Unlike what is the case for physical entities and other types of continuants, few process ontologies exist. This is not only because processes received less attention in the research community, but also because classifying them is challenging. Moreover, upper level categories or classification criteria to help in modelling and integrating lower level process ontologies have thus far not been developed or widely adopted. This paper proposes a basis for further classifying processes in the Basic Formal Ontology. The work is inspired by the aspectual characteristics of verbs such as homeomericity, cumulativity, telicity, atomicity, instantaneity and durativity. But whereas these characteristics have been proposed by linguists and philosophers of language from a linguistic perspective with a focus on how matters are described, our focus is on what is the case in reality thus providing an ontological perspective. This was achieved by first investigating the applicability of these characteristics to the top-level processes in the Gene Ontology, and then, where possible, deriving from the linguistic perspective relationships that are faithful to the ontological principles adhered to by the Basic Formal Ontology.
The importance of process ontologies is rapidly
increasing in several domains such as in event discovery
In BFO
In DOLCE
Although DOLCE and BFO are based on distinct perspectives, it is not unreasonable to view processes and accomplishments in DOLCE as what are processes in BFO. However, how achievements and states are to be interpreted by BFO remains unclear. If achievements are instantaneous happenings with zero-time duration then they correspond to BFO’s process boundaries. If, however, instantaneity is meant to be a short period of time, then they are BFO processes. Additionally, states in DOLCE cannot be BFO processes as they do not involve change.
Early research in philosophy and linguistics provided
different accounts of the so-called events, suggesting various
views and criteria to distinguish between actions, activities,
accomplishments, achievements, processes, performances
states, mental and physical events, bodily movements, and
others
In this paper, we examine the most commonly discussed
criteria in the literature to classify events and verbal
predicates: homeomericity, cumulativity, atomicity, telicity,
durativity and instantaneity. Other aspectual notions, such as
incrementality and structure
To test the applicability of these notions, we attempted to use them to analyze and annotate the top level processes in the Gene Ontology. Table 1 illustrates a sample of these GO processes. The full annotations (and documentation of choices) can be accessed online1. We selected the 35 most top-level processes, that top all other (~45k) processes in the Gene Ontology.
It is worth noting that the top-level processes in the Gene ontology include not only process universals, but also defined classes, such as Single-Organism Process and Multi-Organism Process. Such process types cannot be classified according to criteria such as whether they are e.g., homeomeric or telic because they are collections of different types of processes, defined based on the number of the organisms involved. Therefore, instead of selecting these processes, we selected and annotated their subtypes.
This initial effort turned out to be a challenging task for two reasons: (1) shortcomings in the literature about what precisely is to be understood by these notions, and (2) their focus on how matters are described (the linguistic perspective), rather than on what is the case (the ontological perspective).
In the next sections, we examine the most commonly used aspectual criteria and discuss whether and how they might be used for further classifying processes in the spirit of BFO. 3
Central in our approach is the distinction between
temporal parts of processes, such as the first year of a
toddler’s life, and mere occurrent parts of processes, as
each of the eating processes that are part of a toddler’s life. 2
p occurrent-part-of q (R1)
a primitive relation of parthood holding independently
of time between two process instances when one is a
subprocess of the other
The relation ‘p spans r’ in R3 is shorthand for ‘p occupies spatiotemporal region str which occupies temporal region r’. R3 allows us to define the corresponding universal-level relation: P temporal-part-of Q =def.: (R4) for every particular occurrent p, if p instance-of P, then there is some particular occurrent q such that q instanceof Q and p temporal-part-of q.
Figure 2 depicts an instance of PROCESS PROFILE, a
subtype of PROCESS
DOLCE considers an occurrence to be homeomeric iff all of its temporal parts, in BFO sense, are described by the very expression used for the whole occurrence, a view which is close to the notion of homogeneity as defined in (Dowty 1977:60). An example of a homeomeric process in this DOLCE sense would be what is described by ‘sitting’. Defined in this way, homeomericity is a property of universals, not of instances. Although BFO is a classification of instances and not of universals, we can work around this by recognizing several flavors of this principle at the level of instances: (1) p isotypic-part-of q =def.
p temporal-part-of q & p instance-of all types instantiated by q.
Example (figure 2): p6 isotypic-part-of p5. (R5) (2) p weakly-homeomeric-in P =def. (R6) all temporal parts of p which are not process boundaries are instances of P.
