paolapietrandrea

Representing epistemic constructions as triadic relations

Epistemic constructions are commonly represented as dyadic relations between a marker, often called the ‘modal’, and a scope, often called the ‘prejacent’ of the relation (see among others, von Fintel (2005)). The marker is the element of the construction responsible for attributing a truth value to a propositional content; the scope is the propositional content evaluated by the marker. A formalization of this view of epistemic construction is given in (1):

(1) epc = [probably]m ([it is the postman])s

We prefer to represent epistemic (and, more generally, modal) constructions as complex constructions consisting of a marker, a scope and a relation between the marker and the scope. The relation between the marker and the scope is in our view an essential element of the epistemic construction characterized by precise formal properties, which encodes per se the operation of epistemic (or more generally modal) evaluation. Let us justify our proposition.

It happens quite frequently in the negotiation of epistemic (or more generally modal) commitments that one and the same scope receives more than one evaluation. Consider the following example (2):

(2) C: dovevano venire a leggerla quanto meno A: no anche se non veniva<no> si’ dovevano veni’ a leggerla C: cosi’ almeno si sapeva

C: they were supposed to come and read it at least A: no even if they did not come yes they were supposed to come and read it C: at least as far as we know

It is clear here that the same scope, i.e., the proposition [they come to read it] is attributed five different truth values by the five different markers, the modal dovevano ‘were supposed to’, the pragmatic marker no ‘no’, the pragmatic marker si’, ‘yes’, the list construction dovevano veni’ a leggerla …dovevano veni’ a leggerla defined by the dialogic repetition of dovevano veni’ a leggerla and the utterance cosi’ almeno si sapeva ‘at least this what we were told’. In other words in this exchange, one and the same scope enters five different epistemic constructions. It would not make sense to try to establish the truth-value of the scope independently of the construction in which it is included. It is clear in this case that the epistemic evaluation cannot be conceived as a property of the scope (which indeed receives several evaluations).

It would also be awkward to regard the epistemic evaluation as a property of the marker. As shown in …, for example, one and the same marker, in this case the complement-taking predicate ‘I think’ may or may not attribute a truth value to its scope (in other words may or may not be epistemic) depending on the semantic nature of the scope . The literature, on the other hand, has shown many cases of markers whose epistemic value is only contextually determined (see for example the modal verbs that can be either epistemic or deontic depending on the semantic nature of the predication they scope over).

The epistemic evaluation is therefore better regarded as a function fulfilled by the overall construction consisting of the marker, the scope and the relation holding between them rather than as a function fulfilled by either the marker or the scope. We represent therefore epistemic constructions as in (3):

(3) epc = ([probably]m ([it is the postman]s)) epc

Paola Pietrandrea

See also P. Pietrandrea (submitted) Epistemic constructions at work. A corpus study on spoken Italian dialogues.

A functional definition of epistemicity

Cognitive-functional approaches to modality ground the very definition of epistemicity in the notion of commitment. Let us examine for example the following three quotations:

Any utterance in which the speaker explicitly qualifies his commitment to the truth of the proposition expressed by the sentence he utters…is an epistemically modal, or modalized, utterance. (Lyons, 1977: 797)


The term “epistemic” should apply not simply to modal systems that basically involve the notions of possibility and necessity, but to any modal system that indicates the degree of commitment by the speaker to what he says. In particular, it should include evidentials such as “hearsay” or “report”…or the evidence of the senses. Palmer (1986: 51)


Epistemic forms establish a position of…commitment with respect to the propositional content of the clause. (Verstraete, 2001: 1523)

These quotations show that the notion of commitment to which cognitive-functional approaches refer is regarded as an attitude, of a particular type: it is (i) private, (ii) static and (iii) discourse-independent. (i) Lyons and Palmer speak explicitly of the “speaker’s commitment”. There is no room in their conception of commitment for the idea that commitment may be shared by the participants in discourse. (ii) Lyons and Palmer speak of “the speaker’s commitment”, while Verstraete refers to “A…position of commitment”. In all cases, commitment is viewed as a unique attitude which does not change over discourse. (iii) Lyons and Palmer state that modal forms “indicate”, or “qualify” the commitment of the speaker. In this perspective, commitment is regarded as an attitude that pre-exists discourse and that is merely described by, rather than constructed through, discourse.

