

Dialogue & Discourse 15(2) (2024) 1–35 doi: https://doi.org/10.5210/dad.2024.201

Perspective-taking and protagonist prominence: An empirical
approach to the role of local and global prominence

Sara Meuser MEUSER@UNI-WUPPERTAL.DE
Bergische Universität Wuppertal

Maximilian Hörl MHÖRL1@UNI-KOELN.DE
Universität zu Köln

Stefan Hinterwimmer STEFAN.HINTERWIMMER@UNI-HAMBURG.DE

Universität Hamburg

Editor: Amir Zeldes

Submitted 06/2023; Accepted 06/2024; Published online 08/2024

Abstract
The choice of the perspectival center of a stretch of discourse is crucial for the interpretation of
certain phenomena such as free indirect discourse. It has been argued that the protagonist that is
most prominent compared to competing protagonists gets to be the perspectival center. In this paper
we discuss grammatical function and referential expression as prominence-lending cues and their
impact on perspective-taking. We take the anchoring of free indirect discourse as the indicator for
a shift in perspective as free indirect discourse can only be processed correctly if the reader is able
to ascribe the utterance or thought to a protagonist. Identifying the perspectival center is particu-
larly crucial for the interpretation of a thought or utterance in free indirect discourse mode that can
potentially be ascribed to different protagonists, since in contrast to direct or indirect discourse the
respective speaker or thinker is not explicitly marked as such in free indirect discourse.
In a series of acceptability rating studies, we tested if anchoring of free indirect discourse to the
less prominent of two competing referents is perceived to be unnatural in German. Further, we take
a closer look at the role of subject and object as well as the choice of referential expression (proper
name compared to indefinite noun phrase). We find that a protagonist referred to with a proper
name in subject position is highly preferred as the anchor for free indirect discourse compared to a
protagonist referred to with an indefinite noun phrase in object position.
Building on these findings, we present evidence that the prominence of the referent that is es-
tablished in the sentence preceding a sentence in free indirect discourse mode can be overridden
by discourse prominence. That is, a referent that is repeatedly mentioned in a short discourse is
preferred as the perspectival center regardless of the prominence of a competing referent in the
sentence preceding a sentence in free indirect discourse mode

Keywords: Perspective, Free Indirect Discourse, Prominence, Discourse, Subjecthood

1. Introduction

1.1 Perspective-taking in language

The choice of words in a spoken discourse or a narrative does not only depend on the intention
and the knowledge that a speaker or narrator has but also on their perspective regarding location in
space and time, their relation to other individuals etc. In particular expressions referring to places,

©2024 Sara Meuser, Maximilian Hörl, Stefan Hinterwimmer

This is an open-access article distributed under the terms of a Creative Commons Attribution License
(http ://creativecommons.org/licenses/by/3.0/).


MEUSER, HÖRL AND HINTERWIMMER

points in time or individuals, but also adverbs like luckily or sadly express an involvement in a
given context from a certain point of view – for example, my mother, my daughter or the minister
of defense may refer to the same person depending on the relation between the person uttering
the expression and the person denoted by the expression. A coherent interpretation of a discourse
thus requires the hearer or reader to ascribe context-sensitive utterances to the speaker or narrator
or a certain protagonist. A change in perspective in a narration as well as in spoken discourse
can simply be indicated by means of direct or indirect speech or thought representation; e.g., the
minister of defense can be denoted by the expression my mother if the speaker signals that he or she
is citing a child of the minister of defense – or if the speaker is a child of the minister of defense.
In fiction a certain perspective can also be indicated by narrative mode. A first-person narrator for
example usually reports or recounts events from his or her fixed point of view. A narration from
a third-person perspective, in contrast, may allow for sporadic shifts in perspective for example
in Free Indirect Discourse (FID) or other kinds of perspective-taking such as Viewpoint Shifting
(Hinterwimmer, 2017) or Protagonist Projection (Holton, 1997; Stokke, 2013; Abrusán, 2020). FID
is a form of thought representation where context-sensitive expressions such as deictic expressions
or evaluative terms are interpreted with respect to the perspective of a certain protagonist, while at
the same time the narrative mode of the third-person narrator remains unchanged, as shown in (1a):
The second sentence in (1a) is intuitively interpreted as a thought of Maria, with the deictic adverb
tomorrow being interpreted with respect to Maria’s context, i.e. as referring to the day following
the day on which Maria has the thought. At the same time, Maria is referred to by a third- rather
than a first-person pronoun, and the present rather than the past tense form of the auxiliary will is
used. View Point Shifting and Protagonist Projection, in contrast, do not render conscious thoughts
of protagonists but rather render the contents of their perceptions in a way that is compatible with
their belief state at the time of perceiving, as shown in (1b): The main clause in (1b), for example,
on its most plausible interpretation does not describe an event that is happening in the story world,
but rather a temporary illusion that Mary has a consequence of her sense of balance having been
disturbed by the boat trip.

(1) a. Mary smiled. Tomorrow she would reveal her true identity at the press conference.
(Hinterwimmer, 2019: 79, ex. (1c))
b. When Mary stepped out of the boat, the ground was shaking beneath her feet for a couple
of seconds.
(Hinterwimmer, 2017: 291, ex. (14))

When we approach perspective-taking in language, we have to consider that the common under-
standing of the term perspective refers to a visual experience that is limited according to the ob-
server’s position. The term perspective is metaphorically applied to narration and may be defined
as the linguistic and extralinguistic choices that are limited with respect to a certain point of view
(see Friedman, 1955; Stanzel, 1984). A narration from a certain protagonist’s perspective can thus
only convey information this particular protagonist has according to his or her position in a given
narrative. In spoken language, the speaker is usually the perspectival center of the utterance – un-
less he or she indicates a change in perspective, e.g., by quoting another person. Written discourse
does not allow for such a straightforward ascription of the perspectival center as it is not always as
trivial to pinpoint who is telling the story. In narration, it is not the author who writes down his
or her story, shares memories or perceives a situation, but a narrating instance that is created by an
author and that may be far more abstract than an actual individual and have many properties (such

2


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

as omniscience, for instance) that no actual individual has (see, e.g. Zeman, 2020 for discussion).1

In the tradition of literary studies we will therefore never refer to the author’s perspective but to the
narrator’s perspective.

In this paper, we will give a short overview of perspective-taking in narratives before we discuss
the conditions under which perspective shifts from neutral narration to FID are possible. We will
follow the argumentation outlined in Hinterwimmer (2019) (also Hinterwimmer and Meuser, 2019;
Meuser, 2022) and show that the availability of a protagonist as a possible perspective holder de-
pends on his or her prominence status (see Wiebe, 1990, 1994 for an early discussion of this issue in
a computational framework and Abrusán, 2021 for a sketch of a unified proposal that integrates the
insights of Wiebe, 1990, 1994 and Hinterwimmer, 2019). In Section 2 we report acceptability rating
studies (see also Meuser, 2022) investigating the hypothesis that a prominent referent is more likely
to be the perspectival center of a text segment compared to a competing, less prominent referent.
In our experiments we asked participants to rate the naturalness of German text segments in which
sentences in FID must be interpreted with regard to two discourse referents and found that FID
anchored to the more prominent protagonists received higher ratings.2 The first experiment (n=75)
presented in Section 2.1 will investigate the following hypothesis:

H1: A protagonist that is more prominent in terms of referential expression and grammatical func-
tion is more easily available as the anchor for FID than a competing referent.

In Section 2.2 we will present a follow-up experiment (n=119) that will provide a first attempt to
disentangle the prominence-lending cues grammatical function and referential expression by testing
the following hypotheses:

H2: Protagonists functioning as subjects are more available as anchors for FID than protagonists
functioning as objects.

H3: Protagonists referred to with a proper name are more available as anchors for FID than pro-
tagonists introduced by an indefinite NP.

Furthermore, we will compare the effect investigated with respect to H2 and H3 so that we can
draw conclusions regarding the hierarchies of the two prominence-lending cues. For that purpose,
we raised the rather exploratory question (Q1): Does the type of referential expression play a bigger
role than the grammatical function with respect to the anchoring of FID? A second question (Q2)
that calls for an exploratory investigation is whether the competition of the two protagonists has
an impact on the anchoring preference, i.e., is the effect stronger when the competing referent is
minimally prominent in terms of referential expression?
We conclude that although FID, in contrast to direct and indirect discourse, does not require formal
marking, the correct ascription of FID depends on the availability of a protagonist as the anchor
for the thought or utterance. Identifying the perspectival center is of particular importance when-
ever several protagonists are potentially available to be the anchor for an utterance. Our findings

1. This approach is mostly limited to fictional texts. Imagine a note on the refrigerator saying “I ate the cake”. Here it
is to be assumed that the author of the note ate the cake. An interpretation in which a narrator who is not the author
ate the cake would be absurd.

2. Although the experimental evidence is based on German stimuli rated by German native speakers, we want to argue
that FID follows the same patterns in English. We decided to present English examples in the theoretical part of this
paper to illustrate the issue. All examples were carefully checked to evoke the same effect by English native speakers.

3


MEUSER, HÖRL AND HINTERWIMMER

suggest that the anchoring of FID to a protagonist depends on his or her prominence status and
that grammatical function is a crucial indicator for the availability as a perspectival center while the
referential expression is less important as a prominence-lending cue.
Based on these findings we conducted a third acceptability rating study (n=116) where we test the
effects found in the first two experiments in larger discourse. In particular, we wanted to test if a
referent that is established as the most prominent in terms of grammatical function in a local context,
i.e., the sentence preceding the FID, is still preferred as the perspectival center when a competing
referent is more prominent in the global context, i.e. throughout a short discourse. We investigate
two hypotheses:

H4: Protagonists that are highly activated in a discourse in terms of being repeatedly mentioned
in subject position are just as available as anchors for FID as protagonists that are less often
referred to in a discourse but in subject position in the sentence preceding the FID.

H5: Protagonists that are highly activated in a discourse in terms of being established as the dis-
course topic in a topic-establishing sentence, repeatedly mentioned in subject position, and
mentioned in a title are more available as anchors for FID than protagonists that are less often
referred to in a discourse but in subject position in the sentence preceding the FID.

1.2 Perspective-taking in narratives

We want to approach the term perspective and its connection to fiction by taking a look at how
narrative modes set a certain perspective and allow for changes in perspective.
The most obvious classification of narrating instances holds between the first-person narrator who
tells a story from his or her point of view and a third-person narrator. A second relevant distinc-
tion is the one between homodiegetic and heterodiegetic narrators: Homodiegetic narrators are part
of the story world, while heterodiegetic narrators are not. Consequently, the perspective of a ho-
modiegetic first-person narrator is much more restricted than the perspective of a heterodiegetic
first-person narrator: Since the former is by definition also a protagonist of the story, they perceive
or tell a story they witnessed from a fixed, i.e. their own perspective, and only have access to
their own thoughts and feelings. By contrast, a heterodiegetic narrator has more or less knowledge
about the ongoing story, the past and the future of the story world as well as insights into different
protagonists’ thoughts with varying degrees of omniscience. This holds not only of third-person,
but also of first-person heterodiegetic narrators, which are, however, much rarer than third-person
heterodiegetic narrators, at least in modern fiction (see Saure et al. 2023 for discussion of the differ-
ences between homo- and heterodiegetic as well as first and third-person narrators with respect to
perspective-taking).

Furthermore, literary scholars have done more fine-grained classifications such as the distinction
between first-person, authorial and figural narrators (Stanzel, 1984), between editorial, neutral,
selective and multiple selective omniscience (Friedman, 1955) or between zero focalization, exter-
nal focalization and internal focalization (Genette, 2010). The three latter terms, which have been
very influential in literary studies of narration, refer to modes of narration where the narrator either
has access to the thoughts, feelings and perceptions of all protagonists (zero focalization) or just one
particular, highly prominent protagonist (internal focalization), or where the narrator does not have
access to the thoughts, feelings and perceptions of any protagonist and just describes observable

4


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

events, actions and situations.

