=Paper=
{{Paper
|id=Vol-2969/paper46-FOUST
|storemode=property
|title=Expressions, Utterances and Directive Slots
|pdfUrl=https://ceur-ws.org/Vol-2969/paper46-FOUST.pdf
|volume=Vol-2969
|authors=Adrien Barton,Laure Vieu,Jean-François Éthier
|dblpUrl=https://dblp.org/rec/conf/jowo/BartonVE21
}}
==Expressions, Utterances and Directive Slots==
https://ceur-ws.org/Vol-2969/paper46-FOUST.pdf
Expressions, Utterances and Directive Slots
Adrien Barton 1,2, Laure Vieu1,3 and Jean-François Ethier2
1
Institut de Recherche en Informatique de Toulouse (IRIT), CNRS, France
2
Groupe de recherche interdisciplinaire en informatique de la santé (GRIIS), Université de Sherbrooke, Quebec,
Canada
3
Laboratory for Applied Ontology (LOA), ISTC-CNR, Trento, Italy
Abstract
The same linguistic entity can play a directive or a descriptive role, e.g. “cook dinner” can be
moved from a column “to do” in a Kanban board to a column “done”. We propose to account
for such directive roles by enriching a previous framework of slot mereology for informational
entities with the notion of directive slot. We differentiate expressions from the utterances they
constitute in order to accommodate for the insight from the literature in pragmatics that most
(if not all) expressions can constitute directive utterances in the appropriate context. We
propose axioms involving constitution, slot-having and slot-filling on utterances, and articulate
this theory with former work on directing actions.
Keywords 1
Information content entity, Imperative, Pragmatics, Role, Slot mereology
1. Introduction
Some linguistic entities (abbreviated “LIN-E”) aim at directing an action – a paradigmatic example
being imperative sentences (e.g.: “Eat your vegetables!”). Other LINEs describe an action: either a type
of action (“eating one’s vegetables”), or a particular action that takes place at some moment in time
(“Magalie has eaten her vegetables today.”). Some LIN-Es, however, are ambiguous in that respect.
Consider for example the LIN-E “cook dinner” on a Kanban board. It can instruct me to cook dinner if
placed in a “to do” column, or describe my process of having cooked dinner if placed in a “done”
column. Actually, the very same LIN-E can have those two different roles (directive vs. descriptive) at
different times, if moved from one column to another.
Moreover, sometimes we want to compare not two exactly similar LIN-Es that have different roles,
but two different LIN-Es that have different roles, where one is more specific than the other. Consider
the following example [1]: a doctor writes on a prescription “metoprolol 50 mg bid 1 month” (to instruct
the pharmacist to give to a patient medication containing 50 mg of the active ingredient metoprolol in
enough quantity to take it twice a day for a month), and a pharmacist writes in his records “Apo-
Metoprolol 50 mg – 60 pills” (to register that he distributed to the patient 60 pills of the commercial
drug preparation Apo-Metoprolol, which contains 50 mg of metoprolol by pill). In such a case, we may
want to abstract the directive role from the instruction “metoprolol 50 mg bid 1 month” written by the
doctor; indeed, this would enable to describe the fact that the type of action referred to by the
pharmacist’s utterance (namely the distribution of 60 pills of Apo-Metoprolol 50 mg) is a subtype of
the action type referred to by the doctor’s utterance (namely the distribution of an amount of medication
enabling to take 50 mg of metoprolol twice a day for a month). Thus, it is important to dissociate a
directive role from the content that fills this directive role, in order to compare the contents directly.
Some fields such as Natural Language Processing can aim at disambiguating between directive
and descriptive linguistic entities in context. Here, we are instead interested in a
representational/ontological question, namely: what is the nature of the entities involved in directive
FOUST 2021: 5th Workshop on Foundational Ontology, held at JOWO 2021: Episode VII The Bolzano Summer of Knowledge, September
11-18, 2021, Bolzano, Italy
EMAIL: adrien.barton@irit.fr (A. 1); laure.vieu@irit.fr (A. 2); ethierj@gmail.com (A. 3)
ORCID: 0000-0001-5500-6539 (A. 1); 0000-0003-0303-0531 (A. 2); 0000-0001-9408-0109 (A. 3)
© 2021 Copyright for this paper by its authors.
Use permitted under Creative Commons License Attribution 4.0 International (CC BY 4.0).
CEUR
Wor
Pr
ks
hop
oceedi
ngs
ht
I
tp:
//
ceur
-
SSN1613-
ws
.or
0073
g
CEUR Workshop Proceedings (CEUR-WS.org)
and descriptive utterances? Such an ontological representation is important for work that derive a
database structures from an ontology (see e.g. [2]). Moreover, we want to have systematic
representations not only for supporting the recognition (e.g. by NLP) of the directive or descriptive
nature of an existing linguistic entity, but also so that the author of an utterance can tag it appropriately,
when possible, such as a doctor tagging the statement “metoprolol 50 mg bid 1 month” as directive
when writing it.
For that goal, we will build on former work for representing directive informational entities [3].
More specifically, we will introduce the notion of directive slot, inspired by recent work on a slot
mereology on informational entities [4]. In the following, we will speak of “imperative sentences”
associated with a “directive role”, and “indicative sentences” associated with a “descriptive role”. We
will limit our present investigation to textual informational entities (excluding e.g. pictures, which can
also have directive roles [5]). We will not develop here an ontology of language in the sense defined by
Moltman (“the ontology that a speaker accepts when using a language” [6]), but rather an ontology
describing linguistic entities. Indeed, linguistic entities are, like mental entities, legitimate denizens of
the world, and a fully developed ontology of the world should be able to describe them. Thus, we do
not use here language as a prism to consider the world, but as an object of the world that ontology
should represent.
We present in Section 2 some former work from various disciplines about informational entities,
linguistic entities and imperative sentences. In Section 3, we provide a linguistic and philosophical
analysis of the entities that are relevant for the issue at hand. Section 4 proposes a few formal axioms
relating them. A discussion and a conclusion follow.
