1 Introduction

Consider the two passages below:

(1) The first speaker was her Uncle Tom. She knew the naive candour covering the girding and savage misery of his soul. Who was the other speaker? Whose voice ran on so easy, yet with an inflamed pulse? (The Rainbow, D.H. Lawrence, 1988, cited by Fludernik, 2003: 88)

(2) She gazed back over the sea, at the island. But the leaf was losing its sharpness. It was very small. (To the Lighthouse, Virginia Woolf, 1970, cited by Brinton, 1980: 371)

The final sentences in (1) and (2) are most naturally inferred to be reports of a fictional character’s consciousness: in example (1) they are interpreted as the character’s thoughts, whereas in example (2) they are understood as representing the character’s (visual) perceptions. The two examples illustrate two different levels of representation that are ubiquitous in narratives. One concerns the representation of speech, i.e., a linguistic object, be it an actual utterance or thought/inner speech. The other involves the representation of low-level, non-linguistic mental states, such as perceptual experiences. Crucially, the final sentences in both examples lack an overt embedding construction that would clearly indicate who the thinker/perceiver is (as in ‘She wondered who the other speaker was’ or ‘She saw the leaf losing its sharpness’). Because such a cue is absent, these passages are, in principle, ambiguous: they can be attributed to a story character and therefore receive a ‘shifted’ interpretation, or they can take the standard, speaker-oriented interpretation according to which they are attributed to the narrator, the ‘speaker’ of the story (no actual shifting is involved in this case).1

Even though explicit cues about whose perspective is expressed are often lacking, readers constantly make inferences while engaging with fiction and are able to make sense of a story. How do readers then determine whose consciousness is being reported in such ambiguous statements in narratives and what are the textual and/or pragmatic cues that they rely on during this process?

This paper presents four experiments that explore certain discourse- and narrative-related factors that potentially influence the perspectivisation process in narrative texts. Such factors include different kinds of protagonist prominence, as well as different kinds of narration that are associated with differences in the narrators’ epistemic access. Exploring these factors can give us a better view of the constraints and possibilities of perspective taking in narrative discourse.

2 Theoretical Background

2.1 The distinction between thought and perception reports

Earlier studies in literary research like Brinton (1980) and Banfield (1982) have discussed the linguistic differences between speech and perception, especially in the context of third-person narratives written in so-called ‘free indirect style’. Brinton and Banfield use the terms ‘represented speech and thought’ and ‘reflective consciousness’, and the terms ‘represented perception’ and ‘non-reflective consciousness’ to refer to representations of speech and perception, respectively, of the sort illustrated in examples (1) and (2).2 What is particularly relevant for the current purposes is Banfield’s observation that questions and exclamations are indicators of reflective consciousness, they introduce a thought act, an inner utterance. On the other hand, perception representations are representations of sensory input, i.e., low-level information, and as such they do not entail reflection on the part of the character. As a result, no utterance is involved. The presence of exclamations or question marks thus distinguishes, according to Banfield, represented thought from represented perception. This is illustrated in examples (3a) and (3b):

(3) a. And on the station platform was Clifford on crutches.
  b. And on the station platform — Oh God! — was Clifford on crutches? (Banfield, 1982: 204)

While (3a) can be read as a description of what a protagonist saw on the station platform, the exclamation and question that are present in example (3b) can only be interpreted, according to Banfield, as the protagonist’s inner speech.

Instances of represented thought like (1) and (3b) illustrate the phenomenon of Free Indirect Discourse (FID) which has been the object of extensive study in narratology and linguistics. FID is a vivid style of reporting a character’s thoughts that combines grammatical features of both direct and indirect discourse. More specifically, pronouns and tenses are interpreted relative to the context of utterance, i.e., the narrator’s context, while temporal and locative adverbials indicate the character’s spatio-temporal location. This is shown in example (4), where the indexical adverb here refers to the protagonist Harry’s location rather than the narrator’s:

(4) The reflections did not fade and he looked and looked until a distant noise brought him back to his senses. He couldn’t stay here, he had to find his way back to bed. (Harry Potter and the Sorcerer’s Stone, J. K. Rowling: 167)

Similarly, the character’s perspective is reflected in all other expressions and constructions used in FID that are also found in direct discourse: subjective expressions, questions, exclamations, interjections, idiolects and incomplete sentences (for an exhaustive discussion on the linguistic characteristics of FID, see Banfield, 1982, and Fludernik, 2003). These special grammatical features have recently captured the interest of semanticists who have developed different semantic theories of FID: bicontextual analyses (Eckardt, 2014; Schlenker, 2004; Sharvit, 2008) and mixed quotation theories (Maier, 2015).

Unlike in literary theory where FID is often taken as a special stylistic device of reporting a character’s consciousness in general, be it speech, thoughts, emotions or perceptions, semanticists usually discuss FID as a form of reported (inner) speech.3 In a recent study though, Hinterwimmer (2017b) argues that apart from FID, there is a different kind of perspective shifting which he terms Viewpoint Shifting and involves the representation of the mental or perceptual state of a character. In this respect it contrasts with FID, as Viewpoint Shifting is the reporting of objects that are non-linguistic in nature. Viewpoint Shifting is illustrated in (5):

(5) When Mary stepped out of the boat, the ground was shaking beneath her feet for a couple of seconds. (Hinterwimmer, 2017b: 291)

The second clause represents what Mary sensed when she got out of the boat, without it being necessarily interpreted as a report of her inner thoughts. If readers do not accept that the ground was actually shaking in the story, then the content of the related proposition receives a shifted interpretation: it is interpreted as true only with respect to the protagonist’s mental state.

Importantly, Hinterwimmer argues that Viewpoint Shifting and FID are not only conceptually but also grammatically different. First, he argues that in Viewpoint Shifting non-pronominal indexicals do not shift to the character’s context as is the case in FID. Second, he argues that unlike FID, Viewpoint Shifting is available within the sentence: for instance, in (5), Viewpoint Shifting is introduced after a when-clause. By contrast, FID refers to a speech/thought act; as such it can only be introduced at the root level, and not within complex sentences. This is shown in (6):

(6) *When Mary heard a song by Kendrick Lamar that she liked on the radio on her way home, she would buy his new album tomorrow. (Hinterwimmer, 2017b: 297)

Viewpoint Shifting and FID are licensed pragmatically, via the accommodation of a related event (perceiving event or speech/thought act for Viewpoint Shifting and FID, respectively) on the part of the reader/hearer. According to Hinterwimmer, data like the above motivates the need for two distinct semantic analyses of the two phenomena (see also Maier, 2019).