In spite of using different terminology, all mentioned approaches describe narrators that may al-
low for shifts in perspective, i.e., they are not restricted to one perspective but they may temporally
report a protagonist’s point of view without indicating the shift in perspective by the use of direct or
indirect discourse.

With many terms for perspective-taking in narratives we deal with a phenomenon that is well de-
scribed in literary studies, yet with little attention3 drawn to the restrictions of perspective shift (with
the notable exception of Wiebe, 1990, 1994; see Abrusán, 2021 for discussion) or, as formulated in
Hinterwimmer (2019), the answer to the question: under what conditions do protagonists become
available as perspectival centers?

With these descriptive approaches in mind, we want to point out that narration with a high de-
gree of focalization does not only require the reader to identify the perspective holder in a given
discourse and to be able to follow a shift in perspective from a narrating instance to a protagonist´s
point of view, and vice versa, but also to be able to “integrate confronting perspectives” (Zeman,
2019). While actual physical perspective-taking usually4 limits the observer to only one location
from where a view may be directed, the concept of perspective in language we investigate naturally
allows for multiple points of view. We will briefly elaborate on this idea taking an example from
Zeman that illustrates how the use of propositional attitude verbs commonly licenses (at least) dual
perspectivization.

(2) Little Red Riding Hood believes that the wolf is her grandmother.

In (2) the reader is confronted with a proposition that requires the accommodation of an external
and an internal reading. Here, the information that Little Red Riding Hood believes the creature in
her grandmother´s bed to be her grandmother has to be updated with the external viewpoint that
provides the information that it is in fact the wolf who took the grandmother’s position in bed (i.e.,
the definite description the wolf is interpreted de re). In other words, without the integration of mul-
tiple perspectives we are limited to either Little Red Riding Hood´s point of view (as paraphrased
in (3a) or the narrator´s point of view (as paraphrased in (3b)).

(3) a. Little Red Riding Hood believes she sees her grandmother.
b. The wolf pretends to be Little Red Riding Hood´s grandmother.

We propose a similar multiple perspective reading including the external perspective of the narrator
and the internal perspective of the protagonist for the examples of FID we discuss as indicators of a
perspective shift.

Also, traditionally, in narratology it has been recognized that instances of selective omniscience
or internal focalization are rather two overlapping or mixed perspectives – the one of the narrator

3. Genette considers alterations in focalization as isolated violations that may occur as long as the coherence of the
whole is given (Genette, 2010). However, he does not elaborate to what extent focalization may alter in order for the
text to be coherent and if all protagonists are possible perspective holders.

4. Mirrors and cameras may allow for multiple perspectives.

5


MEUSER, HÖRL AND HINTERWIMMER

and the one of the protagonist (see Pascal 1977 for the “dual voice” discussion). The terminology
shift of perspective used in our discussion may therefore falsely indicate that one perspective is ne-
glected for the sake of another – that is not the case. For reasons of simplicity, however, we will
continue to talk about shifts in perspective – keep in mind that it is rather a shift in narrative mode
from authorial, omniscient or zero focalized narration (henceforth: neutral perspective) to a figural,
multiple omniscient or internally focalized narration (henceforth: protagonist´s perspective).

1.3 Free Indirect Discourse

Although the issue of perspective-taking is well studied in linguistics (for an overview see Eckardt,
2014) one question remains almost untackled: What restricts the choice of the perspectival center in
a given discourse when different protagonists are available as perspectival centers, i.e., are shifts to
the perspective of a certain protagonist that is part of a given discourse always coherent or do they
sometimes lead to incoherence?
This question is particularly interesting with regard to the limitations of FID. Unlike direct or indi-
rect discourse FID renders the thought of a protagonist without using quotation marks or embedding
it under a propositional attitude verb such as say or think. A sentence in FID can thus only be in-
terpreted correctly if the reader is able to take on the perspective of the protagonist whose thoughts
or utterances are reported. Characteristics of FID are, for example, interjections, judgmental state-
ments, exclamatives, discourse particles, rhetorical questions and a partial shift in deixis (Steube,
1985; Banfield, 1982).

(4) a. Thomas looked at the calendar. He thought: “Tomorrow I will finally see Mommy again.”
b. Thomas looked at the calendar. He thought that he was going to see his mother again the
next day.
c. Thomas looked at the calendar. Tomorrow he would finally see his Mommy again.5

Especially the use of deictic expressions, i.e., expressions that depend on a fixed context such as I,
here and now, is notable. In (4a) and (4b) we find a shift from the first-person personal pronoun
(henceforth: PPro) I to the third-person PPro he, the informal expression mommy changes to the
possessive determiner phrase his mother and the adverbial tomorrow shifts to the anaphoric adver-
bial the next day, which refers to the day following the day including the reference time set in the
first sentence, i.e., the time when Thomas looked at the calendar. In (4c), on the other hand, the
use of deictic expressions seems inconsistent. Here the third-person PPro he as well as the posses-
sive determiner his indicate a neutral perspective. The temporal adverbial tomorrow contradicts the
use of past tense in the preceding sentence and thus has to be interpreted with regard to Thomas’
perspective just as the informal address mommy, which likewise expresses Thomas’ perspective. In
order to account for this partial shift, Eckardt (2014) (see also Schlenker, 2004 and Sharvit, 2008
for closely related proposals built on the same basic idea) proposes to take two contexts into con-
sideration: the narrator’s context C and the protagonist’s context c, where c is only introduced in

5. We want to add that this example could be read as a case of objective future-in-the past (see Eckardt, 2017 for the
terminology) if we paraphrase the utterance in the following way:
Unbeknownst to him, he would finally see Mommy again tomorrow.
While this reading is coherent, it is rather odd that the narrator refers to Thomas’ mother as Mommy. We want to
clarify at this point that many utterances in FID mode may receive an objective reading by adding “but he/she was
not aware of it”. We designed all items cautiously to make a future-in-the past reading strikingly odd.

6


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

cases of FID. While the spatial and temporal setting of a narrative usually depends on C, all context
sensitive expressions with the exception of pronouns and tenses have to be interpreted with respect
to c whenever c is available. Pronouns and tenses, in contrast, are always interpreted with respect
to C. Crucially, whenever c is introduced, the proposition denoted by a sentence interpreted with
respect to c as well as C is not interpreted as true with respect to the worlds compatible with the nar-
ration (the story worlds), but only as true in the worlds compatible with the beliefs of the respective
protagonist (Eckardt, 2014). Example (5) can thus be paraphrased as:

(5) There is an event e of Thomas looking at the calendar that is located in the past with respect
to the time of C (= the narration time) and in all worlds that are compatible with the beliefs
of the author of c (= Thomas) at the time of c (= the time of e) there is an event e’ of Thomas
seeing his mother again that is located in the past with respect to the time of C (= the narration
time) and in the future with respect to the time of c and that takes place on the day that follows
the day including the time of c (= the time of e).

Maier (2017), on the other hand, postulates that FID is a special form of mixed quotation in which
everything except pronouns and tenses are quoted – i.e. the latter are systematically unquoted – , as
shown in (6) for our example in (4) c6:

(6) Thomas looked at the calendar. “Tomorrow” he would “finally see” his “Mommy again”.

We will leave open the question of which of the two lines of analysis just sketched is the correct
one, as it is not directly relevant for our purposes in this paper. Our concern is not so much how the
perspective of the narrator interacts with the perspective of a single protagonist, but rather what the
conditions are under which protagonists become available as perspectival centers – a question which
becomes particularly pressing as soon as examples are considered in which several protagonists
interact.

1.4 Prominence and perspective

While the interpretation of perspective-sensitive constructions such as FID as a thought of a pro-
tagonist has been studied extensively in linguistics and narratology, a common concern is whether
it is justified to completely neglect a narrator-oriented interpretation in such cases and ascribe the
propositions to the protagonist, i.e. whether FID represents just the protagonist’s, or both the narra-
tor’s and the protagonist’s perspective. This issue was first tackled empirically by Harris and Potts
(2009), who investigated if epithets and appositives can receive a non-speaker/non-narrator inter-
pretation even in contexts where an overt first-person narrator is competing with a protagonist as
perspectival center. In a series of forced-choice studies they found evidence that epithets and ap-
positives can be interpreted from the perspective of a contextually prominent protagonist despite the
overt narrator, and in the presence of certain contextual clues even favor such an interpretation. In
line with the research of Harris and Potts, Kaiser (2015) confirms that epithets and appositives can
receive non-speaker/non-narrator interpretations especially when they are interpreted as FID. Kaiser
also tested if FID cues – evaluative epithets and adverbs of possibility – trigger the perspective of
the grammatical subject of the preceding sentence rather than a speaker- or narrator-oriented per-
spective. She found that in a sequence of two sentences (Mary looked woefully at Elizabeth. [Poor

6. Note that since it is not possible to unquote the verb without the tense, the whole finite verb is unquoted in (6).

7


MEUSER, HÖRL AND HINTERWIMMER

girl;] she was sick.) the presence of FID cues increases the likelihood that the personal pronoun in
the second sentence refers to the object of the first sentence – despite the well-known preference of
personal pronouns for the subjects of preceding sentences (see, e.g., Crawley and Stevenson, 1990;
Gernsbacher, 1990; Gordon et al., 1993; Stevenson et al., 1994 for discussion). The most plausible
interpretation of this result is that the second sentence is interpreting as expressing Mary’s rather
that the narrator’s perspective.
Kaiser’s approach gives valuable insights on the anchoring of FID. However, her research focusses
on the distinction between speaker/narrator and protagonist perspective without taking into account
the possibility that multiple protagonists might be available as perspectival centers.

Before we get to our experimental approach to the limitations of perspective shifts, we want to
elaborate on the anchoring of FID in contexts where two protagonists are in principle available as
perspectival center. The discussion is based on our intuitions and the argumentation proposed in
Hinterwimmer (2019) regarding different prominence-based accounts as well as a coherence-based
account.
Let us consider a discourse free of any contextual presuppositions or emotional involvement beyond
the two sentences presented in (7) that could act in favor of one perspective or the other. Free of any
background assumptions regarding the two referents Tina and Mike and without any other contex-
tual information we intuitively find differences in the availability of certain protagonists as anchors
for FID:

(7) a. Tina was yelling at Mike. Tonight, that jerk really pushed it too far.
b. Tina was yelling at Mike. Tonight, that bitch really pushed it too far.

While the second sentence in (7a) is most plausibly understood as a thought of Tina about Mike, the
epithet in (7b) can only refer to a female protagonist and thus the FID in (7b) can only be interpreted
as a thought of Mike about Tina. Even though the context provided by the first sentence in which
Mike is being yelled at by Tina gives perfect reason for Mike to have negative thoughts about Tina
the second sentence conveys the impression that it is rather a comment by the narrator than FID
anchored to Mike. Another, presumably rather far-fetched, reading of (7b) would suggest that the
second sentence is a thought by Tina and the female referent that bitch refers to some unmentioned
female person. Regardless of which option is chosen, the second example is harder to interpret than
the first. The uncertainty about the ascription of the second sentence in (7b) illustrates the issue:
linguistic cues in the sentences preceding FID increase or decrease the availability of referents as
perspectival centers and thus anchors for FID.

Hinterwimmer (2019) suggests that it is the most prominent protagonist that is by default the per-
spectival center. Let us narrow down the linguistic notion of prominence with respect to the avail-
ability of referents as the perspectival center of a sentence or short text segment.

