One distinguishing feature of the two types of demonstrative pronouns may be register, with DemPros occurring in more formal contexts than DPros (Wiemer 1996). Patil et al. (2020) followed this line of argumentation and investigated the role of register in a behavioral task. They found that DemPros strongly preferred the formal language register such that contexts involving informal language elicited a sharp decrease in the use of DemPros. For example, in sentences such as (3a) and (3b) where participants had to choose a pronoun from multiple options to fill in the missing word in the sentence, participants chose the DemPro more often in (3a), the formal context, than in (3b), the informal context, where informality was for instance marked by using proper names with determiners and contractions like ‘nen (for the indefinite determiner einen). In contrast, the DPro was almost entirely dispreferred in the formal language environment. Both demonstrative pronouns also preferred the antecedent that was less prominent in terms of grammatical function/thematic role, i.e. the object or patient antecedent was preferred over the subject or agent antecedent.

In a follow-up experiment, using word order variation (canonical vs. non-canonical), Patil et al. (2020) also tested if the DemPro preferred the object antecedent only because it was the last-mentioned antecedent or because it was less prominent being the object. They found that although grammatical function was the main factor influencing the antecedent preference for DemPros, the order of mention of the antecedents had a small effect such that the object antecedent was less strongly preferred when it was mentioned first (see also Fuchs & Schumacher 2020 for an effect of grammatical/thematic role over order of mention).

It has further been suggested that the difference between the two types of demonstrative pronouns may not be due to register but to modality such that DemPros are the preferred choice in the written modality and DPros emerge in the auditory modality (Weinert 2011; Portele & Bader 2016; Voigt 2022). However, register and modality are intertwined, which might account for the mixed findings in rating tasks reported in Patterson et al. (ms.): On the one hand, (unaccented) DPros and DemPros were rated better in spoken modality with an informal register and DPros showed a much stronger effect of modality than DemPros in this case. On the other hand, when the stimuli reflected a formal register, there was no effect of modality for either of the demonstrative pronouns.

Fuchs & Schumacher (2020) compared the referential preference of DPros and DemPros (along with personal pronouns, PPros) and also the referential shift potential of the pronouns through a story continuation task. The task involved completing stories as in (4) where the context sentence had either an active accusative verb or a dative experiencer verb. In the context sentence of the active accusative condition, the first-mentioned referent was the proto-agent subject and the second-mentioned referent was the proto-patient object. In the dative experiencer condition, the first-mentioned referent was the proto-agent object, the experiencer, and the second-mentioned referent was the proto-patient subject, the stimulus (see Dowty 1991; Primus 1999 for proto-roles). Participants were given only one of the three pronouns in each trial and had to complete the stories by writing down six sentences. The continuations with six sentences made it possible to analyze the referential shift potential of these pronouns by looking at the frequency of mentioning a referent over a longer discourse.

As far as the choice of referent is concerned, the story completion task revealed that both DPros and DemPros preferred the second-mentioned referent (for DPros 74% and for DemPros 73% of all cases) and PPros preferred the first-mentioned (65% of all cases) irrespective of the verb type, lending support to the claim that thematic role information and/or order of mention represents a more highly ranked cue than grammatical function. There was no statistical difference in the referential preferences for the two demonstrative pronouns. The results from referential shifts in the story continuation were more nuanced. Based on the frequency of mentions of the two referents from the context sentence in the story continuation and the likelihood of using one of the three pronouns to refer to them, Fuchs & Schumacher (2020) concluded that although both demonstrative pronouns showed referential shift potential (in contrast to referential maintenance conveyed by personal pronouns), shifts triggered by DPros had a more long-lasting effect on the discourse structure than DemPros, which initiated only a momentary, parenthetical shift to their referent.

Patterson & Schumacher (2021) reported three acceptability rating studies using two-sentence items as in (5) and they compared the antecedent preferences of DPros and DemPros (along with PPros). This design of items allowed them to compare the antecedent preference of the pronouns, mentioned in the second sentence, with respect to three referents introduced in the preceding (context) sentence. Since the context sentence was a ditransitive construction, the three referents, which were the arguments of the ditransitive verb, had three different thematic roles – agent, recipient and patient. The example item is from Experiment 1a, but the other two (Experiment 1b and 2) had a comparable design where either the gender of the referents was changed or the order of patient and recipient was reversed. Because of the different thematic roles and different linear positions of the three referents in the context sentence, Patterson & Schumacher (2021) could compare the effect of thematic role and order of mention on pronoun resolution in three variations of the target sentence (5a) – (5c). Crucially, the target sentence licensed only a single referent for the pronoun based on gender congruency and verb semantics.