Abrusán (2018, 2019) proposes a distinct approach.4 She puts forward the view that FID and Viewpoint Shifting are different facets of the same phenomenon (for an account along similar lines, see Stokke, 2013). She argues that their different grammatical characteristics (such as the shifting behavior of indexicals) arise from pragmatic inferences about the distinct properties of speech and perceiving events. More specifically, she suggests that depending on the event that is retrieved in the discourse, different features are salient. According to Abrusán, only speech events have salient space and time coordinates, which as a result licenses the shifting of locative and temporal indexicals as it is observed in FID.

It is important to note that FID and Viewpoint Shifting are discussed in semantics as perspective shifting phenomena and concern cases where speech/thought and perception representations take shifted, character-oriented interpretations. However, as mentioned in the Introduction, speech and perception representations that occur unembedded admit multiple interpretations in terms of perspective. As I am interested not only in shifted, but also in standard, speaker-oriented interpretations, I will use the more straightforward and general term ‘free report’.5 In particular I will use the term ‘free thought report’ to cover phenomena that have previously been discussed as FID, reflective consciousness or represented speech and thought. Similarly, I will use the term ‘free perception report’ to cover phenomena previously referred to as Viewpoint Shifting, non-reflective consciousness or represented perception. The terms ‘free thought’ and ‘free perception report’ emphasise the lack of overt embedding and make clear what kind of representation is expressed, i.e., a speech/thought act as opposed to a perceptual experience. Furthermore, the particular term is neutral with respect to whether the phenomenon termed as ‘Free Indirect Discourse’ is a form of Direct or Indirect Discourse. Additionally, while especially in formal semantics FID is discussed as a phenomenon occurring in third-person narratives, here first-person narratives are also considered. Lastly, ‘shifting’ will be used more broadly to refer to character-oriented interpretations and will not be used in the strict semantic sense where it usually refers to context shifting. Finally, the perceiver/thinker of the content of free reports will be henceforth referred to as the ‘perspectival centre’ or ‘anchor’ for the related proposition. The former term is borrowed from Hinterwimmer (2017a) who uses it specifically for the individual that is taken to be the agent of an (inner) speech event conveyed by a passage in FID.

Speech and perception reports have also been investigated empirically in narratological and linguistic research. FID has received special attention (Bortolussi and Dixon, 2003; Bray, 2007; Dixon and Bortolussi, 2018; Kaiser, 2015; Salem, Weskott, and Holler, 2017; Salem, Weskott, and Holler, 2018; Sotirova, 2006). Among other things, studies on FID have been concerned with the question of whether FID enhances taking a character’s perspective (Abrusán, 2018; Kaiser, 2015; Salem et al., 2017). As for perception reports, they seem to have been studied less extensively but there is recent empirical work on how they are processed and attributed (Dixon and Bortolussi, 2018; Van Krieken, 2018). To my knowledge though, free thought and perception reports have not been directly compared empirically.

2.2 Anchoring free reports: pragmatic factors

Given that free reports are abundant in narratives and that it is often left unspecified as to whose perspective is expressed, it is natural to ask what cues there are in a text to guide readers’ interpretations as to who the perspectival centre for a statement is. Who do readers take to be the thinker of an utterance or the perceiver of a perceptual description?

Hinterwimmer (2017a) focuses on the conditions that make a character a plausible perspectival centre for free thought reports (FID, in his terms). His proposal is that a protagonist has to be ‘locally’ or ‘globally’ prominent in order to be available as perspectival centre. If a protagonist is mentioned as the experiencer of an eventuality in the sentence preceding the free thought report, (s)he is said to be locally prominent. In (7), the experiencer is Susan, the agent of the looking event. That makes her available as the agent of the thinking event that is conveyed by (7a). By contrast, continuation (7b), whose content can only be interpreted as George’s thought, is less felicitous. This shows that George is not available as perspectival centre, which, according to Hinterwimmer, is due to the fact that George is not locally prominent.

(7) Susan looked at George hatefully.
  a. The dumb jerk had managed to make her look like an idiot at the meeting yesterday.
  b. #The mean old hag had managed to make him look like an idiot at the meeting yesterday. (Hinterwimmer, 2017a: 6)

The second condition that makes a protagonist prominent as perspectival centre for free thought reports is global prominence. A protagonist is said to be globally prominent if (s)he is the discourse topic. In (8), George is made globally prominent by virtue of being mentioned in the opening sentence, which makes him available as the anchor for the last sentence.

(8) George entered the room and looked around cautiously. Susan was sitting at a table in the corner with her best friend. Susan looked at George hatefully. The mean old hag had managed to make him look like an idiot at the meeting yesterday. (Hinterwimmer, 2017a: 7)

The prediction that follows from Hinterwimmer’s proposal is that if both a globally and a locally prominent protagonist are present, a free thought report could be ambiguous, as it could be anchored to either protagonist.

The question that arises is whether, in cases of such ambiguity, there is a stronger preference for either globally or locally prominent protagonists as anchors for a free report. Let us first discuss free perception reports. The most natural prediction is that they are anchored to the experiencer of the perceiving event mentioned or accommodated in the discourse (Abrusán, 2018; Abrusán, 2019; Hinterwimmer, 2017b).

Attributing an expression or a statement to some entity is sensitive to additional factors. In a study that investigated the strength of different viewpoint markers (i.e., expressions indicating a character’s perceptual or mental state), Van Krieken (2018) found that readers were more likely to attribute a scene description (‘Outside the sun was shining’) to the character, as opposed to the (impersonal) narrator, when the previous sentence made explicit reference to a perceiving event whose agent was the character, compared to cases where some other event with the same character as agent was mentioned (e.g., ‘Peter looked/stood in front of the window’). This corroborates the view that local prominence, in terms of a protagonist being the experiencer of an event, is important for the attribution of perspective.