In example (7) both Tina and Mike are referred to with a rather common proper name and we
assume that the reader does not have any reason to favor neither Tina nor Mike. Still, Tina can
be regarded as more prominent than Mike on at least two different prominence scales: First, while
Tina is the subject of the first sentence, Mike is the prepositional object. According to the hierarchy
of grammatical functions a subject is more prominent than a direct object, a direct object is more
prominent than an indirect object and so forth (see Himmelmann and Primus, 2015 for an overview

8


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

of prominence hierarchies and von Heusinger and Schumacher, 2019 on prominence in discourse).
Second, regarding the hierarchy of semantic roles, Tina is the agent, which is more prominent than
the patient, Mike (ibid.). Although the FID in (7b) can be regarded as a possible thought of Mike,
who is mad at Tina for yelling at him, Mike is less easily easily available as perspectival center than
Tina in (7a) and thus a thought anchored to Mike seems at least slightly odd and unexpected.7

Having mentioned coherence as one restricting factor for perspective shifts (see Kehler et al., 2008
and Kehler and Rohde, 2019 for an analysis of pronoun resolution in which coherence plays a cru-
cial role)8, we want to point out that coherence alone does not account for the ascription of FID (but
see Abrusán, 2021 for observations showing that coherence relations can have an influence on pro-
tagonists’ availability as perspectival centers). In example (7a), the second sentence can be linked to
the first by providing an explanation – the fact that Mike had done or said something that had pushed
it too far is the reason for Tina to yell at him. In contrast, (7b) leaves room for different more or less
coherent interpretations. One reading suggests that Tina’s yelling is the reason for Mike to have the
thought rendered in FID. In that sense the first sentence may provide the cause for the second one.
Another interpretation may be that Tina has done something else, unmentioned, that night that had
pushed it too far. The sequence of the two sentences in (7b) could then simply be understood to be
a linear order of narrative elements. A third interpretation in which the sentence in FID expresses
a thought of Tina and the epithet refers to someone else provides an explanation for the yelling in
the first place similar to (7a), yet this reading seems rather absurd without any context. Still, there
are at least two ways available to link the proposition denoted by the sentence in FID coherently to
the proposition denoted by the preceding sentence, the first one according to which Tina’s yelling at
Mike causes Mike to have the thought rendered in FID being no less plausible than the one linking
the two sentences in (7a) via providing an explanation. Consequently, the contrast between (7a) and
(7b) cannot be accounted for in terms of coherence exclusively. Rather, the availability of a refer-
ent as the perspectival center crucially depends on its prominence status (cf. (Piwek and Krahmer,
2000) on the interaction of salience, on the one hand, and plausibility/coherence, on the other, as
factors contributing to the resolution of anaphora and presuppositions)9.

While it seems tempting to draw conclusions on the anchoring of FID from pronoun resolution,
Hinterwimmer (2019) points out that pronoun resolution, although it may indicate tendencies for
the availability of perspectival centers, works differently from FID- anchoring. We do not want to
elaborate on the similarities and differences between pronoun resolution and FID-anchoring at this
point. For the investigation presented in this paper, however, it is important to note that preferences
with respect to pronoun resolution do not always coincide with preferences regarding the anchoring
of FID, i.e., the referent that may be preferred as the antecedent of a potentially ambiguous pronoun
is not necessarily preferred as the perspectival center as illustrated in (8).

(8) The new girli delighted Janej.
a. Shei was just so gorgeous! (pronoun: i, perspective: j)

7. We have consulted two native English speakers, who shared our intuitions, but the contrast is admittedly quite subtle
and there might be individual differences between speakers.

8. We can only take linguistic approaches to coherence into account as there is no elaboration on coherence as it is used
in literary studies.

9. We are grateful to an anonymous reviewer for pointing us to this work.

9


MEUSER, HÖRL AND HINTERWIMMER

b. At least, shei hadn’t put on her best dress for nothing. (pronoun: i, perspective: i)
c. But how could shej start a conversation? (pronoun: j, perspective: j)

In line with the classification by Garvey and Caramazza (1974) who classify the verb to delight
as yielding an NP1-bias with respect to the resolution of personal pronouns it seems to be more
expected to pick up the new girl rather than Jane with the personal pronoun she. At the same time,
there seems to be a clear preference to pick up the perspective of the second referent, i.e., Jane, in
(8a) as compared to (8b). Additionally, despite the general preference for the new girl to be picked
up with a pronoun in a subsequent sentence, an utterance in FID mode anchored to Jane as in (8c)
with the pronoun picking up Jane seems to be more acceptable than (8b), where the utterance has to
be anchored to the new girl and the pronoun is resolved to the new girl as well.

With this in mind, let us now continue our discussion of prominence and perspective. The im-
portance of prominence becomes more evident the more unevenly the prominence-lending features
are distributed. In the following example the female protagonist is not only referred to with a proper
name whereas the second referent is introduced with an indefinite article but she is also picked up
by a pronoun.

(9) a. When Lisa was playing in the schoolyard, a boy pushed her into the stinging nettles. Ouch,
that itched!
b. When Lisa was playing in the schoolyard, she pushed a boy into the stinging nettles. Ouch,
that itched!

Here not only the qualitative measures such as type of referring expression but also quantitative
measures such as number of mentions add to the availability of the referents as perspectival centers.
In example (9b) an interpretation in which the boy´s thought is reported is coherent. However,
according to the authors´ intuitions it is slightly awkward and unexpected to take on the boy´s per-
spective, although by no means impossible. In an attempt to anchor the FID this example offers
another possible reading in which Lisa takes on the boy´s perspective and empathically assumes
that it must itch for the boy. This recursive reading again shows that it is the most prominent refer-
ent that is by default the perspectival center.

Intuitively in example (9) Lisa seems to be more prominent than the boy resulting in a strong pref-
erence for (9a) as opposed to (9b). This intuition may best be captured by the notion of topicality.
Following Reinhart (1981), we may regard Lisa as the aboutness topic of the first sentence. If we
assume that every sentence is the answer to an implicit question (van Kuppevelt, 1995; Roberts,
1996; Ginzburg, 2012), example (9) can plausibly be interpreted as the answer to the question what
did Lisa do/experience in the schoolyard?, i.e., the answer to a question about Lisa. A context in
which example (8) is to be understood to answer a question about some boy, i.e., a question such as
what did a boy do/experience in the schoolyard?, on the other hand is rather absurd. This may be
explained by the choice of referring expression. While Lisa is introduced by a proper name, the boy
is introduced with an indefinite description, which indicates that the boy is unknown or at least not
explicitly identifiable amongst other boys on the schoolyard. An implicit question about the boy is
thus rather implausible as a discourse move.

While the assumption that Lisa is the topic in (9) goes well together with our claim that Lisa is

10


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

the most prominent protagonist and consequently the only available perspectival center, we will not
rely on the notion of topicality as it does not lend itself directly to a more fine-grained investigation
of linguistic cues that impact the availability as the perspectival center. For our purpose the notion
of prominence seems more appropriate as it comprises a wide range of linguistic cues and suggests
hierarchies within those cues. Furthermore, the notion of topicality on the level of the sentence it
may not serve well with respect to larger discourse. This becomes more obvious when we make
some syntactic changes to example (9) without affecting the content of our story.

(10) Lisa was playing in the schoolyard. A boy pushed her into the stinging nettles. Ouch, that
itched!

In (10) we still prefer Lisa as the perspectival center, when it is actually the boy who is the subject
and the agent of the sentence preceding the FID while Lisa is the object and the patient. While
the concept of discourse topicality may serve as an indicator for perspectivehood, the notion of dis-
course topicality lacks a clear definition with respect to individual linguistic markers that are crucial
for an empirical investigation (but see van Dijk, 1977 and Roberts, 2012 for relevant discussion).
Here we want to suggest that it is global prominence that favors Lisa as the perspectival center.
Lisa´s prominence features clearly outweigh the features of the boy: She is the subject and agent
of the first sentence, first mentioned, introduced with a proper name, picked up with a personal pro-
noun – consequently mentioned twice –, while the boy is only the subject and agent of the second
sentence.

For the sake of completeness, we want to elaborate briefly on the aspect of competition between
possible perspectival centers. As mentioned above, prominence features as we investigate them
only differentiate between two possible anchors. If FID is preceded by a sentence with only one
protagonist, that protagonist is highly available as the anchor for FID, as in (11) where the little boy
is neither the subject nor referred to with a proper name. Despite presumably low prominence the
FID must be anchored to the protagonist mentioned in the preceding sentence.

(11) The ball hit the little boy right in the face. Ouch, that hurt!

Competition of potential anchors for an utterance in FID may also play a role with respect to pre-
ceding sentences with two interacting protagonists, as in (12). While in (12a) Martin is naturally
perceived to be the perspectival center, in (12b) the subject´s prominence status is raised so that
the availability of Martin as the perspectival center is weakened. Consequently, it is slightly awk-
ward and unexpected (though by no means impossible) to interpret the second sentence in (12b) as
a thought of Martin rendered via FID. In other words, the competition of Lilly, now outweighing
the prominence status of Martin, has an impact on the availability of a referent as the perspectival
center.

(12) a. A young lady asked Martin for the way to the station.
Hmm, didn’t she see the sign?
b. Lilly asked Martin for the way to the station.
Hmm, didn’t she see the sign?

Before we close our discussion of the availability of protagonists as anchors for FID, we want
to consider an example with protagonists of equal prominence. In (13) neither a nor b seem to

11


MEUSER, HÖRL AND HINTERWIMMER

be appropriate. We tentatively suggest an explanation along the following lines: In order to be
available as perspectival center, a single protagonist has to be prominent. In the case of (13), neither
Tina nor Mike is prominent – rather, it is just the plurality referred to by the NP Tina and Mike.
Consequently, both (13a) and (13b) are awkward, since in either case the FID has to be ascribed to
a protagonist that is not maximally prominent.

(13) Tina and Mike were fighting all night.
a. Tonight that jerk had pushed it too far.
b. Tonight that bitch had pushed it too far.

Summarizing the discussion so far, we see that a referent must either be locally prominent in the
sentence preceding the utterance in FID mode, e.g., with regard to the grammatical function, the
thematic role or the type of referential expression, or be globally prominent in a context where at
least two protagonists compete. So far, we have shown that a prominence-based account provides
promising insights in this regard. At the same time, prominence-lending cues comprise a wide range
of linguistic markers – either in the sentence immediately preceding an utterance in FID mode or in a
stretch of discourse – that promote a referent to be the perspectival center. The prominence-lending
cues elaborated above are by no means exhaustive. Yet, in order to gain a better understanding of
how prominence and perspectival centers are related we will present a first attempt at answering that
question by investigating how grammatical function and type of referential expression contribute to
a referent´s potential to serve as the anchor for an utterance in FID mode. While the first experiment
presented in the next section will give preliminary insights that suggest a maximally prominent
referent – in terms of grammatical function, type of referential expression and number of mentions
– is preferred as the anchor for a sentence in FID mode, the experiment presented in Section 2.2
will investigate the impact of grammatical function and referential expression and their interaction.
Finally, in section 2.3 we test these findings in larger contexts providing first empirical evidence for
Hinterwimmer’s (2019) assumption (see also Abrusán, 2021 for discussion) that locally established
prominence interacts with globally established prominence.

2. The experimental studies

2.1 Experiment 1

In our first acceptability rating tasks we asked participants to rate the naturalness of short text seg-
ments such as (14) and (15) in which two utterances in FID mode or a neutral sentence must be
interpreted with regard to the context provided in the first two sentences.

(14) Als die Hochzeit von Prinz William und Kate im Fernsehen übertragen wurde, konnte Robert
seine eigene Hochzeit kaum erwarten.
Auch er hatte seiner Freundin einen Antrag gemacht.
a. FIDpreferred: Schon morgen würde er mit seiner Liebsten vor den Altar treten.
b. FIDdispreferred: Schon morgen würde sie mit ihrem Liebsten vor den Altar treten.
c. Neutral: Sie wollte mit ihm vor den Altar treten.