This take on commitment entails that epistemicity is regarded as a solipsist, static, descriptive category: epistemicity is viewed a category that merely serves to describe (rather than construct) a static (rather than dynamic) attitude held by a single speaker (rather than a shared attitude). Since our initial efforts to develop an annotation scheme for epistemicity, we realized that such a definition of epistemicity would not allow us to account for a variety of constructions that are particularly frequent in discourse and that intuitively have much to do with the expression of truth evaluation.

Let us examine examples (1) through (3):

(1) D: per cui uno ripete le stesse cose no ? C: e' vero[1]
D: so one always repeats the same things don't you agree? C: true

(2) C:   ma secondo me li’ c'e' molto l’istinto materno represso eh? A: lei? C: ah E: dici?
A: non mi sembrava
C: but in my opinion there is a lot of repressed maternal instinct. A: who? her? C: ah E: are you sure? A: I didn't get that impression.

(3) C: Oreste Del Buono sta facendo sti racconti in tv A: ah si’? dove su? C: credo rai uno non lo so
C: Oreste Del Buono is telling these tales on TV A: really? which channel? C: I think Rai Uno I don’t know

These examples challenge respectively the solipsist, the descriptive and the static approach to epistemicity. Example 1 shows that commitment is not a private attitude, rather it is an attitude that speakers aim to share with one another: the markers no? ‘don’t you agree?’ and è vero ‘true’, for example, are used by the speaker respectively to solicit the truth evaluation by her interlocutor and to accept the interlocutor’s evaluation. Example 2 shows that commitment is not a pre-existing attitude which is merely described in discourse: the commitment agreed on, indeed, ‘it is unlikely that there is some repressed maternal instinct’ did not exist before this exchange and it was only reified in discourse through the negotiation between the speakers. Finally, example 3 shows that commitment can change in discourse: while the speaker first considers it probable that the tales were broadcasted by the Rai Uno Channel credo Rai Uno, by saying non lo so ‘I don’t know’ the speaker changes her truth evaluation by diminishing the degree of probability of the truth of the propositional content.

In order to account for the totality of epistemic constructions occurring in dialogues we decided to go back to the notion of commitment originally put forward by Hamblin, and to revisit it in the framework of a communitarian, interactional, dynamic semantics (Ginzburg 2012). As shown by de Brabanter and Dendale (2008), Hamblin’s (1970), original characterization of commitment did not refer to it as an attitude, rather as a propositional token. Hamblin calls “commitment” each of the statements that a speaker adds to the store of statements that he builds as his discourse advances: A speaker who is obliged to maintain consistency needs to keep a store of statements representing his previous commitments, and require of each new statement he makes that it may be added without inconsistency to this store. (Hamblin 1970). By agreeing to consider commitment as a propositional token, discourse can be represented as a succession of commitments. These commitments can correspond to statements, beliefs, claims, assumptions (i.e., truth-bearing semantic objects) as well as to other types of judgments, such as aesthetic or moral judgments of the qualities of a state of affairs, or judgments about the opportunity, or the desirability of the realization of a state of affairs. These different types of commitments realize respectively epistemic (4), appreciative (5), axiological (6), deontic (7), and volitional (8) discourses.

(4) I think he went through a separation with his wife and I think that depressed him. (EnTenTen)


(5) I love your wife, and I think she is beautiful! (EnTenTen)


(6) Quite frankly, I think he has the right to make that decision. (EnTenTen)


(7) I think you are better off fixing the "issues" one by one than going into bankruptcy. (EnTenTen)


(8) I did have a waxing service from one other person here, but I think I will choose to stick with Simona for future waxing services from here. (EnTenTen)

By epistemic commitment we indicate any propositional token that contributes to the construction of an epistemic discourse, i.e., any commitment that can be evaluated in terms of truth value. In order to account for the peculiarities of the construction of commitments in dialogues, we propose to revisit the representation of discourse as a “store of commitments” proposed by Hamblin in the light of the more recent representations of dialogic discourse provided by the communitarian (Stalnaker 1978), dynamic (Groenendijk & Stokhof, 1991) and interactionist (Ginzburg 2012) approaches to semantics. According to Stalnaker 1978 discourse is not merely a stock of commitments produced by a single participant; rather, the participants of a conversation construct a set of shared commitments, the “common ground”. Each of these accepted and shared commitments is intrinsically dynamic, i.e., it has the potential to change the common ground (Groenendijk & Stokhof, 1991) and hence the common ground is better regarded as a dynamic repository of commitments. In this theoretical perspective dialogic discourse is regarded as a process, namely as the process of grounding, i.e., a process by means of which commitments become part of the common ground (Clark & Brennan 1991). As shown by Ginzburg (2012: 8) the process of grounding is an interactive process involving “multiple agents with distinct beliefs, and desires”. This process may entail “misunderstanding, rejection, correction […] as well as success”. We can thus call epistemic grounding the communitarian, dynamic, interactional process that allows for grounding, or attempting to ground, epistemic commitments in discourse.

