discuss properties, methods, their interac-tions and generalizainterac-tions is made.
The concept of properties (or attributes, the preferred term in UML and ER-modeling) in in-formation science is relatively easy to understand. Variables and constants associated with any given class are considered properties of this class. The type of these variables may be simple, in-cluding value types like integer or Boolean, or it may be complex, inin-cluding any other class. Ulti-mately, the composition of object instances is defined through class properties. In UML models, relational attributes are usually not shown in the list of class attributes, but only as end names of associations. However, as soon as programming code is created from these models, the associat-ions are converted to properties or association classes containing the necessary properties.
Table 44. Examples of dictionary and UML definitions for attribute, feature, and property.
Term Collins English Dictionary
(CED 1992) Merriam-Webster’s Collegiate/
New Oxford Dictionary (EB 2001) UML 1.5 definitions (OMG 2003) Attribute
(noun)
[…] 2. a property, quality, or feature belonging to or representative of a person or thing. – 3. an object ac-cepted as belonging to a particular office or position. – 4. Grammar. a.
an adjective or adjectival phrase. b.
an attributive adjective. – 5. Logic.
the property, quality, or feature that is affirmed or denied concerning the subject of a proposition. – [From Latin attribuere to associate with, from tribuere to give]
1. a quality or feature regarded as a characteristic or inherent part of someone or something: flexibility and mobility are the key attributes of Brit-ain’s army. – 2. a material object recognized as symbolic of a person, especially a conventional object used in art to identify a saint or mythical figure. – 3. Grammar: an attributive adjective or noun. – 4. Statistics: a real property which a statistical analysis is attempting to describe.
A feature within a classifier that describes a range of values that instances of the classifier may hold.
(Def. of Classifier is: A mechanism that describes 2. a prominent or distinctive part or aspect, as of a landscape, building, book, etc. […] – 9. Linguistics. a quality of a linguistic unit at some level of description: grammatical feature; semantic feature. – [From Anglo-French feture, from Latin facere to make]
1. a distinctive attribute or aspect of something: impressed with the safety features of the plant. – 2. (usually fea-tures) a part of the face, such as the mouth or eyes, making a significant contribution to its overall appearance.
– 3. Linguistics: a distinctive character-istic of a lingucharacter-istic unit, especially a speech sound or vocabulary item, that serves to distinguish it from others of the same type. […]
A property, like operation or attribute, which is encap-sulated within a classifier, such as an interface, a class, or a data type.
Property (noun)
1. something of value, either tan-gible, such as land, or intantan-gible, such as patents, copyrights, etc.
[…] – 6. a quality, attribute, or distinctive feature of anything, esp.
a characteristic attribute such as the density or strength of a material.
[…] – [From Old French propriété, from Latin proprius one’s own]
[…] an attribute, quality, or character-istic of something: the property of heat to expand metal at uniform rates.
A named value denoting a characteristic of an
Unfortunately, when studying the relation between the physical (e.g., biological) world and their description in information models, the definition of the term “property” (Table 44) in rela-tion to the physical world becomes considerably more difficult. Intuitively, it makes sense to say
“diameter and shape are properties of the head”. This intuition is certainly valuable. However, such a “property” (or “attribute”, “quality”, “distinctive feature”, “characteristic attribute”
accor-ding to the dictionary definitions) may be obtained by various methods, some of which result in comparable, others in non-comparable data. Both the measurement method and the storage me-thod (data type and measurement units) seem to be tightly coupled with the abstract concept of
“properties” themselves.
The discussion will first discuss some aspects of properties under an intuitive concept of “ob-ject property”, then study the dependency of descriptive data on measurement methods, and final-ly attempt to review the relation between property and measurement methods.
Property decomposition
In many cases a property may be complex in the sense that it is possible to express the same in-formation in a combination of other, more atomic properties. It is not possible to generalize that for a given purpose always complex or always atomic properties are preferred. The complex property “inflorescence types” is generally preferred. On the other hand, for object shapes it is customary to decompose the complex shape into a generalized shape plus secondary shape prop-erties. In the case of a square with rounded corners (Fig.18, p.65) most humans attempt to ab-stract the shape as a square for as long as possible, and add the information about the rounded corners separately.
An example from biology is the collection of leaf shapes shown in Fig.75. Technically, all five shapes are different. However, they would all be described as the abstract shape “elliptical”
plus information about the leaf apex (acute, rounded, obtuse to truncate), leaf base (rounded, cordate) and margin (smooth, dentate) described separately.
As discussed already under “Calculated characters” (p.72), it is desirable to provide mapping or calculations to convert complex to atomic properties and vice versa.
Figure 75. Five different leaf shapes, all of which might be described as the abstract shape
“elliptical”, plus separate extension/variant information about tip, base, or margin.
166. What is considered a property is subject to conventions. Complex properties exist that may also be expressed as a set of more atomic properties. A conversion/mapping functionality is desirable.
Pattern versus composition
Any pattern is, by definition, a composition of other elements. In the strict sense, a pattern con-sists of an arrangement or design that is repeated in itself. In biology it is, however, customary to also speak of a pattern when referring to a unique arrangement occurring only once (e.g., a wing pattern), but being repeated on each individual of a species.
When modeling object descriptions, patterns create problems because it is customary (and ef-ficient for human recognition) to give complex patterns “type names” like “striped”, “checkered”,
“hatched”, “dotted”. In biology, even typical wing patterns, e.g., of the moth families “Geometri-dae” or “Noctui“Geometri-dae” may be recognized.