When the wedding of Prince William and Kate was broadcast on TV, Robert could hardly
wait for his own wedding.
He, too, had proposed to his girlfriend.

12


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

a. FIDpreferred: Soon he would walk down the aisle with his darling.
b. FIDdispreferred: Soon she would walk down the aisle with her darling.
c. Neutral: She wanted to walk down the aisle with him.

(15) Als der letzte Band von ”Harry Potter” erschien, kramte Luisa ihr Taschengeld zusammen.
Sofort sagte sie ihrem besten Freund Bescheid.
a. FIDpreferred: Morgen schon würde sie mit diesem Bücherwurm die Buchhandlung stürmen.
b. FIDdispreferred: Morgen schon würde er mit dieser Leseratte die Buchhandlung stürmen.
c. Neutral: Er wollte mit ihr am nächsten Tag in die Buchhandlung gehen.

When the last Harry Potter book was published, Luisa scraped up her money.
She told her best friendmale right away.
a. FIDpreferred: Tomorrow she would hit the bookstore with that bookwormmale.
b. FIDdispreferred: Tomorrow he would hit the bookstore with that bookwormfemale.
c. Neutral: He wanted to go to the bookstore with her the next day.

In our experiment, we will collect evidence for the hypothesis that follows the discussion presented
above:

H1: A protagonist that is more prominent in terms of referential expression and grammatical func-
tion is more available as the anchor for FID than a competing referent.

In order to determine the influence of prominence on the acceptability of FID we distinguished our
two referents by means of maximal and minimal prominence in terms of grammatical function, the
number of occurrences and the type of referring expression. We predict that FID anchored to the
more prominent referent, i.e., the agent in subject position, introduced by a proper name and picked
up by a personal pronoun (compare (14a)/(15a)) will more likely be accepted as the perspectival
center of a sentence in FID than the competing referent mentioned only once in object position
(compare (14b)/(15b)). The neutral condition (compare (14c)/(15c)) is similar to the dispreferred
condition where the less prominent referent is the subject of the target sentence. The neutral con-
dition serves as a control condition: If the neutral condition is rated significantly better than the
dispreferred FID condition we may conclude that the low acceptability of the dispreferred condi-
tion is at least partly due to a low acceptability of the FID regardless of the change of the subject
(another relevant factor might be the presence of rather unusual terms such as bookworm in the FID
conditions as opposed to the neutral condition)10.

2.1.1 MATERIAL

We constructed 22 short stories similar to examples (14) and (15) in 3 conditions. All stories con-
sisted of 3 sentences. The first sentence introduced one referent (R1) in subject position with a
proper name. Also, the first sentence set the context to a specific event in the past (e.g., when Ger-
many won the World Championship, last Valentine’s day). We used an explicit reference to the past
in order to emphasize the contrast of the narrative context and the deictic expression used in the
FID, which in all cases refers to the present (now, today) or an immediate or close future (soon,
tomorrow). In the second sentence the referent was picked up by a personal pronoun in subject

10. We are grateful to an anonymous reviewer for pointing this out to us.

13


MEUSER, HÖRL AND HINTERWIMMER

position11 interacting with a second referent (R2) introduced by a noun phrase anchored to the first
referent with a possessive pronoun (e.g., his friend) in object position. The two referents differ in
gender in order to prevent ambiguity of the target sentence12.

The two FID conditions in the third sentence feature at least 3 indicators of FID: a temporal deictic
expression, a verb in subjunctive II mood (German würde) and kinship, family terms or qualitative
nouns (e.g., her/his darling). In our FID conditions the target sentence varied with respect to R1
and R2. In the first condition the FID can only be anchored to R1, i.e., the referent introduced in the
first sentence by a proper name, by the choice of the respective male or female pronoun and a qual-
itative noun phrase matching the gender of the second referent (compare 14a: he [. . . ] his darling).
The second condition features FID anchored to R2, the referent introduced in the second sentence,
by the choice of a pronoun that matches the second protagonist´s gender whereas the qualitative
expression must refer to the protagonist introduced in the first sentence (compare (14) b: she [. . . ]
her darling). We paid particular attention to the fact that the FID can equally plausibly be linked
to either one of two referents, i.e., the fiancée is just as likely to think the thought regarding their
wedding plans as Robert is. A third, neutral condition (compare (14c)/(15c)) does not feature any
indicators of FID, but it matches the second condition, i.e., R2 is in subject position.

2.1.2 PROCEDURE

We randomly distributed 22 items across 3 lists. Unfortunately, this led to an unbalanced distribu-
tion of the three conditions across the lists – per participant 7 responses for two conditions and 8
for the third13. The 22 items were randomly mixed with 44 fillers on a paper-pencil questionnaire.
Fillers were similar to the items in length, syntax and the number of referents mentioned. In order
to mask our manipulation, we constructed fillers deliberately violating constraints of demonstrative
and personal pronouns resulting in odd, but nevertheless grammatical references to yield low ac-
ceptability. The task was to judge naturalness of the third sentence in the context of the first two on
a scale from 1, entirely unnatural, to 7, entirely natural. In order to prevent participants from judging
according to grammatical acceptability or plausibility of the scenario the instructions stressed that
all sentences are well-formed and the stories are plausible. Participants were encouraged to use the
entire scale from 1 to 7 according to their subjective intuitions and fill in the number in a bracket
behind each item14.

In April 2017, 89 students of the University of Cologne, participated voluntarily in our experi-
ment. 14 Participants got excluded from analysis as they were not monolingual speakers of German
or did not complete the entire questionnaire.

11. We are aware that subjecthood coincides with first mention. However, we decided to keep the design simple and
neglected a potential OVS condition, especially as the OVS syntax is highly marked. Testing an OVS condition may
lead to unintended confounds. We need to point out that the effect that we attribute to subjecthood throughout the
paper may also be the result of first mention – or potentially the interaction of subjecthood and first mention.

12. Gender of the prominent protagonist was equally balanced. Nevertheless, gender stereotypes were avoided in order
to prevent preferences of one referent as anchor for the FID depending on gender.

13. Although this was not an ideal distribution, since we used linear mixed-effects modeling for analyzing the data, we
relied on the robustness of mixed-effects modeling to handle missing data (Baayen et al., 2008) and assume that the
”missing” one response per participant had negligible influence on our effects.

14. In our analysis we treated the scale as an interval scale as the scale was not further labeled, we assume that participants
used the scale continuously.

14


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

2.1.3 DATA ANALYSIS AND RESULTS

Ratings from 75 questionnaires were considered in the analysis. Acceptability judgements were
elicited for 21 items in 3 conditions. In our experiment we tested 22 items, one of which had
to be excluded due to the implausibility of the scenario attentive participants commented on15.
This resulted in a total of 1575 observations. In order to test differences between conditions we
modeled the data using Bayesian mixed effects models 16 estimated in R17 (R Core Team, 2015).
We used a model with acceptability scores as outcome and condition as the sole predictor with
random intercept and random slopes for condition, for participants and items18. Analogously to
frequentist statistics we considered effects to be significant if the 95%-credible interval does not
include 019. The model estimate for the FIDpreferred condition was 4.64 whereas the model estimate
for the FIDdispreferred condition was 3.63. The model revealed a significant difference of 1.01 points
(SE = 0.16, CI: [0.69;1.33]). For the neutral condition the model estimate was 5.04. It served
as a control condition in order to test whether a lower acceptability of the FIDdispreferred version is
due to a change in subject. The comparisons revealed that participants rated the neutral version
significantly better than the FID dispreferred condition (estimate = 1.40; SE = 0.18; CI: [1.06;1.75]).
In our experiment, we tested the claim that FID is more likely to be anchored to a maximally
prominent referent than to a competing referent with rather low prominence (Hinterwimmer, 2019).
The significant difference in the ratings of FID in our preferred condition in comparison to our
FIDdispreferred condition indicates that the hypothesis that all referents are equally available as the
perspectival center as long as the interpretation of the FID is plausible can be rejected.

The comparison of the two FID conditions supports the hypothesis that the maximally prominent
referent is preferred as perspectival center. However, the mediocre ratings, −0.49 below average,
of the FIDdispreferred condition show that the interpretation of the FID as a thought of the second
referent is not entirely impossible while the significant difference indicates a clear preference.

Although we did not test this contrast, the difference of the neutral and the preferred condition
suggests that both our FID conditions are not as acceptable as neutral story continuations. Lower
ratings in the sample of the FIDpreferred condition in which the FID is anchored to the prominent
referent compared to the neutral condition supports concerns that FID in such short stories is rather
unnatural for presumably two – not necessarily unrelated – reasons: FID is a phenomenon typically

15. Participants reported implausibility orally and/or left a comment on the questionnaire, i.e., it was falsely claimed that
the Rock am Ring festival takes place in February not in early summer.

16. In Hinterwimmer and Meuser (2019) we presented the results of this experiment using a non-Bayesian linear mixed
effects models with the lmer function from the lme4 package. With the data of experiment 2 however, this prohibited
the use of a maximal random effects structure as convergence issues occurred. In line with Barr et al. (2013) we
decided to keep the maximal random effects structure by opting for a Bayesian analysis. Contrary to frequentist
mixed models Bayesian mixed models are more stable in the sense that they almost always converge even with
complex random effects structures. We therefore decided to use Bayesian mixed models for the whole analysis to
keep it consistent and maintain the use of a maximal random effects structure in experiment 2. For an introduction to
Bayesian mixed models see: https://www.sciencedirect.com/science/article/abs/pii/S0095447017302310.

17. Using the brms package (Bürkner, 2017). We used weakly informative priors, e.g., for the condition coefficients we
used a normally distributed prior with mean 0 and sd 5.

18. We used treatment contrasts with FIDdispreferred as the reference category. We specified the model as follows:
”acceptability 1 + condition+ (1 + condition|participant) + (1 + condition|item)”

19. If 0 is not included in the 95%- credible interval, then 0 is not included in the range of 95% of the most plausible
values. This means that the posterior probability of the effect being 0 (or less extreme) is maximally 5%.

15


MEUSER, HÖRL AND HINTERWIMMER

found in narratives and can thus possibly not be perceived to sound entirely natural in 3-sentence
stories. Second, if the acceptability of FID depends on prominence, the prominence of a referent
must be sufficiently established. The latter assumption will be subject to further research.

Figure 1: Experiment 1: Acceptability ratings (model estimate and 95%-credible interval)

2.1.4 DISCUSSION

A shortcoming of our acceptability study concerns the comparability of the test items we used. The
task explicitly stated that the third sentence has to be judged with regard to the first two sentences,
although it is possible that participants rated the acceptability of just the third sentence and those
differed across conditions. In order to anchor the FID to one of the two referents we used kinship,
family terms or qualitative nouns that match only one of the two referents in terms of gender. In
cases where we compare minimal pairs such as Liebster (male darling) to Liebste (female darling)
– both commonly used to refer to a loved one – that should not make a difference. However, there
are only few such perfect minimal pairs so that in most cases we ended up comparing Bücherwurm
(m., bookworm) with Leseratte (f., bookworm, literally a reading rat), Schleckermaul (n., sweet
tooth literally a yummy mouth) with Naschkatze (f., again a sweet tooth, literally a nibble cat) and
Tratschtante (f., gossip aunt) with Lästermaul (n., gossip mouth) for which we do not have any
predictions regarding their overall acceptability. In order to make sure that results were not related
to the potentially lower acceptability of the expressions used in the FID dispreferred condition in our
follow-up we improved the comparability of the target sentence.

16


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

2.2 Experiment 2

We regard the significant contrasts found between the FID conditions in the first study presented
above as indicators that acceptability judgements indicate anchoring preferences of FID. In the
previous experiments acceptability varied with respect to the naturalness of FID anchored to a max-
imally prominent protagonist compared to FID anchored to a minimally prominent protagonist20.
Now, in our second experiment, the same methodology was used to measure the differing anchoring
potential of referents depending on their prominence status. Prominence-lending cues we investi-
gated were grammatical function as well as type of referential expression. In an acceptability rating
study items similar to (16) were tested. Again, a short story was followed by an utterance in FID
mode that can only be anchored to one of the two referents in terms of gender.