In a different study, Kaiser (2018) investigated whether readers were more likely to attribute statements including predicates of personal taste to the narrator (who was referred to via first-person indexicals) or to the character, depending on what kind of sensory modality was involved. The items were two-sentence sequences like the following:

(9) When I came into the room, Eliza put/saw/smelled/tasted the muffin on the platter. It was/looked/smelled/tasted disgusting.

She found that readers were significantly more likely to pick the character as the judge of a predicate of personal taste when a sensory modality was specified, that is, when the character was mentioned as the experiencer of the relevant perceiving event, compared to the baseline category where no such modality was explicitly involved. Additionally, the kind of sensory modality also affected who would be chosen as the attitude holder. Overall, the above-mentioned findings underline the importance of the local prominence of a character (as defined in Hinterwimmer, 2017a) as a contextual factor in the process of identifying the anchor for a particular expression of subjective experience.

One additional factor that can affect how the perspectival centre for free reports is determined is the type of narration. For instance, a story can be told from a first- or a third-person point of view. Previous studies have explored questions related to how narrative mode affects engagement with a story, for instance, what effects first- and third-person narration have on the way that readers empathise with a story character (see background discussion and experiments in Salem, Weskott, and Holler, 2017). From a linguistic perspective and regarding the anchoring of free reports, it is logical to assume that when the speaker is explicitly introduced via a pronoun, as in (9), her status becomes quite salient, especially in short stretches of discourse, and it is easier to interpret the perception report “It was/looked/smelled/tasted disgusting” from the speaker’s perspective. If a speaker is not explicitly mentioned, as in (10), Eliza is quite prominent, which may increase the possibility of interpreting the perception report from Eliza’s perspective.6

(10) When Eliza came into the room, she smelled the muffin on the platter. It smelled disgusting.

Regarding free thought reports, to my knowledge there have been no empirical studies taking into account FID in first-person narratives. FID is usually discussed as a phenomenon occurring mainly in third-person narratives and this is also observed in empirical studies on the processing of FID which use third-person stories in their experimental setting or, at least, items where the speaker is not explicitly referred to (see Bortolussi and Dixon, 2003; Bray, 2007; Kaiser, 2015; Salem, Weskott, and Holler, 2017; Salem, Weskott, and Holler, 2018; and Sotirova, 2006). However, FID can also occur in first-person stories, when the character-narrator narrates personal experiences and her corresponding thoughts and reflections as these occurred in some past moment (for more discussion, see Cohn and Cohn, 1978; Fludernik, 2003; Nielsen, 2004; and Stanzel, 1986). See example (11) for an illustration:

(11) I gave a start: the lights had gone on, activated by a photo-electric relay; the sun had set. What would happen next? I was so tense that the sensation of an empty space behind me became unbearable. (Solaris, S. Lem: 15)

However, I am not aware of studies investigating specifically whether FID can occur in first-person narratives to signal the thoughts of a character that is different from the narrator.7

3 Research Questions and Predictions

Based on the above studies, the following questions will be explored. First, how does the type of narration affect how free reports are anchored? Second, do free perception and thought reports differ in the extent to which they trigger shifted interpretations? Third, what role do global and local prominence play when determining the epistemic anchor of free reports?

Regarding the first question, I hypothesise that the type of narration affects how free reports are anchored. Overall, a stronger preference for speaker-oriented interpretations in first-person narratives is expected, as well as a preference for shifted interpretations in third-person narratives with one or multiple salient protagonists. As noted previously, at least in semantics, FID and free perception have been mainly discussed in the context of third-person narratives, i.e., narratives with an impersonal narrator that is not a character in the story and does not participate in the narrated events (for instance, see J.K. Rowling’s Harry Potter series).8 In third-person narratives, there is one or multiple salient protagonists whose perspective is foregrounded in the text, while the narrator is more backgrounded or ‘effaced’ (cf. Chatman, 1980). Additionally, third-person narratives have no or scarce reference to the narrator through first-person indexicals,9 whereas in first-person stories (such as J.D. Salinger’s The Catcher in the Rye), the narrator is a fictional character that participates in the events. For semantic theories of perspective, the interest lies in what contexts allow non-speaker-oriented interpretations, given that the speaker of an utterance is usually taken as the default anchor. Subsequently, much recent work in semantics has investigated shifted interpretations of so-called ‘perspective-sensitive items’ (Bylinina, McCready, and Sudo, 2014) that are standardly considered speaker-oriented, e.g., appositives, expressives and predicates of personal taste, as well as the environments and the pragmatic conditions that license shifts to the perspective of an entity other than the speaker (see Harris and Potts, 2009; Kaiser, 2015; Lasersohn, 2005; and Stephenson, 2007). Fictional narratives provide an ideal context for investigating non-speaker-oriented interpretations—which are characteristic of FID, for instance—since often story characters are more salient than the narrator, i.e., the ‘speaker’ of a story, hence they are better candidates as perspectival centres. In FID, shifted interpretations arise of course not only for particular expressions like the above-mentioned ones, but also for speech acts like exclamations and questions (for more details on exclamatives in FID, see Eckardt, 2014).

The second question has to do with whether free perception and free thought reports differ as to which perspectival centre they are anchored to in different types of narrative. As mentioned above, questions and exclamations occur in FID passages in third-person narratives and can trigger shifting to a character’s perspective. Importantly, according to Banfield (1982), these constructions indicate inner speech. As a result, they are expected to be interpreted as speech acts made by the protagonist. By contrast, I presume that shifting to the third-person protagonist will not occur in first-person narratives to the same extent: if a first-person narrator is construed realistically, i.e., as an entity with human-like properties, then she is not expected to have access to other characters’ thoughts (see Stanzel, 1986, ch. 7, for related discussion). Hence, in first-person narratives exclamations and questions are expected to have the default, speaker-oriented interpretation instead, as in standard communicative discourse, i.e., to function as signals of the narrator’s thoughts.

