Narrative testimony

[S]tyle...is...never something merely “decorative. . Susan Sontag, On Style

Abstract

Epistemologists of testimony have focused almost exclusively on the epistemic dynamics of simple testimony. We do sometimes testify by ways of simple, single sentence assertions. But much of our testimony is narratively structured. I argue that narrative testimony gives rise to a form of epistemic dependence that is far richer and more far reaching than the epistemic dependence characteristic of simple testimony.

Introduction

I am an epistemic dependent. You are too. Both of us rely on the testimony of others, and our reliance is both pervasive and deep. Pervasive because many of our beliefs are testimonially based; deep because such beliefs are often central to our cognitive and practical projects. So far, so platitudinous.

Epistemologists seeking to model epistemic dependence have focused, almost exclusively, on simple testimonial exchanges, that is, on cases in which a speaker assertorically utters a single—most often, very simple—sentence: think ‘Elvis Presley is alive’ (Lackey 2008), ‘I’m travelling to Minsk’ (Moran 2018), or ‘The President is in New York’ (Faulkner 2011).¹

Our speech sometimes fits this pattern. But there are significant chunks of our testimonial practice that this model fails to even roughly approximate. In many cases, we do not testify by offering an audience a single, isolated sentence. Rather, we testify by way of narrative. We offer our hearers not a solitary sentence, but a structured discourse of interlocking claims.²

This occlusion matters. The forms of epistemic dependence characteristic of narrative testimony are—I contend—far richer than the forms of epistemic dependence characteristic of simple testimony.³ A picture of epistemic dependence informed only by the latter will be a picture which is unduly etiolated; a normative epistemology informed by such a picture will be, at best, incomplete and at worst, distorted.

This paper is a corrective. Its project: to model the richer form of epistemic dependence—which I call perspectival dependence—characteristic of narrative testimonial exchange.⁴ When I accept simple testimony, I remain largely responsible for the way in which the accepted content is integrated into my overall system of beliefs. To accept a narrative on your say-so is, by contrast to accept an already-structured bundle of information. To embed content in a narrative is to colour its affective valence and ongoing inferential profile; to accept narrative testimony is to accept these epistemically consequential embeddings.

My model come in three stages. First, I delineate the target phenomenon: perspectival dependence. Second, I offer a detailed picture of the characteristic connection between narrative testimony and perspectival dependence.⁵ Third, I argue that narrative and simple testimony have a different relationships with trust. I close the paper by briefly sketching the implications of my model for a normative epistemology of testimony.

But let me be clear as to scope. This is not a work of normative epistemology: I will defend no claims as to when one ought to accept or offer narrative testimony. The paper is, rather a prolegomenon to a normative project proper: only once the phenomenon of perspectival dependence is brought into view can we begin to theorise its proper regulation. My claim is that the normative questions posed by narrative testimony outstrip those posed by its simple counterpart.

Some further clarifications. First, my focus is testimony, not fiction. To be a narrative is to have certain formal or quasi-formal features; there is no reason texts or utterances with such features cannot have genuine assertoric force. This can be made vivid by example: Gibbon’s The History of the Decline and Fall of the Roman Empire is a narrative, but is supposed to be read as history. Mary Prince’s The History of Mary Prince is a narrative, but it is supposed to tell us the truth about Prince’s life.

Which formal features, then, make for a narrative? I won’t provide anything like a list of necessary and sufficient conditions; rather, I consider narratives to be just those texts and utterances which have the form of a story. I take it that we have a robust pre-theoretic grip as to when that condition is satisfied. Nonetheless, it’s helpful to say a bit about what paradigmatic stories look like. Paradigmatic stories:

• i describe the doings of humans or human-like characters,

• ii which take place over a period of time

• iii encode a goal structure, viz., the characters have aims,

• iv recount an obstacle, viz., the characters are blocked or impeded or face some barrier to the achievement of their aims.⁶

Of course, some stories will lack at least some of these features. For example, science communicators sometimes rely on narratives that do not describe the doings of human or human-like characters (Dahlstrom 2014).⁷ Nonetheless, such stories will often successfully generate perspectival co-ordination. As will become clear, a token of narrative testimony need not have all the features of a paradigm story in order to generate perspectival co-ordination.⁸

Second, ‘narrative’ is often used as a quasi-technical term—as when people talk about ‘the narrative of the good immigrant’—to mean something like schema or stereotype. I am not concerned with narratives in this sense. Third, ‘narrative’ is sometimes understood as an extremely plastic term, on which any recounting of events whatsoever—for example, ‘I had coffee to drink this morning’—counts as a narrative. I ignore such usage: I doubt that it picks out much of interest, and it certainly does not tally with our ordinary, pre-theoretic conception of a ‘story’.

Technologies of co-ordination

One way in which agents might mentally co-ordinate is by sharing opinions. If I believe that Jane Austen first published a novel in 1811, and you believe this too, then our opinions converge. Viewed through a broadly Stalnakerian lens, testimony is a social technology which aims at just this species of mental co-ordination.⁹ When conversations go well, agents co-ordinate on a set of propositions—the ‘common ground’.¹⁰

By contrast, agents co-ordinate on a perspective when they share a way of looking at the world. This cannot be reduced to opinional co-ordination. A perspective, in the sense at issue here, is a suite of interlocking dispositions:

Attentional:

What, in a given scene or object, is inclined to strike me as interesting or noteworthy? Which details draw me in, and which fade into the background?

Inquisitive:

What kinds of inquiry strike me as worthwhile? What sorts of investigative methods am I inclined to use? What sorts of explanations will close my inquiry?

Interpretive:

What inferences am I disposed to draw, and which evaluations am I inclined towards?¹¹

So much for perspectival co-ordination. Whither its connection with narrative testimony? My contention is this. Just as the Stalnakerian model of conversation takes simple testimony to be a conversational technology aimed at opinional co-ordination, I take narrative testimony to be a conversational technology which aims at perspectival co-ordination.¹² Co-operative, trusting hearers seeking to mentally co-ordinate with narrative testifiers will not simply adopt the opinions expressed by the narrative, but will structure and organise information as the narrative suggests. By adopting said structure, the agent allows their perspectival dispositions to be shaped by the speaker. In just the same way that I become opinionally dependent on you when you tell me that p and I come, on your say-so, to adopt your opinion as my own, I become perspectivally dependent upon you when I accept a story on your say-so, and thus come to adopt the perspective encoded in that story.¹³ Once adopted, a perspective may embed itself stably in an agent’s cognitive repertoire, with long-range epistemic effects.