(16) a. Lynn sprach Pablo auf den nächsten Zug an, als eine Durchsage Verspätungen aller Züge
aufgrund starken Schneefalls verkündete.
b. Pablo sprach Lynn auf den nächsten Zug an, als eine Durchsage Verspätungen aller Züge
aufgrund starken Schneefalls verkündete.

c. Eine Reisende sprach Pablo auf den nächsten Zug an, als eine Durchsage Verspätungen
aller Züge aufgrund starken Schneefalls verkündete.
d. Ein Reisender sprach Lynn auf den nächsten Zug an, als eine Durchsage Verspätungen
aller Züge aufgrund starken Schneefalls verkündete.

e. Eine Reisende sprach einen Reisenden auf den nächsten Zug an, als eine Durchsage
Verspätungen aller Züge aufgrund starken Schneefalls verkündete.
f. Ein Reisender sprach eine Reisende auf den nächsten Zug an, als eine Durchsage Verspätungen
aller Züge aufgrund starken Schneefalls verkündete.

g. Lynn sprach einen Reisenden auf den nächsten Zug an, als eine Durchsage Verspätungen
aller Züge aufgrund starken Schneefalls verkündete.
h. Pablo sprach eine Reisende auf den nächsten Zug an, als eine Durchsage Verspätungen
aller Züge aufgrund starken Schneefalls verkündete.

FID: Oh Mann, jetzt würde sie bestimmt ihren Anschlusszug verpassen.

a) Lynn asked Pablo about the train when severe delays due to heavy snowfall were an-
nounced.
b. Pablo asked Lynn about the train when severe delays due to heavy snowfall were an-
nounced.

c. A travelerf asked Pablo about the train when severe delays due to heavy snowfall were
announced.
d. A travelerm asked Lynn about the train when severe delays due to heavy snowfall were
announced.

20. We will refer to a protagonist referred to with a proper name in subject position as maximally prominent while a
protagonist in object position referred to with an indefinite is minimally prominent.

17


MEUSER, HÖRL AND HINTERWIMMER

e. A travelerf asked a travelerm about the train when severe delays due to heavy snowfall
were announced.
f. A travelerm asked a travelerf about the train when severe delays due to heavy snowfall were
announced.

g. Lynn asked a travelerm about the train when severe delays due to heavy snowfall were
announced.
h. Pablo asked a travelerf about the train when severe delays due to heavy snowfall were
announced.

FID: Oh man, now she would miss her connecting train!

In line with the argumentation presented in Section 1.4, the study investigates two main hypotheses:

H2: Protagonists functioning as subjects are more available as anchors for FID than protagonists
functioning as objects.

H3: Protagonists referred to with a proper name are more available as anchors for FID than pro-
tagonists introduced by an indefinite NP.

In the hypotheses we make claims with respect to the two prominence-lending cues, i.e., gram-
matical function and the type of referential expression, in isolation. However, as a referent holds
both prominence-lending cues at the same time, i.e., the referent is referred to in a certain way and
the expression has a certain grammatical function, it is worthwhile to explore the question of how
these different prominence-lending cues interact (Q1): Does the type of referential expression play
a bigger role than the grammatical function with respect to the anchoring of FID? Based on the
discussion presented in Section 1.4, a second question that calls for an exploratory investigation
is whether the competition of the two protagonists has an impact on the anchoring preference. A
similar effect as the one presented in example (11) (”A young lady asked Martin/ Lilly asked Mar-
tin...”), where the prominence status of the object changes as the subject gets more prominent, shall
be investigated by considering the effect of the expression used to refer to the second referent. The
second question of interest (Q2) may thus be phrased as Is it more acceptable to anchor FID to a
referent when the competing referent is less prominent in terms of the expression by which she or he
is referred to?

2.2.1 MATERIAL

For this experiment we constructed 48 test items similar to (16) in 8 conditions. Items consisted of
two sentences. In the first sentence (S1) two referents (R1 & R2), one male and one female, were
introduced as subject and object, respectively. In the second sentence (S2) an utterance in FID mode
was presented that could only be anchored to R1 in terms of gender21. Unlike the first experiment
we varied the gender features of the two referents in the context so that the target sentence would

21. A reading in which the referent who is not referred to with the pronoun is the anchor of the FID, i.e., Pablo thinks
about that poor traveler that she would miss her connecting train, is in theory possible. However, that reading should
yield equally low results as it is rather absurd. If a reader accepts the dispreferred conditions as a recursive reading,
i.e., Pablo thinks that she thinks, that reading is again in line with the hypothesis: the FID is anchored to the more
prominent referent even if that results in a rather absurd interpretation.

18


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

Condition label Subject-Object Ref.-ex R1 Ref.-ex R2
A: R1name – R2name R1-R2 name name
B: R2name – R1name R2-R1 name name
C: R1indef. – R2name R1-R2 indef. name
D: R2indef. – R1name R2-R1 name indef.
E: R1indef. – R2indef. R1-R2 indef. indef.
F: R2indef. – R1indef. R2-R1 indef. indef.
G: R1name – R2indef. R1-R2 name indef.
H: R2name – R1indef. R2-R1 indef. name

Table 1: Experimental conditions in Experiment 2

remain exactly the same throughout conditions and yet varies with respect to the two referents. The
referents were either referred to with a proper name or with an indefinite noun phrase (NP) and
functioned as subjects or objects in S1 resulting in 8 possible combinations presented in Table 1.
We will refer to the different conditions with respect to the subject in S1 and the object in S1. For
half of the items we chose a male referent for R1 while the other half were female in order to prevent
gender biases. In all items the utterance in FID mode featured an interjection, a temporal deictic
expression and a verb in subjunctive II mood (German würde).

To be able to check if participants attentively read the items 16 filler items were included that were
designed to yield low rating while another 16 filler items should be unproblematic and yield high
ratings22. Participants that rate the deliberately bad items better than the good ones will be excluded
from the analyses.

2.2.2 PROCEDURE

All test items were equally distributed across 8 lists so that each participant saw each item in only
one condition. Items were randomized for each participant and mixed with 64 filler items. Fillers
are similar to the test items in terms of length. Due to the rather large number of items – 112 in total
– a 3-minute break was enforced after half of the items were presented.

Participants were asked to rate the naturalness of the item on a 7-point likert scale – 1, entirely
unnatural, to 7, entirely natural. Participants were recruited in class and participated for course
credit. The experiment was programmed using PCIbex (Zehr and Schwarz, 2018). In December
2021, ratings from 129 participants were recorded. Results from 7 participants had to be excluded
from the analysis as they were not native speakers of German. Further, we excluded the results from
3 participants as they rated the fillers that were deliberately designed to trigger low acceptability
better than a set of deliberately good filler items.

22. Example for a good filler item: Marleen und Laurin kochen beide gerne. Marleen mag die argentinische Küche
und Laurin mag auch südamerikanisches Essen. (Engl. Marleen und Laurin both love to cook. Marleen likes the
Argentinian cuisine and Laurin also likes South American food.) Example for a bad filler item: Melina und Thomas
haben beide schöne Haare. Melina hat schwarze Haare und Thomas hat auch blonde Haare. (Engl. Melina and
Thomas both have beautiful hair. Melina has black hair and Thomas also has blond hair.)

19


MEUSER, HÖRL AND HINTERWIMMER

2.2.3 DATA ANALYSIS AND RESULTS

Ratings from 119 questionnaires were considered in the analysis. Acceptability judgements were
elicited for 48 items in 8 conditions. Acceptability judgements were submitted to a Bayesian mixed
effects model. We used a full set of interactions of the factors grammatical function and referen-
tial expression of R1 and R2. We also included maximal random effects structure for subjects and
items23.

There is a significant main effect of grammatical function of 0.57 (SE = 0.07, CI: [0.43;0.72]).
This means that sentences with R1, i.e., the anchor for the utterance in FID mode, in subject posi-
tion were rated better than sentences with R1 in object position. Additionally, there is a significant
main effect of the referential expression used for R1 (estimate = -0.62, SE = 0.07, CI: [-0.75; -0.49])
indicating that referents referred to with a proper name are more acceptable as anchors for FID than
referents referred to with an indefinite NP. There is also a significant main effect of the referential
form of R2 (estimate = 0.37, SE = 0.06, CI: [0.26; 0.47], see Figure 2), i.e., sentences received
higher ratings when R2, the referent that cannot be the anchor for the utterance in FID mode, was
referred to with an indefinite than with a proper name. There was a significant two-way interaction
of the referential form of R1 and R2 (estimate = 0.18, SE = 0.08, CI: [0.03; 0.34], see Figure 2).
This suggest that the effect of R2 depends on the referential expression used to refer to R1, that is
that R2 plays a bigger role if R1 is referred with an indefinite NP. However, an investigation of the
three-way interaction of the referential form of R1 and R2 and the grammatical function reveals a
more nuanced interplay of these factors. The interaction effect of the referential expressions of R1
and R2 varies significantly depending on the grammatical function (estimate = 0.64, SE = 0.18, CI:
[0.29; 1.01]). More specifically, the difference of the effect of R2 between R1name and R1indef. is
higher when R1 is in subject position (compared to R1 in object position). When relating this to
Figure 2, it becomes clear that ’higher’ here means that the effect of R2 seems to differ between
R1name and R1indef. to a noteworthy extent only when R1 is in subject position ((G-A) ¡ (E-C)).
In the case where R1 is in object position the effect of R2 does not seem to differ much between
R1name and R1indef. ((D-B) ≈ (F-H)).

This implies that the two-way interaction between the referential form of R1 and R2 which was
described earlier can be somewhat misleading, because it is (almost exclusively) driven by the ef-
fect of R2 when R1 is in subject position. Therefore, the effect of R2 does not seem to generally
depend on R1 but only when R1 is in subject position. Here, the role of R2 is high(er) when R1 is
referred to with an indefinite NP (E-C) and almost negligible when R1 is referred to with a proper
name (G-A).

23. We specified the model as follows: acceptability 1+ gramfunc ∗R1reftype ∗R2reftype+ (1+ gramfunc ∗
R1reftype ∗R2reftype|participant)+ (1+ gramfunc ∗R1reftype ∗R2reftype|item). In order to interpret
the lower order effects as main effects we used deviation coding for all 3 factors (0.5/+0.5).

20


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

Figure 2: Experiment 2: Acceptability ratings

Condition Model estimate CI
A: R1name – R2name 5.37 [5.15 ; 5.61]
B: R2name – R1name 4.68 [4.44 ; 4.91]
C: R1indef. – R2name 4.55 [4.31 ; 4.79]
D: R2indef. – R1name 5.12 [4.90 ; 5.36]
E: R1indef. – R2indef. 5.16 [4.93 ; 5.38]
F: R2indef. – R1indef. 4.39 [4.14 ; 4.64]
G: R1name – R2indef. 5.48 [5.25 ; 5.72]
H: R2name – R1indef. 4.08 [3.81 ; 4.33]

Table 2: Experiment 2: Acceptability ratings (model estimate and 95%-credible interval)

2.2.4 DISCUSSION

The significant main effect of grammatical function indicates that protagonists in subject position
are more available as anchors for FID than protagonists in object position (H2). Regarding these re-
sults we thus conclude that grammatical function can be considered a prominence-lending cue with
respect to establishing the perspectival center that serves as the anchor for an utterance in FID mode.