As for perception descriptions, the following assumption is made, in line with Kaiser (Kaiser, 2018). Perceptual experiences, especially visual ones, can be shared among individuals to a certain extent. As a consequence, in first-person narration, and in case the narrator is assumed to be part of the described scene, a description can be interpreted as a perception of either the narrator or the protagonist or both, depending on who is actually mentioned as the experiencer of the perceiving event, as findings in Kaiser (2018) and Van Krieken (2018) have shown. In more detail, passages including predicates of personal taste (which allude to a perceiving event, see Bylinina, 2014) but without constructions like questions and exclamations, are expected to be understood by readers merely as perception descriptions and not necessarily as thoughts expressed linguistically (compare ‘She was beautiful’ vs. ‘Was that indeed her? Oh, how beautiful she was!’). Such predicates are expected to have a speaker-oriented interpretation and be epistemically anchored to the narrator in first-person passages at least when the narrator is understood to be a perceiver, otherwise they will be anchored to the experiencer of a previously mentioned perceiving event. In third-person narratives, free perception reports are predicted to be ascribed to the main protagonist due to her prominent status in the story.

Concerning the third question, I aim to investigate if global or local prominence has a stronger effect on readers when they identify the anchor for a given statement, and also what role narration type plays. For instance, if a character other than the narrator is globally salient in a first-person narrative, does this salience override the general preference to take the speaker as the anchor for speech acts or perspective-sensitive expressions? In first-person narratives, when two characters are locally prominent in terms of both being perceivers (e.g., ‘We looked at the floor’), the preferred anchor is predicted to be the narrator, given the general speaker-bias. In third-person narratives with two perceivers (e.g., ‘They looked at the floor’) and an impersonal narrator, the anchor is expected to be the globally prominent character instead, given that the effect of local prominence in this case is ‘neutralised’.10

4 Experimental Studies

Four forced-choice task experiments tested the effect of different factors on readers’ perspectivising free reports. All experiments were made in Qualtrics and had a 2x2 within-subject design. Participants received the same instructions in all experiments. Data was analysed with generalised mixed-effects logistic regression models in R 3.6.1 (R Core Team, 2019), function glmer(): (lme4 package, Bates, Maechler, Bolker and Walker, 2015). The best models were determined via stepwise model comparison on the basis of AIC. In experiments 1, 2 and 3 the dependent variable was the response, with ‘narrator’ as the reference level.11

4.1 Experiment 1: Perception vs. Thought in First- and Third-person narration

Experiments 1 and 2 were conducted in English and were run on Amazon Mechanical Turk, an online crowdsourcing platform. Experiment 1 tested the following questions: do free thought and free perception reports differ with respect to triggering readings in which the character is the anchor? Does this difference depend on the kind of narration in which the story is written, that is, on whether the story is a first-person or a third-person narrative?

4.1.1 Methods: Design, Materials, Participants

The design consisted of the factors Narration Type (First-person/Third-person) and Report Type (Free Perception/Free Thought). The experiment used a two-alternative forced-choice task. In the instructions, participants were told that they would read short passages of fictional stories and that after reading each passage they would have to answer a question related to whose opinion they thought was expressed in a given statement: the narrator’s or the character’s. A test item is shown in (12). The statement in question contained a predicate of personal taste which was included in the free report mentioned in the story.

(12) The whole house was empty and silent.
  First-person: Fred and I stepped into the kitchen.
  Third-person: Fred and Sally stepped into the kitchen.
  Fred looked at the food that was left on the tables.
  Free Perception: It looked disgusting.
  Free Thought: Ugh, it looked disgusting! How many days had it been there?
 
  Whose opinion is it that the food on the tables looked disgusting?
  – The narrator’s
  – Fred’s

The items consisted of 3 to 7 sentences, and had a similar discourse structure as Kaiser (Kaiser, 2018): the free report included a predicate of personal taste, and the preceding sentence referred to a perceiving event whose agent was a character mentioned in the story via the use of a proper name. The items included different sensory modalities (vision (16 items), hearing (8), smell (4), taste (4)) and 27 different subjective predicates (e.g., charming, terrible, delicious). The free report included a perception verb that was always in the past tense (looked, sounded, smelled, tasted) except for 8 items in which using the copula sounded more natural.

For the Free Perception condition simple indicative sentences were used. The Free Thought condition included various FID cues: mainly questions (22 items) and exclamation marks (10), and in fewer cases, expressives (‘Damn’), interjections (‘No’, ‘Wow’) and ellipses.12

A third option indicating that a certain opinion is shared by both the narrator and the character was deliberately not provided in any of the experiments, so that participants had to make a particular choice. A ‘both’ response would be hard to interpret. Although it could mean that both characters are taken as anchors, it could also be a participant’s quick choice in case of uncertainty. Furthermore, even if participants actually have a stronger preference about who the anchor is, they may accept that a different interpretation is also possible. As a result, they may be tempted to choose both the narrator and the character as anchors. However, in this case this choice would not reflect participants’ real preference, which is the point of interest.

A Latin Square design was used to distribute the items over four lists. Participants saw eight target items in each of the four conditions resulting in 32 target items. Ten control items were also used to check participants’ attention. The controls were first-person stories where it was unambiguous whose opinion was expressed: the narrator and the other character held different opinions. Half of them were supposed to elicit narrator responses and the other half character responses. The total of 42 items were presented in random order and each participant was randomly assigned to one of the four lists. The experiment was designed on Qualtrics, an online survey platform. Responses were collected from 40 English native speakers. Participants were compensated with $1.50 for their participation.

4.1.2 Results and Discussion

Data from 28 participants (11 female, mean age 38.6, age range 24-62) were analysed. Twelve participants were excluded from the analysis as more than 25% of their responses to controls were incorrect. Figure 1 shows the percentage of character responses.

The best model was determined via backward stepwise comparison and included Narration Type as fixed factor, by-subject random slopes for Narration Type and random intercepts for items. A main effect of Narration Type was found (β = 3.0683, SE = 0.5684, z = 5.398, p < .001, reference level: First-person). No interaction between Narration Type and Report Type was found. Report Type turned out to be not a significant predictor and was removed from the model.

In line with the predictions, the results show that participants were significantly more likely to choose the character as the perspectival centre of a free report when they read third-person narratives (they chose the character 95% of the time in the Third-person narration condition). This was expected, given the assumption about the narrator’s more backgrounded, less salient status in third-person narratives and the character’s overall prominence. The FID cues in the Free Thought condition in third-person narratives did not increase the likelihood of shifting to the character’s perspective. In general, reading a sentence in the Free Perception or Free Thought condition made no difference as to how readers perspectivised the relevant statements, contrary to what was expected. For both perception and thought reports in first-person narration, readers chose the character approximately 70% of the time (intercept: First-person, β = 1.0764, SE = 0.3241, z = 3.322, p < .001), which reflects a preference for the character.