Examples (figure 2): p2 weakly-homeomeric-in DC1; p3 weakly-homeomeric-in DC1.
From R5 and R6 it follows that all parts of a process that is weakly-homeomeric-in some type are isotypic-part-of that process. (3) p strongly-homeomeric-in P =def. (R7) all temporal parts of p which are not process boundaries are instances of P and there is no such part of p that instantiates a subtype of P.
From R5 and R7 it follows that all parts of a process that is strongly-homeomeric-in some universal are isotypicpart-of that process.
Further to be investigated is whether it is worthwhile to introduce the DC ‘weakly-homeomeric-process’ as the DC formed by all processes which are – or have an occurrent part which is – weakly-homeomeric-in some process type. Also whether it would pay off to implement the notion of ‘strongly-homeomeric-process’ as the DC formed by all processes which are– or have an occurrent part which is – strongly-homeomeric-in some process type or whether it would be sufficient to have an axiom directly at the level of the types to which it applies. In figure 2, this would hold for DC8, 9, 10 and 11. It is also worth noticing that whether any of the relations R5…7 holds, depends on whether (1) a pure metaphysical stance is taken or (2) reality is looked at through what an ontology allows us to see and what is observable at the level of instances. If a restricted ontological commitment would not allow us to observe or describe instances p4…p8, then, under that commitment, e.g., p1 would be strongly-homeomeric-in DC5 and DC6. 3.2
Cumulativity was extensively discussed in the lexical
semantics literature
Looking at it from the BFO perspective, it leaves open the question what sort of mereological sum is intended. Does it include occurrent parthood (see R1), as when my sitting (p1) simultaneously with your sitting (p2) is summed to form p3 which is the sitting of the mereological sum of me and you? Would certain process aggregates such as an orchestra playing a symphony being the mereological sum of the playings of the individual musicians plus the directing of the conductor? Or is it exclusively temporal parthood (R3) such as my sitting during the first 30 minutes of my total sitting and my sitting during the last 30 minutes?
The DOLCE documentation about cumulativity as well as
related proposals advanced in the linguistic community are not
at all clear whether the sort of cumulativity they have in mind
is a property of processes or of how a process is described.
all process types instantiated by p and all process types instantiated by q are instantiated by p, q and p+q.
Example (figure 2): p2 cumulative-with p3; DC1, DC5 and DC6 are the all and only types that are instantiated by both p2 and p3, as well as p1 (the mereological sum of p2 and p3).
P cumulative-in Q =def.
P isa Q & for all p1, p2 instance-of P: (p1+p2) instance-of Q. (R9) (R8) Example (figure 2): DC5 cumulative-in DC1.
Clearly, if some process p is at least weaklyhomeomeric-in P then it is also instance-of a type which is cumulative-in P, but not vice versa. One could assume, for example, that instances of Growth as defined in the Gene ontology would be weakly-homeomeric-in Growth and that all subtypes of Growth would be cumulative-in Growth. Subtypes of Cell Aggregation, however, could be assumed to be cumulative-in Cell Aggregation, but for sure no instance of it would be strongly-homeomeric-in Cell Aggregation: not every temporal part of a cell aggregation process is of the same type (e.g. its sub-process ‘cells coming close to each other’ is not itself an aggregation). However, the mereological sum of two cell aggregations would be a cell aggregation. 3.3
From a lexical semantics perspective, an action predicate
is telic if it refers in one way or another to a terminal point
for the action described as tending ‘towards a goal
envisaged as realized in a perfective tense but contingent in
an imperfective tense’
There are for sure processes that are telic under a specific interpretation of ‘terminal point’, i.e. all those which are finished! This is of course not what we have in mind here. What we do mean is that for a process to be telic, there must be something ‘in it’ what we informally can describe as ‘a change in the process’. However, this is just a matter of speaking as under the perspective of BFO, processes do not change because they are changes!
