1. Introduction
Ontologies are usually constructed to enhance the interoperability and integration of data
that are dispersed in different information systems, including databases. They are thus
designed to provide a general representation of reality that holds independently of
specific circumstances. It is an ontological finding, for instance, that a student can be
classified as a role regardless of which particular student or school we are discussing. In
contrast, our knowledge of the world is mostly context-dependent; and ignorance of the
contextuality of information would yield undesirable consequences. When it is ordered
to go right at the fork to move away, for example, a robot may fail to reach her destination
when it collides with an obstacle that is located in the right road and that she has never
encountered in choosing the same path. The robot should be then reprogrammed to
proceed in an appropriate direction depending on ever-changing situations.
A full ontological exploitation of knowledge therefore requires that its
contextsensitivity be taken with the utmost seriousness. Indeed, classifications of contexts have
been discussed comparatively intensively for the last two decades. For instance, Hayes
[
1
] identifies four kinds of contexts (physical, linguistic, conceptual, and deductive) from
a linguistic point of view. For another example, Borgo et al. [
2
] introduce three types of
contexts (global, local, and internal) to classify an agent’s knowledge in the
manufacturing domain. Contexts are nonetheless notoriously difficult to analyze from an
ontological perspective, in spite of a growing demand for their formal specification.
This short paper aims to take the initial step towards a deeper understanding of the
ontological nature of contexts. Our methodology is to deploy a so-called
‘metaontological’ notion of grounding (which we will delineate below). In ontology
engineering, there is an increasing practical interest in meta-ontology [
3
] as a
secondorder inquiry into ontologies. Among meta-ontological concepts is a truthmaker [
4,5
]:
something that ‘makes true’ (i.e., bears the ‘truthmaking relation’ towards) a proposition.
For instance, Japan can be a truthmaker for the proposition that Japan exists. The idea of
1 E-mail: toyo.fumming@gmail.com, fumiakit@buffalo.edu. Copyright © 2019 for this paper by its
authors. Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
truthmaker or truthmaking has been employed for the last decade to clarify complex
ontological categories and relations. Examples include the formalization of states as
truthmakers for propositions [
6
] and the conceptualization of properties, relations, and
events in terms of ‘truthmaking patterns’ [
7,8
]. As a sibling notion of truthmaking,
grounding has been only recently used, e.g., to analyze the role concept in upper
ontologies [
9, 10
]. Our grounding approach to contexts would therefore contribute to a
new ontological approach to contexts and to heightened awareness of grounding in
foundational ontology research.
The paper is organized as follows. Section 2 presents a general idea of contexts and
related work, or especially Casanave’s [
11
] ‘mediator view’ of contexts. Section 3
explains a meta-ontological concept of grounding. Section 4 sketches out a grounding
analysis of contexts. Section 5 concludes the paper with some remarks on future work.
2. Contexts
2.1. General Ideas
Contexts are doubtlessly elusive concepts, but Baclawski et al. [
12
] articulate their basic
idea: “In general, a context is commonly understood to be the circumstances that form
the setting for an event, statement, process, or idea, and in terms of which the event,
statement, process, or idea can be understood and assessed. Thus for utterance statements
we often talk of the linguistic context of what is being expressed. In addition, there may
be a physical context, circumstance or state of affairs in the real world that provides
context for uttered statements. Some examples of synonyms or alternate terms that have
the flavor of context include circumstances, conditions, factors, perspective, scope, state
of affairs, situation, background, scene, setting, and frame(s) of reference.”
Based on this general idea, we can identify at least two key features of contexts.
First, it is a meta-level entity: a context is always a context for something. To take an
example, the meaning of a sentence “I am hungry” varies from context to context, and
in this case we are speaking of a (linguistic) context for the uttered sentence. Second and
closely relatedly, a context consists in changing our interpretation of something else (for
which the context holds). When Ann sincerely utters “I am hungry,” we are justified in
thinking that Ann is hungry; but we are mistaken in believing so when Bob (but not Ann)
sincerely utters the same sentence.