For the Free Perception condition such findings were expected in both kinds of narrative, given similar results in Kaiser (2018) and Van Krieken (2018) and based on the assumptions about shared perceptual experiences. However, the preference for character responses in the Free Thought condition in first-person narratives was contrary to expectations. Even though the character was the only explicitly mentioned experiencer, the presence of the narrator in the scene, combined with cues like exclamations and questions, was expected to trigger more narrator responses. The results suggest that a character’s local prominence in terms of being the experiencer of the perceiving event influenced readers: readers attributed the subsequent free report to that character, regardless of the report type.13 Results also suggest that considerations related to the first-person narrator’s epistemic access may be less relevant. What seems to be more important instead is which person in the story is explicitly mentioned as having perceived the event and hence whose reactions to this event are reported.

4.2 Experiment 2: Local Prominence and Speaker Preference

Experiment 2 was conducted as a follow-up based on the findings of experiment 1 for the first-person narration condition and tested the effect of local prominence and speaker preference on shifting: if the narrator and the character are both mentioned as agents of a perceiving event, hence if they are equally locally prominent, will readers ascribe the relevant statement to the narrator instead?

4.2.1 Methods: Design, Materials, Participants

The design consisted of the factors Report Type (Free Perception/Free Thought) and Experiencer (Character/Narrator and Character). I used the items from the first-person narration condition of experiment 1 and varied the subject of the perception verb: it was either the character or both the character and the narrator. An example is shown in (13). Data from 38 English native speakers were collected. Participants were compensated with $1.50 for their participation.

(13) The whole house was empty and silent. Fred and I stepped into the kitchen.
  Character: Fred looked at the food that was left on the tables.
  Narrator and Character: We looked at the food that was left on the tables.
  Free Perception: It looked disgusting.
  Free Thought: Ugh, it looked disgusting! How many days had it been there?
 
  Whose opinion is it that the food on the tables looked disgusting?
  – The narrator’s
  – Fred’s

4.2.2 Results and Discussion

The analysis included results from 26 participants (12 female, mean age 43.5, age range 28–72). Twelve participants were excluded as more than 25% of their responses to controls were incorrect. Figure 2 shows the percentage of character responses.

The optimal model was determined via backward stepwise comparison and included the fixed factor Experiencer, by-subject random slopes for Experiencer and random intercepts for items. No interaction between Experiencer and Report Type was found. Report Type did not significantly improve model fit and was removed. There was a main effect of Experiencer (β = –4.2051, SE = 0.5667, z = –7.420, p < .001, reference level: Character) showing that participants were significantly less likely to choose the character as the perspective holder when both the narrator and the character were the explicit experiencers. In this case, participants chose the narrator approximately 95% of the time, in line with the predictions. This indicates that in contexts where the narrator and the character are both equally locally prominent, there is a very strong preference for speaker-oriented interpretations. In other words, when local prominence is factored out, the narrator is the preferred anchor. Contrary to what was expected, participants’ responses once again were not found to be affected by report type.

The mean rates of character responses in the Free Perception–Character condition and in the Free Thought–Character condition are approximately 55% (at chance level). Remember that these conditions are identical with the conditions Free Perception–First-person narration and Free Thought–First-person narration from experiment 1, respectively, whose mean rates were around 70% (see Figure 1).

Figure 1 

The percentage of character responses per condition for Experiment 1 (error bars show +/– 1 SE).

Figure 2 

The percentage of character responses per condition for Experiment 2 (error bars show +/– 1 SE).

To see what strategy led to this pattern of mixed responses, we visualised the mean rates of character responses per participant for the conditions Free Perception–Character and Free Thought–Character (Figure 3). A lot of individual variation is observed in these results: some participants have a general preference for either the narrator or the character in both conditions, while other participants’ means vary around 50%. As an anonymous reviewer points out, such variation may be due to the nature of the experimental task. Participants may have considered both the narrator and the character to be licit anchors but they were forced to make a particular choice. As a result, some of them may have alternated their responses between ‘narrator’ and ‘character’ while others may have consistently chosen one option (see more discussion on the experimental design in section 5). Another possible explanation for the lower means in the respective conditions could be related to factors particular to each experiment, such as the influence of the other conditions. For instance, the design of experiment 1, where the explicit experiencer was always a character other than the speaker, may have led to an overall tendency to choose that character as the anchor. By contrast, all items of experiment 2 were first-person narratives. Potentially then, the narrator had a prominent status in readers’ mental representations during the experiment, which may be responsible for the fewer character responses observed here.

Figure 3 

Mean rates of character responses per subject for the conditions Free Perception (FP)–Character and Free Thought (FT)–Character for Experiment 2. (0 = narrator, 1 = character).

4.3 Experiment 3: Global Prominence and Speaker Preference

The findings of experiment 2 give rise to the following question: will the preference for the narrator be as strong if the other protagonist has a more prominent status in the story, i.e., if the other protagonist is globally prominent? What about third-person narratives with an impersonal narrator and both a globally prominent and a secondary, less salient character? These questions are investigated in experiments 3 and 4 that were conducted in Greek. The experiments were run simultaneously over the internet. Each participant was randomly assigned to either experiment 3 or experiment 4 so that no participant did both experiments.

Given that Amazon Mechanical Turk is sometimes considered a ‘noisy’ tool, experiments 3 and 4 were conducted in Greek mainly because we expected to collect more reliable responses with approximately the same or greater number of participants. This was indeed the case as only two participants were excluded from experiment 3 on the basis of their responses to controls, and none from experiment 4.

Experiment 3 tested the effect of global prominence on perspectivisation in first-person narratives. The goal was to explore if global prominence of another character would suppress speaker preference and if that effect would differ in perception and thought reports.