2. State of the Art
2.1. The Ontology of Informational Entities
Ontologically, directive LIN-Es can be analyzed as informational entities. The Information Artifact
Ontology (IAO) [7] introduces “Information content entities” (abbreviated “ICE”): an ICE can appear
at several places in the physical world, as captured by the relation of concretization between an ICE
and another entity – e.g. a quality such as the shape of an ink pattern or a hard drive configuration, or a
process such as a Morse code signal (for a broader overview of informational entities in ontology, see
[8]). Being part of the OBO Foundry [9], IAO is especially relevant for applications to biomedicine,
such as deriving a database structure from an ontology in order to deal with cases like the metoprolol
example mentioned above. Thus, the ontology we will present here will be largely built as a possible
extension of IAO, but it would certainly be adaptable to other mainstream upper ontologies.
Informational entities are strongly related to the mental states that created them. A specificity of
mental states is that they can carry an “intentionality” or “aboutness”, where this term is taken as “the
power of minds and mental states to be about, to represent, or to stand for, things, properties and states
of affairs.” [10] (in this sense, this refers to a mental attitude different from the “intention to act” that
appears in e.g. belief-desire-intention theories). Informational entities are generally considered as also
carrying an intentionality. IAO endorses Chisholm’s view of “primacy of intentional” [11] according
to which ICEs (and in particular some LIN-Es, for which “intentionality” is often called “reference” in
the semantics field) inherit their intentionality from the intentionality of the ICE’s creator’s mind (see
also the view defended by Haugeland, Searle and Fodor according to which a sentence’s intentionality
is derivative and dependent [10]). Biccheri et al. [12] propose a double view of aboutness as having a
content (that can be expressed by intensional statements, with an “s”) and as being about some
intentional object; those are the respective counterparts of Frege’s sense and reference.
An axiomatic system for a mereology of informational entities inspired by Bennett’s [13] slot
mereology has been proposed [4][14]2. It rests on a distinction between information fillers (IF) and
informational slots (IS): an IF is a proper part of another IF if and only if it fills (relation F) an IS that
is a slot of (relation S) this IF. This enables the same IF to be part several times of another IF, by filling
2
The main differences between this system and Bennett’s is that the former does not accept the notion of improper slot, but allows slots to
have slots, and to be unfilled (to account e.g. for templates).
several slots. For example, consider the string ‘aa’ that has the same letter ‘a’ twice over: in this
framework, this string has two slots, that are both filled by the same information filler ‘a’.
In particular, this theory accepts the following axiom (that will be re-used later):
(AX9) Slot of a filler is filled IF(x) & S(s,x) → ∃y F(y,s)
Former work suggests that slots might not only represent parthood relations, but could also carry
intentionality [14]. This idea will be here explored in the case of directivity, by introducing the notion
of directive slots that will stand for directive roles.
2.2. Speech-Acts, Propositions and Attitudinal Objects
As explained by Moltmann [6], in the standard view in philosophy and natural language semantics,
propositional attitudes and illocutionary acts can be associated with two kinds of entities. The first kind
encompasses mental acts and speech-acts, which we can classify as subkinds of processes. The second
one encompasses propositions, that are the contents towards which propositional attitudes or
illocutionary acts are directed, and whose ontological nature is more complex than the processual nature
of mental acts and speech-acts.
Moltmann argues that we need another kind of entity to account for truth conditions and satisfaction
conditions, which she calls “attitudinal object”. Contrarily to propositions, attitudinal objects’
representational power can be attributed to the intentionality of the mind itself; they have a limited
temporal life span and they are concrete in the sense that they are localized in space-time (even if they
are not material entities). This is a very large category, as it encompasses, among others, judgments,
claims, beliefs, decisions, desires and intentions.
According to Moltmann, attitudinal objects, rather than propositions, are the primary truthbearers
(and propositions would play that role at most derivatively). In this paper, we will not take any stance
on the reality of propositions, but will eschew the category of proposition, and instead introduce a kind
of linguistic entities that share many characteristics with attitudinal objects, namely “utterance”, which
are historically dependent on an author. Before that, we will give a quick overview of the analysis of
imperative sentences and directive linguistic entities in linguistics, philosophy, logic and ontology.
2.3. Imperative Sentences and Directive Linguistic Entities
Various works in ontology have considered the question of directivity by analyzing the directed
processes themselves (e.g. PSL or BPMN – see [3] for a short review). However, it is important to also
integrate in the ontological picture the informational entities directing such processes and their
characteristics (e.g. their author, time of creation, parts, etc.) This is especially important for ontologies
studying documents, e.g. clinical documents [1,15]. A first step in that direction has been proposed [3],
which introduces directive ICEs (DICEs) and the relation “attempting to direct” (a-direct) between a
DICE and an action, defined as: “1) the cognitive system of the agent representing in some way this
DICE […]; 2) this representation leading to an intention to perform an action as described; 3) and this
intention leading to a subsequent action”. This is completed by additional relations such as “successfully
directs”. However, DICEs are intrinsically directive, and thus this previous work does not account for
the fact that the same LIN-E can aim at either describing an aspect of the world or at directing actions,
depending on its context, as illustrated by the above Kanban board and drug prescription examples.
Since the imperative nature of a LIN-E can depend on the context, it might be considered as a role;
indeed, dependence to context is arguably a defining feature of roles [16,17]. Although there is a rich
literature on roles in ontology (e.g. [16–18]), the topic of roles of informational entities has not been
much investigated to our knowledge in applied ontology.
In linguistics and philosophy, two relevant fields of investigation can be distinguished: the study
of imperative sentences, that is, sentences in the imperative mood; and the study of linguistic entities
which have a directive illocutionary force (which can be sentences – but do not need to).
Concerning the first field, several investigations [19–21] have analyzed the logic of how
imperative sentences can be combined to perform inferences. Jørgensen [19] identifies two factors of
an imperative sentence: the imperative factor, “indicating that something is commanded or wished” and
the indicative factor, “describing what it is that is commanded or wished.” He associates to an
imperative sentence (e.g. “Shut the front door.”) a corresponding indicative sentence (e.g. “The front
door is shut.”). Although we will not analyze here the logic of inferences involving imperatives, it will
be important in our analysis to connect imperative LIN-Es with their indicative counterparts.
There is a rich literature in linguistics analyzing imperative sentences. Katz & Postal [22] introduce
an “imperative morpheme” that can distinguish a sentence in the imperative mood from a sentence in
the indicative mood (see also [23,24]). Downes [25] provides further pragmatic analyses of imperatives.