Recall that R1 is the referent that the utterance in FID has to be anchored to in terms of gender.
The significant main effect of referential expression of R1 indicates that protagonists referred to
with a proper name are more available as anchors for FID than protagonists introduced by an indef-
inite NP (H3).
As both factors add to the referent´s prominence status, as discussed in Section 1.4, we also inves-
tigated the impact of each factor and how the two prominence-lending cues interact. The estimates

21


MEUSER, HÖRL AND HINTERWIMMER

for grammatical function and referential expression indicate that the role of subjecthood is nearly
similar to that of the type of referential expression.

We therefore want to make the following claim with respect to Q1: Grammatical function affects
the anchoring of FID just as much as type of referential expression.

While the main effect of the referential expression of R1 indicates that R1 is more available if it
is referred to with a proper name, the main effect of R2 counteracts in a similar way: Just as R1
becomes more or less available as the referential expression increases or decreases its prominence
status, so does R2. Now, if R2 becomes more prominent in virtue of the referential expression, it
becomes more available as the perspectival center and therefore both protagonists compete. With
respect to Q2 we may thus conclude that a high prominence status of the competing referent (R2)
makes the anchoring of FID to R1 particularly difficult.

While we found a main effect for the referential expression, we must also point out that in our
sample the effect of referential expression of R2 does not show in the case of R1 being in sub-
ject position and referred to with a proper name. Conditions A and G received rather similar high
ratings (compared to the oppositional pairs E&C, F&H and D&B). Therefore, we want to suggest
the following: A referent that is maximally prominent in virtue of being referred to with a proper
name in subject position is the default anchor for an utterance in FID mode regardless of the type of
referential expression used to refer to the competing referent.

Complementary to the reporting of the main and interaction effects we would like to provide a
matching interpretation from a slightly different perspective. That is, the three factors increase the
acceptability of R1 as anchor for the FID expression: A proper name as referential form of R1,
subject position of R1 and an indefinite expression that refers to the competing referent R2 enhance
the availability of R1 to serve as the perspectival center. If we now sort the conditions with respect
to the number of these features we get the following pattern: Condition H with no such enhancing
features yielded the lowest acceptability (4.08). Conditions B, C and F with one enhancing feature
have a slightly higher acceptability (from 4.39 to 4.67). Conditions with two features, namely A, D
and E, are even more acceptable (from 5.11 to 5.37) while condition G with all three features has
the highest acceptability (5.47). Figure 3 illustrates this.

Before we forge ahead to the investigation of prominence in larger discourse, we briefly want to
recall the first experiment presented in this chapter and point out how the main effects found in the
second experiment explain the contrast found in the first experiment. In our first experiment we pre-
sented an attempt to verify our intuitions that referents that are prominent with respect to referential
expression and grammatical function are more easily available as perspectival centers than compet-
ing referents, aiming to find a contrast between maximally and minimally prominent protagonists.
Contrasts found for the FID preferred conditions and the FIDdispreferred conditions are parallel to the
contrasts found between the conditions G and H in the second experiment (see (17)). Here again,
the referent that serves as the anchor for the FID is maximally prominent in terms of referential
expression and grammatical function in condition G and minimally prominent in condition H.

22


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

Figure 3: Experiment 2: Acceptability ratings depending on the number of prominence-enhancing
features

(17) Exp.1:
When the wedding of Prince William and Kate was broadcast on TV, Robert could hardly
wait for his own wedding.
He, too, had proposed to his girlfriend.
FIDpreferred: Soon he would walk down the aisle with his darling.
FIDdispreferred: Soon she would walk down the aisle with her darling.

Exp. 2:
G: R1name – R2indef.: Lynn asked a travelerm about the train when severe delays due to heavy
snowfall were announced.
H: R2name – R1indef.: Pablo asked a travelerf about the train when severe delays due to heavy
snowfall were announced.
Oh man, now she would miss her connecting train!

The strong contrast between the acceptability judgements of condition G and H (1.39 points) as well
as the contrasts found for the FIDpreferred conditions and the FIDdispreferred conditions (1.02 points)
may be explained by the main effects reported above: A protagonist is more available as the anchor
for an utterance in FID mode if it is in subject position and referred to with a proper name than when
it is referred to with an indefinite NP and the competing referent is referred to with an indefinite NP
instead of a proper name. The results thus support the intuitions outlined in chapter 1.4 that lead
to the item construction (repeated in (17)), i.e., it feels more natural to anchor an utterance in FID
mode to Robert than to the girlfriend (exp.1) and to Lynn rather than to the traveler (exp.2).

23


MEUSER, HÖRL AND HINTERWIMMER

2.3 Experiment 3

Building on the evidence we found in experiment 1 and 2, which suggests that locally prominent
referents are preferred as perspectival centers, the effect of local prominence in larger discourse is
investigated in Experiment 3. In particular, we want to test if the effect of local prominence prevails
in a context where a competing referent is globally prominent. Recall the discussion presented in
Section 1.4 where it was claimed that a referent that serves as the discourse topic serves as the
perspectival center regardless of any reference to a competing referent in a sentence preceding the
FID, as in (10) (repeated in (18)).

(18) Lisa was playing in the schoolyard. A boy pushed her into the stinging nettles.
Ouch, that itched!

Even though the boy is the subject and the agent of the sentence preceding the FID we have no
difficulties anchoring the FID to Lisa, who is the object and the patient. For this example, however,
it may be claimed that the preference for Lisa is the result of multiple prominence lending cues:
Lisa is the subject of the first sentence, she is introduced with a proper name and she is picked up
with a personal pronoun in the second sentence, i.e., she is mentioned twice – unlike the competitor,
who is only referred to once with an indefinite NP.

In order to investigate the effect of global as opposed to local prominence more systematically,
we conducted an experiment where two referents are introduced with a proper name. However,
they vary with respect to how often they are mentioned as well as with respect to their grammatical
function in the sentence preceding the FID.

In a third acceptability rating study, we presented larger contexts similar to (19). Again, partici-
pants (n = 116) were asked to rate the naturalness of a target sentence in FID mode with respect to
the context.

(19) A: Kein Feierabend in Sicht
(S1) Im Büro gehörten Überstunden zum Alltag. (S2) Da musste ein ruhiger Kopf bewahrt
werden. (S3) Eine gute Arbeitsteilung war notwendig, um alles zu erledigen.
(S4) Trotz des hohen Zeitdrucks duldete das Management keine Verzögerungen.
(S5) Fred gab Caroline die Akten. (S6) Wehe, wenn die ihn heute hängen lassen würde.

B: Kein Feierabend in Sicht
(S1) Im Büro gehörten Überstunden zum Alltag. (S2) Da musste ein ruhiger Kopf bewahrt
werden. (S3) Eine gute Arbeitsteilung war notwendig, um alles zu erledigen.
(S4) Trotz des hohen Zeitdrucks duldete das Management keine Verzögerungen.
(S5) Caroline gab Fred die Akten. (S6) Wehe, wenn die ihn heute hängen lassen würde.

C: Kein Feierabend in Sicht
(S1) Im Büro gehörten Überstunden zum Alltag. (S2) Caroline atmete einmal tief durch. (S3)
In der Frühstückspause hatte sie Fred um Unterstützung gebeten.
(S4) Sie wusste, dass das Management keine weiteren Verzögerungen mehr tolerieren würde.
(S5) Fred gab Caroline die Akten. (S6) Wehe, wenn die ihn heute hängen lassen würde.

24


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

D: Kein Feierabend in Sicht
(S1) Im Büro gehörten Überstunden zum Alltag. (S2) Fred atmete einmal tief durch. (S3) In
der Frühstückspause hatte er Caroline um Unterstützung gebeten.
(S4) Er wusste, dass das Management keine weiteren Verzögerungen mehr tolerieren würde.
(S5) Caroline gab Fred die Akten. (S6) Wehe, wenn die ihn heute hängen lassen würde.

E: Kein Feierabend für Caroline
(S1) Caroline hatte einen langen Arbeitstag vor sich. (S2) Sie atmete einmal tief durch. (S3)
In der Frühstückspause hatte sie Fred um Unterstützung gebeten.
(S4) Sie wusste, dass das Management keine weiteren Verzögerungen mehr tolerieren würde.
(S5) Fred gab Caroline die Akten. (S6) Wehe, wenn die ihn heute hängen lassen würde.

F: Kein Feierabend für Fred
(S1) Fred hatte einen langen Arbeitstag vor sich. (S2) Er atmete einmal tief durch. (S3) In
der Frühstückspause hatte er Caroline um Unterstützung gebeten.
(S4) Er wusste, dass das Management keine weiteren Verzögerungen mehr tolerieren würde.
(S5) Caroline gab Fred die Akten. (S6) Wehe, wenn die ihn heute hängen lassen würde.

A: More work to do
(S1) In the office, working overtime was not unusual. (S2) Keeping a clear head was impor-
tant. (S3) Also dividing work was crucial to get everything done.
(S4) Despite the tremendous time pressure, the management did not tolerate delays.
(S5) Fred handed Caroline the papers. (S6) She better not let him down today.

B: More work to do
(S1) In the office, working overtime was not unusual. (S2) Keeping a clear head was impor-
tant. (S3) Also dividing work was crucial to get everything done.
(S4) Despite the tremendous time pressure, the management did not tolerate delays.
(S5) Caroline handed Fred the papers. (S6) She better not let him down today.

C: More work to do
(S1) In the office, working overtime was not unusual. (S2) Caroline took a deep breath. (S3)
During the morning break she had asked Fred for help.
(S4) She knew that the management did not tolerate delays.
(S5) Fred handed Caroline the papers. (S6) She better not let him down today.

D: More work to do
(S1) In the office, working overtime was not unusual. (S2) Fred took a deep breath. (S3)
During the morning break he had asked Caroline for help.
(S4) He knew that the management did not tolerate delays.
(S5) Caroline handed Fred the papers. (S6) She better not let him down today.

E: No end of work for Caroline
(S1) Caroline was facing a long day at work. (S2) She took a deep breath. (S3) During the
morning break she had asked Fred for help.
(S4) She knew that the management did not tolerate delays.
(S5) Fred handed Caroline the papers. (S6) She better not let him down today.

25


MEUSER, HÖRL AND HINTERWIMMER

F: No end of work for Fred
(S1) Fred was facing a long day at work. (S2) He took a deep breath. (S3) During the morning
break he had asked Caroline for help.
(S4) He knew that the management did not tolerate delays. (S5) Caroline handed Fred the
papers. (S6) She better not let him down today.

The first two conditions serve as a baseline. Based on the results presented above we expect them
to yield results in accordance with H2.

H2: Protagonists functioning as subjects are more available as anchors for FID than protagonists
functioning as objects – if there is no competing referent activated in the context.

For the items similar to condition C and D we expect to find an effect of global prominence, i.e., the
referent that is mentioned in the context should also be available as the perspectival center.

H4: Protagonists that are highly activated in a discourse in terms of being repeatedly mentioned
in subject position are just as available as anchors for FID as protagonists that are less often
referred to in a discourse but in subject position in the sentence preceding the FID.

For items similar to (19E) and (19F) we expect to find an even stronger effect, i.e., they are expected
to be preferred as perspectival centers over the locally prominent referent.

H5: Protagonists that are highly activated in a discourse in terms of being established as the dis-
course topic in a topic-establishing sentence, repeatedly mentioned in subject position, and
mentioned in a title are more available as anchors for FID than protagonists that are less often
referred to in a discourse but in subject position in the sentence preceding the FID.

2.3.1 MATERIAL

For our third experiment we constructed 18 experimental items similar to (19). All items consist
of six-sentence stories. The first four sentences established a neutral context or one where one ref-
erent was (weakly or strongly) activated. The last two sentences should be interpreted as part of
a main storyline. In Sentence 5, the two referents interact, and Sentence 6 can only be interpreted
as a thought from the perspective of one of the two referents (due to the gender of the pronoun it
contains) rendered as FID. Again, we manipulated the gender of the two referents in the context so
that the target sentence in FID mode remains the same in all conditions.