These are bold claims. If I’m right, narrative testimony is a powerful conversational technology. In the next section, I provide a detailed account of how narrative testimony generates perspectival dependence.¹⁴ But first, I want to warm you up to the idea that narrative’s communicative effects outstrip those of simple testimony.

Against sheep shearing

You might think my whole project overblown. Maybe you think something like this:

Sure, we testify by way of narrative. But narratives are just strings of sentences that have been stuck together. So there is no need for a distinct treatment of narrative testimony.

Scepticism that there is anything epistemically distinctive about narrative testimony borrows from the tempting thought that once we have a treatment of the sentence, we have a treatment of discourse. But this is a bad thought. Here’s why.

Suppose I want to tell you that your train leaves from platform three. I will do this by uttering a certain sentence: ‘Your train leaves from platform 3’. For my testimony to be a success, you need to be able to understand the sentence I utter. You must retrieve the content encoded in my utterance. This process of retrieval will generally require, on your part, tacit representation of my utterance’s syntax and an assignment of semantic values to the constituents—whether overt or unarticulated—of my utterance.

Now suppose I want to tell you a simple story. I will do this, typically, by uttering a mini-discourse: a string of sentences. Here’s an example:

Mini:

‘I arrived late at the station today and almost missed my train. As I was running, the guard saw me and held the door for a second, and I made it.’¹⁵

For my narrative testimony to be a success, you must retrieve the content encoded in this string of sentences. Here is a naive, superficially attractive way of thinking about this process of discourse content-retrieval: interpreting the discourse is a matter of interpreting each of its constituent sentences—using the process sketched above—and then sticking the recovered contents together. According to this naive aggregative model, to interpret a discourse is just a matter of interpreting a lot of sentences, in the same way that sheering a flock of sheep is just a matter of sheering a lot of individual sheep.

Such a naive model, though, is not viable. It fails to be viable because it cannot accommodate the interlocking character of discourse sentences. Discourses are cohesive objects: their constituent sentences are knitted together into complex wholes. Models of discourse interpretation must be sensitive to this cohesiveness. An example will help to make this vivid. Consider the following mini-discourse:

1:

Exactly one delegate arrived. She registered.

This mini-discourse contains an example of inter-sentential anaphora: the ‘she’ in the second sentence must be interpreted as referring to the delegate introduced in the first sentence. The ‘sheep sheering’ model could try to handle this data in one of two ways.¹⁶ First they might try to model (1) using a a traditional Montagivian framework. In such a framework, the full stop would be analysed as a conjunction, and sentence (1) treated as semantically equivalent to:

1*:

Exactly one delegate arrived and registered.

But such a treatment fails: it cannot accommodate the fact that ($1*$) but not (1) could be true in a case in which many delegates have arrived, but only one has bothered to register (Kamp et al. 2011). Their second option is to go pragmatic—to argue that the first sentence makes the arrived delegate particularly salient, and thus primes a competent hearer to interpret ‘she’ as referring to the arrived delegate. On this story, the first sentence in (1) acts to constrain our interpretation of the second sentence in much the same way that extra-linguistic features of context act to constrain our assignment of semantic values to unbound pronouns. But this is not an appealing story. Consider the following pair of sentences:

2:

Tilly lost ten marbles, and found all but one of them. It is probably under the couch.

The ‘it’ in the second sentence is naturally interpreted as referring to the one marble under the couch. A pragmatic story about how this resolution of the anaphoric pronoun is effected predicts that there should be little difference between (2) and (3):

$\#3$:

Tilly lost ten marbles, and found exactly nine of them. It is probably under the couch.¹⁷

This prediction fails: (2) sounds good; (3) does not. The asymmetry suggests that the resolution of inter-sentential anaphora is subject to semantic, and not merely pragmatic features of the target sentences.

The sheep-sheering model, then, must be abandoned. It is widely held that, to handle phenomena like inter-sentential anaphora, we must model discourse interpretation as a dynamic process. What does this turn towards dynamism involve? The details are contested; I’ll settle for giving a very low resolution picture. Suppose I am confronted with the mini-discourse in (1). I start off by constructing a mental representation of its first sentence. This will involve the production of something like a mental file—a vehicle for singular thought with associated descriptive content—for the delegate who is mentioned in this first sentence. When I interpret the second sentence, two important things happen. First, I construct a provisional mental representation of the second sentence, in which the anaphoric pronoun ‘she’ is marked as in need of some familiar discourse referent to latch onto. Once I identify the previously mentioned delegate as just such an available discourse referent, I integrate my mental representations of the first and second sentence. The provisional mental representation of the second sentence is merged with the first representation to produce a single integrated mental representation tied to the mental file associated with the delegate.¹⁸

The moral is this. The process of discourse interpretation is one whose mechanics outstrip, by some margin, the mechanics involved in the interpretation a of single, isolated sentence; correspondingly, discourse interpretation cannot be thought of simply as ‘scaled up’ sentence-interpretation. Rather, it involves distinctive dynamic processes which have no clear analogue in our folk model of sentence interpretation.

Obviously, all this falls far short of establishing the bold claims I want to make for narrative testimony. But, on occasion, when I have suggested to epistemologists of testimony that narrative testimony poses questions not posed by its simple analogue, it has been clear that they are inclined towards a ‘reductive’ approach: to assume that the epistemic effects of narrative testimony will be no more than a scaled-up version of the effects of simple testimony. This assumption may turn out to be correct but it should not, I think, be the default. Rather, the default assumption, given recent work in philosophy of language, should be that there are epistemic, as well as linguistic effects, that emerge at the level of discourse.

What to expect

In the coming sections, I identify three characteristic features of narrative testimony: the cuing of representational format, the cuing of gestalt representational features, and the enocding of story schemata. Each of these three features, I argue, cues forms of mental co-ordination characteristic of perspectival dependence. Each mechanism reflects a different way in which narratives organise and structure the information they encode. The perspectival effects of each mechanism flow from the structuring and organising effects of successful narrative exchange.