In this renewed perspective, we can define as epistemic any construction that explicitly signals the process of construction of epistemic grounding, i.e., any construction that serves to explicitly signal the process of shared attribution of a truth value to the propositional tokens that compose a discourse. This reappraisal of the notion of epistemicity has at least two important consequences for the identification of epistemic constructions. First, we can consider as epistemic any construction that attributes a truth value to a propositional content, regardless of whether it serves the function of merely qualifying it, as in (9) or the function of negotiating it (checking, confirming, etc., see section 6.1.5 for a complete typology) as in (10) and (11):

(9) A: forse non non lo so non riesco a spiegamme
A: maybe I don’t know I cannot explain it


(10) a te Giovanna non ti piace l'aglio vero?
you don’t like garlic, Giovanna, right?


(11) A: gli fa comodo la situazione G: ah certo
A: he has his own interest in this situation G: oh yes, it is true

Secondly, by taking up Hamblin’s distinction between epistemic commitment and other types of commitments we also consider truth evaluation as a crucial element for the definition of epistemicity. This allows us to avoid a frequent mistake consisting in considering as epistemic constructions that serve in fact to evaluate a commitment aesthetically, morally or in terms of opportunity and desirability. We have seen for example, that the mere presence of the marker “I think” in (5) through (8) is not sufficient justification for considering the constructions occurring in these utterances as epistemic, since they do not express truth evaluations about the commitments that they express respectively.

[1] All the example in this article are drawn from the VoLip Corpus .

Paola Pietrandrea

See also P. Pietrandrea, submitted. Epistemic Constructions at work. A corpus study on spoken Italian dialogues.

Using the Analec tool

Introduction

Analec is a freeware, multiplatform tool (it runs with all main OS, Windows, iOS, …) for carrying out linguistic annotation of textual data.

Analec uses a Java environment. Click here to download Java.

Getting started

Download Analec from here and install it.

Click on the “Documents” menu, then choose “Ouvrir” to open the corpus and select the .ec file.

Visualization of the corpus

1. Markers and scope portions

You will now be directed to the “Gestion des unités” interface.

Click on “type d’unité” and choose between “marker” and “scope portion” to display the elements in the text (highlighted in yellow);
Click on the element to check the values assigned to each unit.

You can also search for a specific item in the text by clicking on “identifiant” and scroll the results with the up-down arrows next to “identifiant”.

This interface is also available from the “Unités” menu → “Gestion des unites”.

2. Scopes

To view the scopes, i.e. the combinations of two or more scope portions:

Click on the “Schémas” menu;
Click on “Gestions des schémas”;
Select “scope” as “type de schéma”.

3. Relations

To view the relations between the markers and the scopes (in the form of “SCH-scope”):

Click on the “Relations” menu;
Click on “Gestion des relations”;
Select “epistemic relation” as “type de relation”.

Searching the corpus

1. Customization of the vue

The default view of the annotated text shows you the properties and values assigned to each annotated element (e.g. the item ‘okay’) in the way they are defined in the annotation scheme.

By clicking on the “Vue” menu → “Gestion de la vue”, you can customize this default interface by adding to an element type some properties and values assigned to other element types that are directly linked to it via the epistemic relation.
Note that this element type can coincide with the epistemic relation itself.

On the left column you can select the element types you want to take into account (e.g. the ‘marker’ type that includes ‘okay’).
On the right column you can choose, by double-clicking, the new properties that will be transferred to the element type and displayed in the main interface (e.g. the five contextual properties of the epistemic relation in which ‘okay’ is embedded).
In the middle part of the screen these new properties are displayed.

Save this customization by clicking on “Valider” on the bottom left corner.

Now for each item included in the element type both the default properties (here circled in blue) and the new ones (in red) are displayed.

With this tool is also possible to change the way the annotated elements are presented in the text (different colors, font, highlight).
To do so, select “Coloriage d’éléments” in the middle section of the screen.

The view in the screenshot below shows the illocutionary force (assertive in green, interrogative in blue) associated with each epistemic marker.

You can pursue statistical analysis of the corpus by means of three tools available in Analec.
To access to these tools:

Click on the “Statistiques” menu;
Choose between “Fréquences”, “Corrélations” and “Représentations géometriques”.

2. Frequences

“Fréquences” is a tool for quantitative analysis of the units.

Choose the element type (“Elément”) and the field you want to analyze (“Champ à analyser”), as in the screenshot below.

You can filter your search results by selecting a specific property and value associated with the element type in the “Filtre” menu, like in the example below.

It is also possible to export the table by clicking on “Exporter le tableau” and the occurrences by clicking on “Exporter les concordances” (both will be saved in a plain text format).

3. Correlations

The “corrélations” tool allows you to detect significant correlations between properties, calculated through a chi-squared test.

Select the element type of which you want to see the properties in correlation by clicking on the “Eléments” menu;
Choose the properties whose values you want to be displayed in the table horizontally (“Champ 1”) and vertically (“Champ 2”).The most significant correlations are highlighted in dark red, the moderately and mildly significant ones in rose and faded rose, as shown in the example below.

4. Geometric representations

The “Représentations géometriques” tool allows you to carry out factorial analysis of correspondence via geometric representations in the shape of point clouds.
This kind of representation allows to show clusters of items with the same properties, i.e. natural classes of markers or modal constructions.

Choose the element type (“Elément”);
Select the set of properties you want to see represented (“Champs pertinents”). To update the graph at this point, click on “Mettre à jour le graphe”;
Select the specific property you want to see displayed in different colors (“Champ à afficher”);
Click on points in the graph to show the related occurrences and statistical values.

Valerio Cervoni

The annotation scheme

The marker

An epistemic marker is a unary linguistic predicate that takes an element of the context (the scope of the epistemic construction) as its argument and attributes a truth-value to this argument. The marker of an epistemic construction can either attribute a truth value directly, by specifying the degree of truth of the scope or indirectly, by citing a source of evidence for the truth of the scope. For each marker the following properties are annotated: 1. archi-unit, 2. lemma, 3. form 4. morphosyntax, 5. illocution.

1. Archi-Unit
We inherited the notion of Archi-Unit from the EuroEvidMod project. The archi-unit indicates the conceptual domain to which a marker belongs.

2. Lemma
The lemma indicates the dictionary entry under which the marker would appear.

3. Form
The form indicates the concrete realization of the marker in the text.

4. Morphosyntax
Epistemic markers can be realized through the following elements:
(i) morphemes, i.e., affixed element that attributes a truth value to the proposition encoded by the element to which it is affixed.

(ii) modal verb constructions, i.e., constructions characterized by an inflectionally constrained verbal predicate that takes an aspectually incomplete infinitive as a complement and that attributes a truth value to it.

(iii) complement-taking predicates, i.e., verbal predicates that take a sentential predicate as a complement and that attribute a truth value to it.

(iv) adverbials, i.e., a syntactically non obligatory predicates that syntactically depend on another predicate (whether a verb, an adjective, an adverb or an entire clause) and that attribute a truth value to it.

(v) discourse markers, i.e. predicates syntactically independent of the scope to which they attribute a truth value.

(vi) utterances, i.e., syntactically and semantically complete utterances that attribute a truth value to a scope. The reference to the scope is made through a co-reference relation that links an element of the utterance and (an element of) the scope.

(vii) syntactic tag-questions, i.e, interrogative operator+subject pronoun constructions that attach to a main anchor clause to verify its truth value.

(viii) syntactic short answers, i.e., declarative subject pronoun+operator constructions that follow a questioning construction which functions as scope and to which they attribute a truth value.

(vii) syntactic list constructions, i.e., constructions “characterized by the multiple realization of one and the same structural position in a dependency tree, in which the first element of the list represents a scope to which further elements of the list attribute a truth value.

(viii) syntactic interrogative constructions, i.e., syntactically marked interrogative constructions used to check the truth-value of the scope;

(ix) prosodic interrogative constructions; i.e., prosodic profiles added to a scope in order to mark it as an interrogative construction used to check its truth-value .

(x) voice modifications; i.e., modification of the voice used in uttering a scope in order to mark it as a quotation.

5. Illocution
The marker of an epistemic construction can be included in:
(i) an assertive utterance,
(ii) an interrogative utterance,
(iii) an injunctive utterance, or
(iv) an exclamative utterance.

The scope

The scope of an epistemic construction corresponds to the element of the context to which the marker attributes the truth value. As a truth value bearer, the scope of an epistemic construction expresses, by definition, a proposition, i.e., an abstract semantic object capable of being evaluated in terms of truth. From a formal standpoint, the proposition to which the truth value is attributed can be expressed by either a sentence or an utterance. 

1. Illocution
The scope of an epistemic construction can realize (or be included in): (i) an assertive utterance, (ii) an interrogative utterance, (iii) an injunctive utterance, or (iv) an exclamative utterance.

The relation

The relation between the marker and the scope serves the purpose of co-evaluating the truth value of a scope in order to ground (or not) the propositional content of the scope in the common ground.It can be characterized, formally, by : 1. the linear relation between the marker and the scope; 2. the relation between the source that utters the marker and the source that utters the scope; and functionally, by 3. the polarity that it attributes to the scope; 4. the dialogic functions it fulfills; 5. the type of evidential justification (if any) on which the truth evaluation of the scope is based on. 
  
1. Direction
(i)   marker>scope - the marker precedes the scope
(ii)  scope>marker - the scope precedes the marker
(iii) inside - the marker interrupts the scope
(iv)  co-extensive - the marker and the scope are co-extensive

2. Source
(i)  SS -   the marker and the scope are uttered by the same speaker
(ii) OS -   the marker and the scope are uttered by two different   speakers
(iii) SS_OT - the marker and the scope are uttered by the same speaker but in two different speech turns

3. Polarity
(i) positive - the modal construction puts forward the propositional content of the scope as (mostly) true
(ii) negative - the modal construction puts forward the propositional content of the scope as (mostly) false
(iii) neutral - the modal construction puts forward the truth-value of the propositional content of the scope as not decidable.

4. Dialogic Function
(i)   qualification - the epistemic marker is merely used to indicate the knowledge or evidence the speaker has to evaluate the truth of the scope. The scope is added to the common ground within the limits of the validity of the knowledge or evidence provided.
(ii) acceptation - the epistemic marker is used to assent to the scope. The scope is added to the common ground
(iii) non acceptation - the epistemic marker is used to reject the scope. The scope is added to the common ground
(iv) check - the epistemic marker is used to request a confirmation for the validation of the scope.The speaker proposes to add the scope to the common ground.
(v)   confirmation - the epistemic marker is used to confirm the scope whose validity was checked by another speaker. the scope is added to the common ground.
(vi) non confirmation - the epistemic marker is used to reject the scope whose validity was checked by another speaker. the scope is not added to the common ground.
(vii) concession - the epistemic marker is used to either accept or confirm the scope without accepting or confirming the expected argumentative entailments of it. The scope is added to the common ground, its expected argumentative entailments are not. 
(viii) information - the epistemic marker is used by the speaker to inform her interlocutor of the validity of the scope. the scope is added to the common ground, the interlocutor must accept. 

4. Epistemic type
(i)   No evidential justification - The truth evaluation of the scope is not justified on an evidential basis.
(ii) Direct evidential auditory - The truth evaluation of the scope is based on auditory evidence.
(iii) Direct evidential visual - The truth evaluation of the scope is based on visual evidence.
(iv)   Indirect evidential inference - The truth evaluation of the scope is based on an inferential reasoning.
(v) Indirect evidential feeling - The truth evaluation of the scope is based on a non sensory feeling.
(vi) Indirect evidential memory - The truth evaluation of the scope is based on speaker's memory.
(vii) Indirect evidential reportive - The truth evaluation of the scope is based on what was reported to the speaker by other sources.
(viii) Indirect evidential quotative - The scope corresponds to a textual quotation of other sources .

Paola Pietrandrea & Elisa Ghia
(see also Pietrandrea, P. (submitted), Epistemic constructions at work. A corpus study on spoken Italian dialogues.)