After comparing the acceptability ratings across the three experiments, Patterson & Schumacher (2021) concluded that DPros and DemPros showed a clear effect of order of mention such that both demonstrative pronouns, and not just the DemPros (as proposed in Zifonun et al. 1997), preferred the last-mentioned referent over the 1st and 2nd mentioned ones: however, DPros showed weaker last-mentioned referent preference in one of the three experiments. Interestingly, they also found that both demonstratives showed equally graded sensitivity to the prominence-lending cue of thematic role, such that the referent with the proto-agent role was the least preferred antecedent, the referent with the proto-patient role was the most preferred one and the referent with the proto-recipient role had a preference between the two extremes following canonical argument order. This finding is important for our understanding of prominence scales because it reveals that prominence relations are not binary (as suggested by accounts that propose an anti-agent or anti-topic preference of DPros). Rather the two types of demonstrative pronouns show a nuanced sensitivity to the thematic role scale. Overall, DemPros received higher ratings than DPros as expected under the language formality constraint (Patil et al. 2020).

Thematic role biases have also been investigated in the context of implicit causality verbs. Bader et al. (ms.) carried out two sentence completion studies to find out the interaction between semantic biases through implicit causality and structural constraints — order of mention and grammatical function — for DPros and DemPros. In Experiment 1, they showed partial sentences as in (6) and participants were asked to provide continuations for the text. The continuations provided by the participants revealed which of the two referents (e.g. Peter or the fisherman) were chosen as the antecedent of the pronoun. The two conditions (6a) and (6b) varied in terms of the kind of implicit causality verbs they used. In condition (6a), the verb required a subject-experiencer argument and in (6b), the verb required an object-experiencer argument. Since in implicit causality verbs, the stimulus argument (e.g. the object fisherman in (6a) and the subject fisherman in (6b)) has a strong next-mention bias in the subordinate clause starting with weil (because), the pronouns are expected to refer to the antecedent functioning as the stimulus. On the other hand, the structural constraints on the demonstrative pronouns are also expected to have an influence. Essentially, the antecedent preferences should reveal how the interaction between semantic biases through implicit causality and structural constraints pans out. Experiment 2 was similar to Experiment 1 with only a minor difference in the continuation prompt.

The results of the two experiments showed a strong effect of the semantic bias (i.e. implicit causality) such that for both verb types and both demonstrative pronouns, the referent functioning as the stimulus argument of the verb was strikingly the preferred antecedent. Since the stimulus argument had a strong next-mention bias under the implicit causality manipulation, Bader et al. (ms.) concluded that the semantic bias outweighed the structural biases. They further observed that there was still an influence of the structural bias such that the demonstrative pronouns showed preference towards the object antecedent (similar findings were observed by Patterson et al. 2022 for DemPros, where the stimulus bias of implicit causality verbs lowered object preferences of the demonstrative), and interestingly there was no difference between the DPros and DemPros in terms of the influence of either the semantic bias or the structural constraints. Bader et al. (ms.) concluded that the two demonstratives do not differ along structural and semantic biases.

Overall, the existing experimental work comparing DPros and DemPros provides clear evidence that suggests: (i) the two pronouns do not differ in their referential preference — both avoid the most prominent referent, (ii) both show referential shift potential, but (iii) they differ along the scale of language register and modality — DemPros prefer formal language and DPros prefer informal language and the spoken modality. There is also some evidence suggesting that DemPros have a stronger last-mentioned preference than DPros, whereas DPros have a more sustained referential shift potential than DemPros.

Before presenting our own analysis in Section 3.1, we will summarize the analysis of DPros proposed by Hinterwimmer & Bosch (2016; 2017) in the next section, since its basic assumptions are crucial for our account of the contrast between DPros and DemPros.

Hinterwimmer & Bosch (2016; 2017) observe that DPros do not always disprefer maximally prominent referents as antecedents or binders (see also Hinterwimmer et al. 2020 for further empirical evidence; for the purposes of this paper, we will focus on coreference exclusively and disregard binding configurations). Note, however, that the notion of prominence on which this conclusion is based is not clearly defined; on the approach we propose in Section 3.1, the discourse referents that are maximally prominent for Hinterwimmer & Bosch (2016; 2017) are no longer maximally prominent.