In many paradigmatic cases of narrative testimony, these three structuring-and-organising features work in tandem, to reinforce and deepen each others’ effects. I do not claim that each of these features is present in all instances of narrative testimony, and never present in other forms of speech. Nor do I argue that their presence inevitably generates relations of perspectival dependence. Rather, I claim that these features are characteristic of narrative testimony, and characteristically give rise to perspectival dependence.

Representational format and perspectival dependence

In this section of the paper, I argue that narrative testimony cues a representational format. Thus, in successful narrative testimonial exchange—that is, where an audience accepts a speaker’s narrative testimony—speaker and hearer become co-ordinated not only with respect to the informational content of the speaker’s testimony, but also with respect to the content’s representational format. Such co-ordination gives rise to perspectival dependence.

Situation models

Text comprehension is a ‘multilevel’ process (Perrig and Kintsch 1985.)¹⁹ Put differently, text comprehension comes both in ‘thick’ and ‘thin’ varieties. For successful ‘thin’ discourse comprehension, all that is required is a mental representation of the meaning of a discourse. Where the text is something simple—for example, a single sentence like:

• 4 University College is older than Exeter College,

– the mental representation will typically be a sententially structured mental representation with propositional content.

‘Thick’ discourse comprehension, on the other hand, has as its output more than a sententially structured mental representation. Instead, it outputs a mental state in which the information encoded in the text is integrated with an audience’s background knowledge to generate a rich ‘mental model’. Let us call—following standard usage—the output of thick text comprehension a situation model.

So put, the distinction here is not particularly sharp; one way to sharpen it is to identify the different functional profiles of the two outputs. Thin outputs support simple ‘recall’ tasks. For example, someone who has achieved thin comprehension of (2) will be able to correctly answer question (3):

• 5 Which is older: University college or Exeter college?

But this ‘thin’ mental representation may not be enough to support more demanding, ‘inference’ tasks. Consider a second question:

• 6 Which is younger, University college or Keble college?

Someone with only a ‘thin’ mental representation of (4) may find it difficult to answer this question even if they know perfectly well that Keble College is younger than Exeter college. To readily answer (6), the ability to recall the content ‘directly’ associated with (4) is not enough—this content must also be cognitively integrated with a mental representation of Keble College. Someone with only ‘thin’ comprehension of (4), who has not integrated their mental representation of (4)’s content with their background knowledge of Oxford has only scattered and fragmented mental representations of the city, rather than a single, integrated mental representation. The latter, though, is much better placed to facilitate the inferences an agent must make to answer question (6).²⁰ Hence thick and thin comprehension output mental representations that are well-suited to different cognitive tasks.

Situation models may exploit various different representational formats. Suppose you want to represent the layout of a small town. Here are some ways one might do this.

Procedural:

A set of instructions for how to get from one landmark to another.

Spatial:

A map-like representation of the landmarks.

Propositional:

A list of claims specifying the co-ordinates of each landmark.

These different representations will have different functional profiles. In particular, they will support different inference tasks. Compare the following questions:

• 7 Is the church north of the station?

• 8 Must one turn left if travelling from the station to the church?

• 9 As the crow flies, is the church further from the station than it is from Exeter college?

• 10 Will it take me longer to walk to the church from the station or from Exeter college?

Someone with a spatial representation will (ceteris paribus) find (7) and (9) easier to answer than (8) and (10); for someone with a procedural representation, this pattern will be reversed.

Moreover, different styles of mental representation will ‘position’ the agent differently with respect to the town. Someone with a procedural representation will, when using their mental model, tend to imaginatively embed themselves within the town, and first-personally rehearse the relevant movements. When working out how to get from the church to the town hall, they will tend to imagine themselves walking down a certain street, taking a certain turning, and so on. In other words their mental representation will support inferences primarily by exploiting extended first-personal, visualisation. By contrast, someone with a spatial or propositional mental representation of the town will not be inclined to work out how to get from the church to the town hall by way of such imaginative first-personal projection. If their inferences require visualisation at all, it will be visualisation from a ‘bird’s eye’ perspective, rather than from an embedded, first-personal perspective.

Back to narrative

Why does this matter for a discussion of narrative testimony? Answer: narrative testimony allows speakers to cue representational formats. This allows speaker and hearer to perspectivally co-ordinate.

Consider, for example, the following two discourses:

• 11 The little town of Baldwin is an old frontier town in the Midwest. To get there, drive east along the east-west Highway to the Green River which rushes out of some low hills to your left. Among the hills, before you come to the river, you see the high school which is connected to the highway by a small road. To see the town, continue on the highway to Main Street where you can turn either right or left. Going left on Main after a few blocks you see the Lutheran church on your right. Returning on Main Street to the other end, you come to the general store.

• 12 The little town of Baldwin is an old frontier town in the Midwest. It is located where the east-west highway crosses the Green River which rushes out of some low hills north of the highway. Among the hills on the west bank of the river is the high school which is connected to the highway by a small road. The town itself consists of little more than the highway and Main Street which crosses the highway. A few blocks north on Main there is a Lutheran church on its east side. The general store is on the southern end of Main Street.²¹

(11) and (12) communicate roughly the same facts about the geography of Baldwin. But they cue different representational formats: the first, procedural, the second, spatial. This is effected by way of a layered series of cues: in (11), ‘to get there’, ‘you see’, ‘before you come’, ‘going left’, and so on; in (12), ‘north’ ‘a few blocks’, ‘the southern end’, etc. A receptive hearer will pick up on these cues, and respond to them in their choice of representational format.²²

If I hear and accept the first discourse, and you hear and accept the second, we will, in some rough sense, share a representation of Baldwin. Given that the two discourses communicate the same facts about the town, there is a thin sense in which we are mentally coordinated.²³ Nonetheless, we differ in important respects: we use different representational formats. In the next section, I explore the connections between representational format and each of the three dispositions that compose a perspective.

Perspectival impact

First, different representational formats interact in various open-ended ways with attentional dispositions. Consider details like (i) the colour of a building or (ii) the difficulty of walking on the cobblestones between the church and the station. These are far more likely to be salient for someone with a procedural representation than someone with a spatial or a propositional representation of a town.

Second, because the different representations support different inferential tasks, they favour different kinds of inquiry. Opinional convergence in the absence of convergence on representational format is thus likely to be brittle: agents who opinionally converge but whose representational formats differ are likely to expand their opinion sets in different directions, and so become opinionally uncoordinated.