Consider the current laws of physics on Earth, one being that objects such as a rock without support fall down to Earth. The rock cannot keep falling down forever, it naturally stops when reaching a surface. Therefore, any such ‘falling under natural Earth conditions’ is a telic process. This ‘falling’ terminates with (is followed by) a process that can be roughly described as ‘the coming to a stop’ and which can take the form of bouncing a bit (on a hard surface), or penetrating a soft surface (sand, for instance); it is this coming to a stop which is that what informally can be described as ‘the change in the process’. As another example, imagine that by looking through the representational units in figure 2 we do not just see the process profile p1 and the parts p2, p3, … thereof, but the actual beating of somebody’s heart b1 within that timeframe, and the process parts thereof – process parts being processes in their own right – b2, b3, …, such that p1 occurrent-partof b1, p2 occurrent-part-of b2, and so forth. Then b2 did not ‘change’ when b4 terminated and b5 started, and neither did b1. b2 is the change! Remember also that from the BFO perspective it does not make sense to refer to processes ‘at a time’. We believe that telicity is once more not some unary property of an individual process, but rather something in relation to a universal. So we define: p is-telic-in R =def.
p instance-of P & there exists some process q instance-of Q and some process r instance-of R, such that (1) q not instance-of P, (2) p not instance-of Q, (3) p precedes q, and (4) p and q are temporal-part-of r.
Example (figure 2): each of p4 , p5, p7 is-telic-in DC1. P is-telic-in R =def.
for all process p instance-of P, p is-telic-in R.
For example: falling (i.e. under normal earth conditions) is-telic-in moving (p is the falling proper, q is the coming to a stop (bouncing, penetrating, …), r=p+q is the entire motion process). Note that in general language, both p and r might be denoted by ‘falling’ Another example: ‘walking leg swing’ is-telic-in ‘walking’.
We do not follow
Some researchers distinguish between processes that are
instantaneous and others that are durative
Nevertheless, the instantaneous events discussed in the linguistics literature are likely not meant to be time points or process boundaries, but rather, a sort of processes. Thus, we revise the definition of instantaneous processes to be those types of processes that are the peak moments of some longer processes. Processes like, knocking, hitting, departure, arrival, and dying are examples of processes that extend over time, but they have temporal parts that we call peak moments. For example, knocking the door starts when one moves his hand towards the door, then continues by touching and hitting the door for a short moment (the peak moment), and then releasing the hand. Similarly, the arriving of the train may start when the first trolley of the train starts entering the station, and decreases its speed into zero (the peak moment), wait a bit, and then open its doors.
Notice that peak moments are critical parts. For example, if a process started with its agent intending it to be a knocking process but is terminated before its peak moment (i.e., before touching the door) then it is not a knocking, but if it is terminated directly after its peak then it is a knocking. In fact, this shows that instantaneous processes cannot be strongly-homeomeric-in some universal because their temporal parts before and after their peak are not of the same type. One may even slice a process into a higher granularity and ask at what time point a process is considered done, but we believe that this is irrelevant since the boundaries of instantaneous processes are typically fiat boundaries. For example, the exact starting and ending of knocking, and similarly a train’s arrival, are typically a matter of fiat determination. Further to be investigated is whether an entity (continuant or occurrent) is always created at a peak moment, e.g. a sound when knocking, a new individual when giving birth, a hole when drilling a wall, and so on. For this reason, no formalization for instantaneous processes has yet been decided upon. 3.5
Atomicity was adopted by DOLCE to distinguish
between accomplishments (non-atomic) and achievements
(atomic)
We have overviewed some known aspectual notions used to classify verbal phrases and examined their reuse to build process ontologies under BFO. We provided BFOcompatible interpretations of homeomericity, cumulativity and telicity, discussed instantaneity, and explained that atomicity does not seem to be beneficial in classifying processes. We illustrated the use of these notions to analyze and annotate the top levels of the Gene Ontology processes. We plan to extend this work in several directions, including the full ontological analyses of the Gene Ontology processes, in order to detect possible modeling inconsistencies, as, for example, would be the case when a non-homeomeric process is declared a subtype of a homeomeric process, extending BFO process categories, as well as examining the use of these notions at a larger scale, such as restructuring of the process types found in WordNet.
Part of this research was conducted during Prof. Jarrar’s visit to the Ontology Research Center at UB, which was funded by the Fulbright Visiting Scholars Program. This work was supported in part by CTSA NIH 1 UL1 TR001412-01 from the National Institutes of Health. collectivity and Dowty, D. 1977 Toward a Semantic Analysis of Verb Aspect and the
English 'Imperfective' Progressive.Linguistics&Philosophy, 1(1):45-77