4.3.1 Methods: Design, Materials, Participants.

The design consisted of the factors Report Type (Free Perception/Free Thought) and Global Prominence (Global Narrator/Global Character). In all conditions local prominence was kept constant: both the narrator and the character were the agents of the perceiving event in question (looking (12 items), hearing (8), smelling (4)). We manipulated who the globally prominent character (global) was: it was either the narrator or the other protagonist. Global prominence was manipulated by making the relevant character the grammatical subject of the first sentences of the passage. The sentences referred to actions and/or feelings whose agent was that character. The other character was made less prominent (secondary): (s)he was mentioned less in the story and usually in object or adjunct position.

Items consisted of six to 11 sentences. Participants saw six items in each of the four conditions resulting in 24 target items, as well as 16 controls. The free report contained a copula verb in 19 items because in Greek it sounds much more natural compared to perception verbs. The total of 40 items were presented to participants in random order. (14) is an example of a target item. Responses were collected from 27 participants.

(14) Global Narrator: Ξύπνησα πολύ νωρίς εκείνη την ημέρα. Δεν είχα κοιμηθεί καλά όλο το βράδυ, είχα πολύ άγχος όλη την εβδομάδα και δεν μπορούσα να ησυχάσω. Ξύπνησα την Ντίνα καθώς ήθελα παρέα για να πάρω το πρωινό μου.
  Global Character: H Ντίνα ξύπνησε πολύ νωρίς εκείνη την ημέρα. Δεν είχε κοιμηθεί καλά όλο το βράδυ, είχε πολύ άγχος όλη την εβδομάδα και δεν μπορούσε να ησυχάσει. Με ξύπνησε καθώς ήθελε παρέα για να πάρει το πρωινό της.
  Στην κουζίνα υπήρχε μια έντονη μυρωδιά. Κοιτάξαμε τον νεροχύτη και το πάτωμα.
  Free Perception: To θέαμα ήταν αηδιαστικό. Υπήρχε μια στοίβα άπλυτα πιάτα, ποτήρια και αποφάγια από το περασμένο βράδυ.
  Free Thought: Τι αηδιαστικό θέαμα… Αυτά τα παιδιά, πάλι είχαν αφήσει ένα σωρό άπλυτα πιάτα, ποτήρια και αποφάγια από το περασμένο βράδυ!
  Επιπλέον η βρύση έτρεχε και το νερό είχε φτάσει μέχρι και το πάτωμα.
 
  Σύμφωνα με ποιον/-α ήταν αηδιαστικό το θέαμα στην κουζίνα; Σύμφωνα με τον/την αφηγητή/-τρια/την Ντίνα
 
  Global Narrator:I woke up very early that day. I hadn’t slept well all night, I had been very stressed during the whole week and couldn’t rest. I woke up Dina as I wanted some company to take my breakfast.
  Global Character:Dina woke up very early that day. She hadn’t slept well all night, she had been very stressed during the whole week and couldn’t rest. She woke me up as she wanted some company to take her breakfast.
  There was a strong smell in the kitchen. We looked at the sink and the floor.
  Free Perception:The sight was disgusting. There was a pile of unwashed dishes, glasses and leftovers from the previous night.
  Free Thought:What a disgusting sight… Those kids, they had again left a lot of unwashed dishes, glasses and leftovers from the previous night!
  On top of that, the tap was running and the water had reached the floor.
 
  According to whom was the sight in the kitchen disgusting?
  According to the narrator/Dina

4.3.2 Results and Discussion

Figure 4 shows the percentage of character responses. Data from 25 participants was analysed (16 female, mean age 35.3, age range 20-58, Greek native speakers) as two participants were excluded from the analysis (more than 25% of their responses to controls were incorrect).

Figure 4 

The percentage of character responses per condition for Experiment 3 (error bars show +/– 1 SE).

The optimal model was determined via forward stepwise comparison and included Report Type as a fixed factor and random intercepts for subjects and items. A main effect of Report Type was found (β = 1.1892, SE = 0.5064, z = 2.348, p < 0.05, reference level: Free Perception). However, the character responses are very few (only 4% of all responses, given by 10 participants), which does not allow for any generalisations about the effect of the report type. The results are straightforward: the narrator is chosen as the perspectival centre 93% or more of the time across all conditions, even in the presence of a globally more prominent character. This suggests that when the narrator is locally prominent, (s)he becomes the most preferred perspectival centre. The results are not surprising given the findings of experiment 2. The intermediate sentence referring to an event with the narrator being always locally prominent may cancel out any effect of global prominence. It is also possible that the other character’s global prominence could not be established in these items due to their small size. This leaves open the possibility that in longer first-person narratives with a more backgrounded, peripheral narrator that is not the main character in the story, the global character will have a more prominent status and therefore be a more plausible anchor.14

4.4 Experiment 4: Global and Local Prominence

In this experiment I tested the effect of local prominence in third-person narratives with two protagonists while maintaining the status of global prominence constant.

4.4.1 Methods: Design, Materials, Participants

The design consisted of the factors Report Type (Free Perception/Free Thought) and Experiencer (Both Characters/Secondary Character) and used a three-alternative forced-choice task. The items of experiment 3 were converted to third-person stories: first-person pronouns were removed and a proper name was used for the second character. One of the characters was globally prominent across conditions (global). The other character was less prominent (secondary). In the Secondary Character condition, only the secondary character was mentioned as the experiencer of the perceiving event. An example is shown in (15):

(15) Η Ντίνα ξύπνησε πολύ νωρίς εκείνη την ημέρα. Δεν είχε κοιμηθεί καλά όλο το βράδυ, είχε πολύ άγχος όλη την εβδομάδα και δεν μπορούσε να ησυχάσει. Ξύπνησε τον Μιχάλη, καθώς ήθελε παρέα για να πάρει το πρωινό της, και κατέβηκαν στην κουζίνα. Υπήρχε μια έντονη μυρωδιά.
  Both Characters: Κοίταξαν τον νεροχύτη και το πάτωμα.
  Secondary Character: Ο Μιχάλης κοίταξε τον νεροχύτη και το πάτωμα.
  Free Perception: Το θέαμα ήταν αηδιαστικό. Υπήρχε μια στοίβα άπλυτα πιάτα, ποτήρια και αποφάγια από το περασμένο βράδυ.
  Free Thought: Τι αηδιαστικό θέαμα… Αυτά τα παιδιά, πάλι είχαν αφήσει ένα σωρό άπλυτα πιάτα, ποτήρια και αποφάγια από το περασμένο βράδυ!
  Επιπλέον η βρύση έτρεχε και το νερό είχε φτάσει μέχρι και το πάτωμα.
 