Both interpretive and inquisitive dispositions will be influenced by a second feature of representational format. Compare A, whose inferences are supported only by a sparse, bird’s eye representation of a town, B, whose inferences involve the manipulation of propositions, and C, who routinely relies on vivid, embedded first-personal visualisation of the town when engaged in inference or recall tasks.

An agent whose representational formatting requires the latter engages in what is sometimes called ‘dramatic rehearsal’: when they imaginatively navigate the town, they ‘rehearse’ the experience of actually navigating it, or ‘try on’ (Moran 1994) the point of view of a person actually walking through the town.

Why does this matter? Variously. As Camp notes, dramatic rehearsal has much of the ‘phenomenal immediacy’ of perception; it plausibly triggers robust affective responses (Camp 2017a). The inferential mechanisms that A and B rely on are far ‘cooler’, with a correspondingly muted affective profile.

This connects our discussion with one of the most heavily theorised features of narrative fiction. Reading narratives appears to activate experiential representations: readers ‘process information from the spatio-temporal, cognitive, and emotional point of view of narrative protagonists’ (Camp 2017a). Readers will often empathetically re-enact scenes from the perspectives of the characters within the narrative (Currie 1995).²⁴ In other words, the literature on narrative fiction clearly recognises the capacity of narratives to cue a distinctly embedded mode of cognition that relies heavily on simulation.

In my view, though, the capacity of narrative fiction to cue this style of cognition is best conceived as an example of a more general capacity of narrative communication to cue representational formats.²⁵ Consider three different formats we might use in our representation of other agents:

Simulation:

We represent the agent by engaging in experientially rich first-personal simulation.

Theory:

We represent the agent propositionally, paradigmatically by ascribing propositional attitudes.²⁶

Manipulationist:

We represent the agent by way of a set of instructions concerning how to successfully interact with them.

When a narrative—fictional or not—coaxes us into representing its protagonists first-personally, it ‘selects’ a representational format for us. Narratives often, but need not select a simulationist format. Consider the following excerpt from Mailer:

One day, while I was sitting in the tub soaping myself, I noticed that she had forgotten the towels. ‘Ida’, I called, ‘bring me some towels!’ She walked into the bathroom and handed me them. As she stooped over the tub to put the towels on the rack, her bathrobe slid open. I slid to my knees and buried my head in her muff. It happened so quickly that she didn’t have time to rebel or even pretend to rebel. In a moment, I had her in the tub, stockings and all. (Miller 2007).²⁷

This is not a narrative which encourages a first personal simulation of Ida.²⁸ Multiple cues in the text make it clear that we are not invited to see things from her perspective. Instead, the narrative encourages us to think of her primarily in terms of how to manipulate her: she is encoded primarily as a series of actions that must be undertaken to achieve sexual pleasure. The moral is this: if narratives encourage first-personal of its protagonists—and they often do—that is not because any narrative whatsoever cues such a format; rather, specific details of the text are responsible.

How, though, to relate these remarks to perspectives? My preferred route runs through Strawson’s famous distinction between the reactive and the objective stance (Strawson 2008). The reactive stance is partly constituted by certain inquisitive and interpretive dispositions. Part of what it is for me to take up the reactive stance in relation to you just is for me to be interested in acquiring, and disposed to close inquiry only upon receiving, certain kinds of explanation for your behavior—roughly, those which appeal to what Nagel calls your ‘internal point of view’ (Nagel 1989). If you throw a glass at the wall, and I take up the reactive stance towards you, I am interested in making your throwing intelligible from a first personal perspective, and in interpreting it with respect to its first personal significance. Similarly, part of what it is to take up the objective stance towards a fellow creature is to be uninterested in such explanations and interpretations. (‘He’s just mad’, you might say ‘who cares why he thinks he did it! We just need to move away quietly.’)

Narratives which cue a simulationist representation of a given agent incline us to adopt the reactive stance towards said agent, and so to adopt a certain suite of interpretive and inquisitive dispositions. Conversely, manipulationist representations incline us towards the objective stance, with its attendant suite of dispositions.²⁹ Thus representational format influences inquisitive and interpretive dispositions by shaping (i) which kinds of question naturally occur to us concerning an agent, (ii) which sorts of answers close our inquiry, and (iii) which inferences we are inclined to draw when we get new information.

Let us take stock. Narrative testimony cues representational format. This allows speaker and hearer to co-ordinate with respect to representational format. This co-ordination in turn involves co-ordination with respect to the multivalent dispositions constitutive of perspectival co-ordination: attentional, inquisitive, and interpretive. Where the co-ordination is effected as a result of the speaker’s cuing a representational format, we get perspectival dependence.

Narrative versus simple testimony

How much do simple and narrative testimony differ when it comes to such format cuing? Can’t simple testimony also exploit such cues? Yes, but far less powerfully. The two come apart along two important axes: layering and swamping. Let’s take these in turn.

First, narratives allows testifiers to ‘bake into’ their testimony a set of instructions for situation model construction because they allow for a sustained layering of cues. Simple testimony, by contrast, is less amenable to any such layering: a single sentence offers few opportunities for cue-placement.

Second, suppose I offer you some simple testimony about the lay-out of a town—say, ‘The church is north of the library’. Now, consider a question: which representational format—procedural, spatial, or propositional—is most likely to be exploited by the output of your thick comprehension? Presumably, that will depend almost entirely on which representational format is exploited by your pre-existing mental model. If your existing model of the town exploits a procedural format, the output of thick comprehension will probably involve the absorption of my testimonial content into that procedural representation. The cues embedded in the simple testimony will be ‘swamped’ by the demands generated by the formatting of you pre-existing mental representations.

In other words, the process of integrating ‘thin’ representations with background knowledge will often be one of absorbing the testimonially conveyed content into a rich background structure. Given such absorption, there will be little or no meaningful sense in which the resultant product of the ‘thick’ comprehension is as a mental model of the simple testimony. The ‘microworld’ that results from the thick comprehension process will be the product of something like a ‘negotiation’ between speaker and hearer, rather than a more straightforward adoption of the microworld suggested by the speaker’s testimony. But it is the latter which is characteristic of successful narrative exchange.³⁰

We can then add to the stock taken above: there is a meaningful asymmetry between the capacity of simple and the capacity of narrative testimony to cue representational format and so to give rise to perspectival dependence.