Works in philosophy of language have been interested in the meaning of imperative sentences
[26], with a tradition linking imperatives and modals (e.g. [27]). An important subclass of imperatives
expresses an obligation, e.g. “Eat your vegetables!”. However, some may express other deontic
attitudes, such as permission (“Go ahead, take the day off”) or recommendation (“If you feel bad, take
aspirin”). Moreover, some imperative LIN-Es do not express deontics. Some can be bouletic, e.g.
expressing what needs to be done conditionally on a desire of the addressee (“Have a piece of fruit!”)
or teleological, e.g. expressing what needs to be done conditionally on a goal of the addressee (“To get
to Union square, take Broadway”). Other can express a well-wish (“Get well soon!” [26]) or a threat
(“Go on, throw it. Just you dare.” [28]) (see [26] for a more exhaustive list). Although the last two uses
might lie outside our present scope of interest, it is important to keep in mind that the connection
between imperative sentences and deontics is neither straightforward nor systematic. While deontics
are related to the object of our work, they are not essential to structure the proposed approach; therefore,
we will not analyze them further in this paper, but will briefly return to this question in the conclusion.
As explained above, LIN-Es that can have a directive illocutionary force are not limited to
imperative sentences – and not even to sentences. Searle’s [29] taxonomy of illocutionary acts includes
directive acts in which “[t]he speaker tries to get the hearer to act in such a way as to fulfill what is
represented by the propositional content” [30]. Stainton [31] explains how a speech-act can consist not
only in uttering a sentence, but also in uttering merely a word or phrase, and how this can carry an
illocutionary force, such as a directive force. The way directive utterances fit in a discourse is analyzed
in Segmented Discourse Relation Theory (SDRT) [32].
Note that many of the LIN-E classically analyzed in linguistics pertain to spoken discourse or
immutable text, which cannot be moved from one context to another. Thus, although a LIN-E can be
directive or non-directive, in such cases it cannot “lose” or “gain” its directive or non-directive status.
On the contrary, electronic documents offer the possibility to move or duplicate text, and thus to change
the context of a LIN-E. Therefore, we need a theory that can represent the different directive or non-
directive roles that can take a same LIN-E. This is essential to build a database in which a complex
document workflow can be accurately handled, such as a hospital database.
3. Linguistic and Philosophical Analysis
We will look here for an ontological representation of the various informational entities that are
involved in scenarios involving both directive and descriptive LIN-Es, in order to support the
construction of non-ambiguous databases. Consider for example the following clinical scenario, called
“NITRO”. Dr. House writes “nitro” on an electronic drug administration order for a patient Mr. Almásy,
with the aim of instructing a nurse to administer him nitroglycerine. Later, Nurse Ratched copies this
LIN-E “nitro” to a drug administration report, in order to register that she has administered nitro to
Mr. Almásy. How can we capture, in scenario NITRO, what is common and what is different between
the (directive) instruction “Nitro” written by Dr. House and the (descriptive) informational entity
“Nitro” copied by Nurse Ratched in a drug administration report? To account for such a scenario, we
will first distinguish expressions from utterances, then distinguish directive utterances from action
utterances and finally introduce directive slots that will stand for directive roles.
3.1. Expressions and Utterances
We will use here the term “utterance” to refer to enduring entities that are the informational outputs
of some speech-act or mental act3: whereas a speech-act is an occurrent/perdurant, an utterance is here
understood as being a continuant/endurant. Utterances are thus somewhat similar to Kaplan’s [36]
“occurrences” or “sentences-in-context”; however, following IAO, they are not considered as abstract
objects (e.g. propositions), but as informational entities that are created at a given time and are
concretized in the physical world (typically by physical marks or sounds).
Utterances have many commonalities with Moltmann’s category of attitudinal object: their ability
to represent can be attributed to the intentionality of the mind itself, they have a limited temporal life
span and they are concrete (in the sense that as informational entities, they are concretized by some
material features of the world, and cannot exist without such a concretization – although they are
different from those concretizations, and are not themselves material entities). Attitudinal object is a
larger category than our category of utterance though, as it also includes non-linguistic entities.
Apart from utterances, we can introduce another LIN-E named “expressions”. Expressions are
sequences of words (and punctuation signs) on which a syntactic and semantic analysis can be
performed, be they full sentences or smaller linguistic entities (syntagms) like a verb or noun phrase.
An utterance is constituted by an expression (following a suggestion by [37]): intuitively, the expression
constituting an utterance is what remains of this utterance when stripped from all its contextual features
provided by the speech-act that created it. Consider for example three occurrences of the word “nitro”:
one created by Dr. House for Mr. Almásy, another one created by Dr. Zhivago for another patient, and
the last one in a medical dictionary. Those are three different utterances as they have different authors,
have been created at different times, and have different illocutionary forces (the two former aim at
directing actions, although for different patients, whereas the third utterance merely aims at describing
a type of medication); however, they are all constituted by the same expression “nitro”.
Thus, an utterance may have an intentionality beyond or different from the intentionality of the
expression that constitutes it (see Section 5.2 for more considerations on aboutness), and influenced by
the context of utterance (where a context might be analyzed, following Kaplan [36], as composed by
one or several agents, a location, a time and a world, and extended, following SDRT [38], with the
discourse context of the previous utterances). Thus, their study exceeds semantics and involves the field
of pragmatics. On the other hand, an expression constituting an utterance typically pre-exists to this
utterance, as it pre-exists to the speech-act that gave rise to it (and can also constitute other utterances,
uttered by other agents at other times).
We can start introducing the following entities to account for the scenario NITRO. First, let’s call
sa1 the speech-act in which Dr. House prescribed nitro to be administered to Mr. Almásy, which creates
the utterance ut1 (note that sa1 is a bona fide speech-act even if it only consists in the utterance of one
word, as explained by Stainton [31]). We can introduce the word w0 = ‘nitro’ (that is present on this
prescription, but also on many other clinical documents), which is an expression, and could be analyzed
as referring to the type of drug that contain nitro. ut1 is constituted by w0, but it aims to direct a nurse
to administer nitro to Mr. Almásy; that is, ut1 is synonymous4 with ‘Administer nitro!’, whereas w0 is
synonymous with ‘drug type nitro’.