  Σύμφωνα με ποιον/-α ήταν αηδιαστικό το θέαμα στην κουζίνα;
  Σύμφωνα με τον/την αφηγητή/-τρια/την Ντίνα/τον Μιχάλη
 
  Dina woke up very early that day. She hadn’t slept well all night, she had been very stressed during the whole week and couldn’t rest. She woke Michael up, as she wanted some company to take her breakfast, and they went down to the kitchen. There was a strong smell.
  Both Characters:They looked at the sink and the floor.
  Secondary Character:Michael looked at the sink and the floor.
  Free Perception:The sight was disgusting. There was a pile of unwashed dishes, glasses and leftovers from the previous night.
  Free Thought:What a disgusting sight… Those kids, they had again left a pile of unwashed dishes, glasses and leftovers from the previous night!
  On top of that, the tap was running and the water had reached the floor.
 
  According to whom was the sight in the kitchen disgusting?
  According to the narrator/Dina/Michael

Data was collected from 32 Greek native speakers (23 female, mean age 33.5, age range 24-54).

4.4.2 Results and Discussion

We estimated separate models for each binomial contrast (narrator vs. global, narrator vs. secondary, global vs. secondary) by reducing the dataset to two responses each time and performing ordinary binomial logistic regressions. The best model for each contrast was fitted with backward stepwise model comparison.

With respect to the first contrast (narrator vs. global, reference level: narrator), the final model included the fixed factor Report Type (reference level: Free Perception) and random intercepts for subjects and items. In line with our expectations, the findings suggest that readers are significantly more likely (β = 1.3774, SE = 0.2940, z = 4.686, p < .001) to pick the globally prominent character as the anchor when the passage is a free thought report, compared to a free perception report. This indicates that the presence of FID cues in the free thought condition significantly raises the likelihood of shifting to the (main) character’s perspective and is also in line with related experimental work on FID (Kaiser, 2015).

As for the second comparison (narrator vs. secondary, reference level: narrator), the optimal model included the fixed factor Experiencer (reference level: Both Characters), random intercepts for subjects and by-item random slopes for Report Type. In line with the findings from the previous experiments, data suggest that in the Secondary Character condition, readers are significantly more likely (β = 6.2183, SE = 0.6240, z = 9.965, p < .001) to ascribe the subsequent report to that character as opposed to the narrator, irrespectively of the type of report. The results are similar to those of experiment 1 and stress the importance of local prominence in the perspectivisation process of free reports, in line with Hinterwimmer’s (2017a) assumptions.

Similarly, for the third comparison (global vs. secondary, reference level: global), the optimal model included Experiencer as fixed factor (reference level: Both Characters) and random intercepts for subjects and items. No interaction between Experiencer and Report Type was found. Report Type was removed as it did not significantly improve model fit. The results show that the experiencer of the perceiving eventuality is significantly (β = 5.9843, SE = 0.6985, z = 8.567, p < .001) more likely to be picked as the anchor of a free report compared to the globally prominent character. The type of report did not play a role in how that report was anchored.

Figure 5 illustrates the pattern of effects found in the different models.

Figure 5 

The percentage of narrator, global and secondary responses per condition for Experiment 4 (FP: Free Perception, FT: Free Thought).

For the Both Characters condition, significantly more global character responses were expected, compared to secondary ones. Although in principle both characters are plausible anchors, the fact that one of them is more salient was expected to make her the most preferred anchor. Indeed, very few secondary character responses are observed. Remarkably though, narrator responses comprise the majority of the total: participants chose the narrator in 80% of the cases in Free Perception–Both Characters condition and in 60% of the cases in Free Thought–Both Characters condition. Especially for the Free Perception–Both Characters condition, this is not so surprising: the free perception report could be interpreted as true for both characters and therefore as objectively true in the story more generally. For instance, after reading ‘They looked at the sink and the floor. The sight was disgusting’ (example 15), readers may naturally take it that the sight was disgusting for both protagonists, and therefore take this statement as expressing an objective truth in the story. If the narrator is construed as an objective, omniscient observer, participants may have attributed this objective report to the narrator then; alternatively, even if readers did not construct a concept of the narrator in their mental representation, they may have resorted to the ‘narrator’ option due to a ‘both protagonists’ option not being available. However, for the Free Thought condition, FID cues were expected to increase the likelihood of picking the globally prominent character as the anchor, which is what we found. There is a notable increase in global responses in the Free Thought–Both Characters condition (36% global character responses as opposed to 19% in the Free Perception condition). This suggests that FID cues triggered shifting to the global character’s perspective and corroborates previous similar findings (Kaiser, 2015).

For the Secondary Character condition, significantly more secondary character responses were expected especially for perception reports, given the results from the previous experiments. For thought reports, predictions were unclear. Although in terms of local coherence, the secondary character seems to be more plausible as an anchor, it could be that readers attribute the free thought report to the global character because her/his perspective is more foregrounded in the text. Findings are in line with the assumptions. Participants chose the secondary character as the perspective holder around 60% of the time in both kinds of reports. The Free Perception condition elicited narrator and global responses around 39% and 5% of the time, respectively, while Free Thought elicited narrator and global responses in around 33% and 7% of the cases, respectively. The findings are similar to those of experiment 1 and related studies (Kaiser, 2018), indicating that readers tend to attribute a free report (regardless of type) to whoever is mentioned as the experiencer in the preceding sentence. However, unlike experiment 1 that elicited experiencer character responses around 95% of the time in third-person narratives, the proportion of experiencing character’s responses here is lower. This may be due to the presence of the global character. Both kinds of prominence may have made the two characters equally plausible perspectival centres, resulting in uncertainty and thus less robust preference for the locally prominent character.