In the next section, I discuss a second mechanism responsible for perspectival dependence.

Characterization cuing

You and I may gaze at the same picture, and yet see different things: you a rabbit, I, a duck. We differ at the level of gestalt. This is a more fine-grained difference than that of representational format. In this section I argue that narratives offer testifiers the resources to control gestalt features of a hearer’s situation model.

The argument is simple:

• 1 Fictional narratives allow authors to control gestalt features of the audiences’ situation model.

• 2 If fictional narratives allow authors to control gestalt features of the audiences’ situation model ability, narrative testimony allows speakers to do the same for their audiences.

Camp defends premise (1) at length. First, she cashes out the suggestive but imprecise ideology of gestalt in terms of aspectual thought, before showing that fictional narratives cue certain forms of aspectual thought (Camp 2017a).

Aspectual thought involves using one’s characterization of one object to structure one’s thinking about another. A characterization is a structured set of properties. For example, Camp’s characterization of a quarterback includes the following properties: being natural leaders, being affable, dumb, shallow, and having a ready smile Camp (2017a). My characterization of a computer science student includes properties like being shy, being good at maths. Characterizations structure properties along at least two axes: prominence and centrality. A feature is prominent (in a given context) to the extent that it is (i) diagnostic, that is, helpful in classifying an object and (ii) intense, that is, salient, or prominent (Camp 2017a). A feature of a characterization is central to the extent that I take it (implicitly) to explain other features of the characterization:

[C]entrality...connect[s] features into explanatory networks. For instance, I take a quarterback’s being a natural leader to explain more of his other features – why he’s popular and confident, why he smiles so readily, why he’s a quarterback at all – than having a square jaw does (Camp 2017a).³¹

When I use my characterization of a quarterback to structure my thinking about Daniel, the features of Daniel which ‘match’ with the characterization will matter more—and guide my thinking—about Daniel than do those which do not match the characterization (Camp 2017a). Suppose that I use my characterization of a quarterback to structure my thinking about Daniel, who is both shy and affable. His affability will ‘stick out’ more than his shyness. By contrast, if I use my computer scientist characterization of Daniel, his shyness will stick out more than his affability.

Three further features of characterizations are worth noting. First, which characterizations we use is not, in any straightforward sense, under the control of the will. Rather, the tendency to rely on a certain suite of characterizations is typically a matter of ingrained cognitive habit. Second, characterizations may (like stereotypes) be socially dominant and widely shared, or (unlike stereotypes) highly idiosyncratic. Third, they are normatively laden bits of our cognitive apparatus: they ’guide our emotional and evaluative responses’ (Camp 2017a.)

Camp suggests that four ‘stylistic’ mechanisms—ubiquitous within narrative fiction—are responsible for cuing characterizations in its consumers:

Loaded Words:

Words, like “cop”, “brigand”, “debonair” are conventionally associated with particular characterizations.

Figurative Language:

Metaphor, simile, and allusion communicate characterizations using non-conventional pragmatic mechanisms.³²

Level of Complexity and Detail:

Indicate habits of cognitive attention and interest.

Genre:

Genres generate both assumptions and interpretive profiles of concern and response.

We are now in a position to argue for the second premise. Here goes: the stylistic mechanisms Camp identifies as responsible for fictions’ cuing of aspectual thought are also frequent features of narrative testimony. Insofar as such mechanisms cue characterizations in the consumers of narrative fiction, they will also cue characterizations in consumers of narrative testimony.

There are various ways to push back. You might resist the claim that these stylistic features are characteristic of narrative testimony. Or you might hold that they are present in narrative testimony, but that various interpretive conventions make us sensitive to their presence in fiction, but insensitive to their presence in testimony.³³

Both objections, I suspect, point to a genuine asymmetry. Although there are stylistically crude fictions and self-consciously literary instances of narrative testimony (consider literary biography), on the whole, it seems that the metaphors characteristic of narrative fiction are more subtle and sophisticated than those present in everyday narrative testimony, and that the levels of complexity and detail to be found in the former are more finely judged than those typically found in the latter. Similarly, whilst generic expectations do often shape our self-understandings, and our accounts of real lives—as Alasdair McIntrye observes, Thomas Beckett’s life has been written as each of hagiography, as saga, and tragedy (MacIntyre 2013)—the influence of generic expectations is more muted in narrative testimony than in fiction. And so on.

However. That the asymmetry is genuine does not mean that the mechanisms Camp identifies are entirely absent from narrative testimony—they clearly are not. Similarly, our habits of reading may indeed make us more sensitive to the stylistic choices made by producers of fiction than those made by producers of narrative. But that does not establish systematic insensitivity: it is implausible that we are systematically insensitive to the stylistic choices of narrative testifiers. The second premise stands. I now turn to the question of how characterizations relate to perspectives.

Perspectival impact

It should be relatively obvious how these characterization-cuing mechanisms can generate perspectival co-ordination. Which characterizations an agent ‘thinks with’ influences each of the three interlocking dispositional suites that make for a perspective. By making some features of a situation more or less salient, characterizations shape attentional dispositions. Because they direct my attention towards certain features and away from others, and because they position certain features as more explanatorily basic than others, my use of characterizations will shape my inquiries. Lastly, characterizations will have subtle interpretive effects. This is best illustrated by attending to the expectations and interpretive schemata often triggered by generic recognition. Suppose A and B are both listening to narrative testimony from which they learn that John has tripped and fallen. A is listening to a narrative whose genre she has correctly identified as comedy; B’s narrative, by contrast, has been correctly identified as tragic. The inferences A is licensed to draw on learning of John’s fall are quite different from those B is licensed to draw. For example, it would be legitimate for A but not for B to be become relatively highly confident that John was not seriously hurt in the fall. Thus, by embedding the events they recount within different genres, narratives embed events within different inferential networks. This, in turn, alters the significance and affective valence of those events.

Narrative versus simple testimony

A different sort of worry goes like this. Surely simple testimony may exploit at least some of the mechanisms Camp identifies. So can simple testimony also give rise to perspectival dependence?³⁴

First note that my core project is to establish that epistemic dependence is a richer phenomenon than has henceforth been appreciated. If it turns out that many different forms of speech generate these richer forms of perspectival dependence, that’s a result which is favourable to my overall project.