3.2. Action Utterance
We need now to explain what happens when Nurse Ratched copies the instruction “Nitro” to a drug
administration report, creating a new utterance ut2. Something from the original utterance ut1 is retained
while doing so: indeed, both ut1 (created by Dr. House) and ut2 (created by Nurse Ratched) have
3
Note that there are several possible definitions of a speech-act. Some are restrictive (e.g. : “As a first approximation, speech acts are those
acts that can (though need not) be performed by saying that one is doing so.” [33]) and identify speech acts with illocutionary acts [34]. Some
are broader, and encompass all four of the utterance acts, the locutionary acts, the illocutionary acts and the perlocutionary acts [35]. The
definition we endorse here is of the latter type. Note also that sometimes, the term “utterance” is used as a synonym of what we call here a
“speech-act”; but here, “utterance” will strictly refer to an informational entity, never to a process. We will also not enter into the question of
whether all utterances are created by a speech-act, or if some can be created by a mental act that just precedes the speech-act; for simplicity,
we will consider for now that they are the output of a speech-act, and come back to this question in Section 3.5.
4
The notion of synonymy will only be used here at the pre-formal level, to provide an intuitive grasp of the different relevant entities. For
more thoughts on synonymy of directive instructions, see [3].
something in common, namely being about the action type of taking nitro (and not merely referring to
the drug nitro, like the expression “nitro” does). This referential feature of ut1, which is expressed by
the pragmatics of communication, is retained by ut2.
To account for this, we can identify another LIN-E named “ut0” that is an intermediate between w0
and ut1. ut0 also reads “nitro”, and thus is also constituted by w0; however, it refers to the action type of
administering nitro, without any intrinsic directive intention. Note that ut0 has a meaning that exceeds
the meaning of the expression it is constituted by (as w0 does not refer to the action type of administering
nitro, but merely to the drug type nitro). Since ut0 has this meaning because of contextual factors, it
belongs to the level of utterances, rather than to the level of expression: ut0 reads “nitro”, but is
synonymous with “administering nitro”. We will call ut0 an “action utterance”, namely an utterance that
refers to a type of action.
3.3. Directive Utterances and Directive Slots
When Dr. House writes “nitro” in scenario NITRO, he thus creates two utterances ut0 and ut1. What
is the connection between ut0 and ut1? ut0 plays a directive role because of external factors, and it is not
intrinsically directive. In a sense, ut1 can be considered as ut0-qua-directive (see [34] for the notion of
qua-objects). To account for such a role, we introduce the notion of “directive slot”. More precisely, we
postulate that ut0 fills a directive slot s1 of ut1. We adopt thus a two-tier representation of directive
utterances (which will be systematized in Section 4) with the “action-utterances” at one level (ut0) that
are intrinsically merely about a type of action, and the “directive utterances” at another level (ut1) that
aim to direct an action (that is, they have a directive illocutionary force). Both ut0 and ut1 have been
created by the same speech-act sa1, and they are both constituted by the same word w0.
However, as explained above, ut0 and ut1 can be dissociated from each other. Indeed, consider the
speech-act sa2 in which nurse Ratched copies what Dr. House wrote in a database to indicate that nitro
has been administered to Mr. Almásy. sa2 created the utterance ut2, where ut2 is synonymous to “nitro
has been administered”, and is constituted, like ut0 and ut1, by w0. The speech-act sa2 has copied ut0,
but not ut1, since ut2 has no directive intention; instead, ut0 has been incorporated into a new layer of
meaning to form a new utterance ut2. This new layer of meaning will also be represented by a slot,
which we will call s2. Contrarily to s1, s2 does not provide a directive role, but an assertive role; thus,
we will call it a “non-directive slot” (see 5.2.2 for more discussion of the various sub-kinds of non-
directive slots). This explains why it is important to represent this intermediate utterance level ut0,
whose reference is influenced by contextual features (like ut2), but that can be copied or moved from a
directive context to a non-directive one.
Figure 1: Expression, utterances and directive-slot
Let’s come back to the example of the two directive utterances of the word “nitro” created
respectively by Dr. House and by Dr. Zhivago. They each have one action-utterance (the former created
by Dr. House, the latter by Dr. Zhivago). And all those four utterances are constituted by the same
expression “nitro” (which does not have any illocutionary directive force, and does not refer to a class
of actions, but merely refers to the class of nitro drug). Thus, not only action-utterances, but also
expressions are important to capture important aspects of what several directive utterances can have in
common.
3.4. Taxonomy of Relevant Entities
We will use in this paper the taxonomy as pictured on Figure 2.
Entity
Information Entity [IE]
Linguistic entity [LIN-E]
Utterance [UT]
Action-utterance [ACT-UT]
Directive utterance [DIR-UT]
Utterance-slot [UT-S]
Directive slot [DIR-S]
Non-directive slot
Expression [EXP]
Process
Speech-act
Action
Figure 2: Taxonomy of entities and associated unary predicates
In this paper, we will work in FOL (and assume all free variables are universally quantified when
writing axioms and theorems). All the axioms reflecting the taxonomic structure in Figure 2 will be
accepted, that is, if A is a subclass of B, we accept A(x) → B(x). Moreover, we postulate that UT, EXP
and UT-S are all disjoint from each other (so we have e.g. ¬(UT(x) ∧ EXP(x))), and Non-directive slot
is defined as the complement of Directive slot in Utterance-slot.
We introduce the following Aristotelian [39] definitions5:
• Expression: A linguistic entity that is composed by a sequence of words and/or punctuation
signs admissible in a given language (a single word is also an expression).
• Utterance: A linguistic entity that is created by one or several agents in a given context and
constituted by an expression.
• Directive utterance: An utterance that has a directive illocutionary force.
• Action-utterance: An utterance that is about one or several types6 of actions and has no
directive illocutionary force.
• Utterance-slot: A linguistic entity that can only be filled by an utterance.
• Directive slot: An utterance-slot that provides a directive role to the utterance that fills it.
Note that some expressions, similarly to action-utterances, may also denote a type of action, e.g. the
word “eat” (however, no expression is an action-utterance, since EXP and UT are disjoint; expressions
can only constitute utterances, including action-utterances). However, the directive slots that are defined
in this paper appear strictly at the level of utterances, and thus cannot be filled by mere expressions.
Note also that some action-utterances can describe an action by specifying the agent who is supposed
to perform it, the mean used, or other aspects. For example, the action-utterance “meal preparation by
John”, in the right context, is about the type of meal’s preparation by a specific agent, namely John. Let
us now clarify what directive slots stand for.
3.5. Directive Slots Stand for Directive Intentions
We can distinguish two questions concerning the directive (or non-directive) role of an utterance
like ut0 above: first, what provides to an utterance a directive (or non-directive) role; second, how this
role is expressed – that is, the syntactic, semantic or pragmatic features that ensure that the directive (or
non-directive) role of an utterance can be recognized as such by another agent.