In condition A and B the referents only get introduced in S5, i.e., in the sentence preceding the
FID. In condition C, D, E and F both referents are mentioned in the context, yet they differ in the
number of references. That is, in conditions C and D the globally prominent referent is mentioned
in S2, S3 and S4, while the competitor is mentioned once in S3. In conditions E and F the globally
prominent referent is additionally mentioned in the title and in S1 (see example (18) or Table 3).

26


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

Condition Competitor
(not the local subject) Local subject FID

A: X – X – R1 none R1 R1
B: X – X – R2 none R2 R1
C: X – R2 – R1 weak activation R1 R1
D: X – R1 – R2 weak activation R2 R1
E: R2 – R2 – R1 strong activation R1 R1
F: R1 – R1 – R2 strong activation R2 R1

Table 3: Experimental conditions in Experiment 3

2.3.2 PROCEDURE

All conditions were distributed over six lists so that each participant saw each item in only one con-
dition. The 18 experimental items were randomly mixed with 28 fillers (similar to the experimental
items in terms of content, style, and length). In order to assure that participants carefully read the
entire story, rather than just the target sentence, comprehension questions randomly asked for infor-
mation presented in the fillers.
As in experiment 1 and 2, participants’ task was to rate the naturalness of the last sentence in the
context of the preceding sentences on a scale from 1 (entirely unnatural), to 7 (entirely natural).
The experiment was conducted on Qualtrics 24. In May 2020, 122 undergraduate students of the
University of Cologne participated in the experiment for course credit.

2.3.3 DATA ANALYSIS AND RESULTS

Data from six participants had to be excluded from the analysis as they performed poorly on the
comprehension questions (8 or fewer out of 12, mean accuracy = 11.19).
The remainder of the data (n=116) was analyzed using Bayesian mixed effects models estimated in
R (R Core Team, 2015).

The model used acceptability ratings as the outcome and the two factors of local subject and ref-
erence in the context as predictors with random intercept and random slopes for condition, partici-
pants, and items25,26,27. Comparisons of individual conditions were calculated using the emmeans
package (?).

24. https://www.qualtrics.com
25. Treatment contrasts were used with condition A (no competitor/R1) as the reference category. The model was speci-

fied as follows: ( competitor ∗ local + (1 + competitor ∗ local|item) + (1 + competitor ∗ local|participant)
26. The raw data and the script are available online: https://osf.io/wegtn/
27. Analogously to frequentist statistics, effects were considered to be significant if the 95%-credible interval did not

include 0. If 0 is not included in the 95%-credible interval, then 0 is not included in the range of 95% of the most
plausible values. This means that the posterior probability of the effect being 0 (or less extreme) is maximally 5%.

27


MEUSER, HÖRL AND HINTERWIMMER

Condition Model estimate 95%-CI
A: X – X – R1 4.76 [4.35; 5.22]
B: X – X – R2 4.16 [3.71; 4.59]
C: X – R2 – R1 4.49 [4.03; 4.95]
D: X – R1 – R2 4.73 [4.27; 5.19]
E: R2 – R2 – R1 4.53 [4.09; 4.98]
F: R1 – R1 – R2 4.87 [4.42; 5.34]

Table 4: Experiment 3: Acceptability ratings (model estimates and 95%-credible interval)

A comparison of condition A and B confirms H2 and thus replicates the results of Experiments 1
and 2 (Protagonists functioning as subjects are more available as anchors for FID than protago-
nists functioning as objects – if there is no competing referent activated in the context). That is,
the model difference of -0.60 points (SE = 0.13, CI: [-0.86; -0.35]) indicates that items with R1 in
subject position in S5 were rated significantly higher than items with R2 in subject position.

H4 (i.e., protagonists that are highly activated in a discourse in terms of being repeatedly men-
tioned in subject position are more available as anchors for FID than protagonists that are less
often referred to in a discourse but in subject position in the sentence preceding the FID) was tested
in terms of, first, the interaction of conditions A & B and C & D, and, second, the comparison of
the model estimates for condition C and D.

The difference of A and B compared to the difference of C and D reveals a significant two-way
interaction (estimate = 0.83, SE = 0.17, CI: [0.49; 1.17]). This indicates that repeatedly mentioning
a referent in the context overrides the subject preference that shows up in the effect observed for A
and B. While conditions C & D reverse the effect in A & B, the comparison of the estimates for
conditions C and D was not significant (estimate = 0.23, SE = 0.13, CI: [-0.04; 0.48]). That is, the
globally prominent referent was not preferred significantly in comparison to the locally prominent
referent in conditions C and D where the globally referent was weakly activated (compare Table 3).

Likewise, H5 (i.e., Protagonists that are highly activated in a discourse in terms of being estab-
lished as the discourse topic in a topic-establishing sentence, repeatedly mentioned in subject po-
sition, and mentioned in a title are more available as anchors for FID than protagonists that are
less often referred to in a discourse but in subject position in the sentence preceding the FID) was
tested in terms of a comparison of the differences of condition A & B and E & F and a pairwise
comparison of conditions E and F.

Again, the two-way interaction between the difference of condition A & B compared to the dif-
ference of E & F is significant (estimate = 0.94, SE = 0.17, CI: [0.60; 1.29]). Just as for condition
C and D, the presence of a competitor overrides the effect of the local subject.

Unlike the pairwise comparison of condition C and D, with model estimates of 4.49 and 4.73, the
difference between condition E and F, with model estimates of 4.53 and 4.87, is significant (estimate
= 0.34, SE = 0.13, CI: [0.08; 0.60]).

28


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

2.3.4 DISCUSSION

Higher acceptability ratings for condition A compared to condition B once again indicate a pref-
erence for the locally prominent referent as the perspectival center, in particular a preference for
the referent that is the subject of the sentence preceding the FID – if there is no competing referent
activated in the context.

These findings do not only replicate the effect local prominence has on perceptive-taking, rather
they serve as a baseline for the manipulation of the context. That is, without any competing referent
the target sentence was preferably anchored to the subject of the preceding sentence. This prefer-
ence was overridden when the object of the preceding sentence was prominent in the context, i.e.,
globally prominent (compare Figure 4).

Figure 4: The impact of weak and strong globally prominent competitors on local prominence

While the effect of the locally prominent referent is canceled when the competing referent is promi-
nent in the context, as indicated by the difference between condition C & D and A & B, there is
no significant preference for the globally prominent referent, that is, condition D is rated better yet
not significantly higher than condition C. That is, a weak prominence, i.e., three references in the
context does not suffice to promote a referent to be the perspectival center when competing with a
locally prominent referent. Note that the higher activation in the context, in terms of five references
in the context and in the title as in condition E and F, override the preference for the locally promi-
nent referent.

This allows to conclude that generally the preference for a locally prominent referent as the per-

29


MEUSER, HÖRL AND HINTERWIMMER

spectival center can be overridden by the presence of a globally prominent referent. However, the
globally prominent referent is only preferred as the perspectival center when it is highly activated in
terms of repeated mentioning.

The difference between weak (cond. C and D) and strong (cond. E and F) global prominence
calls for further investigation. That is, the question of whether it was merely the higher number of
references or if being mentioned in the title leads to an increase of the acceptability ratings remains
an open question

3. Conclusion

In this paper we have discussed the role of different prominence-lending cues for the anchoring of
sentences in FID mode. While the formal properties distinguishing FID from other forms of speech
or thought representation and its interpretative characteristics have been thoroughly investigated
from a narratological as well as from a linguistic point of view, the question of how protagonists
become available as implicit thinkers or speakers for the anchoring of FID has scarcely been ad-
dressed in previous research. This question becomes particularly pressing in cases where more than
one protagonist is in principle available as perspectival center, a situation that is hardly uncommon
in literary texts. In principle, it is conceivable that the anchoring of FID is resolved on the basis
of content exclusively. Consequently, whenever it makes sense to interpret a sentence in FID mode
as a thought or utterance of a protagonist that has been made available by the preceding linguistic
context, that protagonist should be available as an anchor.

Since antecedent prominence has been shown to play an important role in pronoun resolution, a
competing assumption is that a protagonist’s availability as perspectival center is tied to that pro-
tagonist’s prominence status. Given that not only reference tracking but also the management of
perspective-taking is integral for the comprehension of narrative texts, a detailed and fine- grained
understanding of how perspectival centers are identified is as important for our understanding of the
processes underlying the comprehension of narrative texts as a detailed and fine-grained understand-
ing of how pronoun resolution works. The experimental results reported in this paper constitute a
first step in this direction.

In Section 1.4 we outlined our intuitions with respect to anchoring preferences of FID. We showed
that in short text segments where two protagonists are presented one is usually preferred as the an-
chor for a following sentence in FID mode. We argued that the referent that is more prominent with
respect to different prominence-lending cues is preferred as the perspectival center. In the series of
experiments presented in Section 2 we aimed to test these intuitions empirically. Our results sum-
marized in Section 2.2.4 now support the contrast between example (9a) and (9b) (repeated here as
(20)).

(20) a. When Lisa was playing in the schoolyard, a boy pushed her into the stinging nettles.
Ouch, that itched!
b. When Lisa was playing in the schoolyard, she pushed a boy into the stinging nettles.
Ouch, that itched!

30


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

Based on these results, we suggest that the establishment of a referent as the anchor for FID is the re-
sult of the interaction of various prominence-lending cues that add to a referent’s prominence status.
In order to narrow down the role of different linguistic cues that are involved we provided empirical
evidence that, first, allows us to conclude that a sentence in FID mode is perceived as more natural
when it is anchored to the more prominent referent and less natural when it is anchored to the less
prominent of two competing referents, irrespective of how plausible it is to interpret the sentence as
a thought or utterance of the respective protagonist. As discussed in Section 1.4, an utterance in FID
mode as presented in example (20b) sounds odd or at least surprising. Though the utterance must
be anchored to the physical sensation of the boy on the level of content, linguistic cues press the
reader to step back and ascribe the utterance to a narrating instance – or possibly an act of recursive
perspective-taking, i.e., Lisa thinks that it must itch for the boy.

Secondly, we have shown that locally established prominence can be overridden by the presence
of a globally prominent referent. The third experiment does not only replicate the findings that the
local subject is preferred as the perspectival center by default. Rather, the results of Experiment 3
reveal the impact of global prominence on local prominence. That is, FID anchored to the subject
of the preceding sentence is perceived to be less natural when the competitor was activated in the
context.

This observation allows for at least two explanations, which do not necessarily exclude each other.
Rather, it may well be that both are needed since the results of Experiment 3 turn out to be the
combined effect of two principles that are relevant for protagonists’ perspective taking in narra-
tive discourse. First, perspectival centers may be established in the discourse. Locally established
prominence only comes into play whenever there is no previous activation of any referent that can
serve as the perspectival center. An approach along these lines calls for further investigation of the
role of discourse phenomena such as discourse topicality in perspective taking.

According to the second principle, the availability of a protagonist as the perspectival center is
determined by the sum of prominence- lending cues. That is, if subjecthood promotes a referent to
be the perspectival center, consequently, the referent that occurs more often in subject position is
preferred over the one that occurs less often subject position. The assumption that prominence is
the result of the number of prominence-lending cues adding up is also supported by the results of
Experiment 2, which revealed a correlation between the number of prominence-enhancing features
and the acceptability of a discourse referent as the perspectival center (see Figure 4).

In the research reported in this paper we have relied on acceptability studies exclusively. While
being very useful for testing the empirical validity of linguistic hypotheses in a relatively simple and
straightforward way, it is a general shortcoming of this method that it is rather far removed from the
actual interpretative processes investigated since it cannot be controlled whether participants reflect
consciously on the respective task. In future research on perspective taking in narratives, it would
therefore be fruitful to employ online methods such as reading time measurements or the tracking
of eye movements.