Consider the contrast between (7a) and (7b), on the one hand, and (8a) and (8b), on the other. In all cases, the respective opening sentence introduces a referent that is maximally prominent in terms of order of mention, grammatical function and thematic role: In (7), the proper name Peter occurs in clause-initial position, it is the subject of the matrix clause and its referent is the experiencer of the main clause event as well as the agent of the event introduced by the temporal adjunct clause, where Peter is picked up by a personal pronoun functioning as the subject. Likewise, in the opening sentence of (8) the proper name Paul occurs in clause-initial position, it is the subject of the clause and its referent is the experiencer of the state introduced by the verb wollte (wanted). Nevertheless, the respective maximally prominent referent can be picked up by a DPro more easily in one variant of the continuation than in the other. According to Hinterwimmer & Bosch (2016; 2017), the relevant factor distinguishing the continuations in (7b) and (8b), where the DPro can pick up the maximally prominent referent more easily, from the continuations in (7a) and (8a), where this is not the case, is perspective: Both (7a) and (8a) are most likely interpreted as expressing the perspective of the maximally prominent referent, while the continuations in (7b) and (8b) can only be understood as expressing the speaker’s or narrator’s perspective.

The sentence in (7a) is an instance of Free Indirect Discourse (FID), i.e. it is interpreted as a thought that Peter has when he notices the state his flat is in. As is characteristic of FID, all perspective-dependent linguistic expressions with the exception of pronouns and tenses are interpreted with respect to the perspective of the protagonist whose thought is rendered, while pronouns and tenses are interpreted with respect to the narrator’s perspective. In the case of (7a), verdammt (damn) is most likely understood as expressing Peter’s negative feelings, not the narrator’s, and the deictic temporal adverb gestern (yesterday) is interpreted with respect to Peter’s (fictional) context, i.e. it refers to the day preceding the day on which the events introduced by the opening sentence took place.

Likewise, the modal particles doch and erst (which roughly translate as after all and only in this context) are intuitively understood as expressing Peter’s attitudes, i.e. his surprise that the flat is in a terrible state again in spite of his recent efforts of cleaning up. The second sentence in (8a) is likewise most naturally interpreted as expressing the maximally prominent protagonist’s perspective. On the most plausible interpretation, Peter has a cold and therefore cannot go running. Note that this does not mean that the sentence is an instance of FID rendering a conscious thought of Paul. It simply means that the sentence expresses Paul’s general attitude towards the described situation.

The continuations in (7b) and (8b), in contrast, can only be understood as evaluative comments of the speaker or narrator on the character and dispositions of Peter and Paul, respectively. This is not only due to the content, but also due to the shift from past tense to present tense, which breaks the narrative continuity. Essentially, DPros can pick up referents that are maximally prominent in terms of order of mention, grammatical function and thematic role if the sentence containing the DPro clearly expresses the perspective of a referent that is different from the antecedent. Note, however, that for some speakers the shift from past tense, which suggests standard narration, to present tense, which can only be made sense of if the reader is willing to accommodate a rather unusual, highly ‘involved’ narrator, makes the entire text segment rather awkward. In order to avoid this complication, we have only used text segments in which no such temporal shifts occur in the experiment reported in Section 3.2 below. Based on this observation, Hinterwimmer & Bosch (2016; 2017) draw the conclusion that DPros are anti-logophoric pronouns, i.e. they avoid referents functioning as perspectival centers as their antecedents. They define the term perspectival center (cf. Harris 2012; Harris & Potts 2009) as in (9):

Recall the observation that in neutrally narrated sentences, DPros have a strong tendency to avoid referents that are maximally prominent in terms of order of mention, grammatical function and thematic role. For such cases, Hinterwimmer & Bosch (2017) suggest that these referents usually function as topics. This is based on the assumption that at least in the absence of overriding contextual factors, the topic of a sentence tends to be the referent that is maximally prominent in terms of order of mention, grammatical function and thematic role. Hinterwimmer & Bosch (2017) adopt the notion of aboutness topicality from Reinhart (1981), according to which the topic of a sentence is the referent for which a mental address is created such that the proposition denoted by that sentence is stored under the address. Crucially, they assume that topics are taken as perspectival centers by default in the absence of a perspectivally prominent narrator or protagonist. Consequently, the proposition denoted by the respective sentence is interpreted as the content of a thought or perception of the referent functioning as its aboutness topic by default.