5
See Section 4 for more details on the connection between directive utterances and directive slots, and see Section 5.1 for a discussion of the
nature of utterances.
6
This type could be a universal, or maybe a concept, following Biccheri et al.’s [12] suggestion that this would better account for the
intensional character of aboutness. See Section 5.2 for more discussion.
In the spirit of Chisholm’s primacy of intentionality and IAO’s stance (cf. 2.1), we consider that the
directive role of utterances derives from the intentionality of the mind of the agent who uttered them –
which belongs to the context of the speech-act (contrarily to expressions, whose intentionality does not
depend on the agent using them). Thus, the directive nature of directive slot ultimately stands for such
intentions7. However, an utterance that is directive (because of the intention of its creator) but could not
be recognized as such by anyone else would be linguistically inefficient. Therefore, this directive role
can be publicly expressed in various ways. It could be manifested by syntactic or semantic features of
this utterance, e.g. an imperative mood tense combined with the exclamation mark in ‘Eat your
vegetables!’. It could also be expressed by syntactic or semantic features of other utterances that form
the discourse context of this utterance, e.g. the header “to do” of a column in a Kanban board, or
previous discourse. Finally, it can be expressed by features of the speech-act creating this utterance.
Analyses in pragmatics have indeed shown famous examples of utterances that look superficially
descriptive but can be used with directive purposes – e.g. “This meal is not salted enough.” uttered to
request someone to pass the salt. Note however that the speech-act creating an utterance in an electronic
document is often not perceptible to a reader of this document, which, among other complexities of
pragmatic phenomena, may create significant difficulties for NLP analyses of such documents, and
motivates a system of tagging of the directive or non-directive nature of utterances by their authors.
Epistemically, the extrinsic features have precedence over the intrinsic ones to recognize the
directive nature of an utterance. Indeed, Downes [25] notices that “Practically any sentence can be said
to have the illocutionary potential of commanding in some context.” This means that intrinsic syntactic
or semantic features can at best suggest a directive reading of this utterance, but this (defeasible)
inference can be defeated when considering pragmatic features (Stainton [31] relatedly distinguishes
between the force of an expression and the force of an action, where the latter has precedence). Here,
we are not interested in analyzing which syntactic, semantic or pragmatic features can enable to
recognize the directive nature of an utterance. Instead, we are interested in representing ontologically
the directive nature of utterances. As explained earlier, we identify such directive roles with directive
slots. That is, directive slots encapsulate directive intentions of the creator of the information, and those
intentions might be made visible by syntactic, semantic and/or pragmatic features of the utterance and
its context. Said differently, the directive slot is the reification of the directive illocutionary force
provided by the intention of the author and that can be linguistically expressed by various means.
The informational slot-mereology we build upon ([4,14]) considers that slots can exist even if
unfilled, and this holds in particular for directive slots: consider for example a calendar on which a
manager can write what an employee is expected to do at each hour (but can leave some hour-slots
unfilled); this calendar has several directive slots, which exist independently on whether they are filled
or not – and are actually all unfilled initially.
Now that we have clarified the pre-formal intuitions behind the core notions involved in this paper,
we can propose some elements of a formal theory.
4. Formal Analysis
This section will introduce the binary relations summarized in Table 1, where R has domain X and
range Y means that we accept the axiom R(x,y) → X(x) ∧ Y(y).
Table 1
Binary relations
Binary relation Predicate Domain Range
utterance-slot-of SUT UT-S UT ∨ UT-S
utterance-fills FUT UT UT-S
constitutes CONST EXP UT
7
Another possibility, that will not be discussed here, would be that the aboutness derives from the public commitment of the utterance.
4.1. Constitution
Let’s start by axiomatizing the relation of constitution between an expression and an utterance. We
postulate that every utterance is constituted by exactly one expression:
(AX1DS) UT(x) → ∃!e (EXP(e) ∧ CONST(e,x))
We can trivially deduce from AX1DS that CONST is inverse functional:
(TH1DS) CONST(e,x) ∧ CONST(f,x) → e=f
In our framework, the same expression can constitute several utterances that are uttered by different
agents and in different contexts. For example, the same word w0 = ‘nitro’ can constitute an utterance ut1
by Dr. House prescribing Mr. Almásy to take this medication, as well as another utterance ut3 by the
pharmacist Mr. White stating that he distributed this drug to Mr. Almásy. Thus, CONST is not assumed
to be functional: we can have CONST(w0,ut1) and CONST(w0,ut3) with ut1≠ut3.
4.2. Mereology and Slots
We accept a mereology on utterances based on former works ([4][14]). More precisely, the unary
predicates IF (information filler) and IS (information slot) as well as the binary predicates F (fills) and
S (slot-of) from [4] are here respectively transformed as predicates UT, UT-S, FUT and SUT. In
particular, axiom AX9 reminded in Section 2.1 can be translated here as:
(AX9UT) UT(x) ∧ SUT(s,x) → ∃y FUT(y,s)
We postulate the additional axiom that a directive slot can only be filled by an action-utterance:
(AX2DS) FUT(x,s) ∧ DIR-S(s) → ACT-UT(x)
We could want to impose an axiom stating that the mereological sum of several utterances is
constituted by the mereological sum of the expressions constituting those utterances; but this would
require a theory of mereological sum on informational entities, something that has yet to be developed
(see [4] for some pointers).
4.3. Directing Actions, Revisited
The directive ICEs (DICEs) from [3] correspond to what we called here “directive utterances”, and
we take DIR-UT as a primitive. We then impose that a directive utterance must have a directive-slot:
(AX3DS) DIR-UT(x) → ∃s DIR-S(s) ∧ SUT(s,x)
We refrain however from the converse assumption. For example, the utterance “John said ‘Eat your
vegetables!’” has a directive slot, but it is not a directive utterance (it is rather a descriptive utterance
about John having uttered a directive utterance).
We deduce from the former axioms the following theorem stating that every directive utterance has
a directive slot filled by an action-utterance:
(TH2DS) DIR-UT(x) → ∃s,y DIR-S(s) ∧ SUT(s,x) ∧ FUT(y,s) ∧ ACT-UT(y)
Proof: Suppose that DIR-UT(x). By AX3DS, there is an s such that DIR-S(s) and SUT(s,x). By
AX9UT, there is a y such that FUT(y,s). By AX2DS, we have ACT-UT(y). QED.