Also, it would be very insightful to look at corpus data on the issue. This approach has not been
perused in depth as – to the best of our knowledge – there is no suitable corpus that includes a

31


MEUSER, HÖRL AND HINTERWIMMER

sufficient number of examples that match our hypothesis, i.e. annotated instances of FID and two
referents that are mentioned in the preceding sentence.

A final point worth noting is that it has been pointed out in Abrusán (2021) that the preference
for locally prominent protagonists as anchors for FID only holds if the content of the sentence in
FID mode stands in a subordinating discourse relation such as explanation or elaboration (see Asher
and Vieu, 2005 for an overview over subordinating and coordinating discourse relations) to the con-
tent of the preceding sentence. In general, it is very hard to interpret a sentence as FID that is linked
to the preceding sentence via a discourse relation such as narration or result. If such a reading is
available at all, however, Abrusán (2021) claims that the less prominent of two competing referents
is preferred as perspectival center. Her reasoning is based on examples like the one in (21).

(21) Sam threatened Justin with a knife. He had to defend himself!
(Abrusán, 2021: 335, ex. (18))

On the most likely reading, the second sentence is interpreted as reporting a thought of Sam via
FID that provides an explanation for Sam’s action reported by the first sentence. Since Sam is lo-
cally more prominent than Justin, this is as expected. As observed by Abrusán (2021), however, a
second reading is at least marginally available on which the second sentence reports a thought of
Justin rather than Sam (via FID). On this reading, the reported thought is interpreted as resulting
from Sam’s action rather than explaing it, i.e. Justin thinks that he has to defend himself as a con-
sequence of his being threatened by Sam with a knife. Since Justin is locally less prominent than
Sam and since there is no global context available that could overwrite Sam’s local prominence, this
is unexpected. Abrusán (2021) notes, however, that her observations might actually be compatible
with the analysis of Hinterwimmer (2019): While sentences that are linked via subordinating dis-
course relations usually have the same topic, sentences linked via coordinating discourse relations
move the story forward and hence have the potential to introduce new topics. In the experiments
reported in this paper, test items consisted of sentences linked via subordinating discourse relations
exclusively. It would be fruitful to empirically test Abrusán’s (2021) claims and to investigate how
the potential for coordinating discourse relations to shift the topic interacts with global prominence
in relation to determining the anchor for FID. We have to leave this as a topic for future research,
however.

Funding Information
The research reported in this paper was funded by a research grant from the Deutsche Forschungsge-
sellschaft. The research reported in this paper was funded by the Deutsche Forschungsgemeinschaft
(DFG) for the project C05 Discourse referents as perspectival centers of the Collaborative Research
Center 1252 textitProminence in Language (University of Cologne).

References

Márta Abrusán. The spectrum of perspective shift: Free indirect discourse vs. protagonist projec-
tion. Linguistics and Philosophy, 44, 2020. doi: https://doi.org/10.1007/s10988-020-09300-z.

Márta Abrusán. Computing perspective shift in narratives. In Emar Maier and Andreas Stokke,
editors, The Language of Fiction, pages 325–348. Oxford University Press, Oxford, 2021.

32


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

Nicholas Asher and Laure Vieu. Subordinating and coordinating discourse relations. Lingua,
115(4):591–610, 2005.

Harald Baayen, Douglas Davidson, and Douglas Bates. Mixed-effects modeling with crossed ran-
dom effects for subjects and items. Journal of Memory and Language, 59:390–412, 2008. doi:
https://doi.org/10.1016/j.jml.2007.12.005.

Ann Banfield. Unspeakable sentences: Narration and representation in the language of fiction.
Routledge, Boston, 1982.

Dale J. Barr, Roger Levy, Christoph Scheepers, and Harry J. Tily. Random effects structure for
confirmatory hypothesis testing: Keep it maximal. Journal of Memory and Language, 68(3):
255–278, 2013. doi: https://doi.org/10.1016/j.jml.2012.11.001.

Paul Bürkner. brms: An r package for bayesian multilevel models using stan. Journal of Statistical
Software, 80(1):1–28, 2017. doi: https://doi.org/10.18637/jss.v080.i01.

Rosalind A Crawley and Rosemary J Stevenson. Reference in single sentences and in texts. Journal
of Psycholinguistic Research, 19(3):191–210, 1990. doi: https://doi.org/10.1007/BF01077416.

Regine Eckardt. The Semantics of free indirect discourse: How texts allow us to mind-read and
eavesdrop. Brill, Leiden, 2014.

Regine Eckardt. Perspective and the future-in-the-past. Glossa: A Journal of General Linguistics,
2(1):71, 2017. doi: http://doi.org/10.5334/gjgl.199.

Norman Friedman. Point of view in fiction. the development of a critical concept. PMLA, 70(5):
1160–1184, 1955.

Catherine Garvey and Alfonso Caramazza. Implicit causality in verbs. Linguistic Inquiry, 5(3):
459–464, 1974.

Gérard Genette. Die Erzählung. Fink, Paderborn, 3rd, revised and corrected edition edition, 2010.

Morton Anne Gernsbacher. Language comprehension as structure building. Lawrence Erlbaum
Associates, Hillsdale, 1990.

Jonathan Ginzburg. The interactive stance: Meaning for conversation. Oxford University Press,
Oxford, 2012.

Peter C. Gordon, Barbara J. Grosz, and Laura A. Gilliom. Pronouns, names, and the centering of
attention in discourse. Cognitive Science, 17(3):311–347, 1993. doi: https://doi.org/10.1207/
s15516709cog1703 1.

Jesse A. Harris and Christopher Potts. Perspective-shifting with appositives and expressives. Lin-
guistics and Philosophy, 32(6):523–552, 2009. doi: https://doi.org/10.1007/s10988-010-9070-5.

Nikolaus P. Himmelmann and Beatrice Primus. Prominence beyond prosody: A first approximation.
In Amedeo De Dominicis, editor, pS-prominenceS: Prominences in Linguistics. Proceedings of
the International Conference, pages 38–58. DISUCOM Press, Viterbo, 2015.

33


MEUSER, HÖRL AND HINTERWIMMER

Stefan Hinterwimmer. Two kinds of perspective taking in narrative texts. Proceedings of Semantics
and Linguistic Theory, 27:282–301, 2017. doi: https://doi.org/10.3765/salt.v27i0.4153.

Stefan Hinterwimmer. Prominent protagonists. Journal of Pragmatics, 154:79–91, 2019. doi:
https://doi.org/10.1016/j.pragma.2017.12.003.

Stefan Hinterwimmer and Sara Meuser. Erlebte Rede und Protagonistenprominenz. In Stefan Engel-
berg, Christian Fortmann, and Irene Rapp, editors, Rede- und Gedankenwiedergabe in narrativen
Strukturen - Ambiguitäten und Varianz, volume 27 of Linguistische Berichte Sonderheft, pages
177–200. Buske, Hamburg, 2019.

Richard Holton. Some telling examples: a reply to Tsohatzidis. Journal of Pragmatics, 28:625–628,
1997.

Elsi Kaiser. Perspective-shifting and free indirect discourse: Experimental investigations. Semantics
and Linguistic Theory, 25:346–372, 2015. doi: https://doi.org/10.3765/salt.v25i0.3436.

Andrew Kehler and Hannah Rohde. Prominence and coherence in a bayesian theory of pronoun
interpretation. Journal of Pragmatics, 154:63–78, 2019.

Andrew Kehler, Laura Kertz, Hannah Rohde, and Jeffrey L. Elman. Coherence and coreference
revisited. Journal of Semantics, 25(1):1–44, 2008. doi: https://doi.org/10.1093/jos/ffm018.

Emar Maier. The pragmatics of attraction. explaining unquotation in direct and free indirect dis-
course. In Paul Saka and Michael Johnson, editors, The semantics and pragmatics of quotation,
pages 259–280. Springer, Dordrecht, 2017.

Sara Meuser. How free is free indirect discourse? Empirical approaches to the anchoring mecha-
nisms of perspective-taking. PhD thesis, Universität zu Köln, 2022.

Roy Pascal. The dual voice: Free indirect speech and its functioning in the nineteenth-century
European novel. Manchester University Press, Manchester, 1977.

Paul Piwek and Emiel Krahmer. Presuppositions in context: Constructing bridges. In Pierre Bon-
zon, Marcos Cavalcanti, and Rolf Nossum, editors, Formal Aspects of Context, volume 20 of
Applied Logic Series, pages 85–106. Springer, Dordrecht, 2000. doi: https://doi.org/10.1007/
978-94-015-9397-7 6.

R Core Team. R: A language and environment for statistical computing. 2015. URL https:
//www.R-project.org/.

Tanya Reinhart. Pragmatics and linguistics: An analysis of sentence topics. Philosophica, 27(1):
53–94, 1981.

Craige Roberts. Information structure in discourse: towards an integrated formal theory of prag-
matics. In J.H. Yoon and A. Kathol, editors, Papers in Semantics, pages 91–136. Ohio State
University, Columbus, OH, 1996.

Craige Roberts. Topics. In Klaus von Heusinger, Claudia Maienborn, and Paul Portner, editors, Se-
mantics: An international handbook of natural language meaning, pages 1908–1933. de Gruyter,
Berlin, 2012.

34


PERSPECTIVE-TAKING AND PROTAGONIST PROMINENCE

Christopher Saure, Stefan Hinterwimmer, and Anna Pia Jordan-Bertinelli. An experimental investi-
gation of the interaction of narrators’ and protagonists’ perspectival prominence in narrative texts.
Zeitschrift für Sprachwissenschaft, 42(2):341–372, 2023.

Philippe Schlenker. Context of thought and context of utterance. a note on free indirect discourse
and the historical present. Mind & Language, 19(3):279–304, 2004. doi: https://doi.org/10.1111/
j.1468-0017.2004.00259.x.

Yael Sharvit. The puzzle of free indirect discourse. Linguistics and Philosophy, 31(3):353–395,
2008. doi: https://doi.org/10.1007/s10988-008-9039-9.

Franz K. Stanzel. A theory of narrative (translated by Charlotte Goedsche). Cambridge University
Press, New York, 1984.

Anita Steube. Erlebte Rede aus linguistischer Sicht. Zeitschrift für Germanistik, 6(4):389–406,
1985.

Rosemary J. Stevenson, Rosalind A. Crawley, and David Kleinman. Thematic roles, focus and the
representation of events. Language and Cognitive Processes, 9(4):519–548, 1994. doi: https:
//doi.org/10.1080/01690969408402130.

Andreas Stokke. Protagonist projection. Mind & Language, 28(2):204–232, 2013. doi: https:
//doi.org/10.1111/mila.12016.

Teun van Dijk. Text and context. Longman, London, 1977.

Jan van Kuppevelt. Discourse structure, topicality and questioning. Journal of Linguistics, 31:
109–147, 1995.

Klaus von Heusinger and Petra B. Schumacher. Discourse prominence: Definition and application.
Journal of Pragmatics, 154:117–127, 2019.

Janyce M. Wiebe. Recognizing subjective sentences: A computational investigation of narrative
text. PhD thesis, SUNY Buffalo Dept. of Computer Science, Buffalo, NY, 1990.

Janyce M. Wiebe. Tracking point of view in narrative. Computational Linguistics, 20(2):233–287,
1994.

Jeremy Zehr and Florian Schwarz. PennController for internet based experiments (IBEX), 2018.

Sonja Zeman. Wer spricht? Disambiguierungsfaktoren bei der Perspektivensetzung im narrativen
Diskurs. In Stefan Engelberg, Christian Fortmann, and Irene Rapp, editors, Rede- und Gedanken-
wiedergabe in narrativen Strukturen: Ambiguitäten und Varianz, volume 27 of Linguistische
Berichte Sonderheft, pages 221–251. Buske, Hamburg, 2019.

Sonja Zeman. Parameters of narrative perspectivization: The narrator. Open Library of Humanities,
6(2):28, 2020. doi: https://doi.org/10.16995/olh.502.

35