2.2. Related Work: Casanave’s Mediator View
There are presently few ontological investigations into contexts, notwithstanding several
classifications of (linguistic and/or cognitive) contexts [
1,2
] and rich philosophical
discussion on (linguistic) contexts [
13
]. According to Baclawski et al. [
12
], Casanave
[
11
] can be construed as espousing a ‘mediator view’ of contexts based on the intuition
that a context is “anything that impacts the interpretation or truth value of something
else.” For him, a context acts as a mediator between a set of propositions (or rules) and
the things, where in his terminology, the words therein ‘proposition’ and ‘thing’ mean
the ‘interpretations/truth values’ and the ‘something else’, respectively. In more detail, a
proposition holds within a context, which in turn contextualizes the things; and a true
context implies that the propositions that hold within the context hold for all things that
the context contextualizes (see Figure 1).
Casanave’s mediator (albeit preliminary) account of context is well worth noting.
For one thing, it aims at a general notion of context: “This pattern works for many
contextual dimensions such as time, location and provenance” [
12
]. For another, it
attempts to capture the core idea of context (namely, a meta-level entity that changes our
construal of the world) in terms of the ontological notion of proposition. Contrariwise,
there is some room for improvement in the concepts in his schema, or specifically the
relations used: e.g., the holds-within relation between propositions and a context.
3. Grounding
3.1. Fact-grounding
To move beyond Casanave [
11
], we leverage a meta-ontological tool of grounding
[1416] or especially the most orthodox version of grounding as a primitive relation between
facts [
17,18
]. For instance, the fact (say F1) that a table exists is grounded in the fact (say
F2) that some subatomic particles are arranged table-wise; and informally speaking, F1
holds ‘in virtue of’ F2. The theory of fact-grounding is usually coupled with the claim
that the grounding relation at least entails explanation (see [
16
] for controversy over the
relationship between grounding and explanation). In the table example, F2 grounds, and
ipso facto explains F1. Grounding is also so analogous with causation that one may
sometimes call grounding ‘ontological causation’ [
19
].
Fact-grounding presupposes an ontology of facts. To keep things manageable, we
leave aside the problematic character of facts (see e.g., [
20
]) and simply say that the term
‘fact’ is typically interpreted either as a true proposition or as a state of affairs. On the
one hand, a proposition is standardly taken to play three major roles:
•
•
The semantic content of a (declarative) sentence. E.g., two sentences “Snow is
white” and “La neige est blanche” express the same proposition that snow is
white.
The object of various linguistic and cognitive attitudes (‘propositional
attitudes’) including belief, assertion, and denial. E.g., when she sincerely utters
“Snow is white,” Mary bears the believing attitude towards the proposition that
snow is white.
The truthbearer: the bearer of truth-values (truth and falsehood). E.g., the
proposition that snow is white is true.
On the other hand, a state of affairs is a concrete, non-linguistic portion of reality, but
with a ‘propositional structure’: e.g., a state of affairs of snow being white [
21
]. A fact
can be therefore described as something that ‘holds’ in reality in virtue of its
propositional structure, irrespective of whether it is a true proposition or a state of affairs.
3.2. Full and Partial Grounding
We can distinguish two kinds of grounding relations: full and partial grounding [
18
]. A
plurality of facts F1, F2, … fully ground a fact G (symbolization: F1, F2, … ⇒ G) when
F1, F2, … ground G collectively and completely; and a fact F partly grounds a fact G
(symbolization: F → G) when F grounds G singly and partially. Partial grounding can be
naturally defined in terms of full grounding: F → G if and only if there exist F1, F2, …
such that F, F1, F2, … ⇒ G. To illustrate them, let H and I be the fact that Japan is an
eastern Asian country and the fact that Japan has a population of nearly 130 million,
respectively; and let J the fact that Japan is an eastern Asian country with a population
of nearly 130 million. Then, the following claims hold: H → J, I → J, and H, I ⇒ J.
We can also speak of some formal properties of grounding (whether full or partial)
[
18
]. (i) Irreflexivity: no fact grounds itself. (ii) Transitivity: if a fact F1 grounds a fact
F2, which in turn grounds a fact F3, then F1 grounds F3. From (i) and (ii) follows
straightforwardly the asymmetricity of grounding: if a fact F1 grounds a fact F2, then it
is not the case that F2 grounds F1. To simplify the matter, we set aside a highly debatable
topic of whether grounding is, in some sense, ‘well-founded’: whether it has no infinite
descending chains or not (see e.g., [
22
] for details).