In scenario NITRO, the directive utterance ut1 has a directive slot filled by ut0, which is an action
utterance. On top of this, both ut1 and ut0 are constituted by the same expression. However, the latter
feature does not need to be the case. For example, the directive utterance ‘Eat vegetables!’ has a
directive slot filled by the action-utterance ‘Eat vegetables’, that does not encompass the exclamation
mark; both utterances are thus not constituted by the same expression (the expression constituting the
action-utterance is a part of the expression constituting the directive utterance though, see Section 5.1
below).
Then, the relation a-directs from [3] (as reminded in Section 2.3) will be redefined between a DIR-
UT and an action. Further work should address how to characterize formally this relation in terms of
more basic relations. In particular, step 3) in its definition (cf. Section 2.3) could be represented using
an ontological analysis of intentions as dispositions [40]: an intention leading to a subsequent action is
nothing else than the intention being realized (in the dispositional sense) by this action.
Coming back to scenario NITRO as represented on Figure 1, we have the following facts:
DIR-UT(u1), UT(u2), ACT-UT(ut0), SUT(s1,u1), SUT(s2,u2), FUT(u0,s1), FUT(u0,s2), DIR-S(s1), EXP(w0),
CONST(w0,ut0), CONST(w0,ut1), CONST(w0,ut2). This provides trivially a model of the theory
presented here, proving thus its consistency.
5. Discussion
Our framework raised three main theoretical challenges: first, representing directive informational
entities, including directive utterances, their directive slots and their action-utterances; second,
representing the relation between an action-utterance and what it is about; third, representing the
relation between a directive informational entity and an action that it directs. Here, we have addressed
the first of those challenges and adapted former work [3] to partially address the third one. As we will
see now, some open issues remain concerning each of those three challenges, namely respectively 1) the
structure of expressions and utterances, 2) aboutness and 3) directing not only actions but also
objectives.
5.1. The Structure of Expressions and Utterances
We could build a mereology of expressions on the same basis (namely [3]) as the one we presented
above to build a mereology of utterances. We could then cross-constrain both mereologies. For
example, if an utterance is part of another utterance, then the expression constituting the former is a part
of the expression constituting the latter. But the converse is not true: if Mrs. A utters ut7 = ‘eat’ to
express (in simplified English) that she has eaten today and Mr. B utters ut8 = ‘eat your vegetables!’ to
direct his son to eat his vegetables, then the word w7 = ‘eat’ constitutes ut7, the expression w8 = ‘eat
your vegetables!’ constitutes ut8, w7 is a part of w8, but ut7 is not a part of ut8.
We may want stronger constraints. Indeed, if an utterance x constituted by e contains twice the
utterance y constituted by f (e.g. in scenario NITRO, ut0 filling two different slots of a document
gathering ut1 and ut2), we may want e to have f twice as part. A slot-mereology inspired by Bennett
[4,13] may help in this regard, although it should be amended to enable to count parts adequately [41].
Relations between slots would also need to be added, such as relations of sequential order (slot s is
before or after slot t, to account for the fact that in a text, an utterance is before or after another). The
intended meaning of such relations (e.g. causality, explanation, temporal succession, etc.) might then
be analyzed by a discourse theory, e.g. SDRT [38]. Here, we chose an expressivity level that makes the
theory computationally usable (to structure databases according to this ontology), therefore SDRT
cannot be directly expressed here; however, directive slots might be mapped to SDRT’s “speech-act
labels” (πi) that are linked by rhetorical relations to constitute discourse structure.
We would also need a definition of utterances involving necessary and sufficient conditions (the
definition in 3.1 provided only necessary conditions), in order to account for the fact that in the utterance
‘Eat your meal’, there is no utterance formed by the word ‘your’. This question is related to the debate
about elementary discourse units and fragments in discourse theories [31,42] and would require a more
thorough ontological analysis of speech-acts and their outputs.
5.2. Aboutness
Various questions concerning aboutness are left open in the current work, including specification of
an addressee of an utterance, of the time at which the action specified by a directive utterance needs to
be done, etc. Here, our goal was not to reflect ontologically or logically the internal discursive structure
of statements, a question that we keep for future works. Let us say however a few words about both the
intensionality of directive utterances and the aboutness and illocutionary force of non-directive slots.
5.2.1. The Intensionality of Directive Utterances
Directiveness and aboutness are open to similar challenges. For example, from the fact that the
expressions “evening star” and “morning star” have different intensions despite being co-extensional
in our actual world, the directive utterances “Look at the evening star!” and “Look at the morning star!”
can direct actions differently. This is one of the motivations for distinguishing a directive utterance from
its action-utterance (e.g. in scenario NITRO, distinguishing the directive utterance ut1 from the action-
utterance ut0), where the aboutness of the action utterance might be represented following the lines of
the dual treatment of aboutness proposed by Biccheri et al. [12], and explain how the two corresponding
directive utterances might direct differently. We might thus be able to account for the intensional (with
an “s”) character of a directive utterance as a consequence of the intensional character of its action-
utterance. Since action-utterances are not about a particular action, but about a type of action,
representing their aboutness may give rise to some formal challenges, such as reifying action subtypes.
5.2.2. The Aboutness and Illocutionary Force of Non-directive Slots
As explained earlier, ‘cook dinner’ is about the type of action of cooking dinner, but it also has
the role of referring to a particular dinner cooking action when written in the column “done” of a Kanban
board. That is, although an action-utterance is about a type of action or state of affair, it can have the
role of being about a particular action or state of affair. This (non-directive, assertive) role can then be
represented by a (non-directive) slot. For example, in scenario NITRO, s2 represents a non-directive
role of ut0 to refer to an administration of nitro to Mr. Almásy (and leading to the apparition of utterance
ut2). This slot acts to some extent similarly to an indefinite article (e.g. “a” in English).
We could imagine a large variety of non-directive slots that would add a semantic layer to the
action utterance that fills them (note however that not all slots need to be about something: some might
merely represent a mereological structure). For example, an action-utterance could have the role of
referring to a failed attempt to perform an action of the kind described by the action-utterance if placed
in the column “failed” of a Kanban board. More generally, it could have the role of referring to the fact
that an instruction has been cancelled; that an instruction is permitted, obligatory or recommended; that
the action has been performed by a specific agent, or under a specific mode; etc. We could also imagine
that some slots add both a directive component and an added layer of semantics at the same time.