Already in Fig.43 the triangular pattern on the central object was considered an object compo-sition, whereas the striping was considered a pattern. Fig.76 shows several patterns that can either be described as object compositions or as named pattern categories. The left object is easily described as a composition of a white square with a central gray circle. The second object is more readily described as a square with a regular pattern of gray circles on white background. The use of the pattern approach can still be maintained in the third object (e.g., “alternating rows of gray and black circles on white background”). These named pattern descriptions are more intuitive to humans than an exact enumeration of circles, their color, and which circles are adjacent to each other at which angle (besides that the objects in pattern may easily be too numerous to enumerate them exactly, compare “Categorical multiplicity”, above). However, complex patterns that are a mixture of regular and irregular arrangements (like in the second but last object in Fig.76) are not easily described.
Furthermore, it is difficult to give guidelines about when to use an explicit object composition and when to use a named pattern category. In the last object in Fig.76 the circle object can be described as an ordered composition of black and white horizontal lines of specific length. From this description the fact that these lines appear as a circle with a horizontal striping pattern may still be deducible to computer algorithms, but completely opaque to human recognition.
Figure 76. Patterns that can either be described as object compositions or as named pattern categories. The fourth pattern is not repeated in itself (see text for further information).
Figure 77. The concept of a pattern abstracts from concrete color to the ability to distinguish parts. Distinction may even occur through surface texture of parts of identical color (here illus-trated with a fine line pattern that is intended to symbolize a texture).
A potential solution seems to be to define a library of named pattern types, each of which is defined as an object composition. These object compositions could then be used in the back-ground to allow an algorithm to either give a pattern name for an object composition, or reversely evaluate similarities of patterns. However, a major problem with this approach is that two-dimen-sional patterns are created by objects of different colors, and that it would be impractical to de-fine a type for each color combination. Similarly to the examples discussed in the section “Pro-perty decomposition” above, the concept of “pattern” abstracts a complex reality. The actual reason why components of the pattern are distinguishable (such as color, texture, see Fig.77) and the absolute size and to lesser extent the shape of the pattern components is considered irrelevant, and only the relative arrangement of pattern components isolated. As a consequence, a “pattern
type library” could be a parameterized object composition where the properties of components (color, texture, etc.) are expressed through parameters that are defined in addition to the type.
Although such a solution is conceivable, it seems difficult to implement in OOP languages and requires considerable custom code for analyzing, editing, and reporting descriptions using this concept. Further analysis into this problem is required.
The current approach method of categorizing patterns (used, e.g., in DELTA or NEXUS data sets) has the advantage of not introducing any new concepts or data structures that complicate the information model. It is, however, often unsatisfactory because it creates a dependency between character states of “object pattern” and “object color”. In the presence of a pattern it is customary to list all patterns present in the components of the pattern as object colors, a process in which information is lost (Fig.78).
(A)
(B)
Figure 78. Two populations (clouds) with four patterned objects each. Recording pattern and color as independent properties prevents distinguishing between multiple colors as a result of population variability (A) and co-occurring in a single pattern (B). The class descriptions for populations would be: pattern “dotted”, colors “white, gray, or black”.
167. Patterns are especially problematic situations that may be modeled through properties or compositions. Patterns are highly relevant to the description of biological objects and ad-equate support for them is required.
Property interactions
The previous section already discussed the problems of property interactions in the case of pat-tern and color. Other cases exist:
1. When measuring the spore size (diameter or length/width), some spores may be halonate (they have a low-density gelatinous outer layer) or they may have appendages. The measurement instructions must define whether the halo or the appendages are to be included in the size meas-urement or not.
2. When recording length and width of two-dimensional shapes, two values may be redundant where length and width are by definition identical as in circular or square shapes (Fig.79). The case of globose versus ellipsoidal shapes often occurs in the case of diaspores (seeds, fungal spores, etc.).
Diameter 25 25
25
40
25
40
25
=
Radius 12.5
Figure 79. Example for a dependency of shape and size properties. For most two-dimensional shapes a length and width is recorded, but this is redundant where length and width are by defi-nition identical.
Three models of property definitions are commonly found:
a) 1. Length b) 1. Length or diameter c) 1. Length
2. Width 2. Width 2. Width
3. Diameter
In the case a) a dependency from the shape property to the size property can be created such that diameter is applicable only when shape circular etc. In the case c) the length and width are al-ways entered, regardless whether they are identical or not. No model is fully satisfactory. Pro-vided that a system could rely on a rule that data are always either not coded at all, or coded com-pletely, it is relatively simple to represent either of the above cases using the rule: If the width is missing, replace it with a reference to the length (which would return missing if the length is also missing). This assumption is implicitly made in the case of b). However, whereas in the case of fungal spores it is customary to always record both length and width, this is not necessarily true for all objects. For example, for very small petals only the length might be measured, whereas in larger petals both length and width should be measured, In this case the representation b) and the implementation rule given above would assume that small petals are always round with length = width. Furthermore, length and width may occasionally have an absolute orientation in which case width may become larger than length (see p.169 and Fig.81).
Informing about the conditions for compara-bility of the multiple properties in this case is not trivial. An interesting option is perhaps to introduce “label-dependencies” rather than pro-perty or character dependencies. For example, if shape is circular, property ‘1’ might be labeled
“diameter”, otherwise “length”. Unfortunately, while improving the handling of case b), this does not cover the case of absolute orientation, where it may be desirable to compare a diameter with width as well as length.
3. A related problem occurs where properties or the object composition of sides (e.g., upper versus lower side) of an object may be homo-geneous or heterohomo-geneous. In Fig.80 A) and B)
C) A)
B)
Figure 80. Orientation (upper/lower side) may be significant (A, B) or insignificant (C) when recording object properties (presence and density of circles).
upper and lower side differs and would be separately described. However, in C) both sides are
upper and lower side differs and would be separately described. However, in C) both sides are