4. Towards a Grounding Analysis of Contexts
Going back to contexts, we suggest that contexts be specified in terms of grounding. The
underlying idea is that closer examination of the meta-level concept of context would
necessitate the usage of some meta-ontological tool such as grounding. In particular, the
idea of partial grounding is vital for considering contexts. Baclawski et al. [
12
] contend:
“any reasoning about context is (…) about the methods for finding some implicit
information that should be added to the interpretation of the subject.” Suppose that F1,
F2 ⇒ G. Then, F1 (resp. F2) would seem to deserve a context for G especially when we
tend to believe falsely that F2 ⇒ G (resp. F1 ⇒ G) while neglecting an implicit fact F1
(resp. F2). It is thus reasonable to characterize contexts in terms of partial grounding: a
context for a fact G is a fact that partly grounds G.
Let us illustrate this thesis with the weight/mass example borrowed from Casanave
[
11
]. To do so, we will use the following list of facts:
•
•
•
•
F1: Weight can be calculated by multiplying mass by gravitational acceleration.
F2: Bob’s mass is 100 kilograms.
F3: The gravitational acceleration on the earth is 9.8 m/s2.
F4: The gravitational acceleration on the moon is 1.62 m/s2.
F5: Bob is on the surface of the earth.
F6: Bob is on the surface of the moon.
G1: Bob’s weight is 980 N.
G2: Bob’s weight is 162 N.
Casanave asserts that the surface of the earth provides a ‘location context’ for the rule
about the convertibility of mass into weight. This statement can be interpreted as saying
that F5 is a context for G1 and hence as the partial grounding F5 → G1, which follows
from the full grounding F1, F2, F3, F5 ⇒ G1. The same argument over location context
would apply to the surface of the moon: F6 is a context for G2 in the sense that the partial
grounding F6 → G2 holds, which follows from the full grounding F1, F2, F4, F6 ⇒ G2.
Grounding would enable us to discuss other kinds of contexts than location ones.
One may attribute G1, for instance, to the historical context in which weight became
fundamentally separate from mass in modern physics. Let F0 be the fact that modern
physics brought about a sharp distinction between weight and mass. To say that F0 is a
(historical) context for G1 amounts to the partial grounding F0 → G1, which follows by
the transitivity of partial grounding from F0 → F1 and F1 → G1. Quite importantly, the
same line of reasoning would not mesh with the truthmaking relation because it fails to
preserve the property of chaining [
18
]. Although meta-ontological, truthmaking may be
too restrictive to serve as a conceptual tool for analyzing contexts.
Let us finally compare our grounding perspective on contexts with Casanave’s [
11
]
mediator view of them. First of all, our schema for contexts is simpler than his because
the former is based on one kind of category (fact) and relation (grounding), whereas the
latter on multiple categories (propositions and things) and relations (the holds-within
relation, contextualization, and implication). Seen from a different standpoint, his model
could possibly reduce to ours when our notion of fact is interpreted as a (true) proposition.
One may worry that, unlike Casanave’s, our approach would offer no room for talk of
objects as contexts: e.g., “Oxygen is a physical context for burning.” In ontological
parlance, however, this claim should be construed more meticulously using
factgrounding: the fact that an object burns is grounded in the fact that oxygen is present
around a flammable object. Additionally, Casanave states that a context can be false (e.g.,
“We are not on the surface of the moon”) and says: “If a context is false, propositions do
not hold for what the context contextualizes.” Instead of allowing contexts to have
truthvalues, we think that his alleged falsity of contexts would be explicable in terms of the
failed (partial) grounding relation. Let F*5 be the fact that Bob is not on the surface of
the earth. To say that F*5 is a ‘false context’ for G1 would be equivalent to the failed
partial grounding of G1 in F*5: i.e., it is not the case that F*5 → G1.
5. Conclusion
To summarize, we proposed a novel ontological approach to contexts by exploiting a
meta-ontological notion of grounding, which would be preferable to the truthmaking
relation for our present purpose. The main finding is that a context for a fact F can be
characterized as a fact that partly grounds F. This work would constitute a further step
from Casanave’s relevant mediator view of contexts.
In the future we will deepen our grounding-based conceptualization and
formalization of contexts, e.g., by deploying existing works on the logic of grounding
[
23-25
]. This will contribute to a computational approach to contexts. It would be also a
noteworthy line of research to apply this work to the development of a previous
grounding analysis of the role concept [
9,10
] since content-dependency is one of the
essential features of role [
26
].