However, as explained above with the examples of the Kanban board and the clinical scenario NITRO,
keeping those two layers separated enables to represent how an action-utterance (e.g. ut0) can be copied
in another context while the whole directive utterance (e.g. ut1) is not, even if both are constituted by
the same expression. Finally, note that non-directive slots would also include a variety of slots with
non-directive illocutionary force (assertive, commissive, expressive, declarative, etc.).
5.3. Objectives vs. Actions
The Ontology for Biomedical Investigations (OBI) [43] identifies two different kinds of directive
informational entities: objective specifications and action specifications. Note however that the
distinction between both kinds is not always easy to establish. Consider the instruction ‘STOP’ on a
road sign: is it an objective specification that states that one needs to reach a state of being stopped, or
an action specification that states that one needs to perform the action of stopping?8 Similarly, consider
a medical instruction on a drug prescription that reads ‘Amoxicilin 500 mg’; does it specify an objective
of 500 mg Amoxicilin ending up in the body of the patient, or an action of the patient taking 500 mg of
Amoxicilin? Thus, the category “action-utterance” might have to be replaced by the category
“objective- or action-utterance”. This would require however to clarify the ontology of actions and the
ontology of objectives, and how they relate [44].
8
We are indebted to Alan Ruttenberg for exposing this problem to one of us in a private discussion.
6. Conclusion
We have differentiated expressions from the utterances they constitute, and represented the
directive roles that can take some utterances as directive slots, where this role is provided by the
intention of the creator and can be expressed by syntactic, semantic or pragmatic features of the
utterance or its context. Directive utterances (which can direct actions) are analyzed as having a
directive slot filled by an action-utterance. We can add that this action utterance corresponds roughly
to what is commonly described as being the content of this directive utterance (but contrarily to the
orthodox view, this content is itself an utterance, since its aboutness is determined by pragmatic
features), whereas the directive slot describes its illocutionary directive force. Note that we took the
example of an action-utterance constituted by one word (“Nitro”), but this could be generalized to any
information from which an action description might be generated in the context of discourse.
This action utterance can be moved to other contexts, including non-directive contexts. Thus, this
ontological framework can account for electronic documents (e.g. clinical documents) in which
linguistic entities can be moved around or copied and take either directive or non-directive roles.
Further work should address additional issues concerning the connection between the mereology
on expressions and the mereology on utterances, the distinction between utterances that have an
illocutionary force and those (like action-utterances) that do not, the different kinds of relations that can
hold between slots of utterances, the aboutness of both directive utterances and non-directive slots, and
how the fulfillment of an objective can be directed.
Other possible extensions include analyzing how ambiguous utterances could be misinterpreted,
e.g. misinterpreting utterances that are non-directive as being directive, and vice-versa. For example,
an utterance “nitro” that is intended to instruct to administrate nitro to a patient can be misunderstood
as an utterance stating that nitro has already been administered to this patient. Representing such
misinterpretations is especially important for supporting error-checking and audit process. Note that
structuring unambiguously an information system using the entities introduced here could prevent
subsequent misinterpretations.
A variety of subtypes of directive slots could also be introduced. One could for example adapt
Portner’s work [27] to slots by introducing notions such as deontic slots (particularly important for
documents such as drug prescriptions, cf. Section 4.2 in [45]), bouletic slots and teleological slots, and
providing a modal analysis of those entities. Finally, the current ontology could be integrated into a
more general ontological investigation of attitudinal objects, that could shed some lights on the ontology
of beliefs, desires and intentions [40].
References
[1] A. Barton, P. Fabry, J.-F. Ethier, A classification of instructions in drug prescriptions and
pharmacist documents, in: A. Diehl, W.D. Duncan, G. Sanso (Eds.), Proceedings of the 10th
International Conference on Biomedical Ontology (ICBO 2019), CEUR-WS.org, vol. 2931,
Buffalo, New York, USA, 2021, pp. G.1-7.
[2] J.-F. Ethier, M. McGilchrist, A. Barton, A.-M. Cloutier, V. Curcin, B.C. Delaney, et al., The
TRANSFoRm project: Experience and lessons learned regarding functional and interoperability
requirements to support primary care, Learning Health Systems 2 (2018) e10037.
[3] A. Barton, L. Vieu, J.-F. Ethier, Directing Actions, in: A. Barton, S. Seppälä, D. Porello (Eds.),
Proceedings of the Joint Ontology Workshops 2019 (JOWO 2019), CEUR-WS.org, Graz,
Autriche, 2019, pp. 1–10.
[4] A. Barton, F. Toyoshima, L. Vieu, P. Fabry, J.-F. Ethier, The Mereological Structure of
Informational Entities, in: B. Broadaric, F. Neuhaus (Eds.), Formal Ontology in Information
Systems. Proceedings of the 11th International Conference (FOIS 2020), IOS Press, Bolzano,
Italy, 2020, pp. 201–215.
[5] D.M. Lopes, Directive pictures, The Journal of Aesthetics and Art Criticism 62 (2004) 189–196.
[6] F. Moltmann, Attitudinal objects: their ontology and importance for philosophy and natural
language semantics, in: B. Ball, C. Schuringa (Eds.), The Act and Object of Judgment,
Routledge, 2019, pp. 180–201.
[7] B. Smith, W. Ceusters, Aboutness: Towards foundations for the information artifact ontology, in:
Proceedings of the 6th International Conference on Biomedical Ontology, CEUR-WS.org,
Lisbon, Portugal, 2015, pp. 1–5.
[8] E.M. Sanfilippo, Ontologies for information entities: State of the art and open challenges,
Applied Ontology 16 (2021) 1–25.
[9] B. Smith, M. Ashburner, C. Rosse, J. Bard, W. Bug, W. Ceusters, et al., The OBO Foundry:
coordinated evolution of ontologies to support biomedical data integration, Nature
Biotechnology 25 (2007) 1251–1255.
[10] P. Jacob, Intentionality, The Stanford Encyclopedia of Philosophy, Winter 2019 edition,
Metaphysics Research Lab, Stanford University (2019).
[11] R.M. Chisholm, The primacy of the intentional, Synthese 61 (1984) 89–109.
[12] L. Biccheri, R. Ferrario, D. Porello, Needs and Intentionality, in: Formal Ontology in
Information Systems. Proceedings of the 11th International Conference (FOIS 2020), IOS Press,
Bolzano, Italy, 2020, pp. 125–139.