The account proposed by Hinterwimmer & Bosch (2017) can be applied to explain the acceptability of the DPro in (1a), repeated here as (10a), and its unacceptability in (2), repeated here as (10c). In (10a), the sentence containing the DPro is an evaluative statement about the DPro’s referent, Peter. Consequently, the abstract speaker/narrator is the perspectival center in (10a) and accordingly the DPro is free to pick up the topical referent, Peter. In (10c), in contrast, the sentence containing the DPro is a factual statement describing the next event in the sequence that took place. Consequently, the abstract speaker’s/narrator’s perspective is not prominent, and the topical referent is taken to be the perspectival center by default. Accordingly, the topical referent cannot be picked up by the DPro without violating anti-logophoricity. Finally, in (10b), there is no violation of anti-logophoricity since the DPro does not pick up the topical referent anyway.

An analysis of the pattern exemplified by (10a)–(10c) that simply adopts assumptions from Hinterwimmer & Bosch (2017) about DPros is unsatisfactory insofar as it has to treat DemPros in an entirely different manner: While the behavior of DPros is accounted for via their avoidance of referents functioning as perspectival centers, the behavior of DemPros is accounted for via their avoidance of referents that are maximally prominent in terms of order of mention, grammatical function and thematic role. We, therefore, propose a unified analysis of DPros and DemPros that builds on an assumption already sketched as an alternative account of DPros in Hinterwimmer & Bosch (2017) (see also Hinterwimmer et al. 2020 for discussion). In this analysis, both DPros and DemPros avoid maximally prominent referents as antecedents. They differ with regard to the scales determining the maximally prominent referent, however. On the scale that is relevant for DPros, perspectival centers are more prominent than aboutness topics, which are in turn more prominent than the other referents. The scale that is relevant for DemPros, by contrast, does not include perspectival centers. Consequently, aboutness topics are the maximally prominent referents.

Following Hinterwimmer & Bosch (2017), we assume that referents that are maximally prominent in terms of order of mention, grammatical function and thematic role, function as aboutness topics for both types of pronouns (at least in the absence of overriding factors at the level of global discourse structure). We do not have a conclusive answer to the question of why the prominence scales differ for the two pronouns. However, we do not find this difference very surprising considering the fact that there are many languages that have logophoric pronouns, which are sensitive to perspective taking (see e.g. Sells 1987 for an overview), in addition to pronouns that are not sensitive to perspective taking at all. Hence, perspective taking can be directly relevant for the resolution options of one type of pronoun, and completely irrelevant for another.

Another possible analysis would be in terms of the form-specific multiple constraints approach proposed by Kaiser & Trueswell (2008) for two types of Finnish pronouns. To adopt this proposal, we would have to say that the resolution options of DemPros and DPros are determined with respect to two different sets of constraints, one of which weighs avoidance of the perspectival center higher than avoidance of the aboutness topic whenever the perspectival center is instantiated as a discourse topic, while the other one does not include avoidance of the perspectival center as a constraint. Empirically, this approach would make the same predictions as ours. However, conceptually, we find our approach simpler since it allows us to state clearly what the two types of pronouns have in common — a strong tendency to avoid the most prominent discourse referent.

In order for our analysis to receive empirical support, we have to assume pace Hinterwimmer & Bosch (2017) that aboutness topics are not perspectival centers by default in neutrally narrated sentences. Rather, such sentences simply do not contain perspectival centers at all. For concreteness, we assume, following Altshuler & Maier (2020; 2022), that while all texts are interpreted as being told by an implicit narrator, only narrators whose perspective is made prominent in a text are represented as discourse referents, like speakers in oral conversation. Perspectivally prominent abstract speakers/narrators like those in (10a) (and (7b) and (8b)) are therefore represented as discourse referents that become the highest ranked elements on the scales relevant for the resolution options of DPros. In the absence of such a discourse referent, the second highest ranked element on the scale, the respective aboutness topic, becomes the highest ranked one. For DemPros, in contrast, the discourse referent functioning as the aboutness topic is the highest ranked element, irrespective of whether the semantic representation of the sentence with the DemPro contains a discourse referent for the abstract speaker/narrator or not.

The analysis just proposed makes a clear prediction: Whenever a sentence expresses an evaluation of the referent functioning as aboutness topic, a DPro, but not a DemPro should be able to pick up that referent. In the absence of such an evaluation, both DPros and DemPros should be unable to pick up a topical referent. In Section 3.2, we present two experiments testing this prediction and in Section 4 we turn to the question of what our analysis has to say about the contrast between DPros and DemPros with respect to language formality, modality and referential shift potential.

In this section, we describe two experiments that empirically test the claims of the unified prominence-based account of the two types of demonstrative pronouns. In Experiment 1a, we directly tested the predictions of the hypothesis, and in Experiment 1b, we carried out an exploratory analysis to test if the degree of evaluation — weak to strong — has any influence on the acceptability of either of the demonstrative pronouns.