A second response is less concessive. It grants that speech like metaphors and slurs cue characterizations, but contrasts the ways in which narratives and other forms of speech cue characterizations along three different dimensions.

The first relevant dimension is the conventionality of characterization. As noted above, some characterizations—those associated with stereotypes, for example—are socially dominant and widely shared. Such characterizations are relatively easy to cue. It will be much easier for an extended discourse to build and cue a novel or socially marginal characterization than for a short fragment of speech to do the same. Thus, metaphors et. al, unless they are embedded in extended, coherent discourses, will not be well suited to generating novel or marginal characterizations.

The second, related dimension we ought to attend to is complexity. Camp stresses that individual characterizations can be knitted together into more complicated, open ended interpretive schemas. This knitting together will be more easily accomplished by an extended discourse.

The third relevant dimension is that of embeddedness. Above, I noted that which characterizations we draw on is a matter of cognitive habit. There is an important difference between getting someone to use a characterization on a ‘one off’ basis, and stably embedding that characterization in their cognitive repertoire. Sustained narratives allow speakers to guide and scaffold an audience’s use of a novel, complicated characterisation over a period of time, and so give speakers the opportunity to more-or-less stably embed a characterization in the audience’s cognitive repertoire.

We have now identified two features of narrative testimony responsible for generating perspectival dependence. Onto the third.

Story schemata

Compare the following discourse snippets:

• 13 One sunny day, Jenny ate breakfast later than usual. He grasped the sword, knowing that he might soon have to use it for the first time. A terrible explosion brought them all to their knees. They all lived happily ever after.

• 14 He grasped the sword, knowing that he might soon have to use it for the first time. They all lived happily ever after. A terrible explosion brought them all to their knees. One sunny day, Jenny ate breakfast later than usual.

Neither snippet is cohesive. Neither describes describes a coherent state of affairs. Both frustrate our expectations. Nonetheless, the former snippet is more intelligible than the latter, and much easier to read. Why? Answer: the former, despite having constituents whose semantic contents cannot be woven together into any coherent whole, nonetheless exhibits a recognisable structure: it has a canonical narrative arc. The second does not.

The ideology of story schemata allows this contrast to be made precise and tractable. A story schema is a tool with which we mentally represent events. They are generated by relatively stable expectations about the ways in which stories proceed. A story schema allows us to ‘sort‘ events. When we hear a narrative recounting a series of events, we sort those events by recognising which narrative function each event plays—for an example, an event might help to set the scene, introduce a character, introduce a goal, or introduce an obstacle. Having identified this narrative function, we then assign the event to the corresponding schema constituent.

Story schemata impose two kinds of structure on the events they help represent: vertical and horizontal. The vertical structure consists in a series of exclusive branches: each constituent c is connected to exactly one higher node, namely, the node which specifies a whole of which c is a part. Vertically, the organisation exhibited by a story schema resembles the familiar hierarchic constituent structure by which we represent the grammatical form of a sentence. However, the constituents of a story schema are organised horizontally as well as vertically. The horizontal organisation is serial: each event is related unidirectionally to another. (A good model for serial organisation is our mental representation of the alphabet.) We can now state the difference between (13) and (14). When we read (12) we are able to slot its content into a standard story schema. We cannot do the same with (14).

There is considerable evidence that story schemata are psychologically real. Narratives with recognisable schematic forms are better recalled than narratives missing certain constituents (Mandler 2014). Schematic structure patterns recall in more subtle ways too: central material from story constituents is better recalled than elaborations on said material (Mandler 2014). When aspects of a story are forgotten, confabulated material preserves the expected schematic structure (Mandler 2014). When two stories are presented in an ‘interleaved’ fashion, they tend to be separated out by memory (Mandler 2014). Reading times for sentences at the borders of schematically encoded episodes are longer than for sentences nested within said episodes (Mandler 2014).

Discursive testimony—testimony presented by way of a string of adjacent sentences—may or may not encode a story schemata. Compare the following two presentations of the same events:

• Annal

• 709. Hard winter. Duke Gottfried died.

• 710. Hard year and deficient in crops.

• 711.

• 712. Flood everywhere.

• 713

• 714.

• 715.

• 716.

• 717.

• 718. Charles devastated the Saxon with great destruction.

• 719.

• 720. Charles fought against the Saxons.

• 721.

• 722. Great crops.³⁵

• Narrative

• The years between 709 and 717 saw many hardships: there were floods, poor crops, and hard winters. In 718, the situation further deteriorated: the Saxons invaded. But the year after they invaded, Charles fought against them with great destruction. And after he won, there was a great harvest.

The feel and texture of the two presentations are strikingly different. In the annal:

‘...things happen, but they are neither definitely included nor decisively excluded...There are beginnings and cessations, but no genuine initiations and concluding. One thing replaces another, but does not absorb it and carry it on (Dewey 2005).³⁶

The difference, I suspect, is a product of the different story schemata cued by the two presentations. The annal suggests only a serial organisation: the events recounted follow on from each other as blandly as the letters of the alphabet. With the narrative, things are quite different; as well as being invited to organise the events serially, the reader is invited to organise them vertically, perhaps something like this:

Such an organisation is not forced upon us ‘by the events themselves’. A subtly tweaked narrative suggests a different configuration:

• Narrative Paraphrase

• The century began with a hard winter and Duke Gotfried’s death. Things soon became much worse: in 712, there was a terrible flood. Over the next ten years, Charles fought the Saxons several times, and was victorious at least once. But not until ten years after the flood did the land recover enough to produce great crops once again.³⁷

This suggests a schema more like the following:

This matters: subtle differences in how testimony is presented can suggest that the encoded information should be mentally organised in different ways. And the choices a hearer makes as to how to organise a narrative’s information have significant knock-on effects for their perspective.³⁸

Perspectival impact

By organising narratively encoded information in a particular fashion, story schemata influence each of the dispositions relevant to an agent’s perspective. Let us take these dispositions in turn.

Attentional. Story schemata funnel attention, most obviously towards the events positioned as central within the schema: the victory against the Saxons looms larger given the first narrative presentation; the flood given the second.