[13] K. Bennett, Having a part twice over, Australasian Journal of Philosophy 91 (2013) 83–103.
[14] A. Barton, F. Toyoshima, J.-F. Ethier, Clinical Documents and Their Parts, in: J. Hastings, F.
Loebe (Eds.), ICBO|ODLS 2020 International Conference on Biomedical Ontologies 2020,
CEUR-WS.org, Bolzano, Italy, 2020, pp. A.1-11.
[15] A. Barton, P. Fabry, L. Lavoie, J.-F. Ethier, LABO: An Ontology for Laboratory Test
Prescription and Reporting, in: A. Barton, S. Seppälä, D. Porello (Eds.), Proceedings of the Joint
Ontology Workshops 2019 (JOWO 2019), CEUR-WS.org, Graz, Autriche, 2019, pp. 1–9.
[16] F. Toyoshima, Roles and their three facets: A foundational perspective, Applied Ontology (2021)
1–32.
[17] F. Loebe, Abstract vs. social roles–Towards a general theoretical account of roles, Applied
Ontology 2 (2007) 127–158.
[18] C. Masolo, L. Vieu, E. Bottazzi, C. Catenacci, R. Ferrario, A. Gangemi, et al., Social Roles and
their Descriptions., in: D. Dubois, C. Welty (Eds.), Proceedings of the 9th Int. Conf. on
Principles of Knowledge Representation and Reasoning (KR 2004), AAAI Press, Menlo Park,
CA, 2004, pp. 267–277.
[19] J. Jørgensen, Imperatives and logic, Erkenntnis 7 (1937) 288–296.
[20] P.B. Vranas, New foundations for imperative logic I: Logical connectives, consistency, and
quantifiers, Noûs 42 (2008) 529–572.
[21] P.B. Vranas, In defense of imperative inference, Journal of Philosophical Logic 39 (2010) 59–71.
[22] J.J. Katz, P.M. Postal, An integrated theory of linguistic description, MIT Press, Cambridge,
1964.
[23] J.P. Thorne, English imperative sentences, Journal of Linguistics 2 (1966) 69–78.
[24] R.M. Hare, 1 Imperative Sentences, in: Practical Inferences, University of California Press, 1971,
pp. 1–21.
[25] W. Downes, The imperative and pragmatics, Journal of Linguistics 13 (1977) 77–97.
[26] N. Charlow, The meaning of imperatives, Philosophy Compass 9 (2014) 540–555.
[27] P. Portner, Imperatives and modals, Natural Language Semantics 15 (2007) 351–383.
[28] D. Wilson, D. Sperber, Mood and the analysis of non-declarative sentences, in A. Kasher (Ed.),
Pragmatics: Critical concepts, vol. II, 1998, pp. 262-289.
[29] J.R. Searle, A taxonomy of illocutionary acts, in: K. Gunderson (Ed.), Language, Mind and
Knowledge, University of Minnesota Press, 1975, pp. 344-469.
[30] K. Korta, J. Perry, Pragmatics, The Stanford Encyclopedia of Philosophy, Spring 2020 edition,
Metaphysics Research Lab, Stanford University (2020)
[31] R. Stainton, Words and thoughts: Subsentences, ellipsis, and the philosophy of language, Oxford
University Press, 2006.
[32] A. Lascarides, N. Asher, Imperatives in dialogue, in: P. Kühnlein, Hannes Rieser, Henk Zeevat
(Eds.), Perspectives on Dialogue in the New Millennium (2003) 1–24.
[33] M. Green, Speech Acts, The Stanford Encyclopedia of Philosophy, Winter 2020 edition,
Metaphysics Research Lab, Stanford University (2020).
[34] J.L. Austin, How to do things with words, Oxford university press, 1975.
[35] K. Bach, R.A. Harnish, Linguistic communication and speech acts, MIT Press, Cambridge, 1979.
[36] D. Kaplan, others, Themes from Kaplan, Oxford University Press on Demand, 1989.
[37] C. Masolo, L. Vieu, Graph-Based Approaches to Structural Universals and Complex States of
Affairs, in: S. Borgo, P. Hitzler, O. Kutz (Eds.), Formal Ontology in Information Systems:
Proceedings of the 10th International Conference (FOIS 2018), IOS Press, 2018, pp. 69–82.
[38] N. Asher, A. Lascarides, Logics of conversation, Cambridge University Press, 2003.
[39] S. Seppälä, A. Ruttenberg, B. Smith, The functions of definitions in ontologies, in: R. Ferrario,
W. Kuhn (Eds.), Formal Ontology in Information Systems. Proceedings of the 10th International
Conference (FOIS 2016), IOS Press, 2016, pp. 37–51.
[40] F. Toyoshima, A. Barton, O. Grenier, Foundations for an ontology of belief, desire and intention,
in: B. Broadaric, F. Neuhaus (Eds.), Formal Ontology in Information Systems. Proceedings of
the 11th International Conference (FOIS 2020), IOS Press, Bolzano, Italy, 2020, pp. 140–154.
[41] C. Tarbouriech, A. Barton, L. Vieu, J.-F. Ethier, The Counting Problem of Slot Mereology, in:
T. Sales, E. Sanfilippo (Eds.), Proceedings of the Joint Ontology Workshops 2021 (JOWO
2021), 5th Workshop on Foundational Ontology, Bolzano, Italy, accepted, pp. 1–15.
[42] D. Schlangen, A. Lascarides, Resolving fragments using discourse information, in: Proceedings
of the 6th International Workshop on Formal Semantics and Pragmatics of Dialogue (EDILOG
2002), 2002.
[43] A. Bandrowski, R. Brinkman, M. Brochhausen, M.H. Brush, B. Bug, M.C. Chibucos, et al., The
ontology for biomedical investigations, PloS One 11 (2016) e0154556.
[44] N. Troquard, R. Trypuz, L. Vieu, Towards an ontology of agency and action: From STIT to
OntoSTIT+, in: B. Bennett, C. Fellbaum (Eds.), Formal Ontology in Information Systems.
Proceedings of the 4th International Conference (FOIS 2006), IOS Press, 2006, pp. 179–190.
[45] J.-F. Ethier, A. Barton, R. Taseen, An ontological analysis of drug prescriptions, Applied
Ontology 13 (2018) 273–294.