5 Discussion

I presented four experimental studies in which I explored the effect of different pragmatic factors on perspectivising free reports in narratives, i.e., stretches of discourse interpreted as reports of someone’s perceptions or thoughts that are unmarked with respect to whose perspective is expressed. Taken together, the present findings point to the following tentative hierarchy of perspectival centres in narrative discourse which is illustrated in (16):

(16) locally prominent character (experiencer) > narrator > globally prominent character

Overall, the present study suggests that once a plausible anchor is determined, readers will attribute ambiguous reports of both perception and thought to that entity, usually the character that is mentioned to have the related sensory experience in the preceding discourse, supporting Hinterwimmer (2017a). Given that perception and thought descriptions often intermingle in narratives and thoughts are causally connected to perceptions, it should not be so surprising that free reports are uniformly perspectivised in this way.

Narration type was shown to have a quite robust effect (experiment 1). It was shown that reading third-person narratives, as opposed to first-person narratives, increases the chances of choosing the character as the anchor of free reports as opposed to the narrator, suggesting that the protagonist has a more prominent status in third-person stories, at least when (s)he is the only locally prominent character. Since the manipulation of narration type consisted in the absence/presence of first-person indexicals, these results suggest that at least when first-person indexicals are present, the speaker-narrator becomes more foregrounded and is more likely to serve as perspectival centre. This speaker preference is robust when the narrator is also explicitly mentioned as the experiencer of a perceiving eventuality (experiments 2, 3).

This also stresses the importance of local prominence on the perspectivisation of free reports, in line with Hinterwimmer’s (2017a) hypothesis. As experiments 1 and 4 showed, a free report is usually attributed to the character that is mentioned as the experiencer of a perceiving eventuality in the previous sentence, in line with previous studies (Kaiser, 2018; Van Krieken, 2018). This is the case even when a globally more salient character is present (experiment 4). However, it is possible that global prominence may show an effect in longer narratives.

In third-person narratives with two explicit, equally prominent local experiencers, free reports are attributed to the narrator in most of the cases, which suggests that no actual ‘shifting’ occurs, maybe partly due to both protagonists being equally plausible anchors. However, the presence of FID cues raises the chances to shift to the globally prominent character’s perspective (experiment 4). This supports the generally accepted observation that FID markers trigger shifting to a character’s perspective and corroborates previous findings (Kaiser, 2015).

Except for this case though, no supporting evidence was found that free perception and thought reports are anchored differently. One possible explanation is that the two reports may indeed not differ significantly from each other in this respect. The alternative explanation is that this null finding is due to the experimental items. Questions and exclamations were used in order to trigger an inner speech/thought report reading (Banfield, 1982). This intuitive assumption has not nonetheless been supported empirically in previous work. In fact, other authors like Fludernik (2003) oppose this view. Fludernik argues that questions do not necessarily invoke verbalized reflection on the part of the character (Fludernik, 2003, ch. 8.5). In other words, although the thought report items were intended to be read as such, they may have also been read as perception reports with the extra expressive features and questions being interpreted as conveying emotions, or, they may have been read as combinations of perception and thought reports. In either case, the distinction between the two conditions may not have been as clear-cut as desired. Future experimental work could address what elements make a sentence be interpreted as an utterance or as a perception representation.

A more robust possible means to differentiate thought from perception reports would be the addition of indexical expressions like temporal or locative adverbials. Compare passages (17) and (18):

(17) Fred looked to the right. The envelope was on the table.

(18) Fred looked to the right. The envelope was on the table yesterday.

On an FID reading, yesterday in (18) can have a shifted interpretation. In this case, the second clause is read as a report of Fred’s thoughts, whereas the same sentence without the adverbial in (17) is more likely to be interpreted as a perception report instead.

Another possible limitation of the experiments presented here is the use of the binary forced-choice task. As mentioned earlier, free reports are in principle ambiguous. In many cases it is important for the reader to disambiguate these reports and resolve whose perspective they express as this aids them to make sense of the story and draw the right inferences. If such information cannot be resolved immediately, the reader can proceed and use additional information from later discourse. But it is also possible that such information does not really have to be resolved if it is not purposeful in some way. In other words, in some cases the reader may not need to determine such information in order to make sense of a story. An anonymous reviewer points out that in the items used here there was no particular necessity for the reader to decide upon whose perspective is expressed in each case. Consequently, the particular task may not have captured the readers’ actual interpretations of free reports who, in the actual process of reading, could simply process this information without having to make a choice. Perhaps an additional option such as “I don’t know” would have been useful in this case, as well as an option like “both the narrator and the character/both characters” since in many cases the content of the report could in fact be attributed to both. A direction for future research would be to use stories where, for example, resolution of perspective would be crucial for the correct interpretation of the story. Finally, the question of whether readers determine an anchor at all should also be addressed, albeit with a different methodology. A good starting point would be to look at experimental studies on the closely connected topic of ambiguous pronoun resolution (see, for example, Stewart, Holler and Kidd, 2007).

Related to the above point, although forced-choice tasks have been employed in similar studies on perspective taking (e.g., Harris and Potts, 2009; Kaiser, 2015; Kaiser, 2018), they may be too explicit to capture processes that take place unconsciously while reading. Note that considerations related to a first-person narrator’s epistemic access did not seem to influence reader responses: free thought reports were not necessarily anchored to the speaker in first-person narratives. However, it is natural to assume that such shifts away from the speaker are costlier in first-person narratives and might be even costlier for thought compared to perception reports. Using more sensitive, online methods like response time measurements and eye-tracking may reveal differences in how free perception and thought reports are actually processed in different contexts (Meuser, Patil, and Hinterwimmer, 2018 have conducted a related study on FID with interesting results).

In general, readers are willing to assign the content of a statement to an entity other than the narrator, which is in line with research on the availability of non-speaker-oriented interpretations (e.g., Harris and Potts, 2009; Kaiser, 2018). That this tendency seems to be stronger in third-person stories, as experiment 1 showed, is compatible with Banfield’s (1982) view that the character in third-person narratives is, in her terms, the ‘subject of consciousness’. However, the narrator still functioned as a perspectival centre in third-person stories as well. Future research could further investigate the hierarchy suggested above and further flesh out the conditions under which some character or the narrator become the preferred perspectival centres, by considering more refined distinctions within first- and third-person narratives, such as first-person narratives with narrators as either main or secondary characters. As a final note, it seems reasonable to assume that the less prominent a narrator is in a story, the more possible it is for the reader to shift to a different perspectival centre.