Inquisitive. There are two main ways in which our inquisitive dispositions will be shaped by story schemata. First, and most obviously, how our inquisitive dispositions are be activated depends on where our attention is directed. The second mechanism is more subtle. Consider Dewey’s remarks again—‘[t]here are beginnings and cessations, but no genuine initiations and concluding’. A schematically organised narrative ‘closes’ in way in which an annal does not.³⁹ Put picturesquely, narrative closure—which I analyse as an effect of story schematic structure—invites us to regard the narrative events as a sealed and complete whole. Given a more functional gloss, narrative closure is an erotetic phenomenon: successful narrative closure blocks the question, ‘But what happened next?’

Interpretive. Interpretive dispositions are directed in multiple ways by narrative schemata. Most obviously, the schemata license judgements of comparative importance and interest: the first narrative licenses the judgement that the defeat of the Saxons was a more important event than the flood, and the second licenses the converse.

More subtly, narratives license judgements about which states of affairs are ‘normal’, steady, and normatively unremarkable. Standard story schemata embed an assumption that the starting point of a narrative is as unremarkable background state. This will often license false inferences. Consider Meyers’ work on the narratives of horror and displacement that have emerged since the beginning of the genocide in Darfur (Meyers 2009). Meyers notes that canonical narrative forms license the inference that the pre-genocidal state of affairs was not an unjust or painful one. This is an inference licensed by structural features of such narratives: because the pre-genocidal state of affairs serve a particular narrative function—that, roughly, of scene-setting and character introduction—they are cast as having a particular normative character and affective valence.

The ideology of colouring serves as a useful tag for these dynamics.⁴⁰ Consider the difference between the sentence ‘The dog howled all night’ and ‘The cur howled all night’. The two sentences are often thought to share a semantic content, but differ at a more fine-grained level. Call this the level of colouring. Colouring is sometimes analysed as a matter of mental imagery and aspectual thought—perhaps ‘cur’ makes us imagine a mangier animal than ‘dog’ does. But sometimes a difference in colour is thought to be a matter of two sentences ‘lightly suggesting’ different propositions, as when the assertion that even John passed the test suggests that the test was particularly easy. Either way, my position can be put like this: sentences are coloured by their placement within a narrative. Consider a sentence like ‘Diana woke up and began to sharpen her knives’. Suppose this comes right at the start if a narrative. Given such a placement, it is likely to be identified as playing a scene-setting role, and thus taken to be describing a normal, and normatively unremarkable event. Suppose, on the other hand, that textual cues allow us to identify that sentence not as a scene-setting sentence, but as describing an initiatory event. The event takes on a more sinister, threatening cast: it has been coloured by its placement within a narrative schema.

The ideology of coluring also helps to bring out important differences between the pragmatic impact of narrative structure and more canonical cases of colouring.

Consider the following triple of sentences:

• 13 Even John passed the test.

• 14 John passed the test.

• 15 The test was easy.

One could endorse the semantic content of (13)—represented by ( 14 )—without endorsing (15), and vice versa.⁴¹ A mental representation of one need not be parasitic on a mental representation of the other. If one wants to accept only the uncoloured or ‘neutral’ content of (13), it is easy to do so: there are readily available semantic vehicles which allow for the mental representation of said neutral content.

The relationship between the coloured and the ‘neutral’ content encoded in a narrative is quite different. There is typically no readily available representational vehicle which would smoothly encode only the story’s ‘neutral’ content without also encoding its pragmatic content. One would have to strip from one’s mental representation of the story’s events of its schematic organisation and encode it merely serially instead. But this is a highly unnatural procedure. It is very difficult to find examples of mental representations of events which exhibit ‘pure’ serial structure (Mandler 2014) and likely cognitively costly to impose an entirely new story schema on a set of events. The upshot is that the colourings imposed on narratively recounted events—and so the corresponding perspectival dispositions—by a testifier are extremely ‘sticky’: it is hard for an audience to take on the ‘neutral’ story content without also adopting those colourings.

The upshot is this. Narrative testimony encodes story schemata. When audiences narratively defer to testifiers, they take on the story schemata encoded in the narrative testimony. These schemata have profound effects on the audiences’ ongoing mental representation of events: they pattern attention and inquiry, and they shape the communicated content’s inferential profile and affective valence.

Taking stock

I have identified three different ways in which narrative testimony organises and structures the information it encodes: narrative testimony cues representational formats and aspectual thought, and encodes story schemata. Each of these directs the three interlocking suites of dispositions constitutive of a perspective. By so directing the perspectival dispositions of audiences, narrative testimony effects a form of mental co-ordination far richer, more subtle and far reaching than mere opinional co-ordination. When a hearer structures and organises information as they are cued to by a narrative, they become thickly dependent on the testifier. Not only their opinions, but their way of looking at the world has been adopted from another.

One further point, only touched on so far, deserves further mention. That is the relationship between perspectival co-ordination and long-term opinional co-ordination. The way in which information is structured does not completely determine, but does influence how that information will tend to be extended and retained. Perspectival co-ordination can be thought of as a way to encourage long term opinional convergence: agents who structure information in the same ways are far more likely to remember the same things, and to draw the same inferences, than those who structure the same information differently.⁴²

Trust

When in receipt of simple testimony that p, one can entertain the communicated content without accepting it. With narrative testimony, things are more complicated. Acceptance of narrative testimony involves two things: first, acceptance of the content communicated by the narrative, second, structuring the communicated information in line with the narrative cues. Entertaining a narrative does not require that one accept the narrative’s content, but it does require that one structure the entertained information as the narrative suggests. Thus, when it comes to narrative testimony, the line between entertaining and accepting content is blurrier than it is with simple testimony. This blurriness, I argue, means that narrative and simple testimony interact differently with trust.

For simple testimony to successfully affect an audience’s picture of the world, the audience must trust the testifier prior to the testimonial exchange.⁴³ Narrative testimony is different. Narrative testimony can alter how an agent sees the world even when the audience does not trust the speaker.

A useful model for drawing out how narrative testimony can bypass trust is provided by Camp’s discussion of metaphor (Camp 2017b)⁴⁴. There are three important features of metaphor to note: its irresistibility, its authority, and its world-building character. Let’s explore what Camp calls irresistibility first. For a metaphor to be effective—that is, for a metaphor to produce an ‘image’ in the mind of the hearer—the hearer need not trust the speaker. Rather, ‘merely understanding [a metaphorical utterance] requires a hearer to mold his mind in the speaker’s image’ (Camp 2017b).

Second, the images produced by successful metaphors feel to the hearer like something they themselves have produced, rather than something ‘forced upon’ them by another. Because a successful metaphor really does bring me to ‘see’ its object thus and so, I need not rely on the speaker to find that it may be seen thus and so—rather, I have seen it for myself! Metaphorical images are thus well-placed to inherit authority not from the speaker, but from the hearer themselves. Call this the authority of metaphor.⁴⁵

A third important feature of metaphor is that metaphor cultivates intimacy between speaker and hearer (Camp 2017b; Cohen 1978). As Cohen has it, when a speaker utters a metaphor, the speaker offers her hearer a kind of ‘concealed invitation’ to interpret his remark; once the hearer accepts this invitation, this transaction constitutes the ‘acknowledgement of a community’ (Cohen 1978). Why? In part because metaphors are pragmatically dense: interpreting a metaphor typically requires mutual recognition of a host of nuanced, local conversational assumptions (Camp 2017b). And in part because of metaphors’ irresistibility and authority, which together render the boundary between understanding and acceptance more porous when it comes to metaphor than it is with other forms of speech; after all, even in interpreting a metaphor a hearer has already ‘mold[ed] his mind in the speaker’s image’ (Camp 2017b). It follows that even accepting the invitation to interpret a speaker’s utterance draws the speaker and hearer into a kind of cognitive intimacy. Thus metaphor is world building speech: metaphors draw speaker and hearer into a shared way of looking at the world without relying on pre-existing trustful relations.

My view is that narrative testimony, like metaphor, is world-building speech. Why? Well, like metaphor, narrative testimony is irresistible. Why? Because even entertaining a narrative requires that one actually structure and organise information in the way the narrative suggests. So even entertaining a narrative involves coming to see the world in a particular way. Second, like metaphors, narratives have authority. Once I entertain a narrative, I really have structured and encoded information thus and so.⁴⁶ I have thereby come to really inhabit a certain perspective on the world. Once I inhabit that perspective, the way things strike me from it are ways that they have struck me. The arrangement and patterning of information will feel like the products of my own cognition, and will acquire the corresponding authority. Third, narrative cultivates cognitive intimacy between speaker and hearer in much the same way as metaphor, for they share much of the latter’s pragmatically density as well as its irresistibility authority. And importantly, it generates this intimacy without requring pre-existing relations of trust, for the intimacy is largely generated not by acceptance of the narrative but instead by the cognitive work required for its being fully interpreted and entertained. Thus, narrative testimony is world building speech: it draws speaker and hearer into a shared way of looking at the world without relying on pre-existing trusting relations.

Conclusion

Epistemologists of testimony have thus far restricted their attention to cases of simple testimonial exchange. This paper widens the field of concern. I have argued that the epistemic effects of narrative testimony outstrip those of simple testimony by some margin. To accept narrative testimony is not only to adopt an opinion on the say-so of another, but to organise and structure one’s opinions on their say-so. The effects of such organisational choices are profound: they matter for how one’s attention is directed, how one’s inquisitive energies are inclined, and they shape one’s interpretive habits. In short, they shape one’s epistemic perspective. This insight is particularly significant given the potency of narratives: you can shape my persective even if I don’t trust you.

Once we realise that narrative testimony generates a more far-reaching form of epistemic dependence than that characteristic of simple testimony, new questions for normative epistemology emerge. When, if ever, is it permissible to accept, or to offer narrative testimony? Which, if any, interpersonal relations license making oneself perspectivally dependent on another? Can one ever be required to make oneself so dependent? And so on. Two examples make this especially vivid. First, consider Burge’s influential Acceptance Principle:

Acceptance Principle:

A person is a priori entitled to accept a proposition that is presented as true and that is intelligible to him, unless there are stronger reasons not to do so (Burge 1993).

The Acceptance Principle tells us that we may accept simple testimony so long as we have no special reason to suspect its source. But it does not license similarly generous position with respect to narrative testimony: accepting a narrative involves more than accepting propositions. It also involves structuring them in certain ways that influence the information’s ongoing inferential profile.

One might think it a relatively simple matter to tweak Burge’s principle:

Narrative Acceptance Principle:

A person is a priori entitled to accept a narrative that is presented as true and that is intelligible to him, unless there are stronger reasons not to do so.

But things are not so simple. Consider how Burge motivates his original Acceptance Principle. He argues that there are important conceptual connections between (i) the presentation of an intelligible message, (ii) the rationality of the source presenting the message, and (iii) the truth of the message presented. Obviously there is much to object to in such an argument. But its controversies are not my focus. The important point is that Burge’s argument cannot naturally be extended to generate the Narrative Acceptance Principle. It may be true that ‘prima facie rationality of... source indicate[s] a prima facie source of truth’. To establish the Narrative Acceptance Principle in a similar fashion, it would need to be true that prima facie rationality of... source indicate[s] a prima facie legitimate epistemic perspective. But we have no comparable reason to endorse this stronger claim.⁴⁷

Second, consider the role played by considerations of indispensability in the epistemology of testimony. The broadly anti-sceptical tendency in the epistemology of testimony is largely the product of such considerations. Fricker’s remarks are typical: to refuse testimonially based beliefs would be to give up on ‘most of our beliefs [about] geography, history, the natural and social sciences including medicine, and so forth’ (Fricker 2006). Skepticism about the epistemic status of testimonially based beliefs is taken to be unpalatable because it would beget just too much skepticism: we would turn out to know almost nothing, and be stuck knowing almost nothing. When it comes to narrative testimony, considerations of indispensability look different. It may be true that testimony per se is an indispensable epistemic resource. It is far less plausible that narrative testimony—and its attendant, deep forms of epistemic dependence—are ineradicable features of our epistemic lives.

Burge and Fricker are merely prominent examples: they provide normative frameworks for evaluating simple testimony that cannot naturally be extended to its narrative instances. This should not surprise us. To accept an opinion on the say-so of another is a kind of surrender of one’s epistemic autonomy (Fricker 2006). The surrender of epistemic autonomy involved in accepting a narrative on someone else’s say so is deeper and further reaching. We should not expect norms engineered in the face of the former to help us